diff --git a/parse/train/1GTpBZvNUrk/1GTpBZvNUrk.md b/parse/train/1GTpBZvNUrk/1GTpBZvNUrk.md new file mode 100644 index 0000000000000000000000000000000000000000..ccacd01a964357242ccef83ceb5d35aa8f301d8c --- /dev/null +++ b/parse/train/1GTpBZvNUrk/1GTpBZvNUrk.md @@ -0,0 +1,248 @@ +# TransGAN: Two Pure Transformers Can Make One Strong GAN, and That Can Scale Up + +Yifan Jiang1, Shiyu Chang2,3, Zhangyang Wang1 1University of Texas at Austin 2UC Santa Barbara 3MIT-IBM Watson AI Lab {yifanjiang97,atlaswang}@utexas.edu, chang87@ucsb.edu + +# Abstract + +The recent explosive interest on transformers has suggested their potential to become powerful “universal" models for computer vision tasks, such as classification, detection, and segmentation. While those attempts mainly study the discriminative models, we explore transformers on some more notoriously difficult vision tasks, e.g., generative adversarial networks (GANs). Our goal is to conduct the first pilot study in building a GAN completely free of convolutions, using only pure transformer-based architectures. Our vanilla GAN architecture, dubbed TransGAN, consists of a memory-friendly transformer-based generator that progressively increases feature resolution, and correspondingly a multi-scale discriminator to capture simultaneously semantic contexts and low-level textures. On top of them, we introduce the new module of grid self-attention for alleviating the memory bottleneck further, in order to scale up TransGAN to high-resolution generation. We also develop a unique training recipe including a series of techniques that can mitigate the training instability issues of TransGAN, such as data augmentation, modified normalization, and relative position encoding. Our best architecture achieves highly competitive performance compared to current stateof-the-art GANs using convolutional backbones. Specifically, TransGAN sets the new state-of-the-art inception score of 10.43 and FID of 18.28 on STL-10. It also reaches the inception score of 9.02 and FID of 9.26 on CIFAR-10, and 5.28 FID on CelebA $\mathbf { 1 2 8 \times 1 2 8 }$ , respectively: both on par with the current best results. When it comes to higher-resolution (e.g. $\mathbf { 2 5 6 \times 2 5 6 } )$ ) generation tasks, such as on CelebAHQ and LSUN-Church, TransGAN continues to produce diverse visual examples with high fidelity and reasonable texture details. In addition, we dive deep into the transformer-based generation models to understand how their behaviors differ from convolutional ones, by visualizing training dynamics. The code is available at: https://github.com/VITA-Group/TransGAN. + +# 1 Introduction + +Generative adversarial networks (GANs) have gained considerable success on numerous tasks [1, 2, 3, 4, 5, 6, 7]. Unfortunately, GANs suffer from the notorious training instability, and numerous efforts have been devoted to stabilizing GAN training, introducing various regularization terms [8, 9, 10, 11], better losses [1, 12, 13, 14], and training recipes [15, 16]. Among them, one important route to improving GANs examines their neural architectures. [17, 8] reported a large-scale study of GANs and observed that when serving as (generator) backbones, popular neural architectures perform comparably well across the considered datasets. Their ablation study suggested that most of the variations applied in the ResNet family resulted in very marginal improvements. Nevertheless, neural architecture search (NAS) was later introduced to GANs and suggests enhanced backbone designs are also important for improving GANs, just like for other computer vision tasks. Those works are consistently able to discover stronger GAN architectures beyond the standard ResNet topology [18, 19, 20]. Other efforts include customized modules such as self-attention [21], style-based generator [22], and autoregressive transformer-based part composition [23]. + +![](images/adcb25935a6e25e1b2ab70151735aa65bcbfa9d2b385186c0dc4a0005b6915f2.jpg) +Figure 1: Representative visual examples synthesized by TransGAN, without using convolutional layers. (a) The synthesized visual examples on CelebA-HQ $2 5 6 \times 2 5 6 )$ dataset. (b) The linear interpolation results between two latent vectors, on CelebA-HQ ( $2 5 6 \times 2 5 6 )$ ) dataset. + +However, one last “commonsense" seems to have seldomly been challenged: using convolutional neural networks (CNNs) as GAN backbones. The original GAN [24, 25] used fully-connected networks and can only generate small images. DCGAN [26] was the first to scale up GANs using CNN architectures, which allowed for stable training for higher resolution and deeper generative models. Since then, in the computer vision domain, every successful GAN relies on CNN-based generators and discriminators. Convolutions, with the strong inductive bias for natural images, crucially contribute to the appealing visual results and rich diversity achieved by modern GANs. + +Can we build a strong GAN completely free of convolutions? This is a question not only arising from intellectual curiosity, but also of practical relevance. Fundamentally, a convolution operator has a local receptive field, and hence CNNs cannot process long-range dependencies unless passing through a sufficient number of layers. However, that is inefficient, and could cause the loss of feature resolution and fine details, in addition to the difficulty of optimization. Vanilla CNN-based models are therefore inherently not well suited for capturing an input image’s “global" statistics, as demonstrated by the benefits from adopting self-attention [21] and non-local [27] operations in computer vision. Moreover, the spatial invariance possessed by convolution poses a bottleneck on its ability of adapting to spatially varying/heterogeneous visual patterns, which also motivates the success of relational network [28], dynamic filters [29, 30] and kernel prediction [31] methods. + +# 1.1 Our Contributions + +This paper aims to be the first pilot study to build a GAN completely free of convolutions, using only pure transformer-based architectures. We are inspired by the recent success of transformer architectures in computer vision [32, 33, 34]. Compared to parallel generative modeling works [21, 23, 35] that applied self-attention or transformer encoder in conjunction with CNN-based backbones, our goal is more ambitious and faces several daunting gaps ahead. First and foremost, although a pure transformer architecture applied directly to sequences of image patches can perform very well on image classification tasks [34], it is unclear whether the same way remains effective in generating images, which crucially demands the spatial coherency in structure, color, and texture, as well as the richness of fine details. The handful of existing transformers that output images have unanimously leveraged convolutional part encoders [23] or feature extractors [36, 37]. Moreover, even given well-designed CNN-based architectures, training GANs is notoriously unstable and prone to mode collapse [15]. Training vision transformers are also known to be tedious, heavy, and data-hungry [34]. Combining the two will undoubtedly amplify the challenges of training. + +In view of those challenges, this paper presents a coherent set of efforts and innovations towards building the pure transformer-based GAN architectures, dubbed TransGAN. A naive option may directly stack multiple transformer blocks from raw pixel inputs, but that would scale poorly due to memory explosion. Instead, we start with a memory-friendly transformer-based generator by gradually increasing the feature map resolution in each stage. Correspondingly, we also improve the discriminator with a multi-scale structure that takes patches of varied size as inputs, which balances between capturing global contexts and local details, in addition to enhancing memory efficiency more. Based on the above generator-discriminator design, we introduce a new module called grid self-attention, that alleviates the memory bottleneck further when scaling up TransGAN to high-resolution generation (e.g. $2 5 6 \times 2 5 6$ ). + +To address the aforementioned instability issue brought by both GAN and Transformer, we also develop a unique training recipe in association with our innovative TransGAN architecture, that effectively stabilizes its optimization and generalization. That includes showings the necessity of data augmentation, modifying layer normalization, and replacing absolute token locations with relative position encoding. Our contributions are outlined below: + +• Novel Architecture Design: We build the first GAN using purely transformers and no convolution. TransGAN has customized a memory-friendly generator and a multi-scale discriminator, and is further equipped with a new grid self-attention mechanism. Those architectural components are thoughtfully designed to balance memory efficiency, global feature statistics, and local fine details with spatial variances. +• New Training Recipe: We study a number of techniques to train TransGAN better, including leveraging data augmentation, modifying layer normalization, and adopting relative position encoding, for both generator and discriminator. Extensive ablation studies, discussions, and insights are presented. +Performance and Scalability: TransGAN achieves highly competitive performance compared to current state-of-the-art GANs. Specifically, it sets the new state-of-the-art inception score of 10.43 and FID score of 18.28 on STL-10. It also reaches competitive 9.02 inception score and 9.26 FID on CIFAR-10, and 5.28 FID score on CelebA $1 2 8 \times 1 2 8$ , respectively. Meanwhile, we also evaluate TransGAN on higher-resolution (e.g., $2 5 6 \times 2 5 6 )$ generation tasks, where TransGAN continues to yield diverse and impressive visual examples. + +# 2 Related Works + +Generative Adversarial Networks. After its origin, GANs quickly embraced fully convolutional backbones [26], and inherited most successful designs from CNNs such as batch normalization, pooling, (Leaky) ReLU and more [38, 39, 40, 18]. GANs are widely adopted in image translation [3, 4, 41], image enhancement [7, 42, 43], and image editing [44, 45]. To alleviate its unstable training, a number of techniques have been studied, including the Wasserstein loss [46], the stylebased generator [22], progressive training [16], lottery ticket [47], and spectral normalization [48]. + +Transformers in Computer Vision. The original transformer was built for NLP [49], where the multi-head self-attention and feed-forward MLP layer are stacked to capture the long-term correlation between words. A recent work [34] implements highly competitive ImageNet classification using pure transformers, by treating an image as a sequence of $1 6 \times 1 6$ visual words. It has strong representation capability and is free of human-defined inductive bias. In comparison, CNNs exhibit a strong bias towards feature locality, as well as spatial invariance due to sharing filter weights across all locations. However, the success of original vision transformer relies on pretraining on large-scale external data. [50, 51] improve the data efficiency and address the difficulty of optimizing deeper models. Other works introduce the pyramid/hierarchical structure to transformer [52, 53, 54] or combine it with convolutional layers [55, 56]. Besides image classification task, transformer and its variants are also explored on image processing [37], point cloud [57], semantic segmentation [58], object detection [32, 59] and so on. A comprehensive review is referred to [60]. + +Transformer Modules for Image Generation. There exist several related works combining the transformer modules into image generation models, by replacing certain components of CNNs. [61] firstly formulated image generation as autoregressive sequence generation, for which they adopted a transformer architecture. [62] propose sparse factorization of the attention matrix to reduce its complexity. While those two works did not tackle the GANs, one recent (concurrent) work [23] used a convolutional GAN to learn a codebook of context-rich visual parts, whose composition is subsequently modeled with an autoregressive transformer architecture.The authors demonstrated success in synthesizing high-resolution images. However, the overall CNN architecture remains in place (including CNN encoder/decoder for the generators, and a fully CNN-based discriminator), and the customized designs (e.g, codebook and quantization) also limit their model’s versatility. Another concurrent work [35] employs a bipartite self-attention on StyleGAN and thus it can propagate latent variables to the evolving visual features, yet its main structure is still convolutional, including both the generator and discriminator. To our best knowledge, no other existing work has tried to completely remove convolutions from their generative modeling frameworks. + +![](images/e89dc9eb445115b78109b53440aa8db8c120d23b4e23956ef5aedc971cd11810.jpg) +Figure 2: The pipeline of the pure transform-based generator and discriminator of TransGAN. We take $2 5 6 \times 2 5 6$ resolution image generation task as a typical example to illustrate the main procedure. Here patch size $p$ is set to 32 as an example for the convenience of illustration, while practically the patch size is normally set to be no more than $8 \times 8$ , depending on the specific dataset. Grid Transformer Blocks refers to the transformer blocks with the proposed grid self-attention. Detailed architecture configurations are included in Appendix B. + +# 3 Technical Approach: A Journey Towards GAN with Pure Transformers + +In this section, we start by introducing the memory-friendly generator and multi-scale discriminator, equipped with a novel grid self-attention. We then introduce a series of training techniques to stabilize its training procedure, including data augmentation, the modified normalization, and injecting relative position encoding to self-attention. + +To start with, we choose the transformer encoder [49] as our basic block and try to make minimal changes. An encoder is a composition of two parts. The first part is constructed by a multi-head self-attention module and the second part is a feed-forward MLP with GELU non-linearity. The normalization layer is applied before both of the two parts. Both parts employ residual connection. + +# 3.1 Memory-friendly Generator + +The task of generation poses a high standard for spatial coherency in structure, color, and texture, both globally and locally. The transformer encoders take embedding token words as inputs and calculate the interaction between each token recursively. [63, 34]. The main dilemma here is: what is the right “word" for image generation tasks? If we similarly generate an image in a pixel-by-pixel manner through stacking transformer encoders, even a low-resolution image (e.g. $3 2 \times 3 2$ ) can result in an excessively long sequence (1024), causing the explosive cost of self-attention (quadratic w.r.t. the sequence length) and prohibiting the scalability to higher resolutions. To avoid this daunting cost, we are inspired by a common design philosophy in CNN-based GANs, to iteratively upscale the resolution at multiple stages [25, 16]. Our strategy is hence to increase the input sequence and reduce the embedding dimension gradually . + +Figure 2 (left) illustrates a memory-friendly transformer-based generator that consists of multiple stages. Each stage stacks several transformer blocks. By stages, we gradually increase the feature map resolution until it meets the target resolution $H \times W$ . Specifically, the generator takes the random noise as its input, and passes it through a multiple-layer perceptron (MLP) to a vector of length $H _ { 0 } \times W _ { 0 } \times C$ . The vector is reshaped into a $H _ { 0 } \times W _ { 0 }$ resolution feature map (by default we use $H _ { 0 } = W _ { 0 } = 8$ ), each point a $C$ -dimensional embedding. This “feature map" is next treated as a length-64 sequence of $C$ -dimensional tokens, combined with the learnable positional encoding. + +To scale up to higher-resolution images, we insert an upsampling module after each stage, consisting of a reshaping and resolution-upscaling layer. For lower-resolution stages (resolution lower than $6 4 \times 6 4 )$ ), the upsampling module firstly reshapes the 1D sequence of token embedding back to a 2D feature map $X _ { i } ~ \overset { \cdot } { \in } ~ \mathbb { R } ^ { \cup _ { i } \times W _ { i } \times C }$ and then adopts the bicubic layer to upsample its resolution while the embedded dimension is kept unchanged, resulting in the output $\boldsymbol { X } _ { i } ^ { ' } \in \mathbb { R } ^ { 2 H _ { i } \times 2 W _ { i } \times C }$ . After that, the 2D feature map $X _ { i } ^ { ' }$ is again reshaped into the 1D sequence of embedding tokens. For higher-resolution stages, we replace the bicubic upscaling layer with the pixelshuffle module, which upsamples the resolution of feature map by $2 \times$ ratio and also reduces the embedding dimension to a quarter of the input. This pyramid-structure with modified upscaling layers mitigates the memory and computation explosion. We repeat multiple stages until it reaches the target resolution $( H , W )$ , and then we will project the embedding dimension to 3 and obtain the RGB image $Y \in \mathbb { R } ^ { H \times W \times 3 }$ . + +# 3.2 Multi-scale Discriminator + +Unlike the generator which synthesizes precise pixels, the discriminator is tasked to distinguish between real/fake images. This allows us to treat it as a typical classifier by simply tokenizing the input image in a coarser patch-level [34], where each patch can be regarded as a “word". However, compared to image recognition tasks where classifiers focus on the semantic differences, the discriminator executes a simpler and more detail-oriented task to distinguish between synthesized and real. Therefore, the local visual cues and artifacts will have an important effect on the discriminator. Practically, we observe that the patch splitting rule plays a crucial role, where large patch size sacrifices low-level texture details, and smaller patch size results in a longer sequence that costs more memory. The above dilemma motivates our design of multi-scale discriminator below. + +As shown in Figure 2 (right), a multi-scale discriminator is designed to take varying size of patches as inputs, at its different stages. We firstly split the input images $Y \in R ^ { H \times W \times 3 ^ { \bullet } }$ into three different sequences by choosing different patch sizes $( P , 2 P , 4 P )$ . The longest sequence $\textstyle \left( { \frac { H } { P } } \times { \frac { W } { P } } \right) \times 3$ is linearly transformed to $\begin{array} { r } { \left( \frac { H } { P } \times \frac { W } { P } \right) \times \frac { C } { 4 } } \end{array}$ and then combined with the learnable position encoding to serve as the input of the first stage, where $\frac { C } { 4 }$ is the embedded dimension size. Similarly, the second 4and third sequences are linearly transformed to $\begin{array} { r } { \big ( \frac { H } { 2 P } \times \frac { W } { 2 P } \big ) \times \frac { C } { 4 } } \end{array}$ and $\bigl ( { \frac { H } { 4 P } } \times { \frac { W } { 4 P } } \bigr ) \times { \frac { C } { 2 } }$ , and then separately concatenated into the second and third stages. Thus these three different sequences are able to extract both the semantic structure and texture details. Similar to the generator, we reshape the 1D-sentence to 2D feature map and adopt Average Pooling layer to downsample the feature map resolution, between each stage. By recursively forming the transformer blocks in each stage, we obtain a pyramid architecture where multi-scale representation is extracted. At the end of these blocks, a [cls] token is appended at the beginning of the 1D sequence and then taken by the classification head to output the real/fake prediction. + +# 3.3 Grid Self-Attention: A Scalable Variant of Self-Attention for Image Generation + +Self-attention allows the generator to capture the global correspondence, yet also impedes the efficiency when modeling long sequences/higher resolutions. That motivates many efficient selfattention designs in both language [64, 65] and vision tasks [66, 67]. To adapt self-attention for higher-resolution generative tasks, we propose a simple yet effective strategy, named Grid SelfAttention, tailored for high-resolution image generation. + +As shown in Figure 3, instead of calculating the correspondence between a given token and all other tokens, the grid self-attention partitions the full-size feature map into several non-overlapped grids, and the token interactions are calculated inside each local grid. We add the grid self-attention on high-resolution stages (resolution higher than $3 2 \times 3 2$ ) while still keeping standard self-attention in low-resolution stages, shown as Figure 2, again so as to strategically balance local details and global awareness. The grid self-attention shows surprising effectiveness over other efficient self-attention forms [64, 67] in generative tasks, as compared later in Section 4.1. + +One potential concern might arise with the boundary artifact between each grid. We observe that while the artifact indeed occurs at early training stages, it gradually vanishes given enough training iterations and training data, while producing nicely coherent final results. We think this is owing to the larger, multi-scale receptive field of the discriminator that requires generated image fidelity in different scales. For other cases where the large-scale training data is hard to obtain, we discuss several solutions on Sec. 4.6. + +![](images/5aa30c4dc12e735df83a2803ff77d51d6c674cced188b389797c086e25b8de54.jpg) +Figure 3: Grid Self-Attention across different transformer stages. We replace Standard Self-Attention with Grid Self-Attention when the resolution is higher than $3 2 \times 3 2$ and the grid size is set to be $1 6 \times 1 6$ by default. + +# 3.4 Exploring the Training Recipe + +Data Augmentation. The transformer-based architectures are known to be highly data-hungry due to removing human-designed bias. Particularly in image recognition task [34], they were inferior to CNNs until much larger external data [68] was used for pre-training. To remove this roadblock, data augmentation was revealed as a blessing in [50], which showed that different types of strong data augmentation could lead us to data-efficient training for vision transformers. + +We follow a similar mindset. Traditionally, training CNN-based GANs hardly refers to data augmentation. Recently, there is an interest surge in the few-shot GAN training, aiming to match state-of-the-art GAN results with orders of magnitude fewer real images [69, 70]. Contrary to this “commonsense" in CNNs, data augmentation is found to be crucial in transformer-based architectures, even with $100 \%$ real images being utilized. We show that simply using differential augmentation [69] with three basic operators {T ranslation, Cutout, Color} leads to surprising performance improvement for TransGAN, while CNN-based GANs hardly benefit from it. We conduct a concrete study on the effectiveness of augmentation for both transformer and CNNs: see details in Section 4.2 + +Relative Position Encoding. While classical transformers [49, 34] used deterministic position encoding or learnable position encoding, the relative position encoding [71] gains increasing popularity [72, 28, 52, 73], by exploiting lags instead of absolute positions. Considering a single head of self-attention layer, + +$$ +A t t e n t i o n ( Q , K , V ) = s o f t m a x ( ( \frac { Q K ^ { T } } { \sqrt { d _ { k } } } V ) +$$ + +where $Q , K , V \in \mathbb { R } ^ { ( H \times W ) \times C }$ represent query, key, value matrices, $H , W , C$ denotes the height, width, embedded dimension of the input feature map. The difference in coordinate between each query and key on $H$ axis lies in the range of $[ - ( H - \bar { 1 } ) , H - 1 ]$ , and similar for $W$ axis. By simultaneously considering both $H$ and $W$ axis, the relative position can be represented by a parameterized matrix $M \in \mathbb { R } ^ { ( 2 \hat { H _ { - 1 } } ) \times ( 2 W - 1 ) }$ . Per coordinate, the relative position encoding $E$ is taken from matrix $M$ and added to the attention map $Q K ^ { T }$ as a bias term, shown as following, + +$$ +A t t e n t i o n ( Q , K , V ) = s o f t m a x ( ( ( \frac { Q K ^ { T } } { \sqrt { d _ { k } } } + E ) V ) +$$ + +Compared to its absolute counterpart, relative position encoding learns a stronger “relationship" between local contents, bringing important performance gains in large-scale cases and enjoying widespread use ever since. We also observe it to consistently improve TransGAN, especially on higher-resolution datasets. We hence apply it on top of the learnable absolute positional encoding for both the generator and discriminator. + +Modified Normalization. Normalization layers are known to help stabilize the deep learning training of deep neural networks, sometimes remarkably. While both the original transformer [49] and its variants [52, 54] by default use the layer normalization, we follow previous works [75, 16] and replace it with a token-wise scaling layer to prevent the magnitudes in transformer blocks from being too high, describe as $\begin{array} { r } { Y = X / \sqrt { \frac { 1 } { C } \sum _ { i = 0 } ^ { C - 1 } ( X ^ { i } ) ^ { 2 } + \epsilon } } \end{array}$ , where $\epsilon = 1 e - 8$ by default, $X$ and $Y$ denote the token before and after scaling layer, $C$ represents the embedded dimension. Note that our modified normalization resembles local response normalization that was once used in AlexNet [75]. Unlike other “modern" normalization layers [76, 77, 78] that need affine parameters for both mean and variances, we find that a simple re-scaling without learnable parameters suffices to stabilize TransGAN training – in fact, it makes TransGAN train better and improves the FID. + +Table 1: Unconditional image generation results on CIFAR-10, STl-10, and CelebA $( 1 2 8 \times 1 2 8 )$ dataset. We train the models with their official code if the results are unavailable, denoted as “\*”, others are all reported from references. + +
MethodsCIFAR-10STL-10CelebA
IS↑FID↓IS↑FID↓FID↓
WGAN-GP [1]6.49 ± 0.0939.681=
SN-GAN [48]8.22 ± 0.059.16 ± 0.1240.1
AutoGAN[18]8.55 ±0.1012.429.16 ± 0.1231.01
AdversarialNAS-GAN[18]8.74 ±0.0710.879.63 ± 0.1926.98=
Progressive-GAN[16]8.80 ± 0.0515.52=7.30
COCO-GAN [74]-=-5.74
StyleGAN-V2 [69]9.1811.0710.21* ± 0.1420.84*5.59*
StyleGAN-V2 + DiffAug. [69]9.409.8910.31*± 0.1219.15*5.40*
TransGAN9.02 ± 0.129.2610.43 ± 0.1618.285.28
+ +# 4 Experiments + +Datasets We start by evaluating our methods on three common testbeds: CIFAR-10 [79], STL10 [80], and CelebA [81] dataset. The CIFAR-10 dataset consists of $6 0 \mathbf { k } \ 3 2 \times 3 2$ images, with $5 0 \mathrm { k }$ training and $1 0 \mathrm { k }$ testing images, respectively. We follow the standard setting to use the $5 0 \mathrm { k }$ training images without labels. For the STL-10 dataset, we use both the 5k training images and $1 0 0 \mathrm { k }$ unlabeled images, and all are resized to $4 8 \times 4 8$ resolution. For the CelebA dataset, we use $2 0 0 \mathrm { k }$ unlabeled face images (aligned and cropped version), with each image at $1 2 8 \times 1 2 8$ resolution. We further consider the CelebA-HQ and LSUN Church datasets to scale up TransGAN to higher resolution image generation tasks. We use $3 0 \mathrm { k }$ images for CelebA-HQ [16] dataset and $1 2 5 \mathrm { k }$ images for LSUN Church dataset [82], all at $2 5 6 \times 2 5 6$ resolution. + +Implementation We follow the setting of WGAN [46], and use the WGAN-GP loss [1]. We adopt a learning rate of $1 e - 4$ for both generator and discriminator, an Adam optimizer with $\beta _ { 1 } = 0$ and $\beta _ { 2 } = 0 . 9 9$ , exponential moving average weights for generator, and a batch size of 128 for generator and 64 for discriminator, for all experiments. We choose DiffAug. [69] as basic augmentation strategy during the training process if not specially mentioned, and apply it to our competitors for a fair comparison. Other popular augmentation strategies ([70, 10]) are not discussed here since it is beyond the scope of this work. We use common evaluation metrics Inception Score (IS) [15] and Frechet Inception Distance (FID) [83], both are measured by 50K samples with their official Tensorflow implementations 12. All experiments are set with 16 V100 GPUs, using PyTorch 1.7.0. We include detailed training cost for each dataset in Appendix D. We focus on the unconditional image generation setting for simplicity. + +# 4.1 Comparison with State-of-the-art GANs + +CIFAR-10. We compare TransGAN with recently published results by unconditional CNN-based GANs on the CIFAR-10 dataset, shown in Table 1. Note that some promising conditional GANs [21, 8] are not included, due to the different settings. As shown in Table 1, TransGAN surpasses the strong model of Progressive GAN [16], and many other latest competitors such as SN-GAN [48], AutoGAN [18], and AdversarialNAS-GAN [19], in terms of inception score (IS). It is only next to the huge and heavily engineered StyleGAN-v2 [40]. Once we look at the FID results, TransGAN is even found to outperform StyleGAN-v2 [40] with both applied the same data augmentation [69]. + +STL-10. We then apply TransGAN on another popular benchmark STL-10, which is larger in scale (105k) and higher in resolution (48x48). We compare TransGAN with both the automatic searched and hand-crafted CNN-based GANs, shown in Table 1. Different from the results on CIFAR-10, we find that TransGAN outperforms all current CNN-based GAN models, and sets new state-of-the-art results in terms of both IS and FID score. This is thanks to the fact that the STL-10 dataset size is $2 \times$ larger than CIFAR-10, suggesting that transformer-based architectures benefit much more notably from larger-scale data than CNNs. + +![](images/6e96e44a3e68b78d7018919b2cfbce5508a09fd739e9d26cf4824f8d0c72caee.jpg) +Figure 4: Representative visual results produced by TransGAN on different datasets, as resolution grows from $3 2 \times 3 2$ to $2 5 6 \times 2 5 6$ . More visual examples are included in Appendix F. + +Table 2: The effectiveness of Data Augmentation on both CNN-based GANs and TransGAN. We use the full CIFAR-10 training set and DiffAug [69]. + +
MethodsWGAN-GPAutoGANStyleGAN-V2TransGAN
IS个FID←IS个FID←IS个FID←IS个FID↓
Original6.4939.688.5512.429.1811.078.3622.53
+ DiffAug [69]6.2937.148.6012.729.409.899.029.26
+ +Table 3: The ablation study of proposed techniques in three common dataset CelebA $( 6 4 \times 6 4 )$ , CelebA $( 1 2 8 \times 1 2 8$ , and LSUN Church $( 2 5 6 \times 2 5 6 )$ ). “OOM” represents out-of-momery issue. + +
Training ConfigurationCelebA (64x64)CelebA (128x128)LSUN Church (256x256)
(A). Standard Self-Attention8.9200M0OM
(B). Nystrom Self-Attention [64]13.4717.4239.92
(C). Axis Self-Attention [67]12.3913.9529.30
(D). Grid Self-Attention9.8910.5820.39
+Multi-scale Discriminator9.288.0315.29
+Modified Normalization7.057.1313.27
+Relative Position Encoding6.146.3211.93
(E). Converge5.015.288.94
+ +CelebA (128x128). We continue to examine another common benchmark: CelebA dataset $( 1 2 8 \times 1 2 8$ resolution). As shown in Table 1, TransGAN largely outperforms Progressive-GAN [16] and COCOGAN [74], and is slightly better than the strongest competitor StyleGAN-v2 [40], by reaching a FID score of 5.28. Visual examples generated on CIFAR-10, STL-10, and CelebA $( 1 2 8 \times 1 2 8 )$ are shown in Figure 4, from which we observe pleasing visual details and diversity. + +# 4.2 Scaling Up to Higher-Resolution + +We further scale up TransGAN to higher-resolution $( 2 5 6 \times 2 5 6 )$ generation, including on CelebAHQ [16] and LSUN Church [82]. These high-resolution datasets are significantly more challenging due to their much richer and detailed low-level texture as well as the global composition. Thanks to the proposed multi-scale discriminator, TransGAN produces pleasing visual results, reaching competitive quantitative results with 10.28 FID on CelebA-HQ $2 5 6 \times 2 5 6$ and $8 . 9 4 \mathrm { F I D }$ on LSUN Church dataset, respectively. As shown in Figure 4, diverse examples with rich textures details are produced. We discuss the memory cost reduction brought by the Grid Self-Attention in Appendix E. + +# 4.3 Data Augmentation is Crucial for TransGAN + +We study the effectiveness of data augmentation for both CNN-based GANs and Our TransGAN. We apply the differentiable augmentation [69] to all these methods. As shown in Table 2, for three CNN-based GANs, the performance gains of data augmentation seems to diminish in the full-data regime. Only the largest model, StyleGAN-V2, is improved on both IS and FID. In sharp contrast, TransGAN sees a shockingly large margin of improvement: IS improving from 8.36 to 9.02 and FID improving from 22.53 to 9.26. This phenomenon suggests that CIFAR-10 is still “small-scale " when fitting transformers; it re-confirms our assumption that transformer-based architectures are much more data-hungry than CNNs, and that can be helped by stronger data augmentation. + +![](images/77da0030a49f68b1242125a5ca2cb5670b6ba0577419be181bcc412c753e9bdb.jpg) +Figure 5: Left: training dynamic with training epochs for both TransGAN and MSG-GAN on CelebA-HQ $( 2 5 6 \times 2 5 6 )$ . Right: Interpolation on latent space produced by TransGAN. + +# 4.4 Ablation Study + +To further evaluate the proposed grid self-attention, multi-scale discriminator, and unique training recipe, we conduct the ablation study by separately adding these techniques to the baseline method and report their FID score on different datasets. Due to the fact that most of our contributions are tailored for the challenges brought by higher-resolution tasks, we choose CelebA and LSUN Church as the main testbeds, with details shown in Table 3. We start by constructing our memory-friendly with vanilla discriminator as our baseline method (A), both applied with standard self-attention. The baseline method achieves relatively good results with 8.92 FID on CelebA $( 6 4 \times 6 4 )$ dataset, however, it fail on higher-resolution tasks due to the memory explosion issue brought by self-attention. This motivates us to evaluate two efficient form of self-attention, (B) Nyström Self-Attention [64] and (C) Axis Self-Attention [67] + +By replacing all self-attention layers in high-resolution stages (feature map resolution higher than $3 2 \times 3 2 )$ with these efficient variants, both two methods (B)(C) are able to produce reasonable results. However, they still show to be inferior to standard self-attention, even on the $6 4 \times 6 4$ resolution dataset. By adopting our proposed Grid Self-Attention (D), we observe a significant improvement on both three datasets, reaching 9.89, 10.58, 20.39 FID on CelebA $6 4 \times 6 4$ , $1 2 8 \times 1 2 8$ and LSUN Church $2 5 6 \times 2 5 6$ , respectively. Based on the configuration (D), we continue to add the proposed techniques, including the multi-scale discriminator, modified normalization, and relative position encoding. All these three techniques significantly improve the performance of TransGAN on three datasets. At the end, we train our final configuration (E) until it converges, resulting in the best FID on CelebA $6 4 \times 6 4$ (5.01), CelebA $1 2 8 \times 1 2 8$ (5.28), and LSUN Church $2 5 6 \times 2 5 6$ (8.94). + +# 4.5 Understanding Transformer-based Generative Model + +We dive deep into our transformer-based GAN by conducting interpolation on latent space and comparing its behavior with CNN-based GAN, through visualizing their training dynamics. We choose MSG-GAN [84] for comparison since it extracts multi-scale representation as well. As shown in Figure 5, the CNN-based GAN quickly extracts face representation in the early stage of training process while transformer only produces rough pixels with no meaningful global shape due to missing any inductive bias. However, given enough training iterations, TransGAN gradually learns informative position representation and is able to produce impressive visual examples at convergence. Meanwhile, the boundary artifact also vanishes at the end. For the latent space interpolation, TransGAN continues to show encouraging results where smooth interpolation are maintained on both local and global levels. More high-resolution visual examples will be presented in Appendix F. + +# 4.6 Analyzing the Failure Cases and Improving High-resolution Tasks + +While TransGAN shows competitive or even better results on common low-resolution benchmarks, we still see large improvement room of its performance on high-resolution synthesis tasks, by analyzing the failure cases shown in appendix C. Here we discuss several alternatives tailored for high-resolution synthesis tasks, as potential remedies to address these failure cases. Specifically, we apply the self-modulation [85, 22, 35] to our generator and use cross-attention [53, 86] to map the latent space to the global region. Besides, we replace the current $2 \times$ upsampling layer, and instead firstly upsample it to $4 \times$ lager resolution using bicubic interpolation, and then downsample it back to $2 \times$ larger one. This simple modification not only helps the cross-boundary information interaction, but also help enhances the high-frequency details [87]. Moreover, an overlapped patch splitting strategy for discriminator can slightly improve the FID score. Additionally, we follow the previous work [22, 40] to conduct noise injection before the self-attention layer, which is found to further improve the generation fidelity and diversity of TransGAN. By applying these techniques to our high-resolution GAN frameworks, we observe additional improvement on both qualitative and quantitative results, e.g., the FID score on CelebA $2 5 6 \times 2 5 6$ dataset is further improved from 10.26 to 8.93. + +# 5 Conclusions, Limitation, and Discussions of Broad Impact + +In this work, we provide the first pilot study of building GAN with pure transformers. We have carefully crafted the architectures and thoughtfully designed training techniques. As a result, the proposed TransGAN has achieved state-of-the-art performance across multiple popular datasets, and easily scales up to higher-resolution generative tasks. Although TransGAN provides an encouraging starting point, there is still a large room to explore further, such as achieving state-of-the-art results on $2 5 6 \times 2 5 6$ generation tasks or going towards extremely high resolution generation tasks (e.g., $1 0 2 4 \times 1 0 2 4 ,$ ), which would be our future directions. + +Broader Impact. The proposed generative model can serve as a data engine to alleviate the challenge of data collection. More importantly, using synthesized image examples helps avoid privacy concerns. However, the abuse of advanced generative models may create fake media materials, which demands caution in the future. + +# Acknowledgements + +We would like to express our deepest gratitude to the MIT-IBM Watson AI Lab, in particular John Cohn for generously providing us with the computing resources necessary to conduct this research. Z Wang’s work is in part supported by an IBM Faculty Research Award, and the NSF AI Institute for Foundations of Machine Learning (IFML). + +# References + +[1] Ishaan Gulrajani, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, and Aaron C Courville. Improved training of wasserstein gans. In Advances in neural information processing systems, pages 5767–5777, 2017. +[2] Andrew Brock, Jeff Donahue, and Karen Simonyan. Large scale gan training for high fidelity natural image synthesis. arXiv preprint arXiv:1809.11096, 2018. +[3] Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A Efros. Image-to-image translation with conditional adversarial networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1125–1134, 2017. +[4] Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired image-to-image translation using cycle-consistent adversarial networks. In Proceedings of the IEEE international conference on computer vision, pages 2223–2232, 2017. +[5] Jun-Yan Zhu, Richard Zhang, Deepak Pathak, Trevor Darrell, Alexei A Efros, Oliver Wang, and Eli Shechtman. Toward multimodal image-to-image translation. In Advances in neural information processing systems, pages 465–476, 2017. +[6] Shuai Yang, Zhangyang Wang, Zhaowen Wang, Ning Xu, Jiaying Liu, and Zongming Guo. Controllable artistic text style transfer via shape-matching gan. In Proceedings of the IEEE International Conference on Computer Vision, pages 4442–4451, 2019. +[7] Yifan Jiang, Xinyu Gong, Ding Liu, Yu Cheng, Chen Fang, Xiaohui Shen, Jianchao Yang, Pan Zhou, and Zhangyang Wang. Enlightengan: Deep light enhancement without paired supervision. IEEE Transactions on Image Processing, 30:2340–2349, 2021. +[8] Karol Kurach, Mario Luciˇ c, Xiaohua Zhai, Marcin Michalski, and Sylvain Gelly. A large-scale study ´ on regularization and normalization in gans. In International Conference on Machine Learning, pages 3581–3590. PMLR, 2019. +[9] Kevin Roth, Aurelien Lucchi, Sebastian Nowozin, and Thomas Hofmann. Stabilizing training of generative adversarial networks through regularization. In Advances in neural information processing systems, pages 2018–2028, 2017. +[10] Han Zhang, Zizhao Zhang, Augustus Odena, and Honglak Lee. Consistency regularization for generative adversarial networks. arXiv preprint arXiv:1910.12027, 2019. +[11] Lars Mescheder, Andreas Geiger, and Sebastian Nowozin. Which training methods for gans do actually converge? arXiv preprint arXiv:1801.04406, 2018. +[12] Xudong Mao, Qing Li, Haoran Xie, Raymond YK Lau, Zhen Wang, and Stephen Paul Smolley. Least squares generative adversarial networks. In Proceedings of the IEEE international conference on computer vision, pages 2794–2802, 2017. +[13] Alexia Jolicoeur-Martineau. The relativistic discriminator: a key element missing from standard gan. arXiv preprint arXiv:1807.00734, 2018. +[14] Chun-Liang Li, Wei-Cheng Chang, Yu Cheng, Yiming Yang, and Barnabás Póczos. Mmd gan: Towards deeper understanding of moment matching network. In Advances in neural information processing systems, pages 2203–2213, 2017. +[15] Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen. Improved techniques for training gans. arXiv preprint arXiv:1606.03498, 2016. +[16] Tero Karras, Timo Aila, Samuli Laine, and Jaakko Lehtinen. Progressive growing of gans for improved quality, stability, and variation. arXiv preprint arXiv:1710.10196, 2017. +[17] Mario Lucic, Karol Kurach, Marcin Michalski, Olivier Bousquet, and Sylvain Gelly. Are gans created equal? a large-scale study. In Proceedings of the 32nd International Conference on Neural Information Processing Systems, pages 698–707, 2018. +[18] Xinyu Gong, Shiyu Chang, Yifan Jiang, and Zhangyang Wang. Autogan: Neural architecture search for generative adversarial networks. In Proceedings of the IEEE International Conference on Computer Vision, pages 3224–3234, 2019. +[19] Chen Gao, Yunpeng Chen, Si Liu, Zhenxiong Tan, and Shuicheng Yan. Adversarialnas: Adversarial neural architecture search for gans. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5680–5689, 2020. +[20] Yuan Tian, Qin Wang, Zhiwu Huang, Wen Li, Dengxin Dai, Minghao Yang, Jun Wang, and Olga Fink. Offpolicy reinforcement learning for efficient and effective gan architecture search. In European Conference on Computer Vision, pages 175–192. Springer, 2020. +[21] Han Zhang, Ian Goodfellow, Dimitris Metaxas, and Augustus Odena. Self-attention generative adversarial networks. In International conference on machine learning, pages 7354–7363. PMLR, 2019. +[22] Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative adversarial networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 4401–4410, 2019. +[23] Patrick Esser, Robin Rombach, and Björn Ommer. Taming transformers for high-resolution image synthesis. arXiv preprint arXiv:2012.09841, 2020. +[24] Ian J Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial networks. arXiv preprint arXiv:1406.2661, 2014. +[25] Emily Denton, Soumith Chintala, Arthur Szlam, and Rob Fergus. Deep generative image models using a laplacian pyramid of adversarial networks. arXiv preprint arXiv:1506.05751, 2015. +[26] Alec Radford, Luke Metz, and Soumith Chintala. Unsupervised representation learning with deep convolutional generative adversarial networks. arXiv preprint arXiv:1511.06434, 2015. +[27] Xiaolong Wang, Ross Girshick, Abhinav Gupta, and Kaiming He. Non-local neural networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7794–7803, 2018. +[28] Han Hu, Zheng Zhang, Zhenda Xie, and Stephen Lin. Local relation networks for image recognition. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3464–3473, 2019. +[29] Yu-Syuan Xu, Shou-Yao Roy Tseng, Yu Tseng, Hsien-Kai Kuo, and Yi-Min Tsai. Unified dynamic convolutional network for super-resolution with variational degradations. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12496–12505, 2020. +[30] David Ha, Andrew Dai, and Quoc V Le. Hypernetworks. arXiv preprint arXiv:1609.09106, 2016. +[31] Ben Mildenhall, Jonathan T Barron, Jiawen Chen, Dillon Sharlet, Ren Ng, and Robert Carroll. Burst denoising with kernel prediction networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 2502–2510, 2018. +[32] Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-to-end object detection with transformers. arXiv preprint arXiv:2005.12872, 2020. +[33] Yanhong Zeng, Jianlong Fu, and Hongyang Chao. Learning joint spatial-temporal transformations for video inpainting. In ECCV. Springer, 2020. +[34] Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020. +[35] Drew A Hudson and C Lawrence Zitnick. Generative adversarial transformers. arXiv preprint arXiv:2103.01209, 2021. +[36] Fuzhi Yang, Huan Yang, Jianlong Fu, Hongtao Lu, and Baining Guo. Learning texture transformer network for image super-resolution. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5791–5800, 2020. +[37] Hanting Chen, Yunhe Wang, Tianyu Guo, Chang Xu, Yiping Deng, Zhenhua Liu, Siwei Ma, Chunjing Xu, Chao Xu, and Wen Gao. Pre-trained image processing transformer. arXiv preprint arXiv:2012.00364, 2020. +[38] Jie Gui, Zhenan Sun, Yonggang Wen, Dacheng Tao, and Jieping Ye. A review on generative adversarial networks: Algorithms, theory, and applications. arXiv preprint arXiv:2001.06937, 2020. +[39] Edgar Schonfeld, Bernt Schiele, and Anna Khoreva. A u-net based discriminator for generative adversarial networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8207–8216, 2020. +[40] Tero Karras, Samuli Laine, Miika Aittala, Janne Hellsten, Jaakko Lehtinen, and Timo Aila. Analyzing and improving the image quality of stylegan. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8110–8119, 2020. +[41] Haotao Wang, Shupeng Gui, Haichuan Yang, Ji Liu, and Zhangyang Wang. Gan slimming: All-in-one gan compression by a unified optimization framework. In European Conference on Computer Vision, pages 54–73. Springer, 2020. +[42] Christian Ledig, Lucas Theis, Ferenc Huszár, Jose Caballero, Andrew Cunningham, Alejandro Acosta, Andrew Aitken, Alykhan Tejani, Johannes Totz, Zehan Wang, et al. Photo-realistic single image superresolution using a generative adversarial network. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 4681–4690, 2017. +[43] Orest Kupyn, Tetiana Martyniuk, Junru Wu, and Zhangyang Wang. Deblurgan-v2: Deblurring (ordersof-magnitude) faster and better. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 8878–8887, 2019. +[44] Xi Ouyang, Yu Cheng, Yifan Jiang, Chun-Liang Li, and Pan Zhou. Pedestrian-synthesis-gan: Generating pedestrian data in real scene and beyond. arXiv preprint arXiv:1804.02047, 2018. +[45] Jiahui Yu, Zhe Lin, Jimei Yang, Xiaohui Shen, Xin Lu, and Thomas S Huang. Generative image inpainting with contextual attention. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 5505–5514, 2018. +[46] Martin Arjovsky, Soumith Chintala, and Léon Bottou. Wasserstein gan. arXiv preprint arXiv:1701.07875, 2017. +[47] Tianlong Chen, Yu Cheng, Zhe Gan, Jingjing Liu, and Zhangyang Wang. Ultra-data-efficient gan training: Drawing a lottery ticket first, then training it toughly. arXiv preprint arXiv:2103.00397, 2021. +[48] Takeru Miyato, Toshiki Kataoka, Masanori Koyama, and Yuichi Yoshida. Spectral normalization for generative adversarial networks. arXiv preprint arXiv:1802.05957, 2018. +[49] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in neural information processing systems, pages 5998–6008, 2017. +[50] Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Hervé Jégou. Training data-efficient image transformers & distillation through attention. arXiv preprint arXiv:2012.12877, 2020. +[51] Hugo Touvron, Matthieu Cord, Alexandre Sablayrolles, Gabriel Synnaeve, and Hervé Jégou. Going deeper with image transformers. arXiv preprint arXiv:2103.17239, 2021. +[52] Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. arXiv preprint arXiv:2103.14030, 2021. +[53] Chun-Fu Chen, Quanfu Fan, and Rameswar Panda. Crossvit: Cross-attention multi-scale vision transformer for image classification. arXiv preprint arXiv:2103.14899, 2021. +[54] Pengchuan Zhang, Xiyang Dai, Jianwei Yang, Bin Xiao, Lu Yuan, Lei Zhang, and Jianfeng Gao. Multiscale vision longformer: A new vision transformer for high-resolution image encoding. arXiv preprint arXiv:2103.15358, 2021. +[55] Changlin Li, Tao Tang, Guangrun Wang, Jiefeng Peng, Bing Wang, Xiaodan Liang, and Xiaojun Chang. Bossnas: Exploring hybrid cnn-transformers with block-wisely self-supervised neural architecture search. arXiv preprint arXiv:2103.12424, 2021. +[56] Yutong Xie, Jianpeng Zhang, Chunhua Shen, and Yong Xia. Cotr: Efficiently bridging cnn and transformer for 3d medical image segmentation. arXiv preprint arXiv:2103.03024, 2021. +[57] Hengshuang Zhao, Li Jiang, Jiaya Jia, Philip Torr, and Vladlen Koltun. Point transformer. arXiv preprint arXiv:2012.09164, 2020. +[58] Sixiao Zheng, Jiachen Lu, Hengshuang Zhao, Xiatian Zhu, Zekun Luo, Yabiao Wang, Yanwei Fu, Jianfeng Feng, Tao Xiang, Philip HS Torr, et al. Rethinking semantic segmentation from a sequence-to-sequence perspective with transformers. arXiv preprint arXiv:2012.15840, 2020. +[59] Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, and Jifeng Dai. Deformable detr: Deformable transformers for end-to-end object detection. arXiv preprint arXiv:2010.04159, 2020. +[60] Kai Han, Yunhe Wang, Hanting Chen, Xinghao Chen, Jianyuan Guo, Zhenhua Liu, Yehui Tang, An Xiao, Chunjing Xu, Yixing Xu, et al. A survey on visual transformer. arXiv preprint arXiv:2012.12556, 2020. +[61] Niki Parmar, Ashish Vaswani, Jakob Uszkoreit, Lukasz Kaiser, Noam Shazeer, Alexander Ku, and Dustin Tran. Image transformer. In International Conference on Machine Learning, pages 4055–4064. PMLR, 2018. +[62] Rewon Child, Scott Gray, Alec Radford, and Ilya Sutskever. Generating long sequences with sparse transformers. arXiv preprint arXiv:1904.10509, 2019. +[63] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805, 2018. +[64] Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, and Vikas Singh. Nystromformer: A nystrom-based algorithm for approximating self-attention. arXiv preprint arXiv:2102.03902, 2021. +[65] Iz Beltagy, Matthew E Peters, and Arman Cohan. Longformer: The long-document transformer. arXiv preprint arXiv:2004.05150, 2020. +[66] Gedas Bertasius, Heng Wang, and Lorenzo Torresani. Is space-time attention all you need for video understanding? arXiv preprint arXiv:2102.05095, 2021. +[67] Manoj Kumar, Dirk Weissenborn, and Nal Kalchbrenner. Colorization transformer. arXiv preprint arXiv:2102.04432, 2021. +[68] Chen Sun, Abhinav Shrivastava, Saurabh Singh, and Abhinav Gupta. Revisiting unreasonable effectiveness of data in deep learning era. In Proceedings of the IEEE international conference on computer vision, pages 843–852, 2017. +[69] Shengyu Zhao, Zhijian Liu, Ji Lin, Jun-Yan Zhu, and Song Han. Differentiable augmentation for dataefficient gan training. arXiv preprint arXiv:2006.10738, 2020. +[70] Tero Karras, Miika Aittala, Janne Hellsten, Samuli Laine, Jaakko Lehtinen, and Timo Aila. Training generative adversarial networks with limited data. arXiv preprint arXiv:2006.06676, 2020. +[71] Peter Shaw, Jakob Uszkoreit, and Ashish Vaswani. Self-attention with relative position representations. arXiv preprint arXiv:1803.02155, 2018. +[72] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer. arXiv preprint arXiv:1910.10683, 2019. +[73] Cheng-Zhi Anna Huang, Ashish Vaswani, Jakob Uszkoreit, Noam Shazeer, Curtis Hawthorne, AM Dai, MD Hoffman, and D Eck. Music transformer: Generating music with long-term structure (2018). arXiv preprint arXiv:1809.04281, 2018. +[74] Chieh Hubert Lin, Chia-Che Chang, Yu-Sheng Chen, Da-Cheng Juan, Wei Wei, and Hwann-Tzong Chen. Coco-gan: Generation by parts via conditional coordinating. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 4512–4521, 2019. +[75] Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems, 25:1097–1105, 2012. +[76] Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. Layer normalization. arXiv preprint arXiv:1607.06450, 2016. +[77] Sergey Ioffe and Christian Szegedy. Batch normalization: Accelerating deep network training by reducing internal covariate shift. In International conference on machine learning, pages 448–456. PMLR, 2015. +[78] Dmitry Ulyanov, Andrea Vedaldi, and Victor Lempitsky. Instance normalization: The missing ingredient for fast stylization. arXiv preprint arXiv:1607.08022, 2016. +[79] Alex Krizhevsky, Geoffrey Hinton, et al. Learning multiple layers of features from tiny images. 2009. +[80] Adam Coates, Andrew Ng, and Honglak Lee. An analysis of single-layer networks in unsupervised feature learning. In Proceedings of the fourteenth international conference on artificial intelligence and statistics, pages 215–223. JMLR Workshop and Conference Proceedings, 2011. +[81] Ziwei Liu, Ping Luo, Xiaogang Wang, and Xiaoou Tang. Deep learning face attributes in the wild. In Proceedings of International Conference on Computer Vision (ICCV), 2015. +[82] Fisher Yu, Ari Seff, Yinda Zhang, Shuran Song, Thomas Funkhouser, and Jianxiong Xiao. Lsun: Construction of a large-scale image dataset using deep learning with humans in the loop. arXiv preprint arXiv:1506.03365, 2015. +[83] Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium. arXiv preprint arXiv:1706.08500, 2017. +[84] Animesh Karnewar and Oliver Wang. Msg-gan: Multi-scale gradients for generative adversarial networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7799– 7808, 2020. +[85] Ting Chen, Mario Lucic, Neil Houlsby, and Sylvain Gelly. On self modulation for generative adversarial networks. arXiv preprint arXiv:1810.01365, 2018. +[86] Long Zhao, Zizhao Zhang, Ting Chen, Dimitris N Metaxas, and Han Zhang. Improved transformer for high-resolution gans. arXiv preprint arXiv:2106.07631, 2021. +[87] Tero Karras, Miika Aittala, Samuli Laine, Erik Härkönen, Janne Hellsten, Jaakko Lehtinen, and Timo Aila. Alias-free generative adversarial networks. arXiv preprint arXiv:2106.12423, 2021. \ No newline at end of file diff --git a/parse/train/1GTpBZvNUrk/1GTpBZvNUrk_content_list.json b/parse/train/1GTpBZvNUrk/1GTpBZvNUrk_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..596ea1930591dbe9c02ca4f05dfc3b37a2afe889 --- /dev/null +++ b/parse/train/1GTpBZvNUrk/1GTpBZvNUrk_content_list.json @@ -0,0 +1,996 @@ +[ + { + "type": "text", + "text": "TransGAN: Two Pure Transformers Can Make One Strong GAN, and That Can Scale Up ", + "text_level": 1, + "bbox": [ + 186, + 122, + 815, + 172 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Yifan Jiang1, Shiyu Chang2,3, Zhangyang Wang1 1University of Texas at Austin 2UC Santa Barbara 3MIT-IBM Watson AI Lab {yifanjiang97,atlaswang}@utexas.edu, chang87@ucsb.edu ", + "bbox": [ + 272, + 224, + 727, + 284 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Abstract ", + "text_level": 1, + "bbox": [ + 462, + 306, + 535, + 324 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "The recent explosive interest on transformers has suggested their potential to become powerful “universal\" models for computer vision tasks, such as classification, detection, and segmentation. While those attempts mainly study the discriminative models, we explore transformers on some more notoriously difficult vision tasks, e.g., generative adversarial networks (GANs). Our goal is to conduct the first pilot study in building a GAN completely free of convolutions, using only pure transformer-based architectures. Our vanilla GAN architecture, dubbed TransGAN, consists of a memory-friendly transformer-based generator that progressively increases feature resolution, and correspondingly a multi-scale discriminator to capture simultaneously semantic contexts and low-level textures. On top of them, we introduce the new module of grid self-attention for alleviating the memory bottleneck further, in order to scale up TransGAN to high-resolution generation. We also develop a unique training recipe including a series of techniques that can mitigate the training instability issues of TransGAN, such as data augmentation, modified normalization, and relative position encoding. Our best architecture achieves highly competitive performance compared to current stateof-the-art GANs using convolutional backbones. Specifically, TransGAN sets the new state-of-the-art inception score of 10.43 and FID of 18.28 on STL-10. It also reaches the inception score of 9.02 and FID of 9.26 on CIFAR-10, and 5.28 FID on CelebA $\\mathbf { 1 2 8 \\times 1 2 8 }$ , respectively: both on par with the current best results. When it comes to higher-resolution (e.g. $\\mathbf { 2 5 6 \\times 2 5 6 } )$ ) generation tasks, such as on CelebAHQ and LSUN-Church, TransGAN continues to produce diverse visual examples with high fidelity and reasonable texture details. In addition, we dive deep into the transformer-based generation models to understand how their behaviors differ from convolutional ones, by visualizing training dynamics. The code is available at: https://github.com/VITA-Group/TransGAN. ", + "bbox": [ + 232, + 340, + 766, + 698 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 Introduction ", + "text_level": 1, + "bbox": [ + 174, + 723, + 310, + 739 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Generative adversarial networks (GANs) have gained considerable success on numerous tasks [1, 2, 3, 4, 5, 6, 7]. Unfortunately, GANs suffer from the notorious training instability, and numerous efforts have been devoted to stabilizing GAN training, introducing various regularization terms [8, 9, 10, 11], better losses [1, 12, 13, 14], and training recipes [15, 16]. Among them, one important route to improving GANs examines their neural architectures. [17, 8] reported a large-scale study of GANs and observed that when serving as (generator) backbones, popular neural architectures perform comparably well across the considered datasets. Their ablation study suggested that most of the variations applied in the ResNet family resulted in very marginal improvements. Nevertheless, neural architecture search (NAS) was later introduced to GANs and suggests enhanced backbone designs are also important for improving GANs, just like for other computer vision tasks. Those works are consistently able to discover stronger GAN architectures beyond the standard ResNet topology [18, 19, 20]. Other efforts include customized modules such as self-attention [21], style-based generator [22], and autoregressive transformer-based part composition [23]. ", + "bbox": [ + 174, + 748, + 825, + 900 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/adcb25935a6e25e1b2ab70151735aa65bcbfa9d2b385186c0dc4a0005b6915f2.jpg", + "image_caption": [ + "Figure 1: Representative visual examples synthesized by TransGAN, without using convolutional layers. (a) The synthesized visual examples on CelebA-HQ $2 5 6 \\times 2 5 6 )$ dataset. (b) The linear interpolation results between two latent vectors, on CelebA-HQ ( $2 5 6 \\times 2 5 6 )$ ) dataset. " + ], + "image_footnote": [], + "bbox": [ + 178, + 88, + 820, + 265 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 330, + 823, + 359 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "However, one last “commonsense\" seems to have seldomly been challenged: using convolutional neural networks (CNNs) as GAN backbones. The original GAN [24, 25] used fully-connected networks and can only generate small images. DCGAN [26] was the first to scale up GANs using CNN architectures, which allowed for stable training for higher resolution and deeper generative models. Since then, in the computer vision domain, every successful GAN relies on CNN-based generators and discriminators. Convolutions, with the strong inductive bias for natural images, crucially contribute to the appealing visual results and rich diversity achieved by modern GANs. ", + "bbox": [ + 174, + 364, + 825, + 462 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Can we build a strong GAN completely free of convolutions? This is a question not only arising from intellectual curiosity, but also of practical relevance. Fundamentally, a convolution operator has a local receptive field, and hence CNNs cannot process long-range dependencies unless passing through a sufficient number of layers. However, that is inefficient, and could cause the loss of feature resolution and fine details, in addition to the difficulty of optimization. Vanilla CNN-based models are therefore inherently not well suited for capturing an input image’s “global\" statistics, as demonstrated by the benefits from adopting self-attention [21] and non-local [27] operations in computer vision. Moreover, the spatial invariance possessed by convolution poses a bottleneck on its ability of adapting to spatially varying/heterogeneous visual patterns, which also motivates the success of relational network [28], dynamic filters [29, 30] and kernel prediction [31] methods. ", + "bbox": [ + 174, + 468, + 825, + 607 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "1.1 Our Contributions ", + "text_level": 1, + "bbox": [ + 176, + 621, + 343, + 636 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "This paper aims to be the first pilot study to build a GAN completely free of convolutions, using only pure transformer-based architectures. We are inspired by the recent success of transformer architectures in computer vision [32, 33, 34]. Compared to parallel generative modeling works [21, 23, 35] that applied self-attention or transformer encoder in conjunction with CNN-based backbones, our goal is more ambitious and faces several daunting gaps ahead. First and foremost, although a pure transformer architecture applied directly to sequences of image patches can perform very well on image classification tasks [34], it is unclear whether the same way remains effective in generating images, which crucially demands the spatial coherency in structure, color, and texture, as well as the richness of fine details. The handful of existing transformers that output images have unanimously leveraged convolutional part encoders [23] or feature extractors [36, 37]. Moreover, even given well-designed CNN-based architectures, training GANs is notoriously unstable and prone to mode collapse [15]. Training vision transformers are also known to be tedious, heavy, and data-hungry [34]. Combining the two will undoubtedly amplify the challenges of training. ", + "bbox": [ + 174, + 642, + 825, + 821 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "In view of those challenges, this paper presents a coherent set of efforts and innovations towards building the pure transformer-based GAN architectures, dubbed TransGAN. A naive option may directly stack multiple transformer blocks from raw pixel inputs, but that would scale poorly due to memory explosion. Instead, we start with a memory-friendly transformer-based generator by gradually increasing the feature map resolution in each stage. Correspondingly, we also improve the discriminator with a multi-scale structure that takes patches of varied size as inputs, which balances between capturing global contexts and local details, in addition to enhancing memory efficiency more. Based on the above generator-discriminator design, we introduce a new module called grid self-attention, that alleviates the memory bottleneck further when scaling up TransGAN to high-resolution generation (e.g. $2 5 6 \\times 2 5 6$ ). ", + "bbox": [ + 174, + 828, + 823, + 911 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 90, + 825, + 147 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "To address the aforementioned instability issue brought by both GAN and Transformer, we also develop a unique training recipe in association with our innovative TransGAN architecture, that effectively stabilizes its optimization and generalization. That includes showings the necessity of data augmentation, modifying layer normalization, and replacing absolute token locations with relative position encoding. Our contributions are outlined below: ", + "bbox": [ + 176, + 154, + 825, + 222 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "• Novel Architecture Design: We build the first GAN using purely transformers and no convolution. TransGAN has customized a memory-friendly generator and a multi-scale discriminator, and is further equipped with a new grid self-attention mechanism. Those architectural components are thoughtfully designed to balance memory efficiency, global feature statistics, and local fine details with spatial variances. \n• New Training Recipe: We study a number of techniques to train TransGAN better, including leveraging data augmentation, modifying layer normalization, and adopting relative position encoding, for both generator and discriminator. Extensive ablation studies, discussions, and insights are presented. \nPerformance and Scalability: TransGAN achieves highly competitive performance compared to current state-of-the-art GANs. Specifically, it sets the new state-of-the-art inception score of 10.43 and FID score of 18.28 on STL-10. It also reaches competitive 9.02 inception score and 9.26 FID on CIFAR-10, and 5.28 FID score on CelebA $1 2 8 \\times 1 2 8$ , respectively. Meanwhile, we also evaluate TransGAN on higher-resolution (e.g., $2 5 6 \\times 2 5 6 )$ generation tasks, where TransGAN continues to yield diverse and impressive visual examples. ", + "bbox": [ + 217, + 229, + 825, + 443 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "2 Related Works ", + "text_level": 1, + "bbox": [ + 174, + 457, + 330, + 474 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Generative Adversarial Networks. After its origin, GANs quickly embraced fully convolutional backbones [26], and inherited most successful designs from CNNs such as batch normalization, pooling, (Leaky) ReLU and more [38, 39, 40, 18]. GANs are widely adopted in image translation [3, 4, 41], image enhancement [7, 42, 43], and image editing [44, 45]. To alleviate its unstable training, a number of techniques have been studied, including the Wasserstein loss [46], the stylebased generator [22], progressive training [16], lottery ticket [47], and spectral normalization [48]. ", + "bbox": [ + 174, + 483, + 825, + 568 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Transformers in Computer Vision. The original transformer was built for NLP [49], where the multi-head self-attention and feed-forward MLP layer are stacked to capture the long-term correlation between words. A recent work [34] implements highly competitive ImageNet classification using pure transformers, by treating an image as a sequence of $1 6 \\times 1 6$ visual words. It has strong representation capability and is free of human-defined inductive bias. In comparison, CNNs exhibit a strong bias towards feature locality, as well as spatial invariance due to sharing filter weights across all locations. However, the success of original vision transformer relies on pretraining on large-scale external data. [50, 51] improve the data efficiency and address the difficulty of optimizing deeper models. Other works introduce the pyramid/hierarchical structure to transformer [52, 53, 54] or combine it with convolutional layers [55, 56]. Besides image classification task, transformer and its variants are also explored on image processing [37], point cloud [57], semantic segmentation [58], object detection [32, 59] and so on. A comprehensive review is referred to [60]. ", + "bbox": [ + 173, + 573, + 825, + 738 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Transformer Modules for Image Generation. There exist several related works combining the transformer modules into image generation models, by replacing certain components of CNNs. [61] firstly formulated image generation as autoregressive sequence generation, for which they adopted a transformer architecture. [62] propose sparse factorization of the attention matrix to reduce its complexity. While those two works did not tackle the GANs, one recent (concurrent) work [23] used a convolutional GAN to learn a codebook of context-rich visual parts, whose composition is subsequently modeled with an autoregressive transformer architecture.The authors demonstrated success in synthesizing high-resolution images. However, the overall CNN architecture remains in place (including CNN encoder/decoder for the generators, and a fully CNN-based discriminator), and the customized designs (e.g, codebook and quantization) also limit their model’s versatility. Another concurrent work [35] employs a bipartite self-attention on StyleGAN and thus it can propagate latent variables to the evolving visual features, yet its main structure is still convolutional, including both the generator and discriminator. To our best knowledge, no other existing work has tried to completely remove convolutions from their generative modeling frameworks. ", + "bbox": [ + 174, + 746, + 825, + 911 + ], + "page_idx": 2 + }, + { + "type": "image", + "img_path": "images/e89dc9eb445115b78109b53440aa8db8c120d23b4e23956ef5aedc971cd11810.jpg", + "image_caption": [ + "Figure 2: The pipeline of the pure transform-based generator and discriminator of TransGAN. We take $2 5 6 \\times 2 5 6$ resolution image generation task as a typical example to illustrate the main procedure. Here patch size $p$ is set to 32 as an example for the convenience of illustration, while practically the patch size is normally set to be no more than $8 \\times 8$ , depending on the specific dataset. Grid Transformer Blocks refers to the transformer blocks with the proposed grid self-attention. Detailed architecture configurations are included in Appendix B. " + ], + "image_footnote": [], + "bbox": [ + 186, + 90, + 816, + 429 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 525, + 820, + 553 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3 Technical Approach: A Journey Towards GAN with Pure Transformers ", + "text_level": 1, + "bbox": [ + 173, + 565, + 803, + 583 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "In this section, we start by introducing the memory-friendly generator and multi-scale discriminator, equipped with a novel grid self-attention. We then introduce a series of training techniques to stabilize its training procedure, including data augmentation, the modified normalization, and injecting relative position encoding to self-attention. ", + "bbox": [ + 174, + 592, + 825, + 646 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "To start with, we choose the transformer encoder [49] as our basic block and try to make minimal changes. An encoder is a composition of two parts. The first part is constructed by a multi-head self-attention module and the second part is a feed-forward MLP with GELU non-linearity. The normalization layer is applied before both of the two parts. Both parts employ residual connection. ", + "bbox": [ + 174, + 654, + 825, + 709 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3.1 Memory-friendly Generator ", + "text_level": 1, + "bbox": [ + 174, + 719, + 408, + 734 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "The task of generation poses a high standard for spatial coherency in structure, color, and texture, both globally and locally. The transformer encoders take embedding token words as inputs and calculate the interaction between each token recursively. [63, 34]. The main dilemma here is: what is the right “word\" for image generation tasks? If we similarly generate an image in a pixel-by-pixel manner through stacking transformer encoders, even a low-resolution image (e.g. $3 2 \\times 3 2$ ) can result in an excessively long sequence (1024), causing the explosive cost of self-attention (quadratic w.r.t. the sequence length) and prohibiting the scalability to higher resolutions. To avoid this daunting cost, we are inspired by a common design philosophy in CNN-based GANs, to iteratively upscale the resolution at multiple stages [25, 16]. Our strategy is hence to increase the input sequence and reduce the embedding dimension gradually . ", + "bbox": [ + 173, + 738, + 825, + 877 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Figure 2 (left) illustrates a memory-friendly transformer-based generator that consists of multiple stages. Each stage stacks several transformer blocks. By stages, we gradually increase the feature map resolution until it meets the target resolution $H \\times W$ . Specifically, the generator takes the random noise as its input, and passes it through a multiple-layer perceptron (MLP) to a vector of length $H _ { 0 } \\times W _ { 0 } \\times C$ . The vector is reshaped into a $H _ { 0 } \\times W _ { 0 }$ resolution feature map (by default we use $H _ { 0 } = W _ { 0 } = 8$ ), each point a $C$ -dimensional embedding. This “feature map\" is next treated as a length-64 sequence of $C$ -dimensional tokens, combined with the learnable positional encoding. ", + "bbox": [ + 174, + 883, + 820, + 911 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 90, + 825, + 160 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "To scale up to higher-resolution images, we insert an upsampling module after each stage, consisting of a reshaping and resolution-upscaling layer. For lower-resolution stages (resolution lower than $6 4 \\times 6 4 )$ ), the upsampling module firstly reshapes the 1D sequence of token embedding back to a 2D feature map $X _ { i } ~ \\overset { \\cdot } { \\in } ~ \\mathbb { R } ^ { \\cup _ { i } \\times W _ { i } \\times C }$ and then adopts the bicubic layer to upsample its resolution while the embedded dimension is kept unchanged, resulting in the output $\\boldsymbol { X } _ { i } ^ { ' } \\in \\mathbb { R } ^ { 2 H _ { i } \\times 2 W _ { i } \\times C }$ . After that, the 2D feature map $X _ { i } ^ { ' }$ is again reshaped into the 1D sequence of embedding tokens. For higher-resolution stages, we replace the bicubic upscaling layer with the pixelshuffle module, which upsamples the resolution of feature map by $2 \\times$ ratio and also reduces the embedding dimension to a quarter of the input. This pyramid-structure with modified upscaling layers mitigates the memory and computation explosion. We repeat multiple stages until it reaches the target resolution $( H , W )$ , and then we will project the embedding dimension to 3 and obtain the RGB image $Y \\in \\mathbb { R } ^ { H \\times W \\times 3 }$ . ", + "bbox": [ + 173, + 166, + 826, + 324 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "3.2 Multi-scale Discriminator ", + "text_level": 1, + "bbox": [ + 174, + 335, + 392, + 349 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Unlike the generator which synthesizes precise pixels, the discriminator is tasked to distinguish between real/fake images. This allows us to treat it as a typical classifier by simply tokenizing the input image in a coarser patch-level [34], where each patch can be regarded as a “word\". However, compared to image recognition tasks where classifiers focus on the semantic differences, the discriminator executes a simpler and more detail-oriented task to distinguish between synthesized and real. Therefore, the local visual cues and artifacts will have an important effect on the discriminator. Practically, we observe that the patch splitting rule plays a crucial role, where large patch size sacrifices low-level texture details, and smaller patch size results in a longer sequence that costs more memory. The above dilemma motivates our design of multi-scale discriminator below. ", + "bbox": [ + 173, + 354, + 825, + 479 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "As shown in Figure 2 (right), a multi-scale discriminator is designed to take varying size of patches as inputs, at its different stages. We firstly split the input images $Y \\in R ^ { H \\times W \\times 3 ^ { \\bullet } }$ into three different sequences by choosing different patch sizes $( P , 2 P , 4 P )$ . The longest sequence $\\textstyle \\left( { \\frac { H } { P } } \\times { \\frac { W } { P } } \\right) \\times 3$ is linearly transformed to $\\begin{array} { r } { \\left( \\frac { H } { P } \\times \\frac { W } { P } \\right) \\times \\frac { C } { 4 } } \\end{array}$ and then combined with the learnable position encoding to serve as the input of the first stage, where $\\frac { C } { 4 }$ is the embedded dimension size. Similarly, the second 4and third sequences are linearly transformed to $\\begin{array} { r } { \\big ( \\frac { H } { 2 P } \\times \\frac { W } { 2 P } \\big ) \\times \\frac { C } { 4 } } \\end{array}$ and $\\bigl ( { \\frac { H } { 4 P } } \\times { \\frac { W } { 4 P } } \\bigr ) \\times { \\frac { C } { 2 } }$ , and then separately concatenated into the second and third stages. Thus these three different sequences are able to extract both the semantic structure and texture details. Similar to the generator, we reshape the 1D-sentence to 2D feature map and adopt Average Pooling layer to downsample the feature map resolution, between each stage. By recursively forming the transformer blocks in each stage, we obtain a pyramid architecture where multi-scale representation is extracted. At the end of these blocks, a [cls] token is appended at the beginning of the 1D sequence and then taken by the classification head to output the real/fake prediction. ", + "bbox": [ + 173, + 486, + 825, + 672 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "3.3 Grid Self-Attention: A Scalable Variant of Self-Attention for Image Generation ", + "text_level": 1, + "bbox": [ + 174, + 684, + 763, + 699 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Self-attention allows the generator to capture the global correspondence, yet also impedes the efficiency when modeling long sequences/higher resolutions. That motivates many efficient selfattention designs in both language [64, 65] and vision tasks [66, 67]. To adapt self-attention for higher-resolution generative tasks, we propose a simple yet effective strategy, named Grid SelfAttention, tailored for high-resolution image generation. ", + "bbox": [ + 173, + 704, + 825, + 773 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "As shown in Figure 3, instead of calculating the correspondence between a given token and all other tokens, the grid self-attention partitions the full-size feature map into several non-overlapped grids, and the token interactions are calculated inside each local grid. We add the grid self-attention on high-resolution stages (resolution higher than $3 2 \\times 3 2$ ) while still keeping standard self-attention in low-resolution stages, shown as Figure 2, again so as to strategically balance local details and global awareness. The grid self-attention shows surprising effectiveness over other efficient self-attention forms [64, 67] in generative tasks, as compared later in Section 4.1. ", + "bbox": [ + 174, + 780, + 825, + 877 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "One potential concern might arise with the boundary artifact between each grid. We observe that while the artifact indeed occurs at early training stages, it gradually vanishes given enough training iterations and training data, while producing nicely coherent final results. We think this is owing to the larger, multi-scale receptive field of the discriminator that requires generated image fidelity in different scales. For other cases where the large-scale training data is hard to obtain, we discuss several solutions on Sec. 4.6. ", + "bbox": [ + 174, + 883, + 821, + 911 + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/5aa30c4dc12e735df83a2803ff77d51d6c674cced188b389797c086e25b8de54.jpg", + "image_caption": [ + "Figure 3: Grid Self-Attention across different transformer stages. We replace Standard Self-Attention with Grid Self-Attention when the resolution is higher than $3 2 \\times 3 2$ and the grid size is set to be $1 6 \\times 1 6$ by default. " + ], + "image_footnote": [], + "bbox": [ + 181, + 77, + 820, + 179 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 228, + 825, + 284 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "3.4 Exploring the Training Recipe ", + "text_level": 1, + "bbox": [ + 176, + 294, + 424, + 309 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Data Augmentation. The transformer-based architectures are known to be highly data-hungry due to removing human-designed bias. Particularly in image recognition task [34], they were inferior to CNNs until much larger external data [68] was used for pre-training. To remove this roadblock, data augmentation was revealed as a blessing in [50], which showed that different types of strong data augmentation could lead us to data-efficient training for vision transformers. ", + "bbox": [ + 173, + 313, + 825, + 382 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "We follow a similar mindset. Traditionally, training CNN-based GANs hardly refers to data augmentation. Recently, there is an interest surge in the few-shot GAN training, aiming to match state-of-the-art GAN results with orders of magnitude fewer real images [69, 70]. Contrary to this “commonsense\" in CNNs, data augmentation is found to be crucial in transformer-based architectures, even with $100 \\%$ real images being utilized. We show that simply using differential augmentation [69] with three basic operators {T ranslation, Cutout, Color} leads to surprising performance improvement for TransGAN, while CNN-based GANs hardly benefit from it. We conduct a concrete study on the effectiveness of augmentation for both transformer and CNNs: see details in Section 4.2 ", + "bbox": [ + 173, + 388, + 825, + 501 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Relative Position Encoding. While classical transformers [49, 34] used deterministic position encoding or learnable position encoding, the relative position encoding [71] gains increasing popularity [72, 28, 52, 73], by exploiting lags instead of absolute positions. Considering a single head of self-attention layer, ", + "bbox": [ + 173, + 506, + 825, + 561 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/dbb01f6ced9089b3c2847770df83453bea906785efbac2a8e617ec65bab96ef6.jpg", + "text": "$$\nA t t e n t i o n ( Q , K , V ) = s o f t m a x ( ( \\frac { Q K ^ { T } } { \\sqrt { d _ { k } } } V )\n$$", + "text_format": "latex", + "bbox": [ + 348, + 559, + 650, + 594 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "where $Q , K , V \\in \\mathbb { R } ^ { ( H \\times W ) \\times C }$ represent query, key, value matrices, $H , W , C$ denotes the height, width, embedded dimension of the input feature map. The difference in coordinate between each query and key on $H$ axis lies in the range of $[ - ( H - \\bar { 1 } ) , H - 1 ]$ , and similar for $W$ axis. By simultaneously considering both $H$ and $W$ axis, the relative position can be represented by a parameterized matrix $M \\in \\mathbb { R } ^ { ( 2 \\hat { H _ { - 1 } } ) \\times ( 2 W - 1 ) }$ . Per coordinate, the relative position encoding $E$ is taken from matrix $M$ and added to the attention map $Q K ^ { T }$ as a bias term, shown as following, ", + "bbox": [ + 173, + 597, + 825, + 684 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/d642168023319f11cbc80f8d9df34a9ce6f6ffd07f0e3008d315f28bd6603d49.jpg", + "text": "$$\nA t t e n t i o n ( Q , K , V ) = s o f t m a x ( ( ( \\frac { Q K ^ { T } } { \\sqrt { d _ { k } } } + E ) V )\n$$", + "text_format": "latex", + "bbox": [ + 325, + 689, + 673, + 724 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Compared to its absolute counterpart, relative position encoding learns a stronger “relationship\" between local contents, bringing important performance gains in large-scale cases and enjoying widespread use ever since. We also observe it to consistently improve TransGAN, especially on higher-resolution datasets. We hence apply it on top of the learnable absolute positional encoding for both the generator and discriminator. ", + "bbox": [ + 173, + 728, + 825, + 799 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Modified Normalization. Normalization layers are known to help stabilize the deep learning training of deep neural networks, sometimes remarkably. While both the original transformer [49] and its variants [52, 54] by default use the layer normalization, we follow previous works [75, 16] and replace it with a token-wise scaling layer to prevent the magnitudes in transformer blocks from being too high, describe as $\\begin{array} { r } { Y = X / \\sqrt { \\frac { 1 } { C } \\sum _ { i = 0 } ^ { C - 1 } ( X ^ { i } ) ^ { 2 } + \\epsilon } } \\end{array}$ , where $\\epsilon = 1 e - 8$ by default, $X$ and $Y$ denote the token before and after scaling layer, $C$ represents the embedded dimension. Note that our modified normalization resembles local response normalization that was once used in AlexNet [75]. Unlike other “modern\" normalization layers [76, 77, 78] that need affine parameters for both mean and variances, we find that a simple re-scaling without learnable parameters suffices to stabilize TransGAN training – in fact, it makes TransGAN train better and improves the FID. ", + "bbox": [ + 173, + 804, + 825, + 912 + ], + "page_idx": 5 + }, + { + "type": "table", + "img_path": "images/3736709c036ee2e90a0a961b83e5eadc3a8c00fa0505f62df932fa0420a35521.jpg", + "table_caption": [ + "Table 1: Unconditional image generation results on CIFAR-10, STl-10, and CelebA $( 1 2 8 \\times 1 2 8 )$ dataset. We train the models with their official code if the results are unavailable, denoted as “\\*”, others are all reported from references. " + ], + "table_footnote": [], + "table_body": "
MethodsCIFAR-10STL-10CelebA
IS↑FID↓IS↑FID↓FID↓
WGAN-GP [1]6.49 ± 0.0939.681=
SN-GAN [48]8.22 ± 0.059.16 ± 0.1240.1
AutoGAN[18]8.55 ±0.1012.429.16 ± 0.1231.01
AdversarialNAS-GAN[18]8.74 ±0.0710.879.63 ± 0.1926.98=
Progressive-GAN[16]8.80 ± 0.0515.52=7.30
COCO-GAN [74]-=-5.74
StyleGAN-V2 [69]9.1811.0710.21* ± 0.1420.84*5.59*
StyleGAN-V2 + DiffAug. [69]9.409.8910.31*± 0.1219.15*5.40*
TransGAN9.02 ± 0.129.2610.43 ± 0.1618.285.28
", + "bbox": [ + 183, + 119, + 812, + 280 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "", + "bbox": [ + 176, + 290, + 821, + 332 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4 Experiments ", + "text_level": 1, + "bbox": [ + 173, + 338, + 313, + 356 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Datasets We start by evaluating our methods on three common testbeds: CIFAR-10 [79], STL10 [80], and CelebA [81] dataset. The CIFAR-10 dataset consists of $6 0 \\mathbf { k } \\ 3 2 \\times 3 2$ images, with $5 0 \\mathrm { k }$ training and $1 0 \\mathrm { k }$ testing images, respectively. We follow the standard setting to use the $5 0 \\mathrm { k }$ training images without labels. For the STL-10 dataset, we use both the 5k training images and $1 0 0 \\mathrm { k }$ unlabeled images, and all are resized to $4 8 \\times 4 8$ resolution. For the CelebA dataset, we use $2 0 0 \\mathrm { k }$ unlabeled face images (aligned and cropped version), with each image at $1 2 8 \\times 1 2 8$ resolution. We further consider the CelebA-HQ and LSUN Church datasets to scale up TransGAN to higher resolution image generation tasks. We use $3 0 \\mathrm { k }$ images for CelebA-HQ [16] dataset and $1 2 5 \\mathrm { k }$ images for LSUN Church dataset [82], all at $2 5 6 \\times 2 5 6$ resolution. ", + "bbox": [ + 173, + 359, + 825, + 484 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Implementation We follow the setting of WGAN [46], and use the WGAN-GP loss [1]. We adopt a learning rate of $1 e - 4$ for both generator and discriminator, an Adam optimizer with $\\beta _ { 1 } = 0$ and $\\beta _ { 2 } = 0 . 9 9$ , exponential moving average weights for generator, and a batch size of 128 for generator and 64 for discriminator, for all experiments. We choose DiffAug. [69] as basic augmentation strategy during the training process if not specially mentioned, and apply it to our competitors for a fair comparison. Other popular augmentation strategies ([70, 10]) are not discussed here since it is beyond the scope of this work. We use common evaluation metrics Inception Score (IS) [15] and Frechet Inception Distance (FID) [83], both are measured by 50K samples with their official Tensorflow implementations 12. All experiments are set with 16 V100 GPUs, using PyTorch 1.7.0. We include detailed training cost for each dataset in Appendix D. We focus on the unconditional image generation setting for simplicity. ", + "bbox": [ + 173, + 491, + 825, + 643 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4.1 Comparison with State-of-the-art GANs ", + "text_level": 1, + "bbox": [ + 174, + 654, + 491, + 667 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "CIFAR-10. We compare TransGAN with recently published results by unconditional CNN-based GANs on the CIFAR-10 dataset, shown in Table 1. Note that some promising conditional GANs [21, 8] are not included, due to the different settings. As shown in Table 1, TransGAN surpasses the strong model of Progressive GAN [16], and many other latest competitors such as SN-GAN [48], AutoGAN [18], and AdversarialNAS-GAN [19], in terms of inception score (IS). It is only next to the huge and heavily engineered StyleGAN-v2 [40]. Once we look at the FID results, TransGAN is even found to outperform StyleGAN-v2 [40] with both applied the same data augmentation [69]. ", + "bbox": [ + 173, + 672, + 826, + 770 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "STL-10. We then apply TransGAN on another popular benchmark STL-10, which is larger in scale (105k) and higher in resolution (48x48). We compare TransGAN with both the automatic searched and hand-crafted CNN-based GANs, shown in Table 1. Different from the results on CIFAR-10, we find that TransGAN outperforms all current CNN-based GAN models, and sets new state-of-the-art results in terms of both IS and FID score. This is thanks to the fact that the STL-10 dataset size is $2 \\times$ larger than CIFAR-10, suggesting that transformer-based architectures benefit much more notably from larger-scale data than CNNs. ", + "bbox": [ + 173, + 775, + 825, + 872 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/6e96e44a3e68b78d7018919b2cfbce5508a09fd739e9d26cf4824f8d0c72caee.jpg", + "image_caption": [ + "Figure 4: Representative visual results produced by TransGAN on different datasets, as resolution grows from $3 2 \\times 3 2$ to $2 5 6 \\times 2 5 6$ . More visual examples are included in Appendix F. " + ], + "image_footnote": [], + "bbox": [ + 178, + 82, + 818, + 241 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/f1c4e2b09d5e2f8e5edf0927060448a8a0cc45c22cecb15cb148ce6fe11b716d.jpg", + "table_caption": [ + "Table 2: The effectiveness of Data Augmentation on both CNN-based GANs and TransGAN. We use the full CIFAR-10 training set and DiffAug [69]. " + ], + "table_footnote": [], + "table_body": "
MethodsWGAN-GPAutoGANStyleGAN-V2TransGAN
IS个FID←IS个FID←IS个FID←IS个FID↓
Original6.4939.688.5512.429.1811.078.3622.53
+ DiffAug [69]6.2937.148.6012.729.409.899.029.26
", + "bbox": [ + 194, + 318, + 803, + 391 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/25c0032de9117db6ffe66c1bcb8ba86cc9bfa5f079bdbf7a78a3daa9c1d02d03.jpg", + "table_caption": [ + "Table 3: The ablation study of proposed techniques in three common dataset CelebA $( 6 4 \\times 6 4 )$ , CelebA $( 1 2 8 \\times 1 2 8$ , and LSUN Church $( 2 5 6 \\times 2 5 6 )$ ). “OOM” represents out-of-momery issue. " + ], + "table_footnote": [], + "table_body": "
Training ConfigurationCelebA (64x64)CelebA (128x128)LSUN Church (256x256)
(A). Standard Self-Attention8.9200M0OM
(B). Nystrom Self-Attention [64]13.4717.4239.92
(C). Axis Self-Attention [67]12.3913.9529.30
(D). Grid Self-Attention9.8910.5820.39
+Multi-scale Discriminator9.288.0315.29
+Modified Normalization7.057.1313.27
+Relative Position Encoding6.146.3211.93
(E). Converge5.015.288.94
", + "bbox": [ + 212, + 436, + 782, + 580 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "CelebA (128x128). We continue to examine another common benchmark: CelebA dataset $( 1 2 8 \\times 1 2 8$ resolution). As shown in Table 1, TransGAN largely outperforms Progressive-GAN [16] and COCOGAN [74], and is slightly better than the strongest competitor StyleGAN-v2 [40], by reaching a FID score of 5.28. Visual examples generated on CIFAR-10, STL-10, and CelebA $( 1 2 8 \\times 1 2 8 )$ are shown in Figure 4, from which we observe pleasing visual details and diversity. ", + "bbox": [ + 173, + 593, + 826, + 662 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "4.2 Scaling Up to Higher-Resolution ", + "text_level": 1, + "bbox": [ + 174, + 676, + 439, + 691 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "We further scale up TransGAN to higher-resolution $( 2 5 6 \\times 2 5 6 )$ generation, including on CelebAHQ [16] and LSUN Church [82]. These high-resolution datasets are significantly more challenging due to their much richer and detailed low-level texture as well as the global composition. Thanks to the proposed multi-scale discriminator, TransGAN produces pleasing visual results, reaching competitive quantitative results with 10.28 FID on CelebA-HQ $2 5 6 \\times 2 5 6$ and $8 . 9 4 \\mathrm { F I D }$ on LSUN Church dataset, respectively. As shown in Figure 4, diverse examples with rich textures details are produced. We discuss the memory cost reduction brought by the Grid Self-Attention in Appendix E. ", + "bbox": [ + 173, + 696, + 826, + 795 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "4.3 Data Augmentation is Crucial for TransGAN ", + "text_level": 1, + "bbox": [ + 174, + 808, + 526, + 823 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "We study the effectiveness of data augmentation for both CNN-based GANs and Our TransGAN. We apply the differentiable augmentation [69] to all these methods. As shown in Table 2, for three CNN-based GANs, the performance gains of data augmentation seems to diminish in the full-data regime. Only the largest model, StyleGAN-V2, is improved on both IS and FID. In sharp contrast, TransGAN sees a shockingly large margin of improvement: IS improving from 8.36 to 9.02 and FID improving from 22.53 to 9.26. This phenomenon suggests that CIFAR-10 is still “small-scale \" when fitting transformers; it re-confirms our assumption that transformer-based architectures are much more data-hungry than CNNs, and that can be helped by stronger data augmentation. ", + "bbox": [ + 174, + 827, + 825, + 911 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/77da0030a49f68b1242125a5ca2cb5670b6ba0577419be181bcc412c753e9bdb.jpg", + "image_caption": [ + "Figure 5: Left: training dynamic with training epochs for both TransGAN and MSG-GAN on CelebA-HQ $( 2 5 6 \\times 2 5 6 )$ . Right: Interpolation on latent space produced by TransGAN. " + ], + "image_footnote": [], + "bbox": [ + 179, + 77, + 816, + 174 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 217, + 823, + 246 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "4.4 Ablation Study ", + "text_level": 1, + "bbox": [ + 174, + 256, + 316, + 271 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "To further evaluate the proposed grid self-attention, multi-scale discriminator, and unique training recipe, we conduct the ablation study by separately adding these techniques to the baseline method and report their FID score on different datasets. Due to the fact that most of our contributions are tailored for the challenges brought by higher-resolution tasks, we choose CelebA and LSUN Church as the main testbeds, with details shown in Table 3. We start by constructing our memory-friendly with vanilla discriminator as our baseline method (A), both applied with standard self-attention. The baseline method achieves relatively good results with 8.92 FID on CelebA $( 6 4 \\times 6 4 )$ dataset, however, it fail on higher-resolution tasks due to the memory explosion issue brought by self-attention. This motivates us to evaluate two efficient form of self-attention, (B) Nyström Self-Attention [64] and (C) Axis Self-Attention [67] ", + "bbox": [ + 174, + 276, + 825, + 415 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "By replacing all self-attention layers in high-resolution stages (feature map resolution higher than $3 2 \\times 3 2 )$ with these efficient variants, both two methods (B)(C) are able to produce reasonable results. However, they still show to be inferior to standard self-attention, even on the $6 4 \\times 6 4$ resolution dataset. By adopting our proposed Grid Self-Attention (D), we observe a significant improvement on both three datasets, reaching 9.89, 10.58, 20.39 FID on CelebA $6 4 \\times 6 4$ , $1 2 8 \\times 1 2 8$ and LSUN Church $2 5 6 \\times 2 5 6$ , respectively. Based on the configuration (D), we continue to add the proposed techniques, including the multi-scale discriminator, modified normalization, and relative position encoding. All these three techniques significantly improve the performance of TransGAN on three datasets. At the end, we train our final configuration (E) until it converges, resulting in the best FID on CelebA $6 4 \\times 6 4$ (5.01), CelebA $1 2 8 \\times 1 2 8$ (5.28), and LSUN Church $2 5 6 \\times 2 5 6$ (8.94). ", + "bbox": [ + 174, + 421, + 825, + 559 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "4.5 Understanding Transformer-based Generative Model ", + "text_level": 1, + "bbox": [ + 173, + 570, + 584, + 585 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "We dive deep into our transformer-based GAN by conducting interpolation on latent space and comparing its behavior with CNN-based GAN, through visualizing their training dynamics. We choose MSG-GAN [84] for comparison since it extracts multi-scale representation as well. As shown in Figure 5, the CNN-based GAN quickly extracts face representation in the early stage of training process while transformer only produces rough pixels with no meaningful global shape due to missing any inductive bias. However, given enough training iterations, TransGAN gradually learns informative position representation and is able to produce impressive visual examples at convergence. Meanwhile, the boundary artifact also vanishes at the end. For the latent space interpolation, TransGAN continues to show encouraging results where smooth interpolation are maintained on both local and global levels. More high-resolution visual examples will be presented in Appendix F. ", + "bbox": [ + 173, + 590, + 825, + 729 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "4.6 Analyzing the Failure Cases and Improving High-resolution Tasks ", + "text_level": 1, + "bbox": [ + 176, + 746, + 671, + 762 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "While TransGAN shows competitive or even better results on common low-resolution benchmarks, we still see large improvement room of its performance on high-resolution synthesis tasks, by analyzing the failure cases shown in appendix C. Here we discuss several alternatives tailored for high-resolution synthesis tasks, as potential remedies to address these failure cases. Specifically, we apply the self-modulation [85, 22, 35] to our generator and use cross-attention [53, 86] to map the latent space to the global region. Besides, we replace the current $2 \\times$ upsampling layer, and instead firstly upsample it to $4 \\times$ lager resolution using bicubic interpolation, and then downsample it back to $2 \\times$ larger one. This simple modification not only helps the cross-boundary information interaction, but also help enhances the high-frequency details [87]. Moreover, an overlapped patch splitting strategy for discriminator can slightly improve the FID score. Additionally, we follow the previous work [22, 40] to conduct noise injection before the self-attention layer, which is found to further improve the generation fidelity and diversity of TransGAN. By applying these techniques to our high-resolution GAN frameworks, we observe additional improvement on both qualitative and quantitative results, e.g., the FID score on CelebA $2 5 6 \\times 2 5 6$ dataset is further improved from 10.26 to 8.93. ", + "bbox": [ + 174, + 772, + 825, + 911 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 90, + 825, + 160 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "5 Conclusions, Limitation, and Discussions of Broad Impact ", + "text_level": 1, + "bbox": [ + 174, + 174, + 691, + 191 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "In this work, we provide the first pilot study of building GAN with pure transformers. We have carefully crafted the architectures and thoughtfully designed training techniques. As a result, the proposed TransGAN has achieved state-of-the-art performance across multiple popular datasets, and easily scales up to higher-resolution generative tasks. Although TransGAN provides an encouraging starting point, there is still a large room to explore further, such as achieving state-of-the-art results on $2 5 6 \\times 2 5 6$ generation tasks or going towards extremely high resolution generation tasks (e.g., $1 0 2 4 \\times 1 0 2 4 ,$ ), which would be our future directions. ", + "bbox": [ + 173, + 199, + 825, + 296 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Broader Impact. The proposed generative model can serve as a data engine to alleviate the challenge of data collection. More importantly, using synthesized image examples helps avoid privacy concerns. However, the abuse of advanced generative models may create fake media materials, which demands caution in the future. ", + "bbox": [ + 174, + 303, + 825, + 358 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Acknowledgements ", + "text_level": 1, + "bbox": [ + 174, + 372, + 338, + 388 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "We would like to express our deepest gratitude to the MIT-IBM Watson AI Lab, in particular John Cohn for generously providing us with the computing resources necessary to conduct this research. Z Wang’s work is in part supported by an IBM Faculty Research Award, and the NSF AI Institute for Foundations of Machine Learning (IFML). ", + "bbox": [ + 174, + 397, + 825, + 453 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "References ", + "text_level": 1, + "bbox": [ + 174, + 473, + 266, + 488 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "[1] Ishaan Gulrajani, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, and Aaron C Courville. Improved training of wasserstein gans. In Advances in neural information processing systems, pages 5767–5777, 2017. \n[2] Andrew Brock, Jeff Donahue, and Karen Simonyan. Large scale gan training for high fidelity natural image synthesis. arXiv preprint arXiv:1809.11096, 2018. \n[3] Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A Efros. Image-to-image translation with conditional adversarial networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1125–1134, 2017. \n[4] Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired image-to-image translation using cycle-consistent adversarial networks. In Proceedings of the IEEE international conference on computer vision, pages 2223–2232, 2017. \n[5] Jun-Yan Zhu, Richard Zhang, Deepak Pathak, Trevor Darrell, Alexei A Efros, Oliver Wang, and Eli Shechtman. Toward multimodal image-to-image translation. In Advances in neural information processing systems, pages 465–476, 2017. \n[6] Shuai Yang, Zhangyang Wang, Zhaowen Wang, Ning Xu, Jiaying Liu, and Zongming Guo. Controllable artistic text style transfer via shape-matching gan. In Proceedings of the IEEE International Conference on Computer Vision, pages 4442–4451, 2019. \n[7] Yifan Jiang, Xinyu Gong, Ding Liu, Yu Cheng, Chen Fang, Xiaohui Shen, Jianchao Yang, Pan Zhou, and Zhangyang Wang. Enlightengan: Deep light enhancement without paired supervision. IEEE Transactions on Image Processing, 30:2340–2349, 2021. \n[8] Karol Kurach, Mario Luciˇ c, Xiaohua Zhai, Marcin Michalski, and Sylvain Gelly. A large-scale study ´ on regularization and normalization in gans. In International Conference on Machine Learning, pages 3581–3590. PMLR, 2019. \n[9] Kevin Roth, Aurelien Lucchi, Sebastian Nowozin, and Thomas Hofmann. Stabilizing training of generative adversarial networks through regularization. In Advances in neural information processing systems, pages 2018–2028, 2017. \n[10] Han Zhang, Zizhao Zhang, Augustus Odena, and Honglak Lee. Consistency regularization for generative adversarial networks. arXiv preprint arXiv:1910.12027, 2019. \n[11] Lars Mescheder, Andreas Geiger, and Sebastian Nowozin. Which training methods for gans do actually converge? arXiv preprint arXiv:1801.04406, 2018. \n[12] Xudong Mao, Qing Li, Haoran Xie, Raymond YK Lau, Zhen Wang, and Stephen Paul Smolley. Least squares generative adversarial networks. In Proceedings of the IEEE international conference on computer vision, pages 2794–2802, 2017. \n[13] Alexia Jolicoeur-Martineau. The relativistic discriminator: a key element missing from standard gan. arXiv preprint arXiv:1807.00734, 2018. \n[14] Chun-Liang Li, Wei-Cheng Chang, Yu Cheng, Yiming Yang, and Barnabás Póczos. Mmd gan: Towards deeper understanding of moment matching network. In Advances in neural information processing systems, pages 2203–2213, 2017. \n[15] Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen. Improved techniques for training gans. arXiv preprint arXiv:1606.03498, 2016. \n[16] Tero Karras, Timo Aila, Samuli Laine, and Jaakko Lehtinen. Progressive growing of gans for improved quality, stability, and variation. arXiv preprint arXiv:1710.10196, 2017. \n[17] Mario Lucic, Karol Kurach, Marcin Michalski, Olivier Bousquet, and Sylvain Gelly. Are gans created equal? a large-scale study. In Proceedings of the 32nd International Conference on Neural Information Processing Systems, pages 698–707, 2018. \n[18] Xinyu Gong, Shiyu Chang, Yifan Jiang, and Zhangyang Wang. Autogan: Neural architecture search for generative adversarial networks. In Proceedings of the IEEE International Conference on Computer Vision, pages 3224–3234, 2019. \n[19] Chen Gao, Yunpeng Chen, Si Liu, Zhenxiong Tan, and Shuicheng Yan. Adversarialnas: Adversarial neural architecture search for gans. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5680–5689, 2020. \n[20] Yuan Tian, Qin Wang, Zhiwu Huang, Wen Li, Dengxin Dai, Minghao Yang, Jun Wang, and Olga Fink. Offpolicy reinforcement learning for efficient and effective gan architecture search. In European Conference on Computer Vision, pages 175–192. Springer, 2020. \n[21] Han Zhang, Ian Goodfellow, Dimitris Metaxas, and Augustus Odena. Self-attention generative adversarial networks. In International conference on machine learning, pages 7354–7363. PMLR, 2019. \n[22] Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative adversarial networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 4401–4410, 2019. \n[23] Patrick Esser, Robin Rombach, and Björn Ommer. Taming transformers for high-resolution image synthesis. arXiv preprint arXiv:2012.09841, 2020. \n[24] Ian J Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial networks. arXiv preprint arXiv:1406.2661, 2014. \n[25] Emily Denton, Soumith Chintala, Arthur Szlam, and Rob Fergus. Deep generative image models using a laplacian pyramid of adversarial networks. arXiv preprint arXiv:1506.05751, 2015. \n[26] Alec Radford, Luke Metz, and Soumith Chintala. Unsupervised representation learning with deep convolutional generative adversarial networks. arXiv preprint arXiv:1511.06434, 2015. \n[27] Xiaolong Wang, Ross Girshick, Abhinav Gupta, and Kaiming He. Non-local neural networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7794–7803, 2018. \n[28] Han Hu, Zheng Zhang, Zhenda Xie, and Stephen Lin. Local relation networks for image recognition. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3464–3473, 2019. \n[29] Yu-Syuan Xu, Shou-Yao Roy Tseng, Yu Tseng, Hsien-Kai Kuo, and Yi-Min Tsai. Unified dynamic convolutional network for super-resolution with variational degradations. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12496–12505, 2020. \n[30] David Ha, Andrew Dai, and Quoc V Le. Hypernetworks. arXiv preprint arXiv:1609.09106, 2016. \n[31] Ben Mildenhall, Jonathan T Barron, Jiawen Chen, Dillon Sharlet, Ren Ng, and Robert Carroll. Burst denoising with kernel prediction networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 2502–2510, 2018. \n[32] Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-to-end object detection with transformers. arXiv preprint arXiv:2005.12872, 2020. \n[33] Yanhong Zeng, Jianlong Fu, and Hongyang Chao. Learning joint spatial-temporal transformations for video inpainting. In ECCV. Springer, 2020. \n[34] Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020. \n[35] Drew A Hudson and C Lawrence Zitnick. Generative adversarial transformers. arXiv preprint arXiv:2103.01209, 2021. \n[36] Fuzhi Yang, Huan Yang, Jianlong Fu, Hongtao Lu, and Baining Guo. Learning texture transformer network for image super-resolution. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5791–5800, 2020. \n[37] Hanting Chen, Yunhe Wang, Tianyu Guo, Chang Xu, Yiping Deng, Zhenhua Liu, Siwei Ma, Chunjing Xu, Chao Xu, and Wen Gao. Pre-trained image processing transformer. arXiv preprint arXiv:2012.00364, 2020. \n[38] Jie Gui, Zhenan Sun, Yonggang Wen, Dacheng Tao, and Jieping Ye. A review on generative adversarial networks: Algorithms, theory, and applications. arXiv preprint arXiv:2001.06937, 2020. \n[39] Edgar Schonfeld, Bernt Schiele, and Anna Khoreva. A u-net based discriminator for generative adversarial networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8207–8216, 2020. \n[40] Tero Karras, Samuli Laine, Miika Aittala, Janne Hellsten, Jaakko Lehtinen, and Timo Aila. Analyzing and improving the image quality of stylegan. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8110–8119, 2020. \n[41] Haotao Wang, Shupeng Gui, Haichuan Yang, Ji Liu, and Zhangyang Wang. Gan slimming: All-in-one gan compression by a unified optimization framework. In European Conference on Computer Vision, pages 54–73. Springer, 2020. \n[42] Christian Ledig, Lucas Theis, Ferenc Huszár, Jose Caballero, Andrew Cunningham, Alejandro Acosta, Andrew Aitken, Alykhan Tejani, Johannes Totz, Zehan Wang, et al. Photo-realistic single image superresolution using a generative adversarial network. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 4681–4690, 2017. \n[43] Orest Kupyn, Tetiana Martyniuk, Junru Wu, and Zhangyang Wang. Deblurgan-v2: Deblurring (ordersof-magnitude) faster and better. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 8878–8887, 2019. \n[44] Xi Ouyang, Yu Cheng, Yifan Jiang, Chun-Liang Li, and Pan Zhou. Pedestrian-synthesis-gan: Generating pedestrian data in real scene and beyond. arXiv preprint arXiv:1804.02047, 2018. \n[45] Jiahui Yu, Zhe Lin, Jimei Yang, Xiaohui Shen, Xin Lu, and Thomas S Huang. Generative image inpainting with contextual attention. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 5505–5514, 2018. \n[46] Martin Arjovsky, Soumith Chintala, and Léon Bottou. Wasserstein gan. arXiv preprint arXiv:1701.07875, 2017. \n[47] Tianlong Chen, Yu Cheng, Zhe Gan, Jingjing Liu, and Zhangyang Wang. Ultra-data-efficient gan training: Drawing a lottery ticket first, then training it toughly. arXiv preprint arXiv:2103.00397, 2021. \n[48] Takeru Miyato, Toshiki Kataoka, Masanori Koyama, and Yuichi Yoshida. Spectral normalization for generative adversarial networks. arXiv preprint arXiv:1802.05957, 2018. \n[49] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in neural information processing systems, pages 5998–6008, 2017. \n[50] Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Hervé Jégou. Training data-efficient image transformers & distillation through attention. arXiv preprint arXiv:2012.12877, 2020. \n[51] Hugo Touvron, Matthieu Cord, Alexandre Sablayrolles, Gabriel Synnaeve, and Hervé Jégou. Going deeper with image transformers. arXiv preprint arXiv:2103.17239, 2021. \n[52] Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. arXiv preprint arXiv:2103.14030, 2021. \n[53] Chun-Fu Chen, Quanfu Fan, and Rameswar Panda. Crossvit: Cross-attention multi-scale vision transformer for image classification. arXiv preprint arXiv:2103.14899, 2021. \n[54] Pengchuan Zhang, Xiyang Dai, Jianwei Yang, Bin Xiao, Lu Yuan, Lei Zhang, and Jianfeng Gao. Multiscale vision longformer: A new vision transformer for high-resolution image encoding. arXiv preprint arXiv:2103.15358, 2021. \n[55] Changlin Li, Tao Tang, Guangrun Wang, Jiefeng Peng, Bing Wang, Xiaodan Liang, and Xiaojun Chang. Bossnas: Exploring hybrid cnn-transformers with block-wisely self-supervised neural architecture search. arXiv preprint arXiv:2103.12424, 2021. \n[56] Yutong Xie, Jianpeng Zhang, Chunhua Shen, and Yong Xia. Cotr: Efficiently bridging cnn and transformer for 3d medical image segmentation. arXiv preprint arXiv:2103.03024, 2021. \n[57] Hengshuang Zhao, Li Jiang, Jiaya Jia, Philip Torr, and Vladlen Koltun. Point transformer. arXiv preprint arXiv:2012.09164, 2020. \n[58] Sixiao Zheng, Jiachen Lu, Hengshuang Zhao, Xiatian Zhu, Zekun Luo, Yabiao Wang, Yanwei Fu, Jianfeng Feng, Tao Xiang, Philip HS Torr, et al. Rethinking semantic segmentation from a sequence-to-sequence perspective with transformers. arXiv preprint arXiv:2012.15840, 2020. \n[59] Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, and Jifeng Dai. Deformable detr: Deformable transformers for end-to-end object detection. arXiv preprint arXiv:2010.04159, 2020. \n[60] Kai Han, Yunhe Wang, Hanting Chen, Xinghao Chen, Jianyuan Guo, Zhenhua Liu, Yehui Tang, An Xiao, Chunjing Xu, Yixing Xu, et al. A survey on visual transformer. arXiv preprint arXiv:2012.12556, 2020. \n[61] Niki Parmar, Ashish Vaswani, Jakob Uszkoreit, Lukasz Kaiser, Noam Shazeer, Alexander Ku, and Dustin Tran. Image transformer. In International Conference on Machine Learning, pages 4055–4064. PMLR, 2018. \n[62] Rewon Child, Scott Gray, Alec Radford, and Ilya Sutskever. Generating long sequences with sparse transformers. arXiv preprint arXiv:1904.10509, 2019. \n[63] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805, 2018. \n[64] Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, and Vikas Singh. Nystromformer: A nystrom-based algorithm for approximating self-attention. arXiv preprint arXiv:2102.03902, 2021. \n[65] Iz Beltagy, Matthew E Peters, and Arman Cohan. Longformer: The long-document transformer. arXiv preprint arXiv:2004.05150, 2020. \n[66] Gedas Bertasius, Heng Wang, and Lorenzo Torresani. Is space-time attention all you need for video understanding? arXiv preprint arXiv:2102.05095, 2021. \n[67] Manoj Kumar, Dirk Weissenborn, and Nal Kalchbrenner. Colorization transformer. arXiv preprint arXiv:2102.04432, 2021. \n[68] Chen Sun, Abhinav Shrivastava, Saurabh Singh, and Abhinav Gupta. Revisiting unreasonable effectiveness of data in deep learning era. In Proceedings of the IEEE international conference on computer vision, pages 843–852, 2017. \n[69] Shengyu Zhao, Zhijian Liu, Ji Lin, Jun-Yan Zhu, and Song Han. Differentiable augmentation for dataefficient gan training. arXiv preprint arXiv:2006.10738, 2020. \n[70] Tero Karras, Miika Aittala, Janne Hellsten, Samuli Laine, Jaakko Lehtinen, and Timo Aila. Training generative adversarial networks with limited data. arXiv preprint arXiv:2006.06676, 2020. \n[71] Peter Shaw, Jakob Uszkoreit, and Ashish Vaswani. Self-attention with relative position representations. arXiv preprint arXiv:1803.02155, 2018. \n[72] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer. arXiv preprint arXiv:1910.10683, 2019. \n[73] Cheng-Zhi Anna Huang, Ashish Vaswani, Jakob Uszkoreit, Noam Shazeer, Curtis Hawthorne, AM Dai, MD Hoffman, and D Eck. Music transformer: Generating music with long-term structure (2018). arXiv preprint arXiv:1809.04281, 2018. \n[74] Chieh Hubert Lin, Chia-Che Chang, Yu-Sheng Chen, Da-Cheng Juan, Wei Wei, and Hwann-Tzong Chen. Coco-gan: Generation by parts via conditional coordinating. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 4512–4521, 2019. \n[75] Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems, 25:1097–1105, 2012. \n[76] Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. Layer normalization. arXiv preprint arXiv:1607.06450, 2016. \n[77] Sergey Ioffe and Christian Szegedy. Batch normalization: Accelerating deep network training by reducing internal covariate shift. In International conference on machine learning, pages 448–456. PMLR, 2015. \n[78] Dmitry Ulyanov, Andrea Vedaldi, and Victor Lempitsky. Instance normalization: The missing ingredient for fast stylization. arXiv preprint arXiv:1607.08022, 2016. \n[79] Alex Krizhevsky, Geoffrey Hinton, et al. Learning multiple layers of features from tiny images. 2009. \n[80] Adam Coates, Andrew Ng, and Honglak Lee. An analysis of single-layer networks in unsupervised feature learning. In Proceedings of the fourteenth international conference on artificial intelligence and statistics, pages 215–223. JMLR Workshop and Conference Proceedings, 2011. \n[81] Ziwei Liu, Ping Luo, Xiaogang Wang, and Xiaoou Tang. Deep learning face attributes in the wild. In Proceedings of International Conference on Computer Vision (ICCV), 2015. \n[82] Fisher Yu, Ari Seff, Yinda Zhang, Shuran Song, Thomas Funkhouser, and Jianxiong Xiao. Lsun: Construction of a large-scale image dataset using deep learning with humans in the loop. arXiv preprint arXiv:1506.03365, 2015. \n[83] Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium. arXiv preprint arXiv:1706.08500, 2017. \n[84] Animesh Karnewar and Oliver Wang. Msg-gan: Multi-scale gradients for generative adversarial networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7799– 7808, 2020. \n[85] Ting Chen, Mario Lucic, Neil Houlsby, and Sylvain Gelly. On self modulation for generative adversarial networks. arXiv preprint arXiv:1810.01365, 2018. \n[86] Long Zhao, Zizhao Zhang, Ting Chen, Dimitris N Metaxas, and Han Zhang. Improved transformer for high-resolution gans. arXiv preprint arXiv:2106.07631, 2021. \n[87] Tero Karras, Miika Aittala, Samuli Laine, Erik Härkönen, Janne Hellsten, Jaakko Lehtinen, and Timo Aila. Alias-free generative adversarial networks. arXiv preprint arXiv:2106.12423, 2021. ", + "bbox": [ + 178, + 494, + 826, + 911 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "", + "bbox": [ + 166, + 51, + 828, + 921 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 47, + 828, + 922 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 36, + 828, + 917 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 89, + 828, + 761 + ], + "page_idx": 13 + } +] \ No newline at end of file diff --git a/parse/train/1GTpBZvNUrk/1GTpBZvNUrk_middle.json b/parse/train/1GTpBZvNUrk/1GTpBZvNUrk_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..dc17bf6e7e4cec766a9555cc25bc00d000cdc308 --- /dev/null +++ b/parse/train/1GTpBZvNUrk/1GTpBZvNUrk_middle.json @@ -0,0 +1,36093 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 114, + 97, + 499, + 137 + ], + "lines": [ + { + "bbox": [ + 113, + 97, + 498, + 118 + ], + "spans": [ + { + "bbox": [ + 113, + 97, + 498, + 118 + ], + "score": 1.0, + "content": "TransGAN: Two Pure Transformers Can Make One", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 167, + 116, + 444, + 140 + ], + "spans": [ + { + "bbox": [ + 167, + 116, + 444, + 140 + ], + "score": 1.0, + "content": "Strong GAN, and That Can Scale Up", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 167, + 178, + 445, + 225 + ], + "lines": [ + { + "bbox": [ + 199, + 177, + 410, + 193 + ], + "spans": [ + { + "bbox": [ + 199, + 177, + 410, + 193 + ], + "score": 1.0, + "content": "Yifan Jiang1, Shiyu Chang2,3, Zhangyang Wang1", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 243, + 189, + 368, + 203 + ], + "spans": [ + { + "bbox": [ + 243, + 189, + 368, + 203 + ], + "score": 1.0, + "content": "1University of Texas at Austin", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 207, + 200, + 405, + 215 + ], + "spans": [ + { + "bbox": [ + 207, + 200, + 405, + 215 + ], + "score": 1.0, + "content": "2UC Santa Barbara 3MIT-IBM Watson AI Lab", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 166, + 213, + 445, + 226 + ], + "spans": [ + { + "bbox": [ + 166, + 213, + 445, + 226 + ], + "score": 1.0, + "content": "{yifanjiang97,atlaswang}@utexas.edu, chang87@ucsb.edu", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3.5 + }, + { + "type": "title", + "bbox": [ + 283, + 243, + 328, + 257 + ], + "lines": [ + { + "bbox": [ + 282, + 243, + 330, + 257 + ], + "spans": [ + { + "bbox": [ + 282, + 243, + 330, + 257 + ], + "score": 1.0, + "content": "Abstract", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 142, + 270, + 469, + 553 + ], + "lines": [ + { + "bbox": [ + 142, + 270, + 469, + 281 + ], + "spans": [ + { + "bbox": [ + 142, + 270, + 469, + 281 + ], + "score": 1.0, + "content": "The recent explosive interest on transformers has suggested their potential to", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 280, + 470, + 293 + ], + "spans": [ + { + "bbox": [ + 141, + 280, + 470, + 293 + ], + "score": 1.0, + "content": "become powerful “universal\" models for computer vision tasks, such as classi-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 291, + 469, + 303 + ], + "spans": [ + { + "bbox": [ + 141, + 291, + 469, + 303 + ], + "score": 1.0, + "content": "fication, detection, and segmentation. While those attempts mainly study the", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 302, + 470, + 315 + ], + "spans": [ + { + "bbox": [ + 141, + 302, + 470, + 315 + ], + "score": 1.0, + "content": "discriminative models, we explore transformers on some more notoriously diffi-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 312, + 470, + 326 + ], + "spans": [ + { + "bbox": [ + 141, + 312, + 470, + 326 + ], + "score": 1.0, + "content": "cult vision tasks, e.g., generative adversarial networks (GANs). Our goal is to", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 324, + 470, + 336 + ], + "spans": [ + { + "bbox": [ + 141, + 324, + 470, + 336 + ], + "score": 1.0, + "content": "conduct the first pilot study in building a GAN completely free of convolutions,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 334, + 471, + 347 + ], + "spans": [ + { + "bbox": [ + 141, + 334, + 471, + 347 + ], + "score": 1.0, + "content": "using only pure transformer-based architectures. Our vanilla GAN architecture,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 345, + 470, + 358 + ], + "spans": [ + { + "bbox": [ + 141, + 345, + 470, + 358 + ], + "score": 1.0, + "content": "dubbed TransGAN, consists of a memory-friendly transformer-based generator", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 356, + 469, + 369 + ], + "spans": [ + { + "bbox": [ + 141, + 356, + 469, + 369 + ], + "score": 1.0, + "content": "that progressively increases feature resolution, and correspondingly a multi-scale", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 367, + 470, + 380 + ], + "spans": [ + { + "bbox": [ + 141, + 367, + 470, + 380 + ], + "score": 1.0, + "content": "discriminator to capture simultaneously semantic contexts and low-level textures.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 378, + 470, + 391 + ], + "spans": [ + { + "bbox": [ + 141, + 378, + 470, + 391 + ], + "score": 1.0, + "content": "On top of them, we introduce the new module of grid self-attention for alleviating", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 389, + 469, + 401 + ], + "spans": [ + { + "bbox": [ + 141, + 389, + 469, + 401 + ], + "score": 1.0, + "content": "the memory bottleneck further, in order to scale up TransGAN to high-resolution", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 401, + 470, + 412 + ], + "spans": [ + { + "bbox": [ + 141, + 401, + 470, + 412 + ], + "score": 1.0, + "content": "generation. We also develop a unique training recipe including a series of tech-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 411, + 470, + 424 + ], + "spans": [ + { + "bbox": [ + 141, + 411, + 470, + 424 + ], + "score": 1.0, + "content": "niques that can mitigate the training instability issues of TransGAN, such as data", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 422, + 470, + 434 + ], + "spans": [ + { + "bbox": [ + 141, + 422, + 470, + 434 + ], + "score": 1.0, + "content": "augmentation, modified normalization, and relative position encoding. Our best", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 142, + 433, + 470, + 445 + ], + "spans": [ + { + "bbox": [ + 142, + 433, + 470, + 445 + ], + "score": 1.0, + "content": "architecture achieves highly competitive performance compared to current state-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 142, + 444, + 469, + 456 + ], + "spans": [ + { + "bbox": [ + 142, + 444, + 469, + 456 + ], + "score": 1.0, + "content": "of-the-art GANs using convolutional backbones. Specifically, TransGAN sets the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 455, + 469, + 466 + ], + "spans": [ + { + "bbox": [ + 141, + 455, + 469, + 466 + ], + "score": 1.0, + "content": "new state-of-the-art inception score of 10.43 and FID of 18.28 on STL-10. It also", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 141, + 466, + 469, + 477 + ], + "spans": [ + { + "bbox": [ + 141, + 466, + 469, + 477 + ], + "score": 1.0, + "content": "reaches the inception score of 9.02 and FID of 9.26 on CIFAR-10, and 5.28 FID on", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 142, + 476, + 470, + 488 + ], + "spans": [ + { + "bbox": [ + 142, + 476, + 174, + 488 + ], + "score": 1.0, + "content": "CelebA", + "type": "text" + }, + { + "bbox": [ + 174, + 477, + 221, + 487 + ], + "score": 0.89, + "content": "\\mathbf { 1 2 8 \\times 1 2 8 }", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 476, + 470, + 488 + ], + "score": 1.0, + "content": ", respectively: both on par with the current best results. When it", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 141, + 487, + 470, + 499 + ], + "spans": [ + { + "bbox": [ + 141, + 487, + 271, + 499 + ], + "score": 1.0, + "content": "comes to higher-resolution (e.g.", + "type": "text" + }, + { + "bbox": [ + 271, + 488, + 320, + 498 + ], + "score": 0.89, + "content": "\\mathbf { 2 5 6 \\times 2 5 6 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 487, + 470, + 499 + ], + "score": 1.0, + "content": ") generation tasks, such as on CelebA-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 141, + 498, + 469, + 510 + ], + "spans": [ + { + "bbox": [ + 141, + 498, + 469, + 510 + ], + "score": 1.0, + "content": "HQ and LSUN-Church, TransGAN continues to produce diverse visual examples", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 142, + 509, + 469, + 521 + ], + "spans": [ + { + "bbox": [ + 142, + 509, + 469, + 521 + ], + "score": 1.0, + "content": "with high fidelity and reasonable texture details. In addition, we dive deep into", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 142, + 520, + 469, + 531 + ], + "spans": [ + { + "bbox": [ + 142, + 520, + 469, + 531 + ], + "score": 1.0, + "content": "the transformer-based generation models to understand how their behaviors differ", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 141, + 531, + 469, + 543 + ], + "spans": [ + { + "bbox": [ + 141, + 531, + 469, + 543 + ], + "score": 1.0, + "content": "from convolutional ones, by visualizing training dynamics. The code is available", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 141, + 542, + 360, + 554 + ], + "spans": [ + { + "bbox": [ + 141, + 542, + 360, + 554 + ], + "score": 1.0, + "content": "at: https://github.com/VITA-Group/TransGAN.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 19.5 + }, + { + "type": "title", + "bbox": [ + 107, + 573, + 190, + 586 + ], + "lines": [ + { + "bbox": [ + 105, + 572, + 192, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 192, + 588 + ], + "score": 1.0, + "content": "1 Introduction", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 593, + 505, + 713 + ], + "lines": [ + { + "bbox": [ + 105, + 592, + 507, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 592, + 507, + 607 + ], + "score": 1.0, + "content": "Generative adversarial networks (GANs) have gained considerable success on numerous tasks [1, 2,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 604, + 506, + 616 + ], + "spans": [ + { + "bbox": [ + 106, + 604, + 506, + 616 + ], + "score": 1.0, + "content": "3, 4, 5, 6, 7]. Unfortunately, GANs suffer from the notorious training instability, and numerous efforts", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 615, + 506, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 506, + 628 + ], + "score": 1.0, + "content": "have been devoted to stabilizing GAN training, introducing various regularization terms [8, 9, 10, 11],", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 625, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 506, + 639 + ], + "score": 1.0, + "content": "better losses [1, 12, 13, 14], and training recipes [15, 16]. Among them, one important route to", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 636, + 506, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 506, + 650 + ], + "score": 1.0, + "content": "improving GANs examines their neural architectures. [17, 8] reported a large-scale study of GANs", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "score": 1.0, + "content": "and observed that when serving as (generator) backbones, popular neural architectures perform", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 658, + 506, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 506, + 671 + ], + "score": 1.0, + "content": "comparably well across the considered datasets. Their ablation study suggested that most of the", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 670, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 106, + 670, + 505, + 682 + ], + "score": 1.0, + "content": "variations applied in the ResNet family resulted in very marginal improvements. Nevertheless, neural", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 681, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 681, + 505, + 693 + ], + "score": 1.0, + "content": "architecture search (NAS) was later introduced to GANs and suggests enhanced backbone designs", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 691, + 505, + 704 + ], + "spans": [ + { + "bbox": [ + 105, + 691, + 505, + 704 + ], + "score": 1.0, + "content": "are also important for improving GANs, just like for other computer vision tasks. Those works", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 701, + 505, + 716 + ], + "spans": [ + { + "bbox": [ + 105, + 701, + 505, + 716 + ], + "score": 1.0, + "content": "are consistently able to discover stronger GAN architectures beyond the standard ResNet topology", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 39 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 731, + 385, + 741 + ], + "lines": [ + { + "bbox": [ + 105, + 730, + 387, + 744 + ], + "spans": [ + { + "bbox": [ + 105, + 730, + 387, + 744 + ], + "score": 1.0, + "content": "35th Conference on Neural Information Processing Systems (NeurIPS 2021).", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 114, + 97, + 499, + 137 + ], + "lines": [ + { + "bbox": [ + 113, + 97, + 498, + 118 + ], + "spans": [ + { + "bbox": [ + 113, + 97, + 498, + 118 + ], + "score": 1.0, + "content": "TransGAN: Two Pure Transformers Can Make One", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 167, + 116, + 444, + 140 + ], + "spans": [ + { + "bbox": [ + 167, + 116, + 444, + 140 + ], + "score": 1.0, + "content": "Strong GAN, and That Can Scale Up", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 167, + 178, + 445, + 225 + ], + "lines": [ + { + "bbox": [ + 199, + 177, + 410, + 193 + ], + "spans": [ + { + "bbox": [ + 199, + 177, + 410, + 193 + ], + "score": 1.0, + "content": "Yifan Jiang1, Shiyu Chang2,3, Zhangyang Wang1", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 243, + 189, + 368, + 203 + ], + "spans": [ + { + "bbox": [ + 243, + 189, + 368, + 203 + ], + "score": 1.0, + "content": "1University of Texas at Austin", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 207, + 200, + 405, + 215 + ], + "spans": [ + { + "bbox": [ + 207, + 200, + 405, + 215 + ], + "score": 1.0, + "content": "2UC Santa Barbara 3MIT-IBM Watson AI Lab", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 166, + 213, + 445, + 226 + ], + "spans": [ + { + "bbox": [ + 166, + 213, + 445, + 226 + ], + "score": 1.0, + "content": "{yifanjiang97,atlaswang}@utexas.edu, chang87@ucsb.edu", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3.5, + "bbox_fs": [ + 166, + 177, + 445, + 226 + ] + }, + { + "type": "title", + "bbox": [ + 283, + 243, + 328, + 257 + ], + "lines": [ + { + "bbox": [ + 282, + 243, + 330, + 257 + ], + "spans": [ + { + "bbox": [ + 282, + 243, + 330, + 257 + ], + "score": 1.0, + "content": "Abstract", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 142, + 270, + 469, + 553 + ], + "lines": [ + { + "bbox": [ + 142, + 270, + 469, + 281 + ], + "spans": [ + { + "bbox": [ + 142, + 270, + 469, + 281 + ], + "score": 1.0, + "content": "The recent explosive interest on transformers has suggested their potential to", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 280, + 470, + 293 + ], + "spans": [ + { + "bbox": [ + 141, + 280, + 470, + 293 + ], + "score": 1.0, + "content": "become powerful “universal\" models for computer vision tasks, such as classi-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 291, + 469, + 303 + ], + "spans": [ + { + "bbox": [ + 141, + 291, + 469, + 303 + ], + "score": 1.0, + "content": "fication, detection, and segmentation. While those attempts mainly study the", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 302, + 470, + 315 + ], + "spans": [ + { + "bbox": [ + 141, + 302, + 470, + 315 + ], + "score": 1.0, + "content": "discriminative models, we explore transformers on some more notoriously diffi-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 312, + 470, + 326 + ], + "spans": [ + { + "bbox": [ + 141, + 312, + 470, + 326 + ], + "score": 1.0, + "content": "cult vision tasks, e.g., generative adversarial networks (GANs). Our goal is to", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 324, + 470, + 336 + ], + "spans": [ + { + "bbox": [ + 141, + 324, + 470, + 336 + ], + "score": 1.0, + "content": "conduct the first pilot study in building a GAN completely free of convolutions,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 334, + 471, + 347 + ], + "spans": [ + { + "bbox": [ + 141, + 334, + 471, + 347 + ], + "score": 1.0, + "content": "using only pure transformer-based architectures. Our vanilla GAN architecture,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 345, + 470, + 358 + ], + "spans": [ + { + "bbox": [ + 141, + 345, + 470, + 358 + ], + "score": 1.0, + "content": "dubbed TransGAN, consists of a memory-friendly transformer-based generator", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 356, + 469, + 369 + ], + "spans": [ + { + "bbox": [ + 141, + 356, + 469, + 369 + ], + "score": 1.0, + "content": "that progressively increases feature resolution, and correspondingly a multi-scale", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 367, + 470, + 380 + ], + "spans": [ + { + "bbox": [ + 141, + 367, + 470, + 380 + ], + "score": 1.0, + "content": "discriminator to capture simultaneously semantic contexts and low-level textures.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 378, + 470, + 391 + ], + "spans": [ + { + "bbox": [ + 141, + 378, + 470, + 391 + ], + "score": 1.0, + "content": "On top of them, we introduce the new module of grid self-attention for alleviating", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 389, + 469, + 401 + ], + "spans": [ + { + "bbox": [ + 141, + 389, + 469, + 401 + ], + "score": 1.0, + "content": "the memory bottleneck further, in order to scale up TransGAN to high-resolution", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 401, + 470, + 412 + ], + "spans": [ + { + "bbox": [ + 141, + 401, + 470, + 412 + ], + "score": 1.0, + "content": "generation. We also develop a unique training recipe including a series of tech-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 411, + 470, + 424 + ], + "spans": [ + { + "bbox": [ + 141, + 411, + 470, + 424 + ], + "score": 1.0, + "content": "niques that can mitigate the training instability issues of TransGAN, such as data", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 422, + 470, + 434 + ], + "spans": [ + { + "bbox": [ + 141, + 422, + 470, + 434 + ], + "score": 1.0, + "content": "augmentation, modified normalization, and relative position encoding. Our best", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 142, + 433, + 470, + 445 + ], + "spans": [ + { + "bbox": [ + 142, + 433, + 470, + 445 + ], + "score": 1.0, + "content": "architecture achieves highly competitive performance compared to current state-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 142, + 444, + 469, + 456 + ], + "spans": [ + { + "bbox": [ + 142, + 444, + 469, + 456 + ], + "score": 1.0, + "content": "of-the-art GANs using convolutional backbones. Specifically, TransGAN sets the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 455, + 469, + 466 + ], + "spans": [ + { + "bbox": [ + 141, + 455, + 469, + 466 + ], + "score": 1.0, + "content": "new state-of-the-art inception score of 10.43 and FID of 18.28 on STL-10. It also", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 141, + 466, + 469, + 477 + ], + "spans": [ + { + "bbox": [ + 141, + 466, + 469, + 477 + ], + "score": 1.0, + "content": "reaches the inception score of 9.02 and FID of 9.26 on CIFAR-10, and 5.28 FID on", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 142, + 476, + 470, + 488 + ], + "spans": [ + { + "bbox": [ + 142, + 476, + 174, + 488 + ], + "score": 1.0, + "content": "CelebA", + "type": "text" + }, + { + "bbox": [ + 174, + 477, + 221, + 487 + ], + "score": 0.89, + "content": "\\mathbf { 1 2 8 \\times 1 2 8 }", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 476, + 470, + 488 + ], + "score": 1.0, + "content": ", respectively: both on par with the current best results. When it", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 141, + 487, + 470, + 499 + ], + "spans": [ + { + "bbox": [ + 141, + 487, + 271, + 499 + ], + "score": 1.0, + "content": "comes to higher-resolution (e.g.", + "type": "text" + }, + { + "bbox": [ + 271, + 488, + 320, + 498 + ], + "score": 0.89, + "content": "\\mathbf { 2 5 6 \\times 2 5 6 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 487, + 470, + 499 + ], + "score": 1.0, + "content": ") generation tasks, such as on CelebA-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 141, + 498, + 469, + 510 + ], + "spans": [ + { + "bbox": [ + 141, + 498, + 469, + 510 + ], + "score": 1.0, + "content": "HQ and LSUN-Church, TransGAN continues to produce diverse visual examples", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 142, + 509, + 469, + 521 + ], + "spans": [ + { + "bbox": [ + 142, + 509, + 469, + 521 + ], + "score": 1.0, + "content": "with high fidelity and reasonable texture details. In addition, we dive deep into", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 142, + 520, + 469, + 531 + ], + "spans": [ + { + "bbox": [ + 142, + 520, + 469, + 531 + ], + "score": 1.0, + "content": "the transformer-based generation models to understand how their behaviors differ", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 141, + 531, + 469, + 543 + ], + "spans": [ + { + "bbox": [ + 141, + 531, + 469, + 543 + ], + "score": 1.0, + "content": "from convolutional ones, by visualizing training dynamics. The code is available", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 141, + 542, + 360, + 554 + ], + "spans": [ + { + "bbox": [ + 141, + 542, + 360, + 554 + ], + "score": 1.0, + "content": "at: https://github.com/VITA-Group/TransGAN.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 19.5, + "bbox_fs": [ + 141, + 270, + 471, + 554 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 573, + 190, + 586 + ], + "lines": [ + { + "bbox": [ + 105, + 572, + 192, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 192, + 588 + ], + "score": 1.0, + "content": "1 Introduction", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 593, + 505, + 713 + ], + "lines": [ + { + "bbox": [ + 105, + 592, + 507, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 592, + 507, + 607 + ], + "score": 1.0, + "content": "Generative adversarial networks (GANs) have gained considerable success on numerous tasks [1, 2,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 604, + 506, + 616 + ], + "spans": [ + { + "bbox": [ + 106, + 604, + 506, + 616 + ], + "score": 1.0, + "content": "3, 4, 5, 6, 7]. Unfortunately, GANs suffer from the notorious training instability, and numerous efforts", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 615, + 506, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 506, + 628 + ], + "score": 1.0, + "content": "have been devoted to stabilizing GAN training, introducing various regularization terms [8, 9, 10, 11],", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 625, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 506, + 639 + ], + "score": 1.0, + "content": "better losses [1, 12, 13, 14], and training recipes [15, 16]. Among them, one important route to", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 636, + 506, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 506, + 650 + ], + "score": 1.0, + "content": "improving GANs examines their neural architectures. [17, 8] reported a large-scale study of GANs", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 506, + 661 + ], + "score": 1.0, + "content": "and observed that when serving as (generator) backbones, popular neural architectures perform", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 658, + 506, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 506, + 671 + ], + "score": 1.0, + "content": "comparably well across the considered datasets. Their ablation study suggested that most of the", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 670, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 106, + 670, + 505, + 682 + ], + "score": 1.0, + "content": "variations applied in the ResNet family resulted in very marginal improvements. Nevertheless, neural", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 681, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 681, + 505, + 693 + ], + "score": 1.0, + "content": "architecture search (NAS) was later introduced to GANs and suggests enhanced backbone designs", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 691, + 505, + 704 + ], + "spans": [ + { + "bbox": [ + 105, + 691, + 505, + 704 + ], + "score": 1.0, + "content": "are also important for improving GANs, just like for other computer vision tasks. Those works", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 701, + 505, + 716 + ], + "spans": [ + { + "bbox": [ + 105, + 701, + 505, + 716 + ], + "score": 1.0, + "content": "are consistently able to discover stronger GAN architectures beyond the standard ResNet topology", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 262, + 505, + 275 + ], + "spans": [ + { + "bbox": [ + 106, + 262, + 505, + 275 + ], + "score": 1.0, + "content": "[18, 19, 20]. Other efforts include customized modules such as self-attention [21], style-based", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 273, + 411, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 273, + 411, + 286 + ], + "score": 1.0, + "content": "generator [22], and autoregressive transformer-based part composition [23].", + "type": "text", + "cross_page": true + } + ], + "index": 7 + } + ], + "index": 39, + "bbox_fs": [ + 105, + 592, + 507, + 716 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 109, + 70, + 502, + 210 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 70, + 502, + 210 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 70, + 502, + 210 + ], + "spans": [ + { + "bbox": [ + 109, + 70, + 502, + 210 + ], + "score": 0.973, + "type": "image", + "image_path": "adcb25935a6e25e1b2ab70151735aa65bcbfa9d2b385186c0dc4a0005b6915f2.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 109, + 70, + 502, + 116.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 109, + 116.66666666666666, + 502, + 163.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 109, + 163.33333333333331, + 502, + 209.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 215, + 505, + 248 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 214, + 506, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 506, + 229 + ], + "score": 1.0, + "content": "Figure 1: Representative visual examples synthesized by TransGAN, without using convolutional", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 226, + 505, + 239 + ], + "spans": [ + { + "bbox": [ + 106, + 227, + 360, + 239 + ], + "score": 1.0, + "content": "layers. (a) The synthesized visual examples on CelebA-HQ", + "type": "text" + }, + { + "bbox": [ + 360, + 226, + 407, + 237 + ], + "score": 0.86, + "content": "2 5 6 \\times 2 5 6 )", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 227, + 505, + 239 + ], + "score": 1.0, + "content": "dataset. (b) The linear", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 237, + 446, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 366, + 249 + ], + "score": 1.0, + "content": "interpolation results between two latent vectors, on CelebA-HQ (", + "type": "text" + }, + { + "bbox": [ + 366, + 237, + 411, + 248 + ], + "score": 0.84, + "content": "2 5 6 \\times 2 5 6 )", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 237, + 446, + 249 + ], + "score": 1.0, + "content": ") dataset.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 262, + 504, + 285 + ], + "lines": [ + { + "bbox": [ + 106, + 262, + 505, + 275 + ], + "spans": [ + { + "bbox": [ + 106, + 262, + 505, + 275 + ], + "score": 1.0, + "content": "[18, 19, 20]. Other efforts include customized modules such as self-attention [21], style-based", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 273, + 411, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 273, + 411, + 286 + ], + "score": 1.0, + "content": "generator [22], and autoregressive transformer-based part composition [23].", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 107, + 289, + 505, + 366 + ], + "lines": [ + { + "bbox": [ + 106, + 289, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 106, + 289, + 505, + 302 + ], + "score": 1.0, + "content": "However, one last “commonsense\" seems to have seldomly been challenged: using convolutional", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 300, + 506, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 506, + 313 + ], + "score": 1.0, + "content": "neural networks (CNNs) as GAN backbones. The original GAN [24, 25] used fully-connected", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 104, + 310, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 104, + 310, + 506, + 326 + ], + "score": 1.0, + "content": "networks and can only generate small images. DCGAN [26] was the first to scale up GANs using", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 322, + 506, + 335 + ], + "spans": [ + { + "bbox": [ + 106, + 322, + 506, + 335 + ], + "score": 1.0, + "content": "CNN architectures, which allowed for stable training for higher resolution and deeper generative", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 333, + 506, + 347 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 506, + 347 + ], + "score": 1.0, + "content": "models. Since then, in the computer vision domain, every successful GAN relies on CNN-based", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 344, + 507, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 507, + 357 + ], + "score": 1.0, + "content": "generators and discriminators. Convolutions, with the strong inductive bias for natural images,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 355, + 493, + 368 + ], + "spans": [ + { + "bbox": [ + 105, + 355, + 493, + 368 + ], + "score": 1.0, + "content": "crucially contribute to the appealing visual results and rich diversity achieved by modern GANs.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 107, + 371, + 505, + 481 + ], + "lines": [ + { + "bbox": [ + 105, + 370, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 505, + 385 + ], + "score": 1.0, + "content": "Can we build a strong GAN completely free of convolutions? This is a question not only arising", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 382, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 382, + 506, + 396 + ], + "score": 1.0, + "content": "from intellectual curiosity, but also of practical relevance. Fundamentally, a convolution operator", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 391, + 506, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 506, + 407 + ], + "score": 1.0, + "content": "has a local receptive field, and hence CNNs cannot process long-range dependencies unless passing", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 403, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 505, + 416 + ], + "score": 1.0, + "content": "through a sufficient number of layers. However, that is inefficient, and could cause the loss of feature", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 415, + 505, + 427 + ], + "spans": [ + { + "bbox": [ + 106, + 415, + 505, + 427 + ], + "score": 1.0, + "content": "resolution and fine details, in addition to the difficulty of optimization. Vanilla CNN-based models are", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 426, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 506, + 439 + ], + "score": 1.0, + "content": "therefore inherently not well suited for capturing an input image’s “global\" statistics, as demonstrated", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 437, + 506, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 437, + 506, + 450 + ], + "score": 1.0, + "content": "by the benefits from adopting self-attention [21] and non-local [27] operations in computer vision.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 446, + 506, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 506, + 462 + ], + "score": 1.0, + "content": "Moreover, the spatial invariance possessed by convolution poses a bottleneck on its ability of adapting", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 459, + 506, + 471 + ], + "spans": [ + { + "bbox": [ + 106, + 459, + 506, + 471 + ], + "score": 1.0, + "content": "to spatially varying/heterogeneous visual patterns, which also motivates the success of relational", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 470, + 403, + 481 + ], + "spans": [ + { + "bbox": [ + 106, + 470, + 403, + 481 + ], + "score": 1.0, + "content": "network [28], dynamic filters [29, 30] and kernel prediction [31] methods.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 19.5 + }, + { + "type": "title", + "bbox": [ + 108, + 492, + 210, + 504 + ], + "lines": [ + { + "bbox": [ + 105, + 490, + 212, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 212, + 506 + ], + "score": 1.0, + "content": "1.1 Our Contributions", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 509, + 505, + 651 + ], + "lines": [ + { + "bbox": [ + 105, + 507, + 506, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 507, + 506, + 523 + ], + "score": 1.0, + "content": "This paper aims to be the first pilot study to build a GAN completely free of convolutions, using", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 520, + 506, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 506, + 532 + ], + "score": 1.0, + "content": "only pure transformer-based architectures. We are inspired by the recent success of transformer", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 532, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 505, + 543 + ], + "score": 1.0, + "content": "architectures in computer vision [32, 33, 34]. Compared to parallel generative modeling works", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 541, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 541, + 505, + 554 + ], + "score": 1.0, + "content": "[21, 23, 35] that applied self-attention or transformer encoder in conjunction with CNN-based", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 553, + 506, + 566 + ], + "spans": [ + { + "bbox": [ + 106, + 553, + 506, + 566 + ], + "score": 1.0, + "content": "backbones, our goal is more ambitious and faces several daunting gaps ahead. First and foremost,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 564, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 505, + 576 + ], + "score": 1.0, + "content": "although a pure transformer architecture applied directly to sequences of image patches can perform", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 574, + 506, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 506, + 587 + ], + "score": 1.0, + "content": "very well on image classification tasks [34], it is unclear whether the same way remains effective in", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 585, + 507, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 507, + 599 + ], + "score": 1.0, + "content": "generating images, which crucially demands the spatial coherency in structure, color, and texture,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 596, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 505, + 609 + ], + "score": 1.0, + "content": "as well as the richness of fine details. The handful of existing transformers that output images have", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 606, + 506, + 621 + ], + "spans": [ + { + "bbox": [ + 105, + 606, + 506, + 621 + ], + "score": 1.0, + "content": "unanimously leveraged convolutional part encoders [23] or feature extractors [36, 37]. Moreover,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 618, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 505, + 630 + ], + "score": 1.0, + "content": "even given well-designed CNN-based architectures, training GANs is notoriously unstable and", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 629, + 506, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 629, + 506, + 642 + ], + "score": 1.0, + "content": "prone to mode collapse [15]. Training vision transformers are also known to be tedious, heavy, and", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 639, + 468, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 468, + 653 + ], + "score": 1.0, + "content": "data-hungry [34]. Combining the two will undoubtedly amplify the challenges of training.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 656, + 504, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 505, + 669 + ], + "score": 1.0, + "content": "In view of those challenges, this paper presents a coherent set of efforts and innovations towards", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 666, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 506, + 681 + ], + "score": 1.0, + "content": "building the pure transformer-based GAN architectures, dubbed TransGAN. A naive option may", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "score": 1.0, + "content": "directly stack multiple transformer blocks from raw pixel inputs, but that would scale poorly due", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 689, + 505, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 689, + 505, + 702 + ], + "score": 1.0, + "content": "to memory explosion. Instead, we start with a memory-friendly transformer-based generator by", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 700, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 506, + 713 + ], + "score": 1.0, + "content": "gradually increasing the feature map resolution in each stage. Correspondingly, we also improve", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 710, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 723 + ], + "score": 1.0, + "content": "the discriminator with a multi-scale structure that takes patches of varied size as inputs, which", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 41.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 742, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 310, + 753 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 310, + 753 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 8 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 109, + 70, + 502, + 210 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 70, + 502, + 210 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 70, + 502, + 210 + ], + "spans": [ + { + "bbox": [ + 109, + 70, + 502, + 210 + ], + "score": 0.973, + "type": "image", + "image_path": "adcb25935a6e25e1b2ab70151735aa65bcbfa9d2b385186c0dc4a0005b6915f2.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 109, + 70, + 502, + 116.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 109, + 116.66666666666666, + 502, + 163.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 109, + 163.33333333333331, + 502, + 209.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 215, + 505, + 248 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 214, + 506, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 506, + 229 + ], + "score": 1.0, + "content": "Figure 1: Representative visual examples synthesized by TransGAN, without using convolutional", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 226, + 505, + 239 + ], + "spans": [ + { + "bbox": [ + 106, + 227, + 360, + 239 + ], + "score": 1.0, + "content": "layers. (a) The synthesized visual examples on CelebA-HQ", + "type": "text" + }, + { + "bbox": [ + 360, + 226, + 407, + 237 + ], + "score": 0.86, + "content": "2 5 6 \\times 2 5 6 )", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 227, + 505, + 239 + ], + "score": 1.0, + "content": "dataset. (b) The linear", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 237, + 446, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 366, + 249 + ], + "score": 1.0, + "content": "interpolation results between two latent vectors, on CelebA-HQ (", + "type": "text" + }, + { + "bbox": [ + 366, + 237, + 411, + 248 + ], + "score": 0.84, + "content": "2 5 6 \\times 2 5 6 )", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 237, + 446, + 249 + ], + "score": 1.0, + "content": ") dataset.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 262, + 504, + 285 + ], + "lines": [], + "index": 6.5, + "bbox_fs": [ + 105, + 262, + 505, + 286 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 289, + 505, + 366 + ], + "lines": [ + { + "bbox": [ + 106, + 289, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 106, + 289, + 505, + 302 + ], + "score": 1.0, + "content": "However, one last “commonsense\" seems to have seldomly been challenged: using convolutional", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 300, + 506, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 506, + 313 + ], + "score": 1.0, + "content": "neural networks (CNNs) as GAN backbones. The original GAN [24, 25] used fully-connected", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 104, + 310, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 104, + 310, + 506, + 326 + ], + "score": 1.0, + "content": "networks and can only generate small images. DCGAN [26] was the first to scale up GANs using", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 322, + 506, + 335 + ], + "spans": [ + { + "bbox": [ + 106, + 322, + 506, + 335 + ], + "score": 1.0, + "content": "CNN architectures, which allowed for stable training for higher resolution and deeper generative", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 333, + 506, + 347 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 506, + 347 + ], + "score": 1.0, + "content": "models. Since then, in the computer vision domain, every successful GAN relies on CNN-based", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 344, + 507, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 507, + 357 + ], + "score": 1.0, + "content": "generators and discriminators. Convolutions, with the strong inductive bias for natural images,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 355, + 493, + 368 + ], + "spans": [ + { + "bbox": [ + 105, + 355, + 493, + 368 + ], + "score": 1.0, + "content": "crucially contribute to the appealing visual results and rich diversity achieved by modern GANs.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 11, + "bbox_fs": [ + 104, + 289, + 507, + 368 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 371, + 505, + 481 + ], + "lines": [ + { + "bbox": [ + 105, + 370, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 505, + 385 + ], + "score": 1.0, + "content": "Can we build a strong GAN completely free of convolutions? This is a question not only arising", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 382, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 382, + 506, + 396 + ], + "score": 1.0, + "content": "from intellectual curiosity, but also of practical relevance. Fundamentally, a convolution operator", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 391, + 506, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 506, + 407 + ], + "score": 1.0, + "content": "has a local receptive field, and hence CNNs cannot process long-range dependencies unless passing", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 403, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 505, + 416 + ], + "score": 1.0, + "content": "through a sufficient number of layers. However, that is inefficient, and could cause the loss of feature", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 415, + 505, + 427 + ], + "spans": [ + { + "bbox": [ + 106, + 415, + 505, + 427 + ], + "score": 1.0, + "content": "resolution and fine details, in addition to the difficulty of optimization. Vanilla CNN-based models are", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 426, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 506, + 439 + ], + "score": 1.0, + "content": "therefore inherently not well suited for capturing an input image’s “global\" statistics, as demonstrated", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 437, + 506, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 437, + 506, + 450 + ], + "score": 1.0, + "content": "by the benefits from adopting self-attention [21] and non-local [27] operations in computer vision.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 446, + 506, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 506, + 462 + ], + "score": 1.0, + "content": "Moreover, the spatial invariance possessed by convolution poses a bottleneck on its ability of adapting", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 459, + 506, + 471 + ], + "spans": [ + { + "bbox": [ + 106, + 459, + 506, + 471 + ], + "score": 1.0, + "content": "to spatially varying/heterogeneous visual patterns, which also motivates the success of relational", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 470, + 403, + 481 + ], + "spans": [ + { + "bbox": [ + 106, + 470, + 403, + 481 + ], + "score": 1.0, + "content": "network [28], dynamic filters [29, 30] and kernel prediction [31] methods.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 19.5, + "bbox_fs": [ + 105, + 370, + 506, + 481 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 492, + 210, + 504 + ], + "lines": [ + { + "bbox": [ + 105, + 490, + 212, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 212, + 506 + ], + "score": 1.0, + "content": "1.1 Our Contributions", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 509, + 505, + 651 + ], + "lines": [ + { + "bbox": [ + 105, + 507, + 506, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 507, + 506, + 523 + ], + "score": 1.0, + "content": "This paper aims to be the first pilot study to build a GAN completely free of convolutions, using", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 520, + 506, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 506, + 532 + ], + "score": 1.0, + "content": "only pure transformer-based architectures. We are inspired by the recent success of transformer", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 532, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 505, + 543 + ], + "score": 1.0, + "content": "architectures in computer vision [32, 33, 34]. Compared to parallel generative modeling works", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 541, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 541, + 505, + 554 + ], + "score": 1.0, + "content": "[21, 23, 35] that applied self-attention or transformer encoder in conjunction with CNN-based", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 553, + 506, + 566 + ], + "spans": [ + { + "bbox": [ + 106, + 553, + 506, + 566 + ], + "score": 1.0, + "content": "backbones, our goal is more ambitious and faces several daunting gaps ahead. First and foremost,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 564, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 505, + 576 + ], + "score": 1.0, + "content": "although a pure transformer architecture applied directly to sequences of image patches can perform", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 574, + 506, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 506, + 587 + ], + "score": 1.0, + "content": "very well on image classification tasks [34], it is unclear whether the same way remains effective in", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 585, + 507, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 507, + 599 + ], + "score": 1.0, + "content": "generating images, which crucially demands the spatial coherency in structure, color, and texture,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 596, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 505, + 609 + ], + "score": 1.0, + "content": "as well as the richness of fine details. The handful of existing transformers that output images have", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 606, + 506, + 621 + ], + "spans": [ + { + "bbox": [ + 105, + 606, + 506, + 621 + ], + "score": 1.0, + "content": "unanimously leveraged convolutional part encoders [23] or feature extractors [36, 37]. Moreover,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 618, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 505, + 630 + ], + "score": 1.0, + "content": "even given well-designed CNN-based architectures, training GANs is notoriously unstable and", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 629, + 506, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 629, + 506, + 642 + ], + "score": 1.0, + "content": "prone to mode collapse [15]. Training vision transformers are also known to be tedious, heavy, and", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 639, + 468, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 468, + 653 + ], + "score": 1.0, + "content": "data-hungry [34]. Combining the two will undoubtedly amplify the challenges of training.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 32, + "bbox_fs": [ + 105, + 507, + 507, + 653 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 656, + 504, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 505, + 669 + ], + "score": 1.0, + "content": "In view of those challenges, this paper presents a coherent set of efforts and innovations towards", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 666, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 506, + 681 + ], + "score": 1.0, + "content": "building the pure transformer-based GAN architectures, dubbed TransGAN. A naive option may", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "score": 1.0, + "content": "directly stack multiple transformer blocks from raw pixel inputs, but that would scale poorly due", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 689, + 505, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 689, + 505, + 702 + ], + "score": 1.0, + "content": "to memory explosion. Instead, we start with a memory-friendly transformer-based generator by", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 700, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 506, + 713 + ], + "score": 1.0, + "content": "gradually increasing the feature map resolution in each stage. Correspondingly, we also improve", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 710, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 723 + ], + "score": 1.0, + "content": "the discriminator with a multi-scale structure that takes patches of varied size as inputs, which", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 71, + 506, + 87 + ], + "spans": [ + { + "bbox": [ + 105, + 71, + 506, + 87 + ], + "score": 1.0, + "content": "balances between capturing global contexts and local details, in addition to enhancing memory", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 83, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 505, + 96 + ], + "score": 1.0, + "content": "efficiency more. Based on the above generator-discriminator design, we introduce a new module", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 94, + 505, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 505, + 107 + ], + "score": 1.0, + "content": "called grid self-attention, that alleviates the memory bottleneck further when scaling up TransGAN", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 106, + 295, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 245, + 118 + ], + "score": 1.0, + "content": "to high-resolution generation (e.g.", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 245, + 106, + 290, + 117 + ], + "score": 0.87, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 290, + 106, + 295, + 118 + ], + "score": 1.0, + "content": ").", + "type": "text", + "cross_page": true + } + ], + "index": 3 + } + ], + "index": 41.5, + "bbox_fs": [ + 105, + 654, + 506, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 72, + 505, + 117 + ], + "lines": [ + { + "bbox": [ + 105, + 71, + 506, + 87 + ], + "spans": [ + { + "bbox": [ + 105, + 71, + 506, + 87 + ], + "score": 1.0, + "content": "balances between capturing global contexts and local details, in addition to enhancing memory", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 83, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 505, + 96 + ], + "score": 1.0, + "content": "efficiency more. Based on the above generator-discriminator design, we introduce a new module", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 94, + 505, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 505, + 107 + ], + "score": 1.0, + "content": "called grid self-attention, that alleviates the memory bottleneck further when scaling up TransGAN", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 106, + 295, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 245, + 118 + ], + "score": 1.0, + "content": "to high-resolution generation (e.g.", + "type": "text" + }, + { + "bbox": [ + 245, + 106, + 290, + 117 + ], + "score": 0.87, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 106, + 295, + 118 + ], + "score": 1.0, + "content": ").", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 108, + 122, + 505, + 176 + ], + "lines": [ + { + "bbox": [ + 106, + 120, + 505, + 134 + ], + "spans": [ + { + "bbox": [ + 106, + 120, + 505, + 134 + ], + "score": 1.0, + "content": "To address the aforementioned instability issue brought by both GAN and Transformer, we also", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 131, + 505, + 145 + ], + "spans": [ + { + "bbox": [ + 106, + 131, + 505, + 145 + ], + "score": 1.0, + "content": "develop a unique training recipe in association with our innovative TransGAN architecture, that", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 143, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 106, + 143, + 505, + 156 + ], + "score": 1.0, + "content": "effectively stabilizes its optimization and generalization. That includes showings the necessity of data", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 154, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 154, + 505, + 167 + ], + "score": 1.0, + "content": "augmentation, modifying layer normalization, and replacing absolute token locations with relative", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 165, + 335, + 178 + ], + "spans": [ + { + "bbox": [ + 106, + 165, + 335, + 178 + ], + "score": 1.0, + "content": "position encoding. Our contributions are outlined below:", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 133, + 182, + 505, + 351 + ], + "lines": [ + { + "bbox": [ + 132, + 181, + 506, + 195 + ], + "spans": [ + { + "bbox": [ + 132, + 181, + 506, + 195 + ], + "score": 1.0, + "content": "• Novel Architecture Design: We build the first GAN using purely transformers and no", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 193, + 505, + 206 + ], + "spans": [ + { + "bbox": [ + 141, + 193, + 505, + 206 + ], + "score": 1.0, + "content": "convolution. TransGAN has customized a memory-friendly generator and a multi-scale", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 142, + 205, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 142, + 205, + 505, + 217 + ], + "score": 1.0, + "content": "discriminator, and is further equipped with a new grid self-attention mechanism. Those", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 142, + 215, + 505, + 227 + ], + "spans": [ + { + "bbox": [ + 142, + 215, + 505, + 227 + ], + "score": 1.0, + "content": "architectural components are thoughtfully designed to balance memory efficiency, global", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 226, + 388, + 239 + ], + "spans": [ + { + "bbox": [ + 141, + 226, + 388, + 239 + ], + "score": 1.0, + "content": "feature statistics, and local fine details with spatial variances.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 136, + 240, + 506, + 252 + ], + "spans": [ + { + "bbox": [ + 136, + 240, + 506, + 252 + ], + "score": 1.0, + "content": "• New Training Recipe: We study a number of techniques to train TransGAN better, includ-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 250, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 141, + 250, + 506, + 264 + ], + "score": 1.0, + "content": "ing leveraging data augmentation, modifying layer normalization, and adopting relative", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 262, + 507, + 275 + ], + "spans": [ + { + "bbox": [ + 141, + 262, + 507, + 275 + ], + "score": 1.0, + "content": "position encoding, for both generator and discriminator. Extensive ablation studies, discus-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 273, + 276, + 285 + ], + "spans": [ + { + "bbox": [ + 141, + 273, + 276, + 285 + ], + "score": 1.0, + "content": "sions, and insights are presented.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 138, + 285, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 138, + 285, + 505, + 298 + ], + "score": 1.0, + "content": "Performance and Scalability: TransGAN achieves highly competitive performance com-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 140, + 297, + 505, + 310 + ], + "spans": [ + { + "bbox": [ + 140, + 297, + 505, + 310 + ], + "score": 1.0, + "content": "pared to current state-of-the-art GANs. Specifically, it sets the new state-of-the-art inception", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 140, + 307, + 505, + 321 + ], + "spans": [ + { + "bbox": [ + 140, + 307, + 505, + 321 + ], + "score": 1.0, + "content": "score of 10.43 and FID score of 18.28 on STL-10. It also reaches competitive 9.02 inception", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 317, + 506, + 332 + ], + "spans": [ + { + "bbox": [ + 141, + 317, + 407, + 332 + ], + "score": 1.0, + "content": "score and 9.26 FID on CIFAR-10, and 5.28 FID score on CelebA", + "type": "text" + }, + { + "bbox": [ + 407, + 318, + 450, + 329 + ], + "score": 0.89, + "content": "1 2 8 \\times 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 317, + 506, + 332 + ], + "score": 1.0, + "content": ", respectively.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 329, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 141, + 329, + 412, + 343 + ], + "score": 1.0, + "content": "Meanwhile, we also evaluate TransGAN on higher-resolution (e.g.,", + "type": "text" + }, + { + "bbox": [ + 412, + 330, + 457, + 340 + ], + "score": 0.87, + "content": "2 5 6 \\times 2 5 6 )", + "type": "inline_equation" + }, + { + "bbox": [ + 458, + 329, + 505, + 343 + ], + "score": 1.0, + "content": "generation", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 141, + 340, + 475, + 353 + ], + "spans": [ + { + "bbox": [ + 141, + 340, + 475, + 353 + ], + "score": 1.0, + "content": "tasks, where TransGAN continues to yield diverse and impressive visual examples.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 16 + }, + { + "type": "title", + "bbox": [ + 107, + 362, + 202, + 376 + ], + "lines": [ + { + "bbox": [ + 104, + 362, + 203, + 378 + ], + "spans": [ + { + "bbox": [ + 104, + 362, + 203, + 378 + ], + "score": 1.0, + "content": "2 Related Works", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 383, + 505, + 450 + ], + "lines": [ + { + "bbox": [ + 106, + 383, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 505, + 396 + ], + "score": 1.0, + "content": "Generative Adversarial Networks. After its origin, GANs quickly embraced fully convolutional", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 394, + 506, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 394, + 506, + 407 + ], + "score": 1.0, + "content": "backbones [26], and inherited most successful designs from CNNs such as batch normalization,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 405, + 506, + 418 + ], + "spans": [ + { + "bbox": [ + 104, + 405, + 506, + 418 + ], + "score": 1.0, + "content": "pooling, (Leaky) ReLU and more [38, 39, 40, 18]. GANs are widely adopted in image translation", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 416, + 506, + 429 + ], + "spans": [ + { + "bbox": [ + 106, + 416, + 506, + 429 + ], + "score": 1.0, + "content": "[3, 4, 41], image enhancement [7, 42, 43], and image editing [44, 45]. To alleviate its unstable", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 427, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 427, + 506, + 439 + ], + "score": 1.0, + "content": "training, a number of techniques have been studied, including the Wasserstein loss [46], the style-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 437, + 501, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 437, + 501, + 450 + ], + "score": 1.0, + "content": "based generator [22], progressive training [16], lottery ticket [47], and spectral normalization [48].", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 106, + 454, + 505, + 585 + ], + "lines": [ + { + "bbox": [ + 105, + 454, + 506, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 506, + 467 + ], + "score": 1.0, + "content": "Transformers in Computer Vision. The original transformer was built for NLP [49], where the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "score": 1.0, + "content": "multi-head self-attention and feed-forward MLP layer are stacked to capture the long-term correlation", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 475, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 505, + 489 + ], + "score": 1.0, + "content": "between words. A recent work [34] implements highly competitive ImageNet classification using pure", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 487, + 506, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 308, + 500 + ], + "score": 1.0, + "content": "transformers, by treating an image as a sequence of", + "type": "text" + }, + { + "bbox": [ + 309, + 487, + 341, + 498 + ], + "score": 0.9, + "content": "1 6 \\times 1 6", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 487, + 506, + 500 + ], + "score": 1.0, + "content": "visual words. It has strong representation", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "score": 1.0, + "content": "capability and is free of human-defined inductive bias. In comparison, CNNs exhibit a strong bias", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 509, + 506, + 521 + ], + "spans": [ + { + "bbox": [ + 106, + 509, + 506, + 521 + ], + "score": 1.0, + "content": "towards feature locality, as well as spatial invariance due to sharing filter weights across all locations.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 519, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 505, + 533 + ], + "score": 1.0, + "content": "However, the success of original vision transformer relies on pretraining on large-scale external", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 529, + 506, + 544 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 506, + 544 + ], + "score": 1.0, + "content": "data. [50, 51] improve the data efficiency and address the difficulty of optimizing deeper models.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 542, + 506, + 553 + ], + "spans": [ + { + "bbox": [ + 106, + 542, + 506, + 553 + ], + "score": 1.0, + "content": "Other works introduce the pyramid/hierarchical structure to transformer [52, 53, 54] or combine it", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 553, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 106, + 553, + 505, + 565 + ], + "score": 1.0, + "content": "with convolutional layers [55, 56]. Besides image classification task, transformer and its variants", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 564, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 505, + 576 + ], + "score": 1.0, + "content": "are also explored on image processing [37], point cloud [57], semantic segmentation [58], object", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 574, + 399, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 574, + 399, + 587 + ], + "score": 1.0, + "content": "detection [32, 59] and so on. A comprehensive review is referred to [60].", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 107, + 591, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 591, + 505, + 603 + ], + "spans": [ + { + "bbox": [ + 106, + 591, + 505, + 603 + ], + "score": 1.0, + "content": "Transformer Modules for Image Generation. There exist several related works combining the", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 602, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 602, + 505, + 614 + ], + "score": 1.0, + "content": "transformer modules into image generation models, by replacing certain components of CNNs. [61]", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 612, + 505, + 626 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 505, + 626 + ], + "score": 1.0, + "content": "firstly formulated image generation as autoregressive sequence generation, for which they adopted", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 624, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 505, + 635 + ], + "score": 1.0, + "content": "a transformer architecture. [62] propose sparse factorization of the attention matrix to reduce its", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 635, + 505, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 635, + 505, + 647 + ], + "score": 1.0, + "content": "complexity. While those two works did not tackle the GANs, one recent (concurrent) work [23]", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 645, + 506, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 645, + 506, + 658 + ], + "score": 1.0, + "content": "used a convolutional GAN to learn a codebook of context-rich visual parts, whose composition is", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 656, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 106, + 656, + 506, + 668 + ], + "score": 1.0, + "content": "subsequently modeled with an autoregressive transformer architecture.The authors demonstrated", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 667, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 506, + 680 + ], + "score": 1.0, + "content": "success in synthesizing high-resolution images. However, the overall CNN architecture remains in", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 678, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 506, + 691 + ], + "score": 1.0, + "content": "place (including CNN encoder/decoder for the generators, and a fully CNN-based discriminator), and", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 689, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 689, + 506, + 701 + ], + "score": 1.0, + "content": "the customized designs (e.g, codebook and quantization) also limit their model’s versatility. Another", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 700, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 505, + 712 + ], + "score": 1.0, + "content": "concurrent work [35] employs a bipartite self-attention on StyleGAN and thus it can propagate latent", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "score": 1.0, + "content": "variables to the evolving visual features, yet its main structure is still convolutional, including both the", + "type": "text" + } + ], + "index": 54 + } + ], + "index": 48.5 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 72, + 505, + 117 + ], + "lines": [], + "index": 1.5, + "bbox_fs": [ + 105, + 71, + 506, + 118 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 108, + 122, + 505, + 176 + ], + "lines": [ + { + "bbox": [ + 106, + 120, + 505, + 134 + ], + "spans": [ + { + "bbox": [ + 106, + 120, + 505, + 134 + ], + "score": 1.0, + "content": "To address the aforementioned instability issue brought by both GAN and Transformer, we also", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 131, + 505, + 145 + ], + "spans": [ + { + "bbox": [ + 106, + 131, + 505, + 145 + ], + "score": 1.0, + "content": "develop a unique training recipe in association with our innovative TransGAN architecture, that", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 143, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 106, + 143, + 505, + 156 + ], + "score": 1.0, + "content": "effectively stabilizes its optimization and generalization. That includes showings the necessity of data", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 154, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 154, + 505, + 167 + ], + "score": 1.0, + "content": "augmentation, modifying layer normalization, and replacing absolute token locations with relative", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 165, + 335, + 178 + ], + "spans": [ + { + "bbox": [ + 106, + 165, + 335, + 178 + ], + "score": 1.0, + "content": "position encoding. Our contributions are outlined below:", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6, + "bbox_fs": [ + 105, + 120, + 505, + 178 + ] + }, + { + "type": "list", + "bbox": [ + 133, + 182, + 505, + 351 + ], + "lines": [ + { + "bbox": [ + 132, + 181, + 506, + 195 + ], + "spans": [ + { + "bbox": [ + 132, + 181, + 506, + 195 + ], + "score": 1.0, + "content": "• Novel Architecture Design: We build the first GAN using purely transformers and no", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 193, + 505, + 206 + ], + "spans": [ + { + "bbox": [ + 141, + 193, + 505, + 206 + ], + "score": 1.0, + "content": "convolution. TransGAN has customized a memory-friendly generator and a multi-scale", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 142, + 205, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 142, + 205, + 505, + 217 + ], + "score": 1.0, + "content": "discriminator, and is further equipped with a new grid self-attention mechanism. Those", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 142, + 215, + 505, + 227 + ], + "spans": [ + { + "bbox": [ + 142, + 215, + 505, + 227 + ], + "score": 1.0, + "content": "architectural components are thoughtfully designed to balance memory efficiency, global", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 226, + 388, + 239 + ], + "spans": [ + { + "bbox": [ + 141, + 226, + 388, + 239 + ], + "score": 1.0, + "content": "feature statistics, and local fine details with spatial variances.", + "type": "text" + } + ], + "index": 13, + "is_list_end_line": true + }, + { + "bbox": [ + 136, + 240, + 506, + 252 + ], + "spans": [ + { + "bbox": [ + 136, + 240, + 506, + 252 + ], + "score": 1.0, + "content": "• New Training Recipe: We study a number of techniques to train TransGAN better, includ-", + "type": "text" + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 250, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 141, + 250, + 506, + 264 + ], + "score": 1.0, + "content": "ing leveraging data augmentation, modifying layer normalization, and adopting relative", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 262, + 507, + 275 + ], + "spans": [ + { + "bbox": [ + 141, + 262, + 507, + 275 + ], + "score": 1.0, + "content": "position encoding, for both generator and discriminator. Extensive ablation studies, discus-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 273, + 276, + 285 + ], + "spans": [ + { + "bbox": [ + 141, + 273, + 276, + 285 + ], + "score": 1.0, + "content": "sions, and insights are presented.", + "type": "text" + } + ], + "index": 17, + "is_list_end_line": true + }, + { + "bbox": [ + 138, + 285, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 138, + 285, + 505, + 298 + ], + "score": 1.0, + "content": "Performance and Scalability: TransGAN achieves highly competitive performance com-", + "type": "text" + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 140, + 297, + 505, + 310 + ], + "spans": [ + { + "bbox": [ + 140, + 297, + 505, + 310 + ], + "score": 1.0, + "content": "pared to current state-of-the-art GANs. Specifically, it sets the new state-of-the-art inception", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 140, + 307, + 505, + 321 + ], + "spans": [ + { + "bbox": [ + 140, + 307, + 505, + 321 + ], + "score": 1.0, + "content": "score of 10.43 and FID score of 18.28 on STL-10. It also reaches competitive 9.02 inception", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 317, + 506, + 332 + ], + "spans": [ + { + "bbox": [ + 141, + 317, + 407, + 332 + ], + "score": 1.0, + "content": "score and 9.26 FID on CIFAR-10, and 5.28 FID score on CelebA", + "type": "text" + }, + { + "bbox": [ + 407, + 318, + 450, + 329 + ], + "score": 0.89, + "content": "1 2 8 \\times 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 317, + 506, + 332 + ], + "score": 1.0, + "content": ", respectively.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 329, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 141, + 329, + 412, + 343 + ], + "score": 1.0, + "content": "Meanwhile, we also evaluate TransGAN on higher-resolution (e.g.,", + "type": "text" + }, + { + "bbox": [ + 412, + 330, + 457, + 340 + ], + "score": 0.87, + "content": "2 5 6 \\times 2 5 6 )", + "type": "inline_equation" + }, + { + "bbox": [ + 458, + 329, + 505, + 343 + ], + "score": 1.0, + "content": "generation", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 141, + 340, + 475, + 353 + ], + "spans": [ + { + "bbox": [ + 141, + 340, + 475, + 353 + ], + "score": 1.0, + "content": "tasks, where TransGAN continues to yield diverse and impressive visual examples.", + "type": "text" + } + ], + "index": 23, + "is_list_end_line": true + } + ], + "index": 16, + "bbox_fs": [ + 132, + 181, + 507, + 353 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 362, + 202, + 376 + ], + "lines": [ + { + "bbox": [ + 104, + 362, + 203, + 378 + ], + "spans": [ + { + "bbox": [ + 104, + 362, + 203, + 378 + ], + "score": 1.0, + "content": "2 Related Works", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 383, + 505, + 450 + ], + "lines": [ + { + "bbox": [ + 106, + 383, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 505, + 396 + ], + "score": 1.0, + "content": "Generative Adversarial Networks. After its origin, GANs quickly embraced fully convolutional", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 394, + 506, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 394, + 506, + 407 + ], + "score": 1.0, + "content": "backbones [26], and inherited most successful designs from CNNs such as batch normalization,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 405, + 506, + 418 + ], + "spans": [ + { + "bbox": [ + 104, + 405, + 506, + 418 + ], + "score": 1.0, + "content": "pooling, (Leaky) ReLU and more [38, 39, 40, 18]. GANs are widely adopted in image translation", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 416, + 506, + 429 + ], + "spans": [ + { + "bbox": [ + 106, + 416, + 506, + 429 + ], + "score": 1.0, + "content": "[3, 4, 41], image enhancement [7, 42, 43], and image editing [44, 45]. To alleviate its unstable", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 427, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 427, + 506, + 439 + ], + "score": 1.0, + "content": "training, a number of techniques have been studied, including the Wasserstein loss [46], the style-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 437, + 501, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 437, + 501, + 450 + ], + "score": 1.0, + "content": "based generator [22], progressive training [16], lottery ticket [47], and spectral normalization [48].", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27.5, + "bbox_fs": [ + 104, + 383, + 506, + 450 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 454, + 505, + 585 + ], + "lines": [ + { + "bbox": [ + 105, + 454, + 506, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 506, + 467 + ], + "score": 1.0, + "content": "Transformers in Computer Vision. The original transformer was built for NLP [49], where the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "score": 1.0, + "content": "multi-head self-attention and feed-forward MLP layer are stacked to capture the long-term correlation", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 475, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 505, + 489 + ], + "score": 1.0, + "content": "between words. A recent work [34] implements highly competitive ImageNet classification using pure", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 487, + 506, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 308, + 500 + ], + "score": 1.0, + "content": "transformers, by treating an image as a sequence of", + "type": "text" + }, + { + "bbox": [ + 309, + 487, + 341, + 498 + ], + "score": 0.9, + "content": "1 6 \\times 1 6", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 487, + 506, + 500 + ], + "score": 1.0, + "content": "visual words. It has strong representation", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "score": 1.0, + "content": "capability and is free of human-defined inductive bias. In comparison, CNNs exhibit a strong bias", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 509, + 506, + 521 + ], + "spans": [ + { + "bbox": [ + 106, + 509, + 506, + 521 + ], + "score": 1.0, + "content": "towards feature locality, as well as spatial invariance due to sharing filter weights across all locations.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 519, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 505, + 533 + ], + "score": 1.0, + "content": "However, the success of original vision transformer relies on pretraining on large-scale external", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 529, + 506, + 544 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 506, + 544 + ], + "score": 1.0, + "content": "data. [50, 51] improve the data efficiency and address the difficulty of optimizing deeper models.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 542, + 506, + 553 + ], + "spans": [ + { + "bbox": [ + 106, + 542, + 506, + 553 + ], + "score": 1.0, + "content": "Other works introduce the pyramid/hierarchical structure to transformer [52, 53, 54] or combine it", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 553, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 106, + 553, + 505, + 565 + ], + "score": 1.0, + "content": "with convolutional layers [55, 56]. Besides image classification task, transformer and its variants", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 564, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 505, + 576 + ], + "score": 1.0, + "content": "are also explored on image processing [37], point cloud [57], semantic segmentation [58], object", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 574, + 399, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 574, + 399, + 587 + ], + "score": 1.0, + "content": "detection [32, 59] and so on. A comprehensive review is referred to [60].", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 36.5, + "bbox_fs": [ + 105, + 454, + 506, + 587 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 591, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 591, + 505, + 603 + ], + "spans": [ + { + "bbox": [ + 106, + 591, + 505, + 603 + ], + "score": 1.0, + "content": "Transformer Modules for Image Generation. There exist several related works combining the", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 602, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 602, + 505, + 614 + ], + "score": 1.0, + "content": "transformer modules into image generation models, by replacing certain components of CNNs. [61]", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 612, + 505, + 626 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 505, + 626 + ], + "score": 1.0, + "content": "firstly formulated image generation as autoregressive sequence generation, for which they adopted", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 624, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 505, + 635 + ], + "score": 1.0, + "content": "a transformer architecture. [62] propose sparse factorization of the attention matrix to reduce its", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 635, + 505, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 635, + 505, + 647 + ], + "score": 1.0, + "content": "complexity. While those two works did not tackle the GANs, one recent (concurrent) work [23]", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 645, + 506, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 645, + 506, + 658 + ], + "score": 1.0, + "content": "used a convolutional GAN to learn a codebook of context-rich visual parts, whose composition is", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 656, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 106, + 656, + 506, + 668 + ], + "score": 1.0, + "content": "subsequently modeled with an autoregressive transformer architecture.The authors demonstrated", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 667, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 506, + 680 + ], + "score": 1.0, + "content": "success in synthesizing high-resolution images. However, the overall CNN architecture remains in", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 678, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 506, + 691 + ], + "score": 1.0, + "content": "place (including CNN encoder/decoder for the generators, and a fully CNN-based discriminator), and", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 689, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 689, + 506, + 701 + ], + "score": 1.0, + "content": "the customized designs (e.g, codebook and quantization) also limit their model’s versatility. Another", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 700, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 505, + 712 + ], + "score": 1.0, + "content": "concurrent work [35] employs a bipartite self-attention on StyleGAN and thus it can propagate latent", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "score": 1.0, + "content": "variables to the evolving visual features, yet its main structure is still convolutional, including both the", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 105, + 415, + 504, + 429 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 504, + 429 + ], + "score": 1.0, + "content": "generator and discriminator. To our best knowledge, no other existing work has tried to completely", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 427, + 370, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 427, + 370, + 438 + ], + "score": 1.0, + "content": "remove convolutions from their generative modeling frameworks.", + "type": "text", + "cross_page": true + } + ], + "index": 10 + } + ], + "index": 48.5, + "bbox_fs": [ + 105, + 591, + 506, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 114, + 72, + 500, + 340 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 114, + 72, + 500, + 340 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 114, + 72, + 500, + 340 + ], + "spans": [ + { + "bbox": [ + 114, + 72, + 500, + 340 + ], + "score": 0.934, + "type": "image", + "image_path": "e89dc9eb445115b78109b53440aa8db8c120d23b4e23956ef5aedc971cd11810.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 114, + 72, + 500, + 161.33333333333331 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 114, + 161.33333333333331, + 500, + 250.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 114, + 250.66666666666663, + 500, + 339.99999999999994 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 344, + 505, + 410 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 343, + 505, + 356 + ], + "spans": [ + { + "bbox": [ + 106, + 343, + 505, + 356 + ], + "score": 1.0, + "content": "Figure 2: The pipeline of the pure transform-based generator and discriminator of TransGAN. We take", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 355, + 505, + 368 + ], + "spans": [ + { + "bbox": [ + 106, + 355, + 147, + 366 + ], + "score": 0.9, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 355, + 505, + 368 + ], + "score": 1.0, + "content": "resolution image generation task as a typical example to illustrate the main procedure. Here", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 366, + 505, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 147, + 378 + ], + "score": 1.0, + "content": "patch size", + "type": "text" + }, + { + "bbox": [ + 148, + 368, + 155, + 378 + ], + "score": 0.8, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 366, + 505, + 378 + ], + "score": 1.0, + "content": "is set to 32 as an example for the convenience of illustration, while practically the patch", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 377, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 106, + 377, + 259, + 389 + ], + "score": 1.0, + "content": "size is normally set to be no more than", + "type": "text" + }, + { + "bbox": [ + 259, + 377, + 281, + 387 + ], + "score": 0.9, + "content": "8 \\times 8", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 377, + 505, + 389 + ], + "score": 1.0, + "content": ", depending on the specific dataset. Grid Transformer", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 388, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 505, + 399 + ], + "score": 1.0, + "content": "Blocks refers to the transformer blocks with the proposed grid self-attention. Detailed architecture", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 399, + 281, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 281, + 411 + ], + "score": 1.0, + "content": "configurations are included in Appendix B.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 5.5 + } + ], + "index": 3.25 + }, + { + "type": "text", + "bbox": [ + 107, + 416, + 502, + 438 + ], + "lines": [ + { + "bbox": [ + 105, + 415, + 504, + 429 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 504, + 429 + ], + "score": 1.0, + "content": "generator and discriminator. To our best knowledge, no other existing work has tried to completely", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 427, + 370, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 427, + 370, + 438 + ], + "score": 1.0, + "content": "remove convolutions from their generative modeling frameworks.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + }, + { + "type": "title", + "bbox": [ + 106, + 448, + 492, + 462 + ], + "lines": [ + { + "bbox": [ + 104, + 448, + 493, + 464 + ], + "spans": [ + { + "bbox": [ + 104, + 448, + 493, + 464 + ], + "score": 1.0, + "content": "3 Technical Approach: A Journey Towards GAN with Pure Transformers", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 107, + 469, + 505, + 512 + ], + "lines": [ + { + "bbox": [ + 105, + 468, + 506, + 481 + ], + "spans": [ + { + "bbox": [ + 105, + 468, + 506, + 481 + ], + "score": 1.0, + "content": "In this section, we start by introducing the memory-friendly generator and multi-scale discriminator,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 480, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 505, + 492 + ], + "score": 1.0, + "content": "equipped with a novel grid self-attention. We then introduce a series of training techniques to stabilize", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 490, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 505, + 503 + ], + "score": 1.0, + "content": "its training procedure, including data augmentation, the modified normalization, and injecting relative", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 502, + 247, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 247, + 514 + ], + "score": 1.0, + "content": "position encoding to self-attention.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 107, + 518, + 505, + 562 + ], + "lines": [ + { + "bbox": [ + 105, + 518, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 505, + 530 + ], + "score": 1.0, + "content": "To start with, we choose the transformer encoder [49] as our basic block and try to make minimal", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 528, + 506, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 506, + 542 + ], + "score": 1.0, + "content": "changes. An encoder is a composition of two parts. The first part is constructed by a multi-head", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 540, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 505, + 552 + ], + "score": 1.0, + "content": "self-attention module and the second part is a feed-forward MLP with GELU non-linearity. The", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 551, + 503, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 503, + 563 + ], + "score": 1.0, + "content": "normalization layer is applied before both of the two parts. Both parts employ residual connection.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17.5 + }, + { + "type": "title", + "bbox": [ + 107, + 570, + 250, + 582 + ], + "lines": [ + { + "bbox": [ + 104, + 568, + 253, + 585 + ], + "spans": [ + { + "bbox": [ + 104, + 568, + 253, + 585 + ], + "score": 1.0, + "content": "3.1 Memory-friendly Generator", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 585, + 505, + 695 + ], + "lines": [ + { + "bbox": [ + 105, + 585, + 506, + 597 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 506, + 597 + ], + "score": 1.0, + "content": "The task of generation poses a high standard for spatial coherency in structure, color, and texture, both", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 104, + 596, + 506, + 608 + ], + "spans": [ + { + "bbox": [ + 104, + 596, + 506, + 608 + ], + "score": 1.0, + "content": "globally and locally. The transformer encoders take embedding token words as inputs and calculate", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 606, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 606, + 505, + 620 + ], + "score": 1.0, + "content": "the interaction between each token recursively. [63, 34]. The main dilemma here is: what is the right", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 617, + 506, + 631 + ], + "spans": [ + { + "bbox": [ + 104, + 617, + 506, + 631 + ], + "score": 1.0, + "content": "“word\" for image generation tasks? If we similarly generate an image in a pixel-by-pixel manner", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 628, + 505, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 403, + 642 + ], + "score": 1.0, + "content": "through stacking transformer encoders, even a low-resolution image (e.g.", + "type": "text" + }, + { + "bbox": [ + 403, + 629, + 438, + 640 + ], + "score": 0.88, + "content": "3 2 \\times 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 628, + 505, + 642 + ], + "score": 1.0, + "content": ") can result in an", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 640, + 505, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 505, + 653 + ], + "score": 1.0, + "content": "excessively long sequence (1024), causing the explosive cost of self-attention (quadratic w.r.t. the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 650, + 507, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 507, + 664 + ], + "score": 1.0, + "content": "sequence length) and prohibiting the scalability to higher resolutions. To avoid this daunting cost,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 662, + 505, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 505, + 675 + ], + "score": 1.0, + "content": "we are inspired by a common design philosophy in CNN-based GANs, to iteratively upscale the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 673, + 505, + 685 + ], + "spans": [ + { + "bbox": [ + 106, + 673, + 505, + 685 + ], + "score": 1.0, + "content": "resolution at multiple stages [25, 16]. Our strategy is hence to increase the input sequence and reduce", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 684, + 256, + 696 + ], + "spans": [ + { + "bbox": [ + 106, + 684, + 256, + 696 + ], + "score": 1.0, + "content": "the embedding dimension gradually .", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 107, + 700, + 502, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 700, + 504, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 700, + 504, + 712 + ], + "score": 1.0, + "content": "Figure 2 (left) illustrates a memory-friendly transformer-based generator that consists of multiple", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 710, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 724 + ], + "score": 1.0, + "content": "stages. Each stage stacks several transformer blocks. By stages, we gradually increase the feature", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 742, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 741, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 301, + 741, + 310, + 752 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 11, + "width": 9 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 114, + 72, + 500, + 340 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 114, + 72, + 500, + 340 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 114, + 72, + 500, + 340 + ], + "spans": [ + { + "bbox": [ + 114, + 72, + 500, + 340 + ], + "score": 0.934, + "type": "image", + "image_path": "e89dc9eb445115b78109b53440aa8db8c120d23b4e23956ef5aedc971cd11810.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 114, + 72, + 500, + 161.33333333333331 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 114, + 161.33333333333331, + 500, + 250.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 114, + 250.66666666666663, + 500, + 339.99999999999994 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 344, + 505, + 410 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 343, + 505, + 356 + ], + "spans": [ + { + "bbox": [ + 106, + 343, + 505, + 356 + ], + "score": 1.0, + "content": "Figure 2: The pipeline of the pure transform-based generator and discriminator of TransGAN. We take", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 355, + 505, + 368 + ], + "spans": [ + { + "bbox": [ + 106, + 355, + 147, + 366 + ], + "score": 0.9, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 355, + 505, + 368 + ], + "score": 1.0, + "content": "resolution image generation task as a typical example to illustrate the main procedure. Here", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 366, + 505, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 147, + 378 + ], + "score": 1.0, + "content": "patch size", + "type": "text" + }, + { + "bbox": [ + 148, + 368, + 155, + 378 + ], + "score": 0.8, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 366, + 505, + 378 + ], + "score": 1.0, + "content": "is set to 32 as an example for the convenience of illustration, while practically the patch", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 377, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 106, + 377, + 259, + 389 + ], + "score": 1.0, + "content": "size is normally set to be no more than", + "type": "text" + }, + { + "bbox": [ + 259, + 377, + 281, + 387 + ], + "score": 0.9, + "content": "8 \\times 8", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 377, + 505, + 389 + ], + "score": 1.0, + "content": ", depending on the specific dataset. Grid Transformer", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 388, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 505, + 399 + ], + "score": 1.0, + "content": "Blocks refers to the transformer blocks with the proposed grid self-attention. Detailed architecture", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 399, + 281, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 281, + 411 + ], + "score": 1.0, + "content": "configurations are included in Appendix B.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 5.5 + } + ], + "index": 3.25 + }, + { + "type": "text", + "bbox": [ + 107, + 416, + 502, + 438 + ], + "lines": [], + "index": 9.5, + "bbox_fs": [ + 105, + 415, + 504, + 438 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 106, + 448, + 492, + 462 + ], + "lines": [ + { + "bbox": [ + 104, + 448, + 493, + 464 + ], + "spans": [ + { + "bbox": [ + 104, + 448, + 493, + 464 + ], + "score": 1.0, + "content": "3 Technical Approach: A Journey Towards GAN with Pure Transformers", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 107, + 469, + 505, + 512 + ], + "lines": [ + { + "bbox": [ + 105, + 468, + 506, + 481 + ], + "spans": [ + { + "bbox": [ + 105, + 468, + 506, + 481 + ], + "score": 1.0, + "content": "In this section, we start by introducing the memory-friendly generator and multi-scale discriminator,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 480, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 505, + 492 + ], + "score": 1.0, + "content": "equipped with a novel grid self-attention. We then introduce a series of training techniques to stabilize", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 490, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 505, + 503 + ], + "score": 1.0, + "content": "its training procedure, including data augmentation, the modified normalization, and injecting relative", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 502, + 247, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 247, + 514 + ], + "score": 1.0, + "content": "position encoding to self-attention.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13.5, + "bbox_fs": [ + 105, + 468, + 506, + 514 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 518, + 505, + 562 + ], + "lines": [ + { + "bbox": [ + 105, + 518, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 505, + 530 + ], + "score": 1.0, + "content": "To start with, we choose the transformer encoder [49] as our basic block and try to make minimal", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 528, + 506, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 506, + 542 + ], + "score": 1.0, + "content": "changes. An encoder is a composition of two parts. The first part is constructed by a multi-head", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 540, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 505, + 552 + ], + "score": 1.0, + "content": "self-attention module and the second part is a feed-forward MLP with GELU non-linearity. The", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 551, + 503, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 503, + 563 + ], + "score": 1.0, + "content": "normalization layer is applied before both of the two parts. Both parts employ residual connection.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17.5, + "bbox_fs": [ + 105, + 518, + 506, + 563 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 570, + 250, + 582 + ], + "lines": [ + { + "bbox": [ + 104, + 568, + 253, + 585 + ], + "spans": [ + { + "bbox": [ + 104, + 568, + 253, + 585 + ], + "score": 1.0, + "content": "3.1 Memory-friendly Generator", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 585, + 505, + 695 + ], + "lines": [ + { + "bbox": [ + 105, + 585, + 506, + 597 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 506, + 597 + ], + "score": 1.0, + "content": "The task of generation poses a high standard for spatial coherency in structure, color, and texture, both", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 104, + 596, + 506, + 608 + ], + "spans": [ + { + "bbox": [ + 104, + 596, + 506, + 608 + ], + "score": 1.0, + "content": "globally and locally. The transformer encoders take embedding token words as inputs and calculate", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 606, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 606, + 505, + 620 + ], + "score": 1.0, + "content": "the interaction between each token recursively. [63, 34]. The main dilemma here is: what is the right", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 617, + 506, + 631 + ], + "spans": [ + { + "bbox": [ + 104, + 617, + 506, + 631 + ], + "score": 1.0, + "content": "“word\" for image generation tasks? If we similarly generate an image in a pixel-by-pixel manner", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 628, + 505, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 403, + 642 + ], + "score": 1.0, + "content": "through stacking transformer encoders, even a low-resolution image (e.g.", + "type": "text" + }, + { + "bbox": [ + 403, + 629, + 438, + 640 + ], + "score": 0.88, + "content": "3 2 \\times 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 628, + 505, + 642 + ], + "score": 1.0, + "content": ") can result in an", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 640, + 505, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 505, + 653 + ], + "score": 1.0, + "content": "excessively long sequence (1024), causing the explosive cost of self-attention (quadratic w.r.t. the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 650, + 507, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 507, + 664 + ], + "score": 1.0, + "content": "sequence length) and prohibiting the scalability to higher resolutions. To avoid this daunting cost,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 662, + 505, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 505, + 675 + ], + "score": 1.0, + "content": "we are inspired by a common design philosophy in CNN-based GANs, to iteratively upscale the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 673, + 505, + 685 + ], + "spans": [ + { + "bbox": [ + 106, + 673, + 505, + 685 + ], + "score": 1.0, + "content": "resolution at multiple stages [25, 16]. Our strategy is hence to increase the input sequence and reduce", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 684, + 256, + 696 + ], + "spans": [ + { + "bbox": [ + 106, + 684, + 256, + 696 + ], + "score": 1.0, + "content": "the embedding dimension gradually .", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 25.5, + "bbox_fs": [ + 104, + 585, + 507, + 696 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 700, + 502, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 700, + 504, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 700, + 504, + 712 + ], + "score": 1.0, + "content": "Figure 2 (left) illustrates a memory-friendly transformer-based generator that consists of multiple", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 710, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 724 + ], + "score": 1.0, + "content": "stages. Each stage stacks several transformer blocks. By stages, we gradually increase the feature", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 72, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 315, + 86 + ], + "score": 1.0, + "content": "map resolution until it meets the target resolution", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 315, + 73, + 349, + 83 + ], + "score": 0.88, + "content": "H \\times W", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 349, + 72, + 505, + 86 + ], + "score": 1.0, + "content": ". Specifically, the generator takes the", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 84, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 84, + 505, + 96 + ], + "score": 1.0, + "content": "random noise as its input, and passes it through a multiple-layer perceptron (MLP) to a vector of", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 93, + 505, + 108 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 133, + 108 + ], + "score": 1.0, + "content": "length", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 134, + 95, + 194, + 105 + ], + "score": 0.91, + "content": "H _ { 0 } \\times W _ { 0 } \\times C", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 194, + 93, + 313, + 108 + ], + "score": 1.0, + "content": ". The vector is reshaped into a", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 314, + 95, + 353, + 106 + ], + "score": 0.91, + "content": "H _ { 0 } \\times W _ { 0 }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 354, + 93, + 505, + 108 + ], + "score": 1.0, + "content": "resolution feature map (by default we", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 105, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 122, + 118 + ], + "score": 1.0, + "content": "use", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 123, + 106, + 183, + 117 + ], + "score": 0.89, + "content": "H _ { 0 } = W _ { 0 } = 8", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 183, + 105, + 239, + 118 + ], + "score": 1.0, + "content": "), each point a", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 239, + 106, + 248, + 116 + ], + "score": 0.82, + "content": "C", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 248, + 105, + 505, + 118 + ], + "score": 1.0, + "content": "-dimensional embedding. This “feature map\" is next treated as a", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 115, + 489, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 197, + 130 + ], + "score": 1.0, + "content": "length-64 sequence of", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 197, + 117, + 206, + 126 + ], + "score": 0.83, + "content": "C", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 206, + 115, + 489, + 130 + ], + "score": 1.0, + "content": "-dimensional tokens, combined with the learnable positional encoding.", + "type": "text", + "cross_page": true + } + ], + "index": 4 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 700, + 505, + 724 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 72, + 505, + 127 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 315, + 86 + ], + "score": 1.0, + "content": "map resolution until it meets the target resolution", + "type": "text" + }, + { + "bbox": [ + 315, + 73, + 349, + 83 + ], + "score": 0.88, + "content": "H \\times W", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 72, + 505, + 86 + ], + "score": 1.0, + "content": ". Specifically, the generator takes the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 84, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 84, + 505, + 96 + ], + "score": 1.0, + "content": "random noise as its input, and passes it through a multiple-layer perceptron (MLP) to a vector of", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 93, + 505, + 108 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 133, + 108 + ], + "score": 1.0, + "content": "length", + "type": "text" + }, + { + "bbox": [ + 134, + 95, + 194, + 105 + ], + "score": 0.91, + "content": "H _ { 0 } \\times W _ { 0 } \\times C", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 93, + 313, + 108 + ], + "score": 1.0, + "content": ". The vector is reshaped into a", + "type": "text" + }, + { + "bbox": [ + 314, + 95, + 353, + 106 + ], + "score": 0.91, + "content": "H _ { 0 } \\times W _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 93, + 505, + 108 + ], + "score": 1.0, + "content": "resolution feature map (by default we", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 105, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 122, + 118 + ], + "score": 1.0, + "content": "use", + "type": "text" + }, + { + "bbox": [ + 123, + 106, + 183, + 117 + ], + "score": 0.89, + "content": "H _ { 0 } = W _ { 0 } = 8", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 105, + 239, + 118 + ], + "score": 1.0, + "content": "), each point a", + "type": "text" + }, + { + "bbox": [ + 239, + 106, + 248, + 116 + ], + "score": 0.82, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 105, + 505, + 118 + ], + "score": 1.0, + "content": "-dimensional embedding. This “feature map\" is next treated as a", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 115, + 489, + 130 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 197, + 130 + ], + "score": 1.0, + "content": "length-64 sequence of", + "type": "text" + }, + { + "bbox": [ + 197, + 117, + 206, + 126 + ], + "score": 0.83, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 115, + 489, + 130 + ], + "score": 1.0, + "content": "-dimensional tokens, combined with the learnable positional encoding.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 106, + 132, + 506, + 257 + ], + "lines": [ + { + "bbox": [ + 105, + 132, + 506, + 146 + ], + "spans": [ + { + "bbox": [ + 105, + 132, + 506, + 146 + ], + "score": 1.0, + "content": "To scale up to higher-resolution images, we insert an upsampling module after each stage, consisting", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 144, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 144, + 505, + 156 + ], + "score": 1.0, + "content": "of a reshaping and resolution-upscaling layer. For lower-resolution stages (resolution lower than", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 154, + 506, + 168 + ], + "spans": [ + { + "bbox": [ + 106, + 155, + 142, + 165 + ], + "score": 0.87, + "content": "6 4 \\times 6 4 )", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 154, + 506, + 168 + ], + "score": 1.0, + "content": "), the upsampling module firstly reshapes the 1D sequence of token embedding back to a", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 163, + 506, + 179 + ], + "spans": [ + { + "bbox": [ + 104, + 163, + 174, + 179 + ], + "score": 1.0, + "content": "2D feature map", + "type": "text" + }, + { + "bbox": [ + 174, + 164, + 248, + 176 + ], + "score": 0.91, + "content": "X _ { i } ~ \\overset { \\cdot } { \\in } ~ \\mathbb { R } ^ { \\cup _ { i } \\times W _ { i } \\times C }", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 163, + 506, + 179 + ], + "score": 1.0, + "content": "and then adopts the bicubic layer to upsample its resolution", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 103, + 176, + 507, + 191 + ], + "spans": [ + { + "bbox": [ + 103, + 176, + 397, + 191 + ], + "score": 1.0, + "content": "while the embedded dimension is kept unchanged, resulting in the output", + "type": "text" + }, + { + "bbox": [ + 398, + 176, + 478, + 190 + ], + "score": 0.93, + "content": "\\boldsymbol { X } _ { i } ^ { ' } \\in \\mathbb { R } ^ { 2 H _ { i } \\times 2 W _ { i } \\times C }", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 176, + 507, + 191 + ], + "score": 1.0, + "content": ". After", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 189, + 505, + 203 + ], + "spans": [ + { + "bbox": [ + 105, + 190, + 212, + 203 + ], + "score": 1.0, + "content": "that, the 2D feature map", + "type": "text" + }, + { + "bbox": [ + 212, + 189, + 225, + 203 + ], + "score": 0.9, + "content": "X _ { i } ^ { ' }", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 190, + 505, + 203 + ], + "score": 1.0, + "content": "is again reshaped into the 1D sequence of embedding tokens. For", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 201, + 506, + 214 + ], + "spans": [ + { + "bbox": [ + 105, + 201, + 506, + 214 + ], + "score": 1.0, + "content": "higher-resolution stages, we replace the bicubic upscaling layer with the pixelshuffle module,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 212, + 506, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 301, + 226 + ], + "score": 1.0, + "content": "which upsamples the resolution of feature map by", + "type": "text" + }, + { + "bbox": [ + 301, + 213, + 315, + 223 + ], + "score": 0.87, + "content": "2 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 212, + 506, + 226 + ], + "score": 1.0, + "content": "ratio and also reduces the embedding dimension", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 104, + 223, + 506, + 237 + ], + "spans": [ + { + "bbox": [ + 104, + 223, + 506, + 237 + ], + "score": 1.0, + "content": "to a quarter of the input. This pyramid-structure with modified upscaling layers mitigates the memory", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 234, + 506, + 247 + ], + "spans": [ + { + "bbox": [ + 106, + 234, + 470, + 247 + ], + "score": 1.0, + "content": "and computation explosion. We repeat multiple stages until it reaches the target resolution", + "type": "text" + }, + { + "bbox": [ + 470, + 234, + 502, + 246 + ], + "score": 0.87, + "content": "( H , W )", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 234, + 506, + 247 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 244, + 504, + 258 + ], + "spans": [ + { + "bbox": [ + 104, + 244, + 437, + 258 + ], + "score": 1.0, + "content": "and then we will project the embedding dimension to 3 and obtain the RGB image", + "type": "text" + }, + { + "bbox": [ + 437, + 245, + 498, + 256 + ], + "score": 0.89, + "content": "Y \\in \\mathbb { R } ^ { H \\times W \\times 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 498, + 244, + 504, + 258 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 10 + }, + { + "type": "title", + "bbox": [ + 107, + 266, + 240, + 277 + ], + "lines": [ + { + "bbox": [ + 105, + 264, + 241, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 241, + 279 + ], + "score": 1.0, + "content": "3.2 Multi-scale Discriminator", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 281, + 505, + 380 + ], + "lines": [ + { + "bbox": [ + 106, + 281, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 106, + 281, + 505, + 294 + ], + "score": 1.0, + "content": "Unlike the generator which synthesizes precise pixels, the discriminator is tasked to distinguish", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 292, + 504, + 304 + ], + "spans": [ + { + "bbox": [ + 106, + 292, + 504, + 304 + ], + "score": 1.0, + "content": "between real/fake images. This allows us to treat it as a typical classifier by simply tokenizing the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 303, + 506, + 316 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 506, + 316 + ], + "score": 1.0, + "content": "input image in a coarser patch-level [34], where each patch can be regarded as a “word\". However,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 314, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 314, + 506, + 326 + ], + "score": 1.0, + "content": "compared to image recognition tasks where classifiers focus on the semantic differences, the dis-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 325, + 506, + 338 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 506, + 338 + ], + "score": 1.0, + "content": "criminator executes a simpler and more detail-oriented task to distinguish between synthesized and", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 335, + 506, + 349 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 506, + 349 + ], + "score": 1.0, + "content": "real. Therefore, the local visual cues and artifacts will have an important effect on the discriminator.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 346, + 505, + 360 + ], + "spans": [ + { + "bbox": [ + 106, + 346, + 505, + 360 + ], + "score": 1.0, + "content": "Practically, we observe that the patch splitting rule plays a crucial role, where large patch size", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 357, + 505, + 371 + ], + "spans": [ + { + "bbox": [ + 106, + 357, + 505, + 371 + ], + "score": 1.0, + "content": "sacrifices low-level texture details, and smaller patch size results in a longer sequence that costs more", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 368, + 453, + 381 + ], + "spans": [ + { + "bbox": [ + 106, + 368, + 453, + 381 + ], + "score": 1.0, + "content": "memory. The above dilemma motivates our design of multi-scale discriminator below.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 385, + 505, + 533 + ], + "lines": [ + { + "bbox": [ + 106, + 384, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 505, + 397 + ], + "score": 1.0, + "content": "As shown in Figure 2 (right), a multi-scale discriminator is designed to take varying size of patches", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 393, + 506, + 409 + ], + "spans": [ + { + "bbox": [ + 104, + 393, + 365, + 409 + ], + "score": 1.0, + "content": "as inputs, at its different stages. We firstly split the input images", + "type": "text" + }, + { + "bbox": [ + 365, + 395, + 426, + 406 + ], + "score": 0.92, + "content": "Y \\in R ^ { H \\times W \\times 3 ^ { \\bullet } }", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 393, + 506, + 409 + ], + "score": 1.0, + "content": "into three different", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 403, + 507, + 422 + ], + "spans": [ + { + "bbox": [ + 104, + 403, + 288, + 422 + ], + "score": 1.0, + "content": "sequences by choosing different patch sizes", + "type": "text" + }, + { + "bbox": [ + 288, + 407, + 339, + 418 + ], + "score": 0.79, + "content": "( P , 2 P , 4 P )", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 403, + 435, + 422 + ], + "score": 1.0, + "content": ". The longest sequence", + "type": "text" + }, + { + "bbox": [ + 436, + 405, + 494, + 420 + ], + "score": 0.93, + "content": "\\textstyle \\left( { \\frac { H } { P } } \\times { \\frac { W } { P } } \\right) \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 403, + 507, + 422 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 416, + 508, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 202, + 437 + ], + "score": 1.0, + "content": "linearly transformed to", + "type": "text" + }, + { + "bbox": [ + 203, + 418, + 264, + 433 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\left( \\frac { H } { P } \\times \\frac { W } { P } \\right) \\times \\frac { C } { 4 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 416, + 508, + 437 + ], + "score": 1.0, + "content": "and then combined with the learnable position encoding to", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 431, + 506, + 446 + ], + "spans": [ + { + "bbox": [ + 104, + 431, + 275, + 446 + ], + "score": 1.0, + "content": "serve as the input of the first stage, where", + "type": "text" + }, + { + "bbox": [ + 275, + 432, + 284, + 446 + ], + "score": 0.88, + "content": "\\frac { C } { 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 431, + 506, + 446 + ], + "score": 1.0, + "content": "is the embedded dimension size. Similarly, the second", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 102, + 439, + 509, + 464 + ], + "spans": [ + { + "bbox": [ + 102, + 439, + 305, + 464 + ], + "score": 1.0, + "content": "4and third sequences are linearly transformed to", + "type": "text" + }, + { + "bbox": [ + 306, + 444, + 373, + 459 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\big ( \\frac { H } { 2 P } \\times \\frac { W } { 2 P } \\big ) \\times \\frac { C } { 4 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 439, + 394, + 464 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 394, + 444, + 462, + 459 + ], + "score": 0.91, + "content": "\\bigl ( { \\frac { H } { 4 P } } \\times { \\frac { W } { 4 P } } \\bigr ) \\times { \\frac { C } { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 439, + 509, + 464 + ], + "score": 1.0, + "content": ", and then", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 456, + 506, + 470 + ], + "spans": [ + { + "bbox": [ + 104, + 456, + 506, + 470 + ], + "score": 1.0, + "content": "separately concatenated into the second and third stages. Thus these three different sequences are", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 467, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 505, + 480 + ], + "score": 1.0, + "content": "able to extract both the semantic structure and texture details. Similar to the generator, we reshape", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 478, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 505, + 491 + ], + "score": 1.0, + "content": "the 1D-sentence to 2D feature map and adopt Average Pooling layer to downsample the feature", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 490, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 505, + 502 + ], + "score": 1.0, + "content": "map resolution, between each stage. By recursively forming the transformer blocks in each stage, we", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 500, + 506, + 513 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 506, + 513 + ], + "score": 1.0, + "content": "obtain a pyramid architecture where multi-scale representation is extracted. At the end of these blocks,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 104, + 510, + 506, + 524 + ], + "spans": [ + { + "bbox": [ + 104, + 510, + 506, + 524 + ], + "score": 1.0, + "content": "a [cls] token is appended at the beginning of the 1D sequence and then taken by the classification", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 523, + 263, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 263, + 534 + ], + "score": 1.0, + "content": "head to output the real/fake prediction.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 32 + }, + { + "type": "title", + "bbox": [ + 107, + 542, + 467, + 554 + ], + "lines": [ + { + "bbox": [ + 104, + 540, + 469, + 558 + ], + "spans": [ + { + "bbox": [ + 104, + 540, + 469, + 558 + ], + "score": 1.0, + "content": "3.3 Grid Self-Attention: A Scalable Variant of Self-Attention for Image Generation", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 106, + 558, + 505, + 613 + ], + "lines": [ + { + "bbox": [ + 106, + 558, + 505, + 571 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 505, + 571 + ], + "score": 1.0, + "content": "Self-attention allows the generator to capture the global correspondence, yet also impedes the", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 568, + 506, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 506, + 581 + ], + "score": 1.0, + "content": "efficiency when modeling long sequences/higher resolutions. That motivates many efficient self-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 579, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 579, + 506, + 592 + ], + "score": 1.0, + "content": "attention designs in both language [64, 65] and vision tasks [66, 67]. To adapt self-attention for", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 591, + 506, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 591, + 506, + 604 + ], + "score": 1.0, + "content": "higher-resolution generative tasks, we propose a simple yet effective strategy, named Grid Self-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 104, + 601, + 332, + 615 + ], + "spans": [ + { + "bbox": [ + 104, + 601, + 332, + 615 + ], + "score": 1.0, + "content": "Attention, tailored for high-resolution image generation.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 42 + }, + { + "type": "text", + "bbox": [ + 107, + 618, + 505, + 695 + ], + "lines": [ + { + "bbox": [ + 105, + 618, + 506, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 506, + 630 + ], + "score": 1.0, + "content": "As shown in Figure 3, instead of calculating the correspondence between a given token and all other", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 629, + 506, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 629, + 506, + 642 + ], + "score": 1.0, + "content": "tokens, the grid self-attention partitions the full-size feature map into several non-overlapped grids,", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 639, + 506, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 506, + 652 + ], + "score": 1.0, + "content": "and the token interactions are calculated inside each local grid. We add the grid self-attention on", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 651, + 504, + 662 + ], + "spans": [ + { + "bbox": [ + 106, + 651, + 289, + 662 + ], + "score": 1.0, + "content": "high-resolution stages (resolution higher than", + "type": "text" + }, + { + "bbox": [ + 290, + 651, + 324, + 661 + ], + "score": 0.88, + "content": "3 2 \\times 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 651, + 504, + 662 + ], + "score": 1.0, + "content": ") while still keeping standard self-attention in", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 662, + 505, + 674 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 505, + 674 + ], + "score": 1.0, + "content": "low-resolution stages, shown as Figure 2, again so as to strategically balance local details and global", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 672, + 505, + 685 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 505, + 685 + ], + "score": 1.0, + "content": "awareness. The grid self-attention shows surprising effectiveness over other efficient self-attention", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 684, + 379, + 696 + ], + "spans": [ + { + "bbox": [ + 106, + 684, + 379, + 696 + ], + "score": 1.0, + "content": "forms [64, 67] in generative tasks, as compared later in Section 4.1.", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 48 + }, + { + "type": "text", + "bbox": [ + 107, + 700, + 503, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "score": 1.0, + "content": "One potential concern might arise with the boundary artifact between each grid. We observe that", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 710, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 724 + ], + "score": 1.0, + "content": "while the artifact indeed occurs at early training stages, it gradually vanishes given enough training", + "type": "text" + } + ], + "index": 53 + } + ], + "index": 52.5 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 740, + 309, + 753 + ], + "spans": [ + { + "bbox": [ + 302, + 740, + 309, + 753 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 72, + 505, + 127 + ], + "lines": [], + "index": 2, + "bbox_fs": [ + 105, + 72, + 505, + 130 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 132, + 506, + 257 + ], + "lines": [ + { + "bbox": [ + 105, + 132, + 506, + 146 + ], + "spans": [ + { + "bbox": [ + 105, + 132, + 506, + 146 + ], + "score": 1.0, + "content": "To scale up to higher-resolution images, we insert an upsampling module after each stage, consisting", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 144, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 144, + 505, + 156 + ], + "score": 1.0, + "content": "of a reshaping and resolution-upscaling layer. For lower-resolution stages (resolution lower than", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 154, + 506, + 168 + ], + "spans": [ + { + "bbox": [ + 106, + 155, + 142, + 165 + ], + "score": 0.87, + "content": "6 4 \\times 6 4 )", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 154, + 506, + 168 + ], + "score": 1.0, + "content": "), the upsampling module firstly reshapes the 1D sequence of token embedding back to a", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 163, + 506, + 179 + ], + "spans": [ + { + "bbox": [ + 104, + 163, + 174, + 179 + ], + "score": 1.0, + "content": "2D feature map", + "type": "text" + }, + { + "bbox": [ + 174, + 164, + 248, + 176 + ], + "score": 0.91, + "content": "X _ { i } ~ \\overset { \\cdot } { \\in } ~ \\mathbb { R } ^ { \\cup _ { i } \\times W _ { i } \\times C }", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 163, + 506, + 179 + ], + "score": 1.0, + "content": "and then adopts the bicubic layer to upsample its resolution", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 103, + 176, + 507, + 191 + ], + "spans": [ + { + "bbox": [ + 103, + 176, + 397, + 191 + ], + "score": 1.0, + "content": "while the embedded dimension is kept unchanged, resulting in the output", + "type": "text" + }, + { + "bbox": [ + 398, + 176, + 478, + 190 + ], + "score": 0.93, + "content": "\\boldsymbol { X } _ { i } ^ { ' } \\in \\mathbb { R } ^ { 2 H _ { i } \\times 2 W _ { i } \\times C }", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 176, + 507, + 191 + ], + "score": 1.0, + "content": ". After", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 189, + 505, + 203 + ], + "spans": [ + { + "bbox": [ + 105, + 190, + 212, + 203 + ], + "score": 1.0, + "content": "that, the 2D feature map", + "type": "text" + }, + { + "bbox": [ + 212, + 189, + 225, + 203 + ], + "score": 0.9, + "content": "X _ { i } ^ { ' }", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 190, + 505, + 203 + ], + "score": 1.0, + "content": "is again reshaped into the 1D sequence of embedding tokens. For", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 201, + 506, + 214 + ], + "spans": [ + { + "bbox": [ + 105, + 201, + 506, + 214 + ], + "score": 1.0, + "content": "higher-resolution stages, we replace the bicubic upscaling layer with the pixelshuffle module,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 212, + 506, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 301, + 226 + ], + "score": 1.0, + "content": "which upsamples the resolution of feature map by", + "type": "text" + }, + { + "bbox": [ + 301, + 213, + 315, + 223 + ], + "score": 0.87, + "content": "2 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 212, + 506, + 226 + ], + "score": 1.0, + "content": "ratio and also reduces the embedding dimension", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 104, + 223, + 506, + 237 + ], + "spans": [ + { + "bbox": [ + 104, + 223, + 506, + 237 + ], + "score": 1.0, + "content": "to a quarter of the input. This pyramid-structure with modified upscaling layers mitigates the memory", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 234, + 506, + 247 + ], + "spans": [ + { + "bbox": [ + 106, + 234, + 470, + 247 + ], + "score": 1.0, + "content": "and computation explosion. We repeat multiple stages until it reaches the target resolution", + "type": "text" + }, + { + "bbox": [ + 470, + 234, + 502, + 246 + ], + "score": 0.87, + "content": "( H , W )", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 234, + 506, + 247 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 244, + 504, + 258 + ], + "spans": [ + { + "bbox": [ + 104, + 244, + 437, + 258 + ], + "score": 1.0, + "content": "and then we will project the embedding dimension to 3 and obtain the RGB image", + "type": "text" + }, + { + "bbox": [ + 437, + 245, + 498, + 256 + ], + "score": 0.89, + "content": "Y \\in \\mathbb { R } ^ { H \\times W \\times 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 498, + 244, + 504, + 258 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 10, + "bbox_fs": [ + 103, + 132, + 507, + 258 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 266, + 240, + 277 + ], + "lines": [ + { + "bbox": [ + 105, + 264, + 241, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 241, + 279 + ], + "score": 1.0, + "content": "3.2 Multi-scale Discriminator", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 281, + 505, + 380 + ], + "lines": [ + { + "bbox": [ + 106, + 281, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 106, + 281, + 505, + 294 + ], + "score": 1.0, + "content": "Unlike the generator which synthesizes precise pixels, the discriminator is tasked to distinguish", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 292, + 504, + 304 + ], + "spans": [ + { + "bbox": [ + 106, + 292, + 504, + 304 + ], + "score": 1.0, + "content": "between real/fake images. This allows us to treat it as a typical classifier by simply tokenizing the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 303, + 506, + 316 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 506, + 316 + ], + "score": 1.0, + "content": "input image in a coarser patch-level [34], where each patch can be regarded as a “word\". However,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 314, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 314, + 506, + 326 + ], + "score": 1.0, + "content": "compared to image recognition tasks where classifiers focus on the semantic differences, the dis-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 325, + 506, + 338 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 506, + 338 + ], + "score": 1.0, + "content": "criminator executes a simpler and more detail-oriented task to distinguish between synthesized and", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 335, + 506, + 349 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 506, + 349 + ], + "score": 1.0, + "content": "real. Therefore, the local visual cues and artifacts will have an important effect on the discriminator.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 346, + 505, + 360 + ], + "spans": [ + { + "bbox": [ + 106, + 346, + 505, + 360 + ], + "score": 1.0, + "content": "Practically, we observe that the patch splitting rule plays a crucial role, where large patch size", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 357, + 505, + 371 + ], + "spans": [ + { + "bbox": [ + 106, + 357, + 505, + 371 + ], + "score": 1.0, + "content": "sacrifices low-level texture details, and smaller patch size results in a longer sequence that costs more", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 368, + 453, + 381 + ], + "spans": [ + { + "bbox": [ + 106, + 368, + 453, + 381 + ], + "score": 1.0, + "content": "memory. The above dilemma motivates our design of multi-scale discriminator below.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 21, + "bbox_fs": [ + 105, + 281, + 506, + 381 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 385, + 505, + 533 + ], + "lines": [ + { + "bbox": [ + 106, + 384, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 505, + 397 + ], + "score": 1.0, + "content": "As shown in Figure 2 (right), a multi-scale discriminator is designed to take varying size of patches", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 393, + 506, + 409 + ], + "spans": [ + { + "bbox": [ + 104, + 393, + 365, + 409 + ], + "score": 1.0, + "content": "as inputs, at its different stages. We firstly split the input images", + "type": "text" + }, + { + "bbox": [ + 365, + 395, + 426, + 406 + ], + "score": 0.92, + "content": "Y \\in R ^ { H \\times W \\times 3 ^ { \\bullet } }", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 393, + 506, + 409 + ], + "score": 1.0, + "content": "into three different", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 403, + 507, + 422 + ], + "spans": [ + { + "bbox": [ + 104, + 403, + 288, + 422 + ], + "score": 1.0, + "content": "sequences by choosing different patch sizes", + "type": "text" + }, + { + "bbox": [ + 288, + 407, + 339, + 418 + ], + "score": 0.79, + "content": "( P , 2 P , 4 P )", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 403, + 435, + 422 + ], + "score": 1.0, + "content": ". The longest sequence", + "type": "text" + }, + { + "bbox": [ + 436, + 405, + 494, + 420 + ], + "score": 0.93, + "content": "\\textstyle \\left( { \\frac { H } { P } } \\times { \\frac { W } { P } } \\right) \\times 3", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 403, + 507, + 422 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 416, + 508, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 202, + 437 + ], + "score": 1.0, + "content": "linearly transformed to", + "type": "text" + }, + { + "bbox": [ + 203, + 418, + 264, + 433 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\left( \\frac { H } { P } \\times \\frac { W } { P } \\right) \\times \\frac { C } { 4 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 416, + 508, + 437 + ], + "score": 1.0, + "content": "and then combined with the learnable position encoding to", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 431, + 506, + 446 + ], + "spans": [ + { + "bbox": [ + 104, + 431, + 275, + 446 + ], + "score": 1.0, + "content": "serve as the input of the first stage, where", + "type": "text" + }, + { + "bbox": [ + 275, + 432, + 284, + 446 + ], + "score": 0.88, + "content": "\\frac { C } { 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 431, + 506, + 446 + ], + "score": 1.0, + "content": "is the embedded dimension size. Similarly, the second", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 102, + 439, + 509, + 464 + ], + "spans": [ + { + "bbox": [ + 102, + 439, + 305, + 464 + ], + "score": 1.0, + "content": "4and third sequences are linearly transformed to", + "type": "text" + }, + { + "bbox": [ + 306, + 444, + 373, + 459 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\big ( \\frac { H } { 2 P } \\times \\frac { W } { 2 P } \\big ) \\times \\frac { C } { 4 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 439, + 394, + 464 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 394, + 444, + 462, + 459 + ], + "score": 0.91, + "content": "\\bigl ( { \\frac { H } { 4 P } } \\times { \\frac { W } { 4 P } } \\bigr ) \\times { \\frac { C } { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 439, + 509, + 464 + ], + "score": 1.0, + "content": ", and then", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 456, + 506, + 470 + ], + "spans": [ + { + "bbox": [ + 104, + 456, + 506, + 470 + ], + "score": 1.0, + "content": "separately concatenated into the second and third stages. Thus these three different sequences are", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 467, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 505, + 480 + ], + "score": 1.0, + "content": "able to extract both the semantic structure and texture details. Similar to the generator, we reshape", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 478, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 505, + 491 + ], + "score": 1.0, + "content": "the 1D-sentence to 2D feature map and adopt Average Pooling layer to downsample the feature", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 490, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 505, + 502 + ], + "score": 1.0, + "content": "map resolution, between each stage. By recursively forming the transformer blocks in each stage, we", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 500, + 506, + 513 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 506, + 513 + ], + "score": 1.0, + "content": "obtain a pyramid architecture where multi-scale representation is extracted. At the end of these blocks,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 104, + 510, + 506, + 524 + ], + "spans": [ + { + "bbox": [ + 104, + 510, + 506, + 524 + ], + "score": 1.0, + "content": "a [cls] token is appended at the beginning of the 1D sequence and then taken by the classification", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 523, + 263, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 263, + 534 + ], + "score": 1.0, + "content": "head to output the real/fake prediction.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 32, + "bbox_fs": [ + 102, + 384, + 509, + 534 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 542, + 467, + 554 + ], + "lines": [ + { + "bbox": [ + 104, + 540, + 469, + 558 + ], + "spans": [ + { + "bbox": [ + 104, + 540, + 469, + 558 + ], + "score": 1.0, + "content": "3.3 Grid Self-Attention: A Scalable Variant of Self-Attention for Image Generation", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 106, + 558, + 505, + 613 + ], + "lines": [ + { + "bbox": [ + 106, + 558, + 505, + 571 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 505, + 571 + ], + "score": 1.0, + "content": "Self-attention allows the generator to capture the global correspondence, yet also impedes the", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 568, + 506, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 506, + 581 + ], + "score": 1.0, + "content": "efficiency when modeling long sequences/higher resolutions. That motivates many efficient self-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 579, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 579, + 506, + 592 + ], + "score": 1.0, + "content": "attention designs in both language [64, 65] and vision tasks [66, 67]. To adapt self-attention for", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 591, + 506, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 591, + 506, + 604 + ], + "score": 1.0, + "content": "higher-resolution generative tasks, we propose a simple yet effective strategy, named Grid Self-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 104, + 601, + 332, + 615 + ], + "spans": [ + { + "bbox": [ + 104, + 601, + 332, + 615 + ], + "score": 1.0, + "content": "Attention, tailored for high-resolution image generation.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 42, + "bbox_fs": [ + 104, + 558, + 506, + 615 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 618, + 505, + 695 + ], + "lines": [ + { + "bbox": [ + 105, + 618, + 506, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 618, + 506, + 630 + ], + "score": 1.0, + "content": "As shown in Figure 3, instead of calculating the correspondence between a given token and all other", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 629, + 506, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 629, + 506, + 642 + ], + "score": 1.0, + "content": "tokens, the grid self-attention partitions the full-size feature map into several non-overlapped grids,", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 639, + 506, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 506, + 652 + ], + "score": 1.0, + "content": "and the token interactions are calculated inside each local grid. We add the grid self-attention on", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 651, + 504, + 662 + ], + "spans": [ + { + "bbox": [ + 106, + 651, + 289, + 662 + ], + "score": 1.0, + "content": "high-resolution stages (resolution higher than", + "type": "text" + }, + { + "bbox": [ + 290, + 651, + 324, + 661 + ], + "score": 0.88, + "content": "3 2 \\times 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 651, + 504, + 662 + ], + "score": 1.0, + "content": ") while still keeping standard self-attention in", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 662, + 505, + 674 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 505, + 674 + ], + "score": 1.0, + "content": "low-resolution stages, shown as Figure 2, again so as to strategically balance local details and global", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 672, + 505, + 685 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 505, + 685 + ], + "score": 1.0, + "content": "awareness. The grid self-attention shows surprising effectiveness over other efficient self-attention", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 684, + 379, + 696 + ], + "spans": [ + { + "bbox": [ + 106, + 684, + 379, + 696 + ], + "score": 1.0, + "content": "forms [64, 67] in generative tasks, as compared later in Section 4.1.", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 48, + "bbox_fs": [ + 105, + 618, + 506, + 696 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 700, + 503, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "score": 1.0, + "content": "One potential concern might arise with the boundary artifact between each grid. We observe that", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 710, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 724 + ], + "score": 1.0, + "content": "while the artifact indeed occurs at early training stages, it gradually vanishes given enough training", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 105, + 180, + 506, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 180, + 506, + 194 + ], + "score": 1.0, + "content": "iterations and training data, while producing nicely coherent final results. We think this is owing", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 192, + 504, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 192, + 504, + 204 + ], + "score": 1.0, + "content": "to the larger, multi-scale receptive field of the discriminator that requires generated image fidelity", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 203, + 505, + 216 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 505, + 216 + ], + "score": 1.0, + "content": "in different scales. For other cases where the large-scale training data is hard to obtain, we discuss", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 215, + 224, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 224, + 225 + ], + "score": 1.0, + "content": "several solutions on Sec. 4.6.", + "type": "text", + "cross_page": true + } + ], + "index": 9 + } + ], + "index": 52.5, + "bbox_fs": [ + 105, + 700, + 505, + 724 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 111, + 61, + 502, + 142 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 111, + 61, + 502, + 142 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 111, + 61, + 502, + 142 + ], + "spans": [ + { + "bbox": [ + 111, + 61, + 502, + 142 + ], + "score": 0.954, + "type": "image", + "image_path": "5aa30c4dc12e735df83a2803ff77d51d6c674cced188b389797c086e25b8de54.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 111, + 61, + 502, + 88.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 111, + 88.0, + 502, + 115.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 111, + 115.0, + 502, + 142.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 144, + 505, + 176 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 144, + 504, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 144, + 504, + 155 + ], + "score": 1.0, + "content": "Figure 3: Grid Self-Attention across different transformer stages. We replace Standard Self-Attention", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 154, + 505, + 166 + ], + "spans": [ + { + "bbox": [ + 106, + 154, + 352, + 166 + ], + "score": 1.0, + "content": "with Grid Self-Attention when the resolution is higher than", + "type": "text" + }, + { + "bbox": [ + 352, + 154, + 387, + 165 + ], + "score": 0.9, + "content": "3 2 \\times 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 154, + 505, + 166 + ], + "score": 1.0, + "content": "and the grid size is set to be", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 165, + 185, + 176 + ], + "spans": [ + { + "bbox": [ + 106, + 165, + 140, + 176 + ], + "score": 0.88, + "content": "1 6 \\times 1 6", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 165, + 185, + 176 + ], + "score": 1.0, + "content": "by default.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 107, + 181, + 505, + 225 + ], + "lines": [ + { + "bbox": [ + 105, + 180, + 506, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 180, + 506, + 194 + ], + "score": 1.0, + "content": "iterations and training data, while producing nicely coherent final results. We think this is owing", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 192, + 504, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 192, + 504, + 204 + ], + "score": 1.0, + "content": "to the larger, multi-scale receptive field of the discriminator that requires generated image fidelity", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 203, + 505, + 216 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 505, + 216 + ], + "score": 1.0, + "content": "in different scales. For other cases where the large-scale training data is hard to obtain, we discuss", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 215, + 224, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 224, + 225 + ], + "score": 1.0, + "content": "several solutions on Sec. 4.6.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 7.5 + }, + { + "type": "title", + "bbox": [ + 108, + 233, + 260, + 245 + ], + "lines": [ + { + "bbox": [ + 105, + 231, + 260, + 248 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 260, + 248 + ], + "score": 1.0, + "content": "3.4 Exploring the Training Recipe", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 248, + 505, + 303 + ], + "lines": [ + { + "bbox": [ + 106, + 248, + 505, + 261 + ], + "spans": [ + { + "bbox": [ + 106, + 248, + 505, + 261 + ], + "score": 1.0, + "content": "Data Augmentation. The transformer-based architectures are known to be highly data-hungry due", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 258, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 258, + 505, + 272 + ], + "score": 1.0, + "content": "to removing human-designed bias. Particularly in image recognition task [34], they were inferior to", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 271, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 505, + 282 + ], + "score": 1.0, + "content": "CNNs until much larger external data [68] was used for pre-training. To remove this roadblock, data", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 281, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 505, + 294 + ], + "score": 1.0, + "content": "augmentation was revealed as a blessing in [50], which showed that different types of strong data", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 291, + 413, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 413, + 305 + ], + "score": 1.0, + "content": "augmentation could lead us to data-efficient training for vision transformers.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 106, + 308, + 505, + 397 + ], + "lines": [ + { + "bbox": [ + 105, + 308, + 506, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 308, + 506, + 321 + ], + "score": 1.0, + "content": "We follow a similar mindset. Traditionally, training CNN-based GANs hardly refers to data augmenta-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 319, + 505, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 319, + 505, + 331 + ], + "score": 1.0, + "content": "tion. Recently, there is an interest surge in the few-shot GAN training, aiming to match state-of-the-art", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 330, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 330, + 505, + 343 + ], + "score": 1.0, + "content": "GAN results with orders of magnitude fewer real images [69, 70]. Contrary to this “commonsense\"", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 341, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 341, + 505, + 353 + ], + "score": 1.0, + "content": "in CNNs, data augmentation is found to be crucial in transformer-based architectures, even with", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 351, + 506, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 131, + 363 + ], + "score": 0.87, + "content": "100 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 131, + 351, + 506, + 366 + ], + "score": 1.0, + "content": "real images being utilized. We show that simply using differential augmentation [69] with", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 362, + 506, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 506, + 376 + ], + "score": 1.0, + "content": "three basic operators {T ranslation, Cutout, Color} leads to surprising performance improvement", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 374, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 505, + 386 + ], + "score": 1.0, + "content": "for TransGAN, while CNN-based GANs hardly benefit from it. We conduct a concrete study on the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 385, + 461, + 397 + ], + "spans": [ + { + "bbox": [ + 106, + 385, + 461, + 397 + ], + "score": 1.0, + "content": "effectiveness of augmentation for both transformer and CNNs: see details in Section 4.2", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 106, + 401, + 505, + 445 + ], + "lines": [ + { + "bbox": [ + 105, + 401, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 505, + 414 + ], + "score": 1.0, + "content": "Relative Position Encoding. While classical transformers [49, 34] used deterministic position", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 412, + 506, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 506, + 425 + ], + "score": 1.0, + "content": "encoding or learnable position encoding, the relative position encoding [71] gains increasing popular-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 423, + 506, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 506, + 436 + ], + "score": 1.0, + "content": "ity [72, 28, 52, 73], by exploiting lags instead of absolute positions. Considering a single head of", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 433, + 186, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 186, + 447 + ], + "score": 1.0, + "content": "self-attention layer,", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5 + }, + { + "type": "interline_equation", + "bbox": [ + 213, + 443, + 398, + 471 + ], + "lines": [ + { + "bbox": [ + 213, + 443, + 398, + 471 + ], + "spans": [ + { + "bbox": [ + 213, + 443, + 398, + 471 + ], + "score": 0.92, + "content": "A t t e n t i o n ( Q , K , V ) = s o f t m a x ( ( \\frac { Q K ^ { T } } { \\sqrt { d _ { k } } } V )", + "type": "interline_equation", + "image_path": "dbb01f6ced9089b3c2847770df83453bea906785efbac2a8e617ec65bab96ef6.jpg" + } + ] + } + ], + "index": 28.5, + "virtual_lines": [ + { + "bbox": [ + 213, + 443, + 398, + 457.0 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 213, + 457.0, + 398, + 471.0 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 473, + 505, + 542 + ], + "lines": [ + { + "bbox": [ + 104, + 471, + 507, + 489 + ], + "spans": [ + { + "bbox": [ + 104, + 471, + 132, + 489 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 474, + 223, + 486 + ], + "score": 0.91, + "content": "Q , K , V \\in \\mathbb { R } ^ { ( H \\times W ) \\times C }", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 471, + 368, + 489 + ], + "score": 1.0, + "content": "represent query, key, value matrices,", + "type": "text" + }, + { + "bbox": [ + 369, + 475, + 402, + 486 + ], + "score": 0.84, + "content": "H , W , C", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 471, + 507, + 489 + ], + "score": 1.0, + "content": "denotes the height, width,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 485, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 106, + 485, + 505, + 497 + ], + "score": 1.0, + "content": "embedded dimension of the input feature map. The difference in coordinate between each query and", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 496, + 505, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 136, + 509 + ], + "score": 1.0, + "content": "key on", + "type": "text" + }, + { + "bbox": [ + 136, + 497, + 147, + 506 + ], + "score": 0.8, + "content": "H", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 496, + 245, + 509 + ], + "score": 1.0, + "content": "axis lies in the range of", + "type": "text" + }, + { + "bbox": [ + 246, + 496, + 325, + 509 + ], + "score": 0.91, + "content": "[ - ( H - \\bar { 1 } ) , H - 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 496, + 391, + 509 + ], + "score": 1.0, + "content": ", and similar for", + "type": "text" + }, + { + "bbox": [ + 391, + 497, + 403, + 507 + ], + "score": 0.58, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 496, + 505, + 509 + ], + "score": 1.0, + "content": "axis. By simultaneously", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 507, + 505, + 520 + ], + "spans": [ + { + "bbox": [ + 106, + 508, + 176, + 520 + ], + "score": 1.0, + "content": "considering both", + "type": "text" + }, + { + "bbox": [ + 176, + 508, + 186, + 517 + ], + "score": 0.84, + "content": "H", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 508, + 204, + 520 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 205, + 507, + 217, + 517 + ], + "score": 0.71, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 508, + 505, + 520 + ], + "score": 1.0, + "content": "axis, the relative position can be represented by a parameterized matrix", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 107, + 516, + 507, + 534 + ], + "spans": [ + { + "bbox": [ + 107, + 518, + 200, + 530 + ], + "score": 0.9, + "content": "M \\in \\mathbb { R } ^ { ( 2 \\hat { H _ { - 1 } } ) \\times ( 2 W - 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 516, + 383, + 534 + ], + "score": 1.0, + "content": ". Per coordinate, the relative position encoding", + "type": "text" + }, + { + "bbox": [ + 384, + 520, + 393, + 529 + ], + "score": 0.83, + "content": "E", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 516, + 475, + 534 + ], + "score": 1.0, + "content": "is taken from matrix", + "type": "text" + }, + { + "bbox": [ + 476, + 520, + 487, + 529 + ], + "score": 0.77, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 488, + 516, + 507, + 534 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 528, + 383, + 544 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 214, + 544 + ], + "score": 1.0, + "content": "added to the attention map", + "type": "text" + }, + { + "bbox": [ + 215, + 530, + 239, + 542 + ], + "score": 0.92, + "content": "Q K ^ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 528, + 383, + 544 + ], + "score": 1.0, + "content": "as a bias term, shown as following,", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 32.5 + }, + { + "type": "interline_equation", + "bbox": [ + 199, + 546, + 412, + 574 + ], + "lines": [ + { + "bbox": [ + 199, + 546, + 412, + 574 + ], + "spans": [ + { + "bbox": [ + 199, + 546, + 412, + 574 + ], + "score": 0.93, + "content": "A t t e n t i o n ( Q , K , V ) = s o f t m a x ( ( ( \\frac { Q K ^ { T } } { \\sqrt { d _ { k } } } + E ) V )", + "type": "interline_equation", + "image_path": "d642168023319f11cbc80f8d9df34a9ce6f6ffd07f0e3008d315f28bd6603d49.jpg" + } + ] + } + ], + "index": 36.5, + "virtual_lines": [ + { + "bbox": [ + 199, + 546, + 412, + 560.0 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 199, + 560.0, + 412, + 574.0 + ], + "spans": [], + "index": 37 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 577, + 505, + 633 + ], + "lines": [ + { + "bbox": [ + 106, + 577, + 504, + 591 + ], + "spans": [ + { + "bbox": [ + 106, + 577, + 504, + 591 + ], + "score": 1.0, + "content": "Compared to its absolute counterpart, relative position encoding learns a stronger “relationship\"", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 586, + 505, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 505, + 603 + ], + "score": 1.0, + "content": "between local contents, bringing important performance gains in large-scale cases and enjoying", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 598, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 505, + 613 + ], + "score": 1.0, + "content": "widespread use ever since. We also observe it to consistently improve TransGAN, especially on", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 609, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 505, + 624 + ], + "score": 1.0, + "content": "higher-resolution datasets. We hence apply it on top of the learnable absolute positional encoding for", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 620, + 256, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 620, + 256, + 635 + ], + "score": 1.0, + "content": "both the generator and discriminator.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 106, + 637, + 505, + 723 + ], + "lines": [ + { + "bbox": [ + 104, + 635, + 506, + 652 + ], + "spans": [ + { + "bbox": [ + 104, + 635, + 506, + 652 + ], + "score": 1.0, + "content": "Modified Normalization. Normalization layers are known to help stabilize the deep learning training", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 648, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 106, + 648, + 505, + 661 + ], + "score": 1.0, + "content": "of deep neural networks, sometimes remarkably. While both the original transformer [49] and its", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 659, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 505, + 672 + ], + "score": 1.0, + "content": "variants [52, 54] by default use the layer normalization, we follow previous works [75, 16] and", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 104, + 669, + 506, + 684 + ], + "spans": [ + { + "bbox": [ + 104, + 669, + 506, + 684 + ], + "score": 1.0, + "content": "replace it with a token-wise scaling layer to prevent the magnitudes in transformer blocks from", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 101, + 678, + 504, + 705 + ], + "spans": [ + { + "bbox": [ + 101, + 678, + 218, + 705 + ], + "score": 1.0, + "content": "being too high, describe as", + "type": "text" + }, + { + "bbox": [ + 218, + 682, + 342, + 702 + ], + "score": 0.93, + "content": "\\begin{array} { r } { Y = X / \\sqrt { \\frac { 1 } { C } \\sum _ { i = 0 } ^ { C - 1 } ( X ^ { i } ) ^ { 2 } + \\epsilon } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 678, + 373, + 705 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 373, + 687, + 419, + 697 + ], + "score": 0.89, + "content": "\\epsilon = 1 e - 8", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 678, + 465, + 705 + ], + "score": 1.0, + "content": "by default,", + "type": "text" + }, + { + "bbox": [ + 466, + 687, + 476, + 697 + ], + "score": 0.8, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 678, + 494, + 705 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 495, + 687, + 504, + 697 + ], + "score": 0.68, + "content": "Y", + "type": "inline_equation" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 699, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 302, + 713 + ], + "score": 1.0, + "content": "denote the token before and after scaling layer,", + "type": "text" + }, + { + "bbox": [ + 302, + 701, + 311, + 710 + ], + "score": 0.75, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 699, + 506, + 713 + ], + "score": 1.0, + "content": "represents the embedded dimension. Note that", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "score": 1.0, + "content": "our modified normalization resembles local response normalization that was once used in AlexNet", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 46 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 742, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 310, + 752 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 111, + 61, + 502, + 142 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 111, + 61, + 502, + 142 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 111, + 61, + 502, + 142 + ], + "spans": [ + { + "bbox": [ + 111, + 61, + 502, + 142 + ], + "score": 0.954, + "type": "image", + "image_path": "5aa30c4dc12e735df83a2803ff77d51d6c674cced188b389797c086e25b8de54.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 111, + 61, + 502, + 88.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 111, + 88.0, + 502, + 115.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 111, + 115.0, + 502, + 142.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 144, + 505, + 176 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 144, + 504, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 144, + 504, + 155 + ], + "score": 1.0, + "content": "Figure 3: Grid Self-Attention across different transformer stages. We replace Standard Self-Attention", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 154, + 505, + 166 + ], + "spans": [ + { + "bbox": [ + 106, + 154, + 352, + 166 + ], + "score": 1.0, + "content": "with Grid Self-Attention when the resolution is higher than", + "type": "text" + }, + { + "bbox": [ + 352, + 154, + 387, + 165 + ], + "score": 0.9, + "content": "3 2 \\times 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 154, + 505, + 166 + ], + "score": 1.0, + "content": "and the grid size is set to be", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 165, + 185, + 176 + ], + "spans": [ + { + "bbox": [ + 106, + 165, + 140, + 176 + ], + "score": 0.88, + "content": "1 6 \\times 1 6", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 165, + 185, + 176 + ], + "score": 1.0, + "content": "by default.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 107, + 181, + 505, + 225 + ], + "lines": [], + "index": 7.5, + "bbox_fs": [ + 105, + 180, + 506, + 225 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 108, + 233, + 260, + 245 + ], + "lines": [ + { + "bbox": [ + 105, + 231, + 260, + 248 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 260, + 248 + ], + "score": 1.0, + "content": "3.4 Exploring the Training Recipe", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 248, + 505, + 303 + ], + "lines": [ + { + "bbox": [ + 106, + 248, + 505, + 261 + ], + "spans": [ + { + "bbox": [ + 106, + 248, + 505, + 261 + ], + "score": 1.0, + "content": "Data Augmentation. The transformer-based architectures are known to be highly data-hungry due", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 258, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 258, + 505, + 272 + ], + "score": 1.0, + "content": "to removing human-designed bias. Particularly in image recognition task [34], they were inferior to", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 271, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 505, + 282 + ], + "score": 1.0, + "content": "CNNs until much larger external data [68] was used for pre-training. To remove this roadblock, data", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 281, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 505, + 294 + ], + "score": 1.0, + "content": "augmentation was revealed as a blessing in [50], which showed that different types of strong data", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 291, + 413, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 413, + 305 + ], + "score": 1.0, + "content": "augmentation could lead us to data-efficient training for vision transformers.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13, + "bbox_fs": [ + 105, + 248, + 505, + 305 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 308, + 505, + 397 + ], + "lines": [ + { + "bbox": [ + 105, + 308, + 506, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 308, + 506, + 321 + ], + "score": 1.0, + "content": "We follow a similar mindset. Traditionally, training CNN-based GANs hardly refers to data augmenta-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 319, + 505, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 319, + 505, + 331 + ], + "score": 1.0, + "content": "tion. Recently, there is an interest surge in the few-shot GAN training, aiming to match state-of-the-art", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 330, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 330, + 505, + 343 + ], + "score": 1.0, + "content": "GAN results with orders of magnitude fewer real images [69, 70]. Contrary to this “commonsense\"", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 341, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 341, + 505, + 353 + ], + "score": 1.0, + "content": "in CNNs, data augmentation is found to be crucial in transformer-based architectures, even with", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 351, + 506, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 131, + 363 + ], + "score": 0.87, + "content": "100 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 131, + 351, + 506, + 366 + ], + "score": 1.0, + "content": "real images being utilized. We show that simply using differential augmentation [69] with", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 362, + 506, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 506, + 376 + ], + "score": 1.0, + "content": "three basic operators {T ranslation, Cutout, Color} leads to surprising performance improvement", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 374, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 505, + 386 + ], + "score": 1.0, + "content": "for TransGAN, while CNN-based GANs hardly benefit from it. We conduct a concrete study on the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 385, + 461, + 397 + ], + "spans": [ + { + "bbox": [ + 106, + 385, + 461, + 397 + ], + "score": 1.0, + "content": "effectiveness of augmentation for both transformer and CNNs: see details in Section 4.2", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 19.5, + "bbox_fs": [ + 105, + 308, + 506, + 397 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 401, + 505, + 445 + ], + "lines": [ + { + "bbox": [ + 105, + 401, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 505, + 414 + ], + "score": 1.0, + "content": "Relative Position Encoding. While classical transformers [49, 34] used deterministic position", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 412, + 506, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 506, + 425 + ], + "score": 1.0, + "content": "encoding or learnable position encoding, the relative position encoding [71] gains increasing popular-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 423, + 506, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 506, + 436 + ], + "score": 1.0, + "content": "ity [72, 28, 52, 73], by exploiting lags instead of absolute positions. Considering a single head of", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 433, + 186, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 186, + 447 + ], + "score": 1.0, + "content": "self-attention layer,", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 401, + 506, + 447 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 213, + 443, + 398, + 471 + ], + "lines": [ + { + "bbox": [ + 213, + 443, + 398, + 471 + ], + "spans": [ + { + "bbox": [ + 213, + 443, + 398, + 471 + ], + "score": 0.92, + "content": "A t t e n t i o n ( Q , K , V ) = s o f t m a x ( ( \\frac { Q K ^ { T } } { \\sqrt { d _ { k } } } V )", + "type": "interline_equation", + "image_path": "dbb01f6ced9089b3c2847770df83453bea906785efbac2a8e617ec65bab96ef6.jpg" + } + ] + } + ], + "index": 28.5, + "virtual_lines": [ + { + "bbox": [ + 213, + 443, + 398, + 457.0 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 213, + 457.0, + 398, + 471.0 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 473, + 505, + 542 + ], + "lines": [ + { + "bbox": [ + 104, + 471, + 507, + 489 + ], + "spans": [ + { + "bbox": [ + 104, + 471, + 132, + 489 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 474, + 223, + 486 + ], + "score": 0.91, + "content": "Q , K , V \\in \\mathbb { R } ^ { ( H \\times W ) \\times C }", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 471, + 368, + 489 + ], + "score": 1.0, + "content": "represent query, key, value matrices,", + "type": "text" + }, + { + "bbox": [ + 369, + 475, + 402, + 486 + ], + "score": 0.84, + "content": "H , W , C", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 471, + 507, + 489 + ], + "score": 1.0, + "content": "denotes the height, width,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 485, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 106, + 485, + 505, + 497 + ], + "score": 1.0, + "content": "embedded dimension of the input feature map. The difference in coordinate between each query and", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 496, + 505, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 136, + 509 + ], + "score": 1.0, + "content": "key on", + "type": "text" + }, + { + "bbox": [ + 136, + 497, + 147, + 506 + ], + "score": 0.8, + "content": "H", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 496, + 245, + 509 + ], + "score": 1.0, + "content": "axis lies in the range of", + "type": "text" + }, + { + "bbox": [ + 246, + 496, + 325, + 509 + ], + "score": 0.91, + "content": "[ - ( H - \\bar { 1 } ) , H - 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 496, + 391, + 509 + ], + "score": 1.0, + "content": ", and similar for", + "type": "text" + }, + { + "bbox": [ + 391, + 497, + 403, + 507 + ], + "score": 0.58, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 496, + 505, + 509 + ], + "score": 1.0, + "content": "axis. By simultaneously", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 507, + 505, + 520 + ], + "spans": [ + { + "bbox": [ + 106, + 508, + 176, + 520 + ], + "score": 1.0, + "content": "considering both", + "type": "text" + }, + { + "bbox": [ + 176, + 508, + 186, + 517 + ], + "score": 0.84, + "content": "H", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 508, + 204, + 520 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 205, + 507, + 217, + 517 + ], + "score": 0.71, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 508, + 505, + 520 + ], + "score": 1.0, + "content": "axis, the relative position can be represented by a parameterized matrix", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 107, + 516, + 507, + 534 + ], + "spans": [ + { + "bbox": [ + 107, + 518, + 200, + 530 + ], + "score": 0.9, + "content": "M \\in \\mathbb { R } ^ { ( 2 \\hat { H _ { - 1 } } ) \\times ( 2 W - 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 516, + 383, + 534 + ], + "score": 1.0, + "content": ". Per coordinate, the relative position encoding", + "type": "text" + }, + { + "bbox": [ + 384, + 520, + 393, + 529 + ], + "score": 0.83, + "content": "E", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 516, + 475, + 534 + ], + "score": 1.0, + "content": "is taken from matrix", + "type": "text" + }, + { + "bbox": [ + 476, + 520, + 487, + 529 + ], + "score": 0.77, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 488, + 516, + 507, + 534 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 528, + 383, + 544 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 214, + 544 + ], + "score": 1.0, + "content": "added to the attention map", + "type": "text" + }, + { + "bbox": [ + 215, + 530, + 239, + 542 + ], + "score": 0.92, + "content": "Q K ^ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 528, + 383, + 544 + ], + "score": 1.0, + "content": "as a bias term, shown as following,", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 32.5, + "bbox_fs": [ + 104, + 471, + 507, + 544 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 199, + 546, + 412, + 574 + ], + "lines": [ + { + "bbox": [ + 199, + 546, + 412, + 574 + ], + "spans": [ + { + "bbox": [ + 199, + 546, + 412, + 574 + ], + "score": 0.93, + "content": "A t t e n t i o n ( Q , K , V ) = s o f t m a x ( ( ( \\frac { Q K ^ { T } } { \\sqrt { d _ { k } } } + E ) V )", + "type": "interline_equation", + "image_path": "d642168023319f11cbc80f8d9df34a9ce6f6ffd07f0e3008d315f28bd6603d49.jpg" + } + ] + } + ], + "index": 36.5, + "virtual_lines": [ + { + "bbox": [ + 199, + 546, + 412, + 560.0 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 199, + 560.0, + 412, + 574.0 + ], + "spans": [], + "index": 37 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 577, + 505, + 633 + ], + "lines": [ + { + "bbox": [ + 106, + 577, + 504, + 591 + ], + "spans": [ + { + "bbox": [ + 106, + 577, + 504, + 591 + ], + "score": 1.0, + "content": "Compared to its absolute counterpart, relative position encoding learns a stronger “relationship\"", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 586, + 505, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 505, + 603 + ], + "score": 1.0, + "content": "between local contents, bringing important performance gains in large-scale cases and enjoying", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 598, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 505, + 613 + ], + "score": 1.0, + "content": "widespread use ever since. We also observe it to consistently improve TransGAN, especially on", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 609, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 505, + 624 + ], + "score": 1.0, + "content": "higher-resolution datasets. We hence apply it on top of the learnable absolute positional encoding for", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 620, + 256, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 620, + 256, + 635 + ], + "score": 1.0, + "content": "both the generator and discriminator.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40, + "bbox_fs": [ + 105, + 577, + 505, + 635 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 637, + 505, + 723 + ], + "lines": [ + { + "bbox": [ + 104, + 635, + 506, + 652 + ], + "spans": [ + { + "bbox": [ + 104, + 635, + 506, + 652 + ], + "score": 1.0, + "content": "Modified Normalization. Normalization layers are known to help stabilize the deep learning training", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 648, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 106, + 648, + 505, + 661 + ], + "score": 1.0, + "content": "of deep neural networks, sometimes remarkably. While both the original transformer [49] and its", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 659, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 505, + 672 + ], + "score": 1.0, + "content": "variants [52, 54] by default use the layer normalization, we follow previous works [75, 16] and", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 104, + 669, + 506, + 684 + ], + "spans": [ + { + "bbox": [ + 104, + 669, + 506, + 684 + ], + "score": 1.0, + "content": "replace it with a token-wise scaling layer to prevent the magnitudes in transformer blocks from", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 101, + 678, + 504, + 705 + ], + "spans": [ + { + "bbox": [ + 101, + 678, + 218, + 705 + ], + "score": 1.0, + "content": "being too high, describe as", + "type": "text" + }, + { + "bbox": [ + 218, + 682, + 342, + 702 + ], + "score": 0.93, + "content": "\\begin{array} { r } { Y = X / \\sqrt { \\frac { 1 } { C } \\sum _ { i = 0 } ^ { C - 1 } ( X ^ { i } ) ^ { 2 } + \\epsilon } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 678, + 373, + 705 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 373, + 687, + 419, + 697 + ], + "score": 0.89, + "content": "\\epsilon = 1 e - 8", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 678, + 465, + 705 + ], + "score": 1.0, + "content": "by default,", + "type": "text" + }, + { + "bbox": [ + 466, + 687, + 476, + 697 + ], + "score": 0.8, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 678, + 494, + 705 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 495, + 687, + 504, + 697 + ], + "score": 0.68, + "content": "Y", + "type": "inline_equation" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 699, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 302, + 713 + ], + "score": 1.0, + "content": "denote the token before and after scaling layer,", + "type": "text" + }, + { + "bbox": [ + 302, + 701, + 311, + 710 + ], + "score": 0.75, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 699, + 506, + 713 + ], + "score": 1.0, + "content": "represents the embedded dimension. Note that", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "score": 1.0, + "content": "our modified normalization resembles local response normalization that was once used in AlexNet", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 228, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 505, + 243 + ], + "score": 1.0, + "content": "[75]. Unlike other “modern\" normalization layers [76, 77, 78] that need affine parameters for both", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 241, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 106, + 241, + 505, + 253 + ], + "score": 1.0, + "content": "mean and variances, we find that a simple re-scaling without learnable parameters suffices to stabilize", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 107, + 252, + 443, + 264 + ], + "spans": [ + { + "bbox": [ + 107, + 252, + 443, + 264 + ], + "score": 1.0, + "content": "TransGAN training – in fact, it makes TransGAN train better and improves the FID.", + "type": "text", + "cross_page": true + } + ], + "index": 8 + } + ], + "index": 46, + "bbox_fs": [ + 101, + 635, + 506, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 112, + 95, + 497, + 222 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 59, + 506, + 92 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 59, + 506, + 72 + ], + "spans": [ + { + "bbox": [ + 105, + 59, + 455, + 72 + ], + "score": 1.0, + "content": "Table 1: Unconditional image generation results on CIFAR-10, STl-10, and CelebA", + "type": "text" + }, + { + "bbox": [ + 455, + 60, + 503, + 70 + ], + "score": 0.81, + "content": "( 1 2 8 \\times 1 2 8 )", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 59, + 506, + 72 + ], + "score": 0.0, + "content": "", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 104, + 69, + 506, + 82 + ], + "spans": [ + { + "bbox": [ + 104, + 69, + 506, + 82 + ], + "score": 1.0, + "content": "dataset. We train the models with their official code if the results are unavailable, denoted as “*”,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 81, + 264, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 264, + 94 + ], + "score": 1.0, + "content": "others are all reported from references.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "table_body", + "bbox": [ + 112, + 95, + 497, + 222 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 95, + 497, + 222 + ], + "spans": [ + { + "bbox": [ + 112, + 95, + 497, + 222 + ], + "score": 0.982, + "html": "
MethodsCIFAR-10STL-10CelebA
IS↑FID↓IS↑FID↓FID↓
WGAN-GP [1]6.49 ± 0.0939.681=
SN-GAN [48]8.22 ± 0.059.16 ± 0.1240.1
AutoGAN[18]8.55 ±0.1012.429.16 ± 0.1231.01
AdversarialNAS-GAN[18]8.74 ±0.0710.879.63 ± 0.1926.98=
Progressive-GAN[16]8.80 ± 0.0515.52=7.30
COCO-GAN [74]-=-5.74
StyleGAN-V2 [69]9.1811.0710.21* ± 0.1420.84*5.59*
StyleGAN-V2 + DiffAug. [69]9.409.8910.31*± 0.1219.15*5.40*
TransGAN9.02 ± 0.129.2610.43 ± 0.1618.285.28
", + "type": "table", + "image_path": "3736709c036ee2e90a0a961b83e5eadc3a8c00fa0505f62df932fa0420a35521.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 112, + 95, + 497, + 137.33333333333334 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 112, + 137.33333333333334, + 497, + 179.66666666666669 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 112, + 179.66666666666669, + 497, + 222.00000000000003 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 108, + 230, + 503, + 263 + ], + "lines": [ + { + "bbox": [ + 105, + 228, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 505, + 243 + ], + "score": 1.0, + "content": "[75]. Unlike other “modern\" normalization layers [76, 77, 78] that need affine parameters for both", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 241, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 106, + 241, + 505, + 253 + ], + "score": 1.0, + "content": "mean and variances, we find that a simple re-scaling without learnable parameters suffices to stabilize", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 107, + 252, + 443, + 264 + ], + "spans": [ + { + "bbox": [ + 107, + 252, + 443, + 264 + ], + "score": 1.0, + "content": "TransGAN training – in fact, it makes TransGAN train better and improves the FID.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "title", + "bbox": [ + 106, + 268, + 192, + 282 + ], + "lines": [ + { + "bbox": [ + 104, + 266, + 193, + 286 + ], + "spans": [ + { + "bbox": [ + 104, + 266, + 193, + 286 + ], + "score": 1.0, + "content": "4 Experiments", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 106, + 285, + 505, + 384 + ], + "lines": [ + { + "bbox": [ + 106, + 286, + 506, + 297 + ], + "spans": [ + { + "bbox": [ + 106, + 286, + 506, + 297 + ], + "score": 1.0, + "content": "Datasets We start by evaluating our methods on three common testbeds: CIFAR-10 [79], STL-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 296, + 506, + 309 + ], + "spans": [ + { + "bbox": [ + 105, + 296, + 395, + 309 + ], + "score": 1.0, + "content": "10 [80], and CelebA [81] dataset. The CIFAR-10 dataset consists of", + "type": "text" + }, + { + "bbox": [ + 396, + 297, + 448, + 307 + ], + "score": 0.86, + "content": "6 0 \\mathbf { k } \\ 3 2 \\times 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 296, + 506, + 309 + ], + "score": 1.0, + "content": "images, with", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 307, + 505, + 320 + ], + "spans": [ + { + "bbox": [ + 106, + 308, + 124, + 318 + ], + "score": 0.54, + "content": "5 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 307, + 177, + 320 + ], + "score": 1.0, + "content": "training and", + "type": "text" + }, + { + "bbox": [ + 178, + 308, + 194, + 318 + ], + "score": 0.3, + "content": "1 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 307, + 487, + 320 + ], + "score": 1.0, + "content": "testing images, respectively. We follow the standard setting to use the", + "type": "text" + }, + { + "bbox": [ + 487, + 308, + 505, + 318 + ], + "score": 0.46, + "content": "5 0 \\mathrm { k }", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 318, + 506, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 506, + 331 + ], + "score": 1.0, + "content": "training images without labels. For the STL-10 dataset, we use both the 5k training images and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 328, + 506, + 342 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 128, + 339 + ], + "score": 0.27, + "content": "1 0 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 128, + 328, + 293, + 342 + ], + "score": 1.0, + "content": "unlabeled images, and all are resized to", + "type": "text" + }, + { + "bbox": [ + 293, + 329, + 327, + 340 + ], + "score": 0.9, + "content": "4 8 \\times 4 8", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 328, + 506, + 342 + ], + "score": 1.0, + "content": "resolution. For the CelebA dataset, we use", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 339, + 506, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 340, + 128, + 351 + ], + "score": 0.29, + "content": "2 0 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 128, + 339, + 418, + 352 + ], + "score": 1.0, + "content": "unlabeled face images (aligned and cropped version), with each image at", + "type": "text" + }, + { + "bbox": [ + 418, + 340, + 461, + 351 + ], + "score": 0.89, + "content": "1 2 8 \\times 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 339, + 506, + 352 + ], + "score": 1.0, + "content": "resolution.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 351, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 505, + 364 + ], + "score": 1.0, + "content": "We further consider the CelebA-HQ and LSUN Church datasets to scale up TransGAN to higher", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 360, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 104, + 360, + 275, + 376 + ], + "score": 1.0, + "content": "resolution image generation tasks. We use", + "type": "text" + }, + { + "bbox": [ + 275, + 362, + 291, + 372 + ], + "score": 0.43, + "content": "3 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 360, + 452, + 376 + ], + "score": 1.0, + "content": "images for CelebA-HQ [16] dataset and", + "type": "text" + }, + { + "bbox": [ + 453, + 362, + 474, + 372 + ], + "score": 0.36, + "content": "1 2 5 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 360, + 505, + 376 + ], + "score": 1.0, + "content": "images", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 372, + 343, + 384 + ], + "spans": [ + { + "bbox": [ + 106, + 372, + 255, + 384 + ], + "score": 1.0, + "content": "for LSUN Church dataset [82], all at", + "type": "text" + }, + { + "bbox": [ + 255, + 373, + 299, + 383 + ], + "score": 0.9, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 372, + 343, + 384 + ], + "score": 1.0, + "content": "resolution.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 106, + 389, + 505, + 510 + ], + "lines": [ + { + "bbox": [ + 106, + 389, + 506, + 402 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 506, + 402 + ], + "score": 1.0, + "content": "Implementation We follow the setting of WGAN [46], and use the WGAN-GP loss [1]. We adopt a", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 400, + 506, + 412 + ], + "spans": [ + { + "bbox": [ + 105, + 400, + 171, + 412 + ], + "score": 1.0, + "content": "learning rate of", + "type": "text" + }, + { + "bbox": [ + 172, + 400, + 200, + 411 + ], + "score": 0.87, + "content": "1 e - 4", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 400, + 456, + 412 + ], + "score": 1.0, + "content": "for both generator and discriminator, an Adam optimizer with", + "type": "text" + }, + { + "bbox": [ + 456, + 400, + 487, + 411 + ], + "score": 0.92, + "content": "\\beta _ { 1 } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 400, + 506, + 412 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 107, + 411, + 505, + 424 + ], + "spans": [ + { + "bbox": [ + 107, + 411, + 149, + 423 + ], + "score": 0.9, + "content": "\\beta _ { 2 } = 0 . 9 9", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 411, + 505, + 424 + ], + "score": 1.0, + "content": ", exponential moving average weights for generator, and a batch size of 128 for generator", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 421, + 506, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 506, + 436 + ], + "score": 1.0, + "content": "and 64 for discriminator, for all experiments. We choose DiffAug. [69] as basic augmentation strategy", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 432, + 506, + 446 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 506, + 446 + ], + "score": 1.0, + "content": "during the training process if not specially mentioned, and apply it to our competitors for a fair", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 443, + 506, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 506, + 456 + ], + "score": 1.0, + "content": "comparison. Other popular augmentation strategies ([70, 10]) are not discussed here since it is beyond", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 455, + 506, + 467 + ], + "spans": [ + { + "bbox": [ + 106, + 455, + 506, + 467 + ], + "score": 1.0, + "content": "the scope of this work. We use common evaluation metrics Inception Score (IS) [15] and Frechet", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 505, + 478 + ], + "score": 1.0, + "content": "Inception Distance (FID) [83], both are measured by 50K samples with their official Tensorflow", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 474, + 506, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 506, + 489 + ], + "score": 1.0, + "content": "implementations 12. All experiments are set with 16 V100 GPUs, using PyTorch 1.7.0. We include", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 487, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 505, + 500 + ], + "score": 1.0, + "content": "detailed training cost for each dataset in Appendix D. We focus on the unconditional image generation", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 498, + 194, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 498, + 194, + 513 + ], + "score": 1.0, + "content": "setting for simplicity.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 24 + }, + { + "type": "title", + "bbox": [ + 107, + 518, + 301, + 529 + ], + "lines": [ + { + "bbox": [ + 104, + 516, + 303, + 532 + ], + "spans": [ + { + "bbox": [ + 104, + 516, + 303, + 532 + ], + "score": 1.0, + "content": "4.1 Comparison with State-of-the-art GANs", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 106, + 533, + 506, + 610 + ], + "lines": [ + { + "bbox": [ + 106, + 533, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 533, + 505, + 546 + ], + "score": 1.0, + "content": "CIFAR-10. We compare TransGAN with recently published results by unconditional CNN-based", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 543, + 506, + 557 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 506, + 557 + ], + "score": 1.0, + "content": "GANs on the CIFAR-10 dataset, shown in Table 1. Note that some promising conditional GANs [21,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 554, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 506, + 567 + ], + "score": 1.0, + "content": "8] are not included, due to the different settings. As shown in Table 1, TransGAN surpasses the", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 566, + 506, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 566, + 506, + 578 + ], + "score": 1.0, + "content": "strong model of Progressive GAN [16], and many other latest competitors such as SN-GAN [48],", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 577, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 506, + 589 + ], + "score": 1.0, + "content": "AutoGAN [18], and AdversarialNAS-GAN [19], in terms of inception score (IS). It is only next to", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 588, + 506, + 600 + ], + "spans": [ + { + "bbox": [ + 106, + 588, + 506, + 600 + ], + "score": 1.0, + "content": "the huge and heavily engineered StyleGAN-v2 [40]. Once we look at the FID results, TransGAN is", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 599, + 495, + 611 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 495, + 611 + ], + "score": 1.0, + "content": "even found to outperform StyleGAN-v2 [40] with both applied the same data augmentation [69].", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 106, + 614, + 505, + 691 + ], + "lines": [ + { + "bbox": [ + 106, + 614, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 505, + 627 + ], + "score": 1.0, + "content": "STL-10. We then apply TransGAN on another popular benchmark STL-10, which is larger in scale", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 626, + 505, + 638 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 505, + 638 + ], + "score": 1.0, + "content": "(105k) and higher in resolution (48x48). We compare TransGAN with both the automatic searched", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 637, + 506, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 506, + 649 + ], + "score": 1.0, + "content": "and hand-crafted CNN-based GANs, shown in Table 1. Different from the results on CIFAR-10, we", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 648, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 106, + 648, + 505, + 659 + ], + "score": 1.0, + "content": "find that TransGAN outperforms all current CNN-based GAN models, and sets new state-of-the-art", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 658, + 504, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 489, + 671 + ], + "score": 1.0, + "content": "results in terms of both IS and FID score. This is thanks to the fact that the STL-10 dataset size is", + "type": "text" + }, + { + "bbox": [ + 489, + 659, + 504, + 669 + ], + "score": 0.85, + "content": "2 \\times", + "type": "inline_equation" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 670, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 670, + 505, + 682 + ], + "score": 1.0, + "content": "larger than CIFAR-10, suggesting that transformer-based architectures benefit much more notably", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 680, + 246, + 693 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 246, + 693 + ], + "score": 1.0, + "content": "from larger-scale data than CNNs.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 41 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 118, + 701, + 370, + 722 + ], + "lines": [ + { + "bbox": [ + 119, + 699, + 371, + 713 + ], + "spans": [ + { + "bbox": [ + 119, + 699, + 371, + 713 + ], + "score": 1.0, + "content": "1https://github.com/openai/improved-gan/tree/master/inception_score", + "type": "text" + } + ] + }, + { + "bbox": [ + 118, + 710, + 255, + 723 + ], + "spans": [ + { + "bbox": [ + 118, + 710, + 255, + 723 + ], + "score": 1.0, + "content": "2https://github.com/bioinf-jku/TTUR", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 741, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 112, + 95, + 497, + 222 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 59, + 506, + 92 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 59, + 506, + 72 + ], + "spans": [ + { + "bbox": [ + 105, + 59, + 455, + 72 + ], + "score": 1.0, + "content": "Table 1: Unconditional image generation results on CIFAR-10, STl-10, and CelebA", + "type": "text" + }, + { + "bbox": [ + 455, + 60, + 503, + 70 + ], + "score": 0.81, + "content": "( 1 2 8 \\times 1 2 8 )", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 59, + 506, + 72 + ], + "score": 0.0, + "content": "", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 104, + 69, + 506, + 82 + ], + "spans": [ + { + "bbox": [ + 104, + 69, + 506, + 82 + ], + "score": 1.0, + "content": "dataset. We train the models with their official code if the results are unavailable, denoted as “*”,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 81, + 264, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 264, + 94 + ], + "score": 1.0, + "content": "others are all reported from references.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "table_body", + "bbox": [ + 112, + 95, + 497, + 222 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 95, + 497, + 222 + ], + "spans": [ + { + "bbox": [ + 112, + 95, + 497, + 222 + ], + "score": 0.982, + "html": "
MethodsCIFAR-10STL-10CelebA
IS↑FID↓IS↑FID↓FID↓
WGAN-GP [1]6.49 ± 0.0939.681=
SN-GAN [48]8.22 ± 0.059.16 ± 0.1240.1
AutoGAN[18]8.55 ±0.1012.429.16 ± 0.1231.01
AdversarialNAS-GAN[18]8.74 ±0.0710.879.63 ± 0.1926.98=
Progressive-GAN[16]8.80 ± 0.0515.52=7.30
COCO-GAN [74]-=-5.74
StyleGAN-V2 [69]9.1811.0710.21* ± 0.1420.84*5.59*
StyleGAN-V2 + DiffAug. [69]9.409.8910.31*± 0.1219.15*5.40*
TransGAN9.02 ± 0.129.2610.43 ± 0.1618.285.28
", + "type": "table", + "image_path": "3736709c036ee2e90a0a961b83e5eadc3a8c00fa0505f62df932fa0420a35521.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 112, + 95, + 497, + 137.33333333333334 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 112, + 137.33333333333334, + 497, + 179.66666666666669 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 112, + 179.66666666666669, + 497, + 222.00000000000003 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 108, + 230, + 503, + 263 + ], + "lines": [], + "index": 7, + "bbox_fs": [ + 105, + 228, + 505, + 264 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 106, + 268, + 192, + 282 + ], + "lines": [ + { + "bbox": [ + 104, + 266, + 193, + 286 + ], + "spans": [ + { + "bbox": [ + 104, + 266, + 193, + 286 + ], + "score": 1.0, + "content": "4 Experiments", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 106, + 285, + 505, + 384 + ], + "lines": [ + { + "bbox": [ + 106, + 286, + 506, + 297 + ], + "spans": [ + { + "bbox": [ + 106, + 286, + 506, + 297 + ], + "score": 1.0, + "content": "Datasets We start by evaluating our methods on three common testbeds: CIFAR-10 [79], STL-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 296, + 506, + 309 + ], + "spans": [ + { + "bbox": [ + 105, + 296, + 395, + 309 + ], + "score": 1.0, + "content": "10 [80], and CelebA [81] dataset. The CIFAR-10 dataset consists of", + "type": "text" + }, + { + "bbox": [ + 396, + 297, + 448, + 307 + ], + "score": 0.86, + "content": "6 0 \\mathbf { k } \\ 3 2 \\times 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 296, + 506, + 309 + ], + "score": 1.0, + "content": "images, with", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 307, + 505, + 320 + ], + "spans": [ + { + "bbox": [ + 106, + 308, + 124, + 318 + ], + "score": 0.54, + "content": "5 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 307, + 177, + 320 + ], + "score": 1.0, + "content": "training and", + "type": "text" + }, + { + "bbox": [ + 178, + 308, + 194, + 318 + ], + "score": 0.3, + "content": "1 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 307, + 487, + 320 + ], + "score": 1.0, + "content": "testing images, respectively. We follow the standard setting to use the", + "type": "text" + }, + { + "bbox": [ + 487, + 308, + 505, + 318 + ], + "score": 0.46, + "content": "5 0 \\mathrm { k }", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 318, + 506, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 506, + 331 + ], + "score": 1.0, + "content": "training images without labels. For the STL-10 dataset, we use both the 5k training images and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 328, + 506, + 342 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 128, + 339 + ], + "score": 0.27, + "content": "1 0 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 128, + 328, + 293, + 342 + ], + "score": 1.0, + "content": "unlabeled images, and all are resized to", + "type": "text" + }, + { + "bbox": [ + 293, + 329, + 327, + 340 + ], + "score": 0.9, + "content": "4 8 \\times 4 8", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 328, + 506, + 342 + ], + "score": 1.0, + "content": "resolution. For the CelebA dataset, we use", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 339, + 506, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 340, + 128, + 351 + ], + "score": 0.29, + "content": "2 0 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 128, + 339, + 418, + 352 + ], + "score": 1.0, + "content": "unlabeled face images (aligned and cropped version), with each image at", + "type": "text" + }, + { + "bbox": [ + 418, + 340, + 461, + 351 + ], + "score": 0.89, + "content": "1 2 8 \\times 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 339, + 506, + 352 + ], + "score": 1.0, + "content": "resolution.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 351, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 505, + 364 + ], + "score": 1.0, + "content": "We further consider the CelebA-HQ and LSUN Church datasets to scale up TransGAN to higher", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 360, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 104, + 360, + 275, + 376 + ], + "score": 1.0, + "content": "resolution image generation tasks. We use", + "type": "text" + }, + { + "bbox": [ + 275, + 362, + 291, + 372 + ], + "score": 0.43, + "content": "3 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 360, + 452, + 376 + ], + "score": 1.0, + "content": "images for CelebA-HQ [16] dataset and", + "type": "text" + }, + { + "bbox": [ + 453, + 362, + 474, + 372 + ], + "score": 0.36, + "content": "1 2 5 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 360, + 505, + 376 + ], + "score": 1.0, + "content": "images", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 372, + 343, + 384 + ], + "spans": [ + { + "bbox": [ + 106, + 372, + 255, + 384 + ], + "score": 1.0, + "content": "for LSUN Church dataset [82], all at", + "type": "text" + }, + { + "bbox": [ + 255, + 373, + 299, + 383 + ], + "score": 0.9, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 372, + 343, + 384 + ], + "score": 1.0, + "content": "resolution.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 14, + "bbox_fs": [ + 104, + 286, + 506, + 384 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 389, + 505, + 510 + ], + "lines": [ + { + "bbox": [ + 106, + 389, + 506, + 402 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 506, + 402 + ], + "score": 1.0, + "content": "Implementation We follow the setting of WGAN [46], and use the WGAN-GP loss [1]. We adopt a", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 400, + 506, + 412 + ], + "spans": [ + { + "bbox": [ + 105, + 400, + 171, + 412 + ], + "score": 1.0, + "content": "learning rate of", + "type": "text" + }, + { + "bbox": [ + 172, + 400, + 200, + 411 + ], + "score": 0.87, + "content": "1 e - 4", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 400, + 456, + 412 + ], + "score": 1.0, + "content": "for both generator and discriminator, an Adam optimizer with", + "type": "text" + }, + { + "bbox": [ + 456, + 400, + 487, + 411 + ], + "score": 0.92, + "content": "\\beta _ { 1 } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 400, + 506, + 412 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 107, + 411, + 505, + 424 + ], + "spans": [ + { + "bbox": [ + 107, + 411, + 149, + 423 + ], + "score": 0.9, + "content": "\\beta _ { 2 } = 0 . 9 9", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 411, + 505, + 424 + ], + "score": 1.0, + "content": ", exponential moving average weights for generator, and a batch size of 128 for generator", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 421, + 506, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 506, + 436 + ], + "score": 1.0, + "content": "and 64 for discriminator, for all experiments. We choose DiffAug. [69] as basic augmentation strategy", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 432, + 506, + 446 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 506, + 446 + ], + "score": 1.0, + "content": "during the training process if not specially mentioned, and apply it to our competitors for a fair", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 443, + 506, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 506, + 456 + ], + "score": 1.0, + "content": "comparison. Other popular augmentation strategies ([70, 10]) are not discussed here since it is beyond", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 455, + 506, + 467 + ], + "spans": [ + { + "bbox": [ + 106, + 455, + 506, + 467 + ], + "score": 1.0, + "content": "the scope of this work. We use common evaluation metrics Inception Score (IS) [15] and Frechet", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 505, + 478 + ], + "score": 1.0, + "content": "Inception Distance (FID) [83], both are measured by 50K samples with their official Tensorflow", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 474, + 506, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 506, + 489 + ], + "score": 1.0, + "content": "implementations 12. All experiments are set with 16 V100 GPUs, using PyTorch 1.7.0. We include", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 487, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 505, + 500 + ], + "score": 1.0, + "content": "detailed training cost for each dataset in Appendix D. We focus on the unconditional image generation", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 498, + 194, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 498, + 194, + 513 + ], + "score": 1.0, + "content": "setting for simplicity.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 389, + 506, + 513 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 518, + 301, + 529 + ], + "lines": [ + { + "bbox": [ + 104, + 516, + 303, + 532 + ], + "spans": [ + { + "bbox": [ + 104, + 516, + 303, + 532 + ], + "score": 1.0, + "content": "4.1 Comparison with State-of-the-art GANs", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 106, + 533, + 506, + 610 + ], + "lines": [ + { + "bbox": [ + 106, + 533, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 533, + 505, + 546 + ], + "score": 1.0, + "content": "CIFAR-10. We compare TransGAN with recently published results by unconditional CNN-based", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 543, + 506, + 557 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 506, + 557 + ], + "score": 1.0, + "content": "GANs on the CIFAR-10 dataset, shown in Table 1. Note that some promising conditional GANs [21,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 554, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 506, + 567 + ], + "score": 1.0, + "content": "8] are not included, due to the different settings. As shown in Table 1, TransGAN surpasses the", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 566, + 506, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 566, + 506, + 578 + ], + "score": 1.0, + "content": "strong model of Progressive GAN [16], and many other latest competitors such as SN-GAN [48],", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 577, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 506, + 589 + ], + "score": 1.0, + "content": "AutoGAN [18], and AdversarialNAS-GAN [19], in terms of inception score (IS). It is only next to", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 588, + 506, + 600 + ], + "spans": [ + { + "bbox": [ + 106, + 588, + 506, + 600 + ], + "score": 1.0, + "content": "the huge and heavily engineered StyleGAN-v2 [40]. Once we look at the FID results, TransGAN is", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 599, + 495, + 611 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 495, + 611 + ], + "score": 1.0, + "content": "even found to outperform StyleGAN-v2 [40] with both applied the same data augmentation [69].", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 533, + 506, + 611 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 614, + 505, + 691 + ], + "lines": [ + { + "bbox": [ + 106, + 614, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 614, + 505, + 627 + ], + "score": 1.0, + "content": "STL-10. We then apply TransGAN on another popular benchmark STL-10, which is larger in scale", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 626, + 505, + 638 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 505, + 638 + ], + "score": 1.0, + "content": "(105k) and higher in resolution (48x48). We compare TransGAN with both the automatic searched", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 637, + 506, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 506, + 649 + ], + "score": 1.0, + "content": "and hand-crafted CNN-based GANs, shown in Table 1. Different from the results on CIFAR-10, we", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 648, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 106, + 648, + 505, + 659 + ], + "score": 1.0, + "content": "find that TransGAN outperforms all current CNN-based GAN models, and sets new state-of-the-art", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 658, + 504, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 489, + 671 + ], + "score": 1.0, + "content": "results in terms of both IS and FID score. This is thanks to the fact that the STL-10 dataset size is", + "type": "text" + }, + { + "bbox": [ + 489, + 659, + 504, + 669 + ], + "score": 0.85, + "content": "2 \\times", + "type": "inline_equation" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 670, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 670, + 505, + 682 + ], + "score": 1.0, + "content": "larger than CIFAR-10, suggesting that transformer-based architectures benefit much more notably", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 680, + 246, + 693 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 246, + 693 + ], + "score": 1.0, + "content": "from larger-scale data than CNNs.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 41, + "bbox_fs": [ + 105, + 614, + 506, + 693 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 109, + 65, + 501, + 191 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 65, + 501, + 191 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 65, + 501, + 191 + ], + "spans": [ + { + "bbox": [ + 109, + 65, + 501, + 191 + ], + "score": 0.968, + "type": "image", + "image_path": "6e96e44a3e68b78d7018919b2cfbce5508a09fd739e9d26cf4824f8d0c72caee.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 109, + 65, + 501, + 107.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 109, + 107.0, + 501, + 149.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 109, + 149.0, + 501, + 191.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 194, + 504, + 217 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 194, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 106, + 194, + 505, + 207 + ], + "score": 1.0, + "content": "Figure 4: Representative visual results produced by TransGAN on different datasets, as resolution", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 204, + 452, + 218 + ], + "spans": [ + { + "bbox": [ + 105, + 204, + 155, + 218 + ], + "score": 1.0, + "content": "grows from", + "type": "text" + }, + { + "bbox": [ + 155, + 205, + 189, + 216 + ], + "score": 0.89, + "content": "3 2 \\times 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 204, + 200, + 218 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 200, + 205, + 244, + 216 + ], + "score": 0.89, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 204, + 452, + 218 + ], + "score": 1.0, + "content": ". More visual examples are included in Appendix F.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "table", + "bbox": [ + 119, + 252, + 492, + 310 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 228, + 502, + 251 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 227, + 504, + 241 + ], + "spans": [ + { + "bbox": [ + 105, + 227, + 504, + 241 + ], + "score": 1.0, + "content": "Table 2: The effectiveness of Data Augmentation on both CNN-based GANs and TransGAN. We use", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 239, + 303, + 252 + ], + "spans": [ + { + "bbox": [ + 106, + 239, + 303, + 252 + ], + "score": 1.0, + "content": "the full CIFAR-10 training set and DiffAug [69].", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "table_body", + "bbox": [ + 119, + 252, + 492, + 310 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 119, + 252, + 492, + 310 + ], + "spans": [ + { + "bbox": [ + 119, + 252, + 492, + 310 + ], + "score": 0.977, + "html": "
MethodsWGAN-GPAutoGANStyleGAN-V2TransGAN
IS个FID←IS个FID←IS个FID←IS个FID↓
Original6.4939.688.5512.429.1811.078.3622.53
+ DiffAug [69]6.2937.148.6012.729.409.899.029.26
", + "type": "table", + "image_path": "f1c4e2b09d5e2f8e5edf0927060448a8a0cc45c22cecb15cb148ce6fe11b716d.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 119, + 252, + 492, + 271.3333333333333 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 119, + 271.3333333333333, + 492, + 290.66666666666663 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 119, + 290.66666666666663, + 492, + 309.99999999999994 + ], + "spans": [], + "index": 9 + } + ] + } + ], + "index": 6.75 + }, + { + "type": "table", + "bbox": [ + 130, + 346, + 479, + 460 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 109, + 322, + 505, + 345 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 322, + 506, + 335 + ], + "spans": [ + { + "bbox": [ + 106, + 322, + 463, + 335 + ], + "score": 1.0, + "content": "Table 3: The ablation study of proposed techniques in three common dataset CelebA", + "type": "text" + }, + { + "bbox": [ + 464, + 322, + 501, + 334 + ], + "score": 0.84, + "content": "( 6 4 \\times 6 4 )", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 322, + 506, + 335 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 107, + 333, + 485, + 346 + ], + "spans": [ + { + "bbox": [ + 107, + 333, + 139, + 346 + ], + "score": 1.0, + "content": "CelebA", + "type": "text" + }, + { + "bbox": [ + 139, + 333, + 184, + 344 + ], + "score": 0.83, + "content": "( 1 2 8 \\times 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 333, + 264, + 346 + ], + "score": 1.0, + "content": ", and LSUN Church", + "type": "text" + }, + { + "bbox": [ + 264, + 333, + 312, + 344 + ], + "score": 0.83, + "content": "( 2 5 6 \\times 2 5 6 )", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 333, + 485, + 346 + ], + "score": 1.0, + "content": "). “OOM” represents out-of-momery issue.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + }, + { + "type": "table_body", + "bbox": [ + 130, + 346, + 479, + 460 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 130, + 346, + 479, + 460 + ], + "spans": [ + { + "bbox": [ + 130, + 346, + 479, + 460 + ], + "score": 0.983, + "html": "
Training ConfigurationCelebA (64x64)CelebA (128x128)LSUN Church (256x256)
(A). Standard Self-Attention8.9200M0OM
(B). Nystrom Self-Attention [64]13.4717.4239.92
(C). Axis Self-Attention [67]12.3913.9529.30
(D). Grid Self-Attention9.8910.5820.39
+Multi-scale Discriminator9.288.0315.29
+Modified Normalization7.057.1313.27
+Relative Position Encoding6.146.3211.93
(E). Converge5.015.288.94
", + "type": "table", + "image_path": "25c0032de9117db6ffe66c1bcb8ba86cc9bfa5f079bdbf7a78a3daa9c1d02d03.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 130, + 346, + 479, + 384.0 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 130, + 384.0, + 479, + 422.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 130, + 422.0, + 479, + 460.0 + ], + "spans": [], + "index": 14 + } + ] + } + ], + "index": 11.75 + }, + { + "type": "text", + "bbox": [ + 106, + 470, + 506, + 525 + ], + "lines": [ + { + "bbox": [ + 106, + 470, + 505, + 483 + ], + "spans": [ + { + "bbox": [ + 106, + 470, + 462, + 483 + ], + "score": 1.0, + "content": "CelebA (128x128). We continue to examine another common benchmark: CelebA dataset", + "type": "text" + }, + { + "bbox": [ + 462, + 470, + 505, + 481 + ], + "score": 0.87, + "content": "( 1 2 8 \\times 1 2 8", + "type": "inline_equation" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 480, + 506, + 494 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 506, + 494 + ], + "score": 1.0, + "content": "resolution). As shown in Table 1, TransGAN largely outperforms Progressive-GAN [16] and COCO-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 491, + 505, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 491, + 505, + 505 + ], + "score": 1.0, + "content": "GAN [74], and is slightly better than the strongest competitor StyleGAN-v2 [40], by reaching a FID", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 502, + 505, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 414, + 516 + ], + "score": 1.0, + "content": "score of 5.28. Visual examples generated on CIFAR-10, STL-10, and CelebA", + "type": "text" + }, + { + "bbox": [ + 414, + 504, + 461, + 514 + ], + "score": 0.87, + "content": "( 1 2 8 \\times 1 2 8 )", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 502, + 505, + 516 + ], + "score": 1.0, + "content": "are shown", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 513, + 397, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 397, + 527 + ], + "score": 1.0, + "content": "in Figure 4, from which we observe pleasing visual details and diversity.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17 + }, + { + "type": "title", + "bbox": [ + 107, + 536, + 269, + 548 + ], + "lines": [ + { + "bbox": [ + 105, + 535, + 269, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 269, + 550 + ], + "score": 1.0, + "content": "4.2 Scaling Up to Higher-Resolution", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 552, + 506, + 630 + ], + "lines": [ + { + "bbox": [ + 105, + 551, + 507, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 319, + 565 + ], + "score": 1.0, + "content": "We further scale up TransGAN to higher-resolution", + "type": "text" + }, + { + "bbox": [ + 319, + 552, + 366, + 563 + ], + "score": 0.86, + "content": "( 2 5 6 \\times 2 5 6 )", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 551, + 507, + 565 + ], + "score": 1.0, + "content": "generation, including on CelebA-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 104, + 561, + 506, + 577 + ], + "spans": [ + { + "bbox": [ + 104, + 561, + 506, + 577 + ], + "score": 1.0, + "content": "HQ [16] and LSUN Church [82]. These high-resolution datasets are significantly more challenging", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 573, + 506, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 506, + 586 + ], + "score": 1.0, + "content": "due to their much richer and detailed low-level texture as well as the global composition. Thanks", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 585, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 506, + 599 + ], + "score": 1.0, + "content": "to the proposed multi-scale discriminator, TransGAN produces pleasing visual results, reaching", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 596, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 363, + 608 + ], + "score": 1.0, + "content": "competitive quantitative results with 10.28 FID on CelebA-HQ", + "type": "text" + }, + { + "bbox": [ + 363, + 596, + 407, + 607 + ], + "score": 0.92, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 596, + 424, + 608 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 425, + 596, + 463, + 607 + ], + "score": 0.26, + "content": "8 . 9 4 \\mathrm { F I D }", + "type": "inline_equation" + }, + { + "bbox": [ + 463, + 596, + 505, + 608 + ], + "score": 1.0, + "content": "on LSUN", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 606, + 506, + 619 + ], + "spans": [ + { + "bbox": [ + 106, + 606, + 506, + 619 + ], + "score": 1.0, + "content": "Church dataset, respectively. As shown in Figure 4, diverse examples with rich textures details are", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 617, + 507, + 631 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 507, + 631 + ], + "score": 1.0, + "content": "produced. We discuss the memory cost reduction brought by the Grid Self-Attention in Appendix E.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 24 + }, + { + "type": "title", + "bbox": [ + 107, + 640, + 322, + 652 + ], + "lines": [ + { + "bbox": [ + 105, + 638, + 324, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 324, + 653 + ], + "score": 1.0, + "content": "4.3 Data Augmentation is Crucial for TransGAN", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 655, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 655, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 506, + 668 + ], + "score": 1.0, + "content": "We study the effectiveness of data augmentation for both CNN-based GANs and Our TransGAN.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 667, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 505, + 680 + ], + "score": 1.0, + "content": "We apply the differentiable augmentation [69] to all these methods. As shown in Table 2, for three", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 677, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 690 + ], + "score": 1.0, + "content": "CNN-based GANs, the performance gains of data augmentation seems to diminish in the full-data", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "regime. Only the largest model, StyleGAN-V2, is improved on both IS and FID. In sharp contrast,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "score": 1.0, + "content": "TransGAN sees a shockingly large margin of improvement: IS improving from 8.36 to 9.02 and FID", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 711, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 505, + 723 + ], + "score": 1.0, + "content": "improving from 22.53 to 9.26. This phenomenon suggests that CIFAR-10 is still “small-scale \" when", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 31.5 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 742, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 9 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 109, + 65, + 501, + 191 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 109, + 65, + 501, + 191 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 109, + 65, + 501, + 191 + ], + "spans": [ + { + "bbox": [ + 109, + 65, + 501, + 191 + ], + "score": 0.968, + "type": "image", + "image_path": "6e96e44a3e68b78d7018919b2cfbce5508a09fd739e9d26cf4824f8d0c72caee.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 109, + 65, + 501, + 107.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 109, + 107.0, + 501, + 149.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 109, + 149.0, + 501, + 191.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 194, + 504, + 217 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 194, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 106, + 194, + 505, + 207 + ], + "score": 1.0, + "content": "Figure 4: Representative visual results produced by TransGAN on different datasets, as resolution", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 204, + 452, + 218 + ], + "spans": [ + { + "bbox": [ + 105, + 204, + 155, + 218 + ], + "score": 1.0, + "content": "grows from", + "type": "text" + }, + { + "bbox": [ + 155, + 205, + 189, + 216 + ], + "score": 0.89, + "content": "3 2 \\times 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 204, + 200, + 218 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 200, + 205, + 244, + 216 + ], + "score": 0.89, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 204, + 452, + 218 + ], + "score": 1.0, + "content": ". More visual examples are included in Appendix F.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "table", + "bbox": [ + 119, + 252, + 492, + 310 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 228, + 502, + 251 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 227, + 504, + 241 + ], + "spans": [ + { + "bbox": [ + 105, + 227, + 504, + 241 + ], + "score": 1.0, + "content": "Table 2: The effectiveness of Data Augmentation on both CNN-based GANs and TransGAN. We use", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 239, + 303, + 252 + ], + "spans": [ + { + "bbox": [ + 106, + 239, + 303, + 252 + ], + "score": 1.0, + "content": "the full CIFAR-10 training set and DiffAug [69].", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "table_body", + "bbox": [ + 119, + 252, + 492, + 310 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 119, + 252, + 492, + 310 + ], + "spans": [ + { + "bbox": [ + 119, + 252, + 492, + 310 + ], + "score": 0.977, + "html": "
MethodsWGAN-GPAutoGANStyleGAN-V2TransGAN
IS个FID←IS个FID←IS个FID←IS个FID↓
Original6.4939.688.5512.429.1811.078.3622.53
+ DiffAug [69]6.2937.148.6012.729.409.899.029.26
", + "type": "table", + "image_path": "f1c4e2b09d5e2f8e5edf0927060448a8a0cc45c22cecb15cb148ce6fe11b716d.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 119, + 252, + 492, + 271.3333333333333 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 119, + 271.3333333333333, + 492, + 290.66666666666663 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 119, + 290.66666666666663, + 492, + 309.99999999999994 + ], + "spans": [], + "index": 9 + } + ] + } + ], + "index": 6.75 + }, + { + "type": "table", + "bbox": [ + 130, + 346, + 479, + 460 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 109, + 322, + 505, + 345 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 322, + 506, + 335 + ], + "spans": [ + { + "bbox": [ + 106, + 322, + 463, + 335 + ], + "score": 1.0, + "content": "Table 3: The ablation study of proposed techniques in three common dataset CelebA", + "type": "text" + }, + { + "bbox": [ + 464, + 322, + 501, + 334 + ], + "score": 0.84, + "content": "( 6 4 \\times 6 4 )", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 322, + 506, + 335 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 107, + 333, + 485, + 346 + ], + "spans": [ + { + "bbox": [ + 107, + 333, + 139, + 346 + ], + "score": 1.0, + "content": "CelebA", + "type": "text" + }, + { + "bbox": [ + 139, + 333, + 184, + 344 + ], + "score": 0.83, + "content": "( 1 2 8 \\times 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 333, + 264, + 346 + ], + "score": 1.0, + "content": ", and LSUN Church", + "type": "text" + }, + { + "bbox": [ + 264, + 333, + 312, + 344 + ], + "score": 0.83, + "content": "( 2 5 6 \\times 2 5 6 )", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 333, + 485, + 346 + ], + "score": 1.0, + "content": "). “OOM” represents out-of-momery issue.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + }, + { + "type": "table_body", + "bbox": [ + 130, + 346, + 479, + 460 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 130, + 346, + 479, + 460 + ], + "spans": [ + { + "bbox": [ + 130, + 346, + 479, + 460 + ], + "score": 0.983, + "html": "
Training ConfigurationCelebA (64x64)CelebA (128x128)LSUN Church (256x256)
(A). Standard Self-Attention8.9200M0OM
(B). Nystrom Self-Attention [64]13.4717.4239.92
(C). Axis Self-Attention [67]12.3913.9529.30
(D). Grid Self-Attention9.8910.5820.39
+Multi-scale Discriminator9.288.0315.29
+Modified Normalization7.057.1313.27
+Relative Position Encoding6.146.3211.93
(E). Converge5.015.288.94
", + "type": "table", + "image_path": "25c0032de9117db6ffe66c1bcb8ba86cc9bfa5f079bdbf7a78a3daa9c1d02d03.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 130, + 346, + 479, + 384.0 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 130, + 384.0, + 479, + 422.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 130, + 422.0, + 479, + 460.0 + ], + "spans": [], + "index": 14 + } + ] + } + ], + "index": 11.75 + }, + { + "type": "text", + "bbox": [ + 106, + 470, + 506, + 525 + ], + "lines": [ + { + "bbox": [ + 106, + 470, + 505, + 483 + ], + "spans": [ + { + "bbox": [ + 106, + 470, + 462, + 483 + ], + "score": 1.0, + "content": "CelebA (128x128). We continue to examine another common benchmark: CelebA dataset", + "type": "text" + }, + { + "bbox": [ + 462, + 470, + 505, + 481 + ], + "score": 0.87, + "content": "( 1 2 8 \\times 1 2 8", + "type": "inline_equation" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 480, + 506, + 494 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 506, + 494 + ], + "score": 1.0, + "content": "resolution). As shown in Table 1, TransGAN largely outperforms Progressive-GAN [16] and COCO-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 491, + 505, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 491, + 505, + 505 + ], + "score": 1.0, + "content": "GAN [74], and is slightly better than the strongest competitor StyleGAN-v2 [40], by reaching a FID", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 502, + 505, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 414, + 516 + ], + "score": 1.0, + "content": "score of 5.28. Visual examples generated on CIFAR-10, STL-10, and CelebA", + "type": "text" + }, + { + "bbox": [ + 414, + 504, + 461, + 514 + ], + "score": 0.87, + "content": "( 1 2 8 \\times 1 2 8 )", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 502, + 505, + 516 + ], + "score": 1.0, + "content": "are shown", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 513, + 397, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 397, + 527 + ], + "score": 1.0, + "content": "in Figure 4, from which we observe pleasing visual details and diversity.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17, + "bbox_fs": [ + 105, + 470, + 506, + 527 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 536, + 269, + 548 + ], + "lines": [ + { + "bbox": [ + 105, + 535, + 269, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 269, + 550 + ], + "score": 1.0, + "content": "4.2 Scaling Up to Higher-Resolution", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 552, + 506, + 630 + ], + "lines": [ + { + "bbox": [ + 105, + 551, + 507, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 319, + 565 + ], + "score": 1.0, + "content": "We further scale up TransGAN to higher-resolution", + "type": "text" + }, + { + "bbox": [ + 319, + 552, + 366, + 563 + ], + "score": 0.86, + "content": "( 2 5 6 \\times 2 5 6 )", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 551, + 507, + 565 + ], + "score": 1.0, + "content": "generation, including on CelebA-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 104, + 561, + 506, + 577 + ], + "spans": [ + { + "bbox": [ + 104, + 561, + 506, + 577 + ], + "score": 1.0, + "content": "HQ [16] and LSUN Church [82]. These high-resolution datasets are significantly more challenging", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 573, + 506, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 506, + 586 + ], + "score": 1.0, + "content": "due to their much richer and detailed low-level texture as well as the global composition. Thanks", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 585, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 506, + 599 + ], + "score": 1.0, + "content": "to the proposed multi-scale discriminator, TransGAN produces pleasing visual results, reaching", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 596, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 363, + 608 + ], + "score": 1.0, + "content": "competitive quantitative results with 10.28 FID on CelebA-HQ", + "type": "text" + }, + { + "bbox": [ + 363, + 596, + 407, + 607 + ], + "score": 0.92, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 596, + 424, + 608 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 425, + 596, + 463, + 607 + ], + "score": 0.26, + "content": "8 . 9 4 \\mathrm { F I D }", + "type": "inline_equation" + }, + { + "bbox": [ + 463, + 596, + 505, + 608 + ], + "score": 1.0, + "content": "on LSUN", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 606, + 506, + 619 + ], + "spans": [ + { + "bbox": [ + 106, + 606, + 506, + 619 + ], + "score": 1.0, + "content": "Church dataset, respectively. As shown in Figure 4, diverse examples with rich textures details are", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 617, + 507, + 631 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 507, + 631 + ], + "score": 1.0, + "content": "produced. We discuss the memory cost reduction brought by the Grid Self-Attention in Appendix E.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 24, + "bbox_fs": [ + 104, + 551, + 507, + 631 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 640, + 322, + 652 + ], + "lines": [ + { + "bbox": [ + 105, + 638, + 324, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 324, + 653 + ], + "score": 1.0, + "content": "4.3 Data Augmentation is Crucial for TransGAN", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 655, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 655, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 506, + 668 + ], + "score": 1.0, + "content": "We study the effectiveness of data augmentation for both CNN-based GANs and Our TransGAN.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 667, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 505, + 680 + ], + "score": 1.0, + "content": "We apply the differentiable augmentation [69] to all these methods. As shown in Table 2, for three", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 677, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 690 + ], + "score": 1.0, + "content": "CNN-based GANs, the performance gains of data augmentation seems to diminish in the full-data", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "regime. Only the largest model, StyleGAN-V2, is improved on both IS and FID. In sharp contrast,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 505, + 712 + ], + "score": 1.0, + "content": "TransGAN sees a shockingly large margin of improvement: IS improving from 8.36 to 9.02 and FID", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 711, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 505, + 723 + ], + "score": 1.0, + "content": "improving from 22.53 to 9.26. This phenomenon suggests that CIFAR-10 is still “small-scale \" when", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 172, + 505, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 172, + 505, + 185 + ], + "score": 1.0, + "content": "fitting transformers; it re-confirms our assumption that transformer-based architectures are much", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 183, + 447, + 196 + ], + "spans": [ + { + "bbox": [ + 106, + 183, + 447, + 196 + ], + "score": 1.0, + "content": "more data-hungry than CNNs, and that can be helped by stronger data augmentation.", + "type": "text", + "cross_page": true + } + ], + "index": 6 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 655, + 506, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 61, + 500, + 138 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 61, + 500, + 138 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 61, + 500, + 138 + ], + "spans": [ + { + "bbox": [ + 110, + 61, + 500, + 138 + ], + "score": 0.951, + "type": "image", + "image_path": "77da0030a49f68b1242125a5ca2cb5670b6ba0577419be181bcc412c753e9bdb.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 61, + 500, + 86.66666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 86.66666666666667, + 500, + 112.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 112.33333333333334, + 500, + 138.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 143, + 504, + 166 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 143, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 106, + 143, + 505, + 156 + ], + "score": 1.0, + "content": "Figure 5: Left: training dynamic with training epochs for both TransGAN and MSG-GAN on", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 154, + 456, + 167 + ], + "spans": [ + { + "bbox": [ + 106, + 154, + 159, + 167 + ], + "score": 1.0, + "content": "CelebA-HQ", + "type": "text" + }, + { + "bbox": [ + 159, + 154, + 206, + 165 + ], + "score": 0.86, + "content": "( 2 5 6 \\times 2 5 6 )", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 154, + 456, + 167 + ], + "score": 1.0, + "content": ". Right: Interpolation on latent space produced by TransGAN.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 106, + 172, + 504, + 195 + ], + "lines": [ + { + "bbox": [ + 106, + 172, + 505, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 172, + 505, + 185 + ], + "score": 1.0, + "content": "fitting transformers; it re-confirms our assumption that transformer-based architectures are much", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 183, + 447, + 196 + ], + "spans": [ + { + "bbox": [ + 106, + 183, + 447, + 196 + ], + "score": 1.0, + "content": "more data-hungry than CNNs, and that can be helped by stronger data augmentation.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "title", + "bbox": [ + 107, + 203, + 194, + 215 + ], + "lines": [ + { + "bbox": [ + 105, + 201, + 196, + 218 + ], + "spans": [ + { + "bbox": [ + 105, + 201, + 196, + 218 + ], + "score": 1.0, + "content": "4.4 Ablation Study", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 107, + 219, + 505, + 329 + ], + "lines": [ + { + "bbox": [ + 105, + 218, + 506, + 233 + ], + "spans": [ + { + "bbox": [ + 105, + 218, + 506, + 233 + ], + "score": 1.0, + "content": "To further evaluate the proposed grid self-attention, multi-scale discriminator, and unique training", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 230, + 506, + 242 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 506, + 242 + ], + "score": 1.0, + "content": "recipe, we conduct the ablation study by separately adding these techniques to the baseline method", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 242, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 106, + 242, + 505, + 253 + ], + "score": 1.0, + "content": "and report their FID score on different datasets. Due to the fact that most of our contributions are", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 252, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 506, + 264 + ], + "score": 1.0, + "content": "tailored for the challenges brought by higher-resolution tasks, we choose CelebA and LSUN Church", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 263, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 505, + 276 + ], + "score": 1.0, + "content": "as the main testbeds, with details shown in Table 3. We start by constructing our memory-friendly", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 275, + 505, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 275, + 505, + 286 + ], + "score": 1.0, + "content": "with vanilla discriminator as our baseline method (A), both applied with standard self-attention. The", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 282, + 507, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 282, + 399, + 299 + ], + "score": 1.0, + "content": "baseline method achieves relatively good results with 8.92 FID on CelebA", + "type": "text" + }, + { + "bbox": [ + 399, + 285, + 435, + 296 + ], + "score": 0.87, + "content": "( 6 4 \\times 6 4 )", + "type": "inline_equation" + }, + { + "bbox": [ + 436, + 282, + 507, + 299 + ], + "score": 1.0, + "content": "dataset, however,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 295, + 506, + 308 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 506, + 308 + ], + "score": 1.0, + "content": "it fail on higher-resolution tasks due to the memory explosion issue brought by self-attention. This", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 306, + 505, + 319 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 505, + 319 + ], + "score": 1.0, + "content": "motivates us to evaluate two efficient form of self-attention, (B) Nyström Self-Attention [64]", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 317, + 260, + 330 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 260, + 330 + ], + "score": 1.0, + "content": "and (C) Axis Self-Attention [67]", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 107, + 334, + 505, + 443 + ], + "lines": [ + { + "bbox": [ + 106, + 333, + 505, + 346 + ], + "spans": [ + { + "bbox": [ + 106, + 333, + 505, + 346 + ], + "score": 1.0, + "content": "By replacing all self-attention layers in high-resolution stages (feature map resolution higher than", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 345, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 140, + 356 + ], + "score": 0.87, + "content": "3 2 \\times 3 2 )", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 345, + 506, + 356 + ], + "score": 1.0, + "content": "with these efficient variants, both two methods (B)(C) are able to produce reasonable results.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 355, + 506, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 355, + 402, + 369 + ], + "score": 1.0, + "content": "However, they still show to be inferior to standard self-attention, even on the", + "type": "text" + }, + { + "bbox": [ + 403, + 356, + 433, + 366 + ], + "score": 0.9, + "content": "6 4 \\times 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 355, + 506, + 369 + ], + "score": 1.0, + "content": "resolution dataset.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 367, + 506, + 379 + ], + "spans": [ + { + "bbox": [ + 106, + 367, + 506, + 379 + ], + "score": 1.0, + "content": "By adopting our proposed Grid Self-Attention (D), we observe a significant improvement on", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 378, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 372, + 389 + ], + "score": 1.0, + "content": "both three datasets, reaching 9.89, 10.58, 20.39 FID on CelebA", + "type": "text" + }, + { + "bbox": [ + 372, + 378, + 407, + 388 + ], + "score": 0.87, + "content": "6 4 \\times 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 378, + 411, + 389 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 412, + 378, + 456, + 388 + ], + "score": 0.87, + "content": "1 2 8 \\times 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 378, + 505, + 389 + ], + "score": 1.0, + "content": "and LSUN", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 387, + 506, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 138, + 402 + ], + "score": 1.0, + "content": "Church", + "type": "text" + }, + { + "bbox": [ + 139, + 389, + 182, + 399 + ], + "score": 0.89, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 387, + 506, + 402 + ], + "score": 1.0, + "content": ", respectively. Based on the configuration (D), we continue to add the proposed", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 400, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 106, + 400, + 505, + 411 + ], + "score": 1.0, + "content": "techniques, including the multi-scale discriminator, modified normalization, and relative position", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 410, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 505, + 422 + ], + "score": 1.0, + "content": "encoding. All these three techniques significantly improve the performance of TransGAN on three", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 420, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 506, + 434 + ], + "score": 1.0, + "content": "datasets. At the end, we train our final configuration (E) until it converges, resulting in the best FID", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 432, + 473, + 444 + ], + "spans": [ + { + "bbox": [ + 106, + 432, + 151, + 444 + ], + "score": 1.0, + "content": "on CelebA", + "type": "text" + }, + { + "bbox": [ + 152, + 432, + 186, + 443 + ], + "score": 0.92, + "content": "6 4 \\times 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 432, + 248, + 444 + ], + "score": 1.0, + "content": "(5.01), CelebA", + "type": "text" + }, + { + "bbox": [ + 249, + 432, + 292, + 443 + ], + "score": 0.89, + "content": "1 2 8 \\times 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 432, + 399, + 444 + ], + "score": 1.0, + "content": "(5.28), and LSUN Church", + "type": "text" + }, + { + "bbox": [ + 399, + 432, + 443, + 443 + ], + "score": 0.88, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + }, + { + "bbox": [ + 443, + 432, + 473, + 444 + ], + "score": 1.0, + "content": "(8.94).", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 22.5 + }, + { + "type": "title", + "bbox": [ + 106, + 452, + 358, + 464 + ], + "lines": [ + { + "bbox": [ + 105, + 452, + 358, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 452, + 358, + 465 + ], + "score": 1.0, + "content": "4.5 Understanding Transformer-based Generative Model", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 468, + 505, + 578 + ], + "lines": [ + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "We dive deep into our transformer-based GAN by conducting interpolation on latent space and", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 479, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 506, + 492 + ], + "score": 1.0, + "content": "comparing its behavior with CNN-based GAN, through visualizing their training dynamics. We", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 489, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 505, + 503 + ], + "score": 1.0, + "content": "choose MSG-GAN [84] for comparison since it extracts multi-scale representation as well. As shown", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 500, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 506, + 515 + ], + "score": 1.0, + "content": "in Figure 5, the CNN-based GAN quickly extracts face representation in the early stage of training", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 511, + 506, + 527 + ], + "spans": [ + { + "bbox": [ + 104, + 511, + 506, + 527 + ], + "score": 1.0, + "content": "process while transformer only produces rough pixels with no meaningful global shape due to missing", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 523, + 506, + 536 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 506, + 536 + ], + "score": 1.0, + "content": "any inductive bias. However, given enough training iterations, TransGAN gradually learns informative", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 533, + 506, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 506, + 546 + ], + "score": 1.0, + "content": "position representation and is able to produce impressive visual examples at convergence. Meanwhile,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 545, + 506, + 557 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 506, + 557 + ], + "score": 1.0, + "content": "the boundary artifact also vanishes at the end. For the latent space interpolation, TransGAN continues", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 555, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 506, + 567 + ], + "score": 1.0, + "content": "to show encouraging results where smooth interpolation are maintained on both local and global", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 566, + 421, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 566, + 421, + 579 + ], + "score": 1.0, + "content": "levels. More high-resolution visual examples will be presented in Appendix F.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 33.5 + }, + { + "type": "title", + "bbox": [ + 108, + 591, + 411, + 604 + ], + "lines": [ + { + "bbox": [ + 106, + 591, + 412, + 605 + ], + "spans": [ + { + "bbox": [ + 106, + 591, + 412, + 605 + ], + "score": 1.0, + "content": "4.6 Analyzing the Failure Cases and Improving High-resolution Tasks", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 107, + 612, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 612, + 506, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 506, + 625 + ], + "score": 1.0, + "content": "While TransGAN shows competitive or even better results on common low-resolution benchmarks,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 623, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 623, + 505, + 637 + ], + "score": 1.0, + "content": "we still see large improvement room of its performance on high-resolution synthesis tasks, by", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 634, + 506, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 634, + 506, + 647 + ], + "score": 1.0, + "content": "analyzing the failure cases shown in appendix C. Here we discuss several alternatives tailored for", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 645, + 506, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 645, + 506, + 658 + ], + "score": 1.0, + "content": "high-resolution synthesis tasks, as potential remedies to address these failure cases. Specifically,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 656, + 506, + 669 + ], + "spans": [ + { + "bbox": [ + 105, + 656, + 506, + 669 + ], + "score": 1.0, + "content": "we apply the self-modulation [85, 22, 35] to our generator and use cross-attention [53, 86] to map", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 667, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 395, + 680 + ], + "score": 1.0, + "content": "the latent space to the global region. Besides, we replace the current", + "type": "text" + }, + { + "bbox": [ + 395, + 667, + 410, + 678 + ], + "score": 0.86, + "content": "2 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 667, + 506, + 680 + ], + "score": 1.0, + "content": "upsampling layer, and", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 678, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 222, + 691 + ], + "score": 1.0, + "content": "instead firstly upsample it to", + "type": "text" + }, + { + "bbox": [ + 223, + 678, + 237, + 689 + ], + "score": 0.86, + "content": "4 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 678, + 505, + 691 + ], + "score": 1.0, + "content": "lager resolution using bicubic interpolation, and then downsample", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 688, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 147, + 701 + ], + "score": 1.0, + "content": "it back to", + "type": "text" + }, + { + "bbox": [ + 147, + 689, + 161, + 699 + ], + "score": 0.86, + "content": "2 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 688, + 505, + 701 + ], + "score": 1.0, + "content": "larger one. This simple modification not only helps the cross-boundary information", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 700, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 505, + 713 + ], + "score": 1.0, + "content": "interaction, but also help enhances the high-frequency details [87]. Moreover, an overlapped patch", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "score": 1.0, + "content": "splitting strategy for discriminator can slightly improve the FID score. Additionally, we follow the", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 44.5 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 742, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 110, + 61, + 500, + 138 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 110, + 61, + 500, + 138 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 61, + 500, + 138 + ], + "spans": [ + { + "bbox": [ + 110, + 61, + 500, + 138 + ], + "score": 0.951, + "type": "image", + "image_path": "77da0030a49f68b1242125a5ca2cb5670b6ba0577419be181bcc412c753e9bdb.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 110, + 61, + 500, + 86.66666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 110, + 86.66666666666667, + 500, + 112.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 110, + 112.33333333333334, + 500, + 138.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 143, + 504, + 166 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 143, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 106, + 143, + 505, + 156 + ], + "score": 1.0, + "content": "Figure 5: Left: training dynamic with training epochs for both TransGAN and MSG-GAN on", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 154, + 456, + 167 + ], + "spans": [ + { + "bbox": [ + 106, + 154, + 159, + 167 + ], + "score": 1.0, + "content": "CelebA-HQ", + "type": "text" + }, + { + "bbox": [ + 159, + 154, + 206, + 165 + ], + "score": 0.86, + "content": "( 2 5 6 \\times 2 5 6 )", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 154, + 456, + 167 + ], + "score": 1.0, + "content": ". Right: Interpolation on latent space produced by TransGAN.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "text", + "bbox": [ + 106, + 172, + 504, + 195 + ], + "lines": [], + "index": 5.5, + "bbox_fs": [ + 106, + 172, + 505, + 196 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 107, + 203, + 194, + 215 + ], + "lines": [ + { + "bbox": [ + 105, + 201, + 196, + 218 + ], + "spans": [ + { + "bbox": [ + 105, + 201, + 196, + 218 + ], + "score": 1.0, + "content": "4.4 Ablation Study", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 107, + 219, + 505, + 329 + ], + "lines": [ + { + "bbox": [ + 105, + 218, + 506, + 233 + ], + "spans": [ + { + "bbox": [ + 105, + 218, + 506, + 233 + ], + "score": 1.0, + "content": "To further evaluate the proposed grid self-attention, multi-scale discriminator, and unique training", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 230, + 506, + 242 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 506, + 242 + ], + "score": 1.0, + "content": "recipe, we conduct the ablation study by separately adding these techniques to the baseline method", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 242, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 106, + 242, + 505, + 253 + ], + "score": 1.0, + "content": "and report their FID score on different datasets. Due to the fact that most of our contributions are", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 252, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 506, + 264 + ], + "score": 1.0, + "content": "tailored for the challenges brought by higher-resolution tasks, we choose CelebA and LSUN Church", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 263, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 505, + 276 + ], + "score": 1.0, + "content": "as the main testbeds, with details shown in Table 3. We start by constructing our memory-friendly", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 275, + 505, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 275, + 505, + 286 + ], + "score": 1.0, + "content": "with vanilla discriminator as our baseline method (A), both applied with standard self-attention. The", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 282, + 507, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 282, + 399, + 299 + ], + "score": 1.0, + "content": "baseline method achieves relatively good results with 8.92 FID on CelebA", + "type": "text" + }, + { + "bbox": [ + 399, + 285, + 435, + 296 + ], + "score": 0.87, + "content": "( 6 4 \\times 6 4 )", + "type": "inline_equation" + }, + { + "bbox": [ + 436, + 282, + 507, + 299 + ], + "score": 1.0, + "content": "dataset, however,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 295, + 506, + 308 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 506, + 308 + ], + "score": 1.0, + "content": "it fail on higher-resolution tasks due to the memory explosion issue brought by self-attention. This", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 306, + 505, + 319 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 505, + 319 + ], + "score": 1.0, + "content": "motivates us to evaluate two efficient form of self-attention, (B) Nyström Self-Attention [64]", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 317, + 260, + 330 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 260, + 330 + ], + "score": 1.0, + "content": "and (C) Axis Self-Attention [67]", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 218, + 507, + 330 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 334, + 505, + 443 + ], + "lines": [ + { + "bbox": [ + 106, + 333, + 505, + 346 + ], + "spans": [ + { + "bbox": [ + 106, + 333, + 505, + 346 + ], + "score": 1.0, + "content": "By replacing all self-attention layers in high-resolution stages (feature map resolution higher than", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 345, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 140, + 356 + ], + "score": 0.87, + "content": "3 2 \\times 3 2 )", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 345, + 506, + 356 + ], + "score": 1.0, + "content": "with these efficient variants, both two methods (B)(C) are able to produce reasonable results.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 355, + 506, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 355, + 402, + 369 + ], + "score": 1.0, + "content": "However, they still show to be inferior to standard self-attention, even on the", + "type": "text" + }, + { + "bbox": [ + 403, + 356, + 433, + 366 + ], + "score": 0.9, + "content": "6 4 \\times 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 355, + 506, + 369 + ], + "score": 1.0, + "content": "resolution dataset.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 367, + 506, + 379 + ], + "spans": [ + { + "bbox": [ + 106, + 367, + 506, + 379 + ], + "score": 1.0, + "content": "By adopting our proposed Grid Self-Attention (D), we observe a significant improvement on", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 378, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 372, + 389 + ], + "score": 1.0, + "content": "both three datasets, reaching 9.89, 10.58, 20.39 FID on CelebA", + "type": "text" + }, + { + "bbox": [ + 372, + 378, + 407, + 388 + ], + "score": 0.87, + "content": "6 4 \\times 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 378, + 411, + 389 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 412, + 378, + 456, + 388 + ], + "score": 0.87, + "content": "1 2 8 \\times 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 378, + 505, + 389 + ], + "score": 1.0, + "content": "and LSUN", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 387, + 506, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 138, + 402 + ], + "score": 1.0, + "content": "Church", + "type": "text" + }, + { + "bbox": [ + 139, + 389, + 182, + 399 + ], + "score": 0.89, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 387, + 506, + 402 + ], + "score": 1.0, + "content": ", respectively. Based on the configuration (D), we continue to add the proposed", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 400, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 106, + 400, + 505, + 411 + ], + "score": 1.0, + "content": "techniques, including the multi-scale discriminator, modified normalization, and relative position", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 410, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 505, + 422 + ], + "score": 1.0, + "content": "encoding. All these three techniques significantly improve the performance of TransGAN on three", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 420, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 506, + 434 + ], + "score": 1.0, + "content": "datasets. At the end, we train our final configuration (E) until it converges, resulting in the best FID", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 432, + 473, + 444 + ], + "spans": [ + { + "bbox": [ + 106, + 432, + 151, + 444 + ], + "score": 1.0, + "content": "on CelebA", + "type": "text" + }, + { + "bbox": [ + 152, + 432, + 186, + 443 + ], + "score": 0.92, + "content": "6 4 \\times 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 432, + 248, + 444 + ], + "score": 1.0, + "content": "(5.01), CelebA", + "type": "text" + }, + { + "bbox": [ + 249, + 432, + 292, + 443 + ], + "score": 0.89, + "content": "1 2 8 \\times 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 432, + 399, + 444 + ], + "score": 1.0, + "content": "(5.28), and LSUN Church", + "type": "text" + }, + { + "bbox": [ + 399, + 432, + 443, + 443 + ], + "score": 0.88, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + }, + { + "bbox": [ + 443, + 432, + 473, + 444 + ], + "score": 1.0, + "content": "(8.94).", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 333, + 506, + 444 + ] + }, + { + "type": "title", + "bbox": [ + 106, + 452, + 358, + 464 + ], + "lines": [ + { + "bbox": [ + 105, + 452, + 358, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 452, + 358, + 465 + ], + "score": 1.0, + "content": "4.5 Understanding Transformer-based Generative Model", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 468, + 505, + 578 + ], + "lines": [ + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "We dive deep into our transformer-based GAN by conducting interpolation on latent space and", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 479, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 506, + 492 + ], + "score": 1.0, + "content": "comparing its behavior with CNN-based GAN, through visualizing their training dynamics. We", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 489, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 505, + 503 + ], + "score": 1.0, + "content": "choose MSG-GAN [84] for comparison since it extracts multi-scale representation as well. As shown", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 500, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 506, + 515 + ], + "score": 1.0, + "content": "in Figure 5, the CNN-based GAN quickly extracts face representation in the early stage of training", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 511, + 506, + 527 + ], + "spans": [ + { + "bbox": [ + 104, + 511, + 506, + 527 + ], + "score": 1.0, + "content": "process while transformer only produces rough pixels with no meaningful global shape due to missing", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 523, + 506, + 536 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 506, + 536 + ], + "score": 1.0, + "content": "any inductive bias. However, given enough training iterations, TransGAN gradually learns informative", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 533, + 506, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 506, + 546 + ], + "score": 1.0, + "content": "position representation and is able to produce impressive visual examples at convergence. Meanwhile,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 545, + 506, + 557 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 506, + 557 + ], + "score": 1.0, + "content": "the boundary artifact also vanishes at the end. For the latent space interpolation, TransGAN continues", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 555, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 506, + 567 + ], + "score": 1.0, + "content": "to show encouraging results where smooth interpolation are maintained on both local and global", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 566, + 421, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 566, + 421, + 579 + ], + "score": 1.0, + "content": "levels. More high-resolution visual examples will be presented in Appendix F.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 33.5, + "bbox_fs": [ + 104, + 467, + 506, + 579 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 591, + 411, + 604 + ], + "lines": [ + { + "bbox": [ + 106, + 591, + 412, + 605 + ], + "spans": [ + { + "bbox": [ + 106, + 591, + 412, + 605 + ], + "score": 1.0, + "content": "4.6 Analyzing the Failure Cases and Improving High-resolution Tasks", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 107, + 612, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 612, + 506, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 506, + 625 + ], + "score": 1.0, + "content": "While TransGAN shows competitive or even better results on common low-resolution benchmarks,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 623, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 623, + 505, + 637 + ], + "score": 1.0, + "content": "we still see large improvement room of its performance on high-resolution synthesis tasks, by", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 634, + 506, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 634, + 506, + 647 + ], + "score": 1.0, + "content": "analyzing the failure cases shown in appendix C. Here we discuss several alternatives tailored for", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 645, + 506, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 645, + 506, + 658 + ], + "score": 1.0, + "content": "high-resolution synthesis tasks, as potential remedies to address these failure cases. Specifically,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 656, + 506, + 669 + ], + "spans": [ + { + "bbox": [ + 105, + 656, + 506, + 669 + ], + "score": 1.0, + "content": "we apply the self-modulation [85, 22, 35] to our generator and use cross-attention [53, 86] to map", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 667, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 395, + 680 + ], + "score": 1.0, + "content": "the latent space to the global region. Besides, we replace the current", + "type": "text" + }, + { + "bbox": [ + 395, + 667, + 410, + 678 + ], + "score": 0.86, + "content": "2 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 667, + 506, + 680 + ], + "score": 1.0, + "content": "upsampling layer, and", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 678, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 222, + 691 + ], + "score": 1.0, + "content": "instead firstly upsample it to", + "type": "text" + }, + { + "bbox": [ + 223, + 678, + 237, + 689 + ], + "score": 0.86, + "content": "4 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 678, + 505, + 691 + ], + "score": 1.0, + "content": "lager resolution using bicubic interpolation, and then downsample", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 688, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 147, + 701 + ], + "score": 1.0, + "content": "it back to", + "type": "text" + }, + { + "bbox": [ + 147, + 689, + 161, + 699 + ], + "score": 0.86, + "content": "2 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 688, + 505, + 701 + ], + "score": 1.0, + "content": "larger one. This simple modification not only helps the cross-boundary information", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 700, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 505, + 713 + ], + "score": 1.0, + "content": "interaction, but also help enhances the high-frequency details [87]. Moreover, an overlapped patch", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "score": 1.0, + "content": "splitting strategy for discriminator can slightly improve the FID score. Additionally, we follow the", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 73, + 506, + 85 + ], + "spans": [ + { + "bbox": [ + 105, + 73, + 506, + 85 + ], + "score": 1.0, + "content": "previous work [22, 40] to conduct noise injection before the self-attention layer, which is found to", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 84, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 84, + 505, + 96 + ], + "score": 1.0, + "content": "further improve the generation fidelity and diversity of TransGAN. By applying these techniques to", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 94, + 505, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 505, + 107 + ], + "score": 1.0, + "content": "our high-resolution GAN frameworks, we observe additional improvement on both qualitative and", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 105, + 506, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 306, + 119 + ], + "score": 1.0, + "content": "quantitative results, e.g., the FID score on CelebA", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 306, + 106, + 350, + 117 + ], + "score": 0.89, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 350, + 105, + 506, + 119 + ], + "score": 1.0, + "content": "dataset is further improved from 10.26", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 116, + 139, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 116, + 139, + 128 + ], + "score": 1.0, + "content": "to 8.93.", + "type": "text", + "cross_page": true + } + ], + "index": 4 + } + ], + "index": 44.5, + "bbox_fs": [ + 105, + 612, + 506, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 72, + 505, + 127 + ], + "lines": [ + { + "bbox": [ + 105, + 73, + 506, + 85 + ], + "spans": [ + { + "bbox": [ + 105, + 73, + 506, + 85 + ], + "score": 1.0, + "content": "previous work [22, 40] to conduct noise injection before the self-attention layer, which is found to", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 84, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 84, + 505, + 96 + ], + "score": 1.0, + "content": "further improve the generation fidelity and diversity of TransGAN. By applying these techniques to", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 94, + 505, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 505, + 107 + ], + "score": 1.0, + "content": "our high-resolution GAN frameworks, we observe additional improvement on both qualitative and", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 105, + 506, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 306, + 119 + ], + "score": 1.0, + "content": "quantitative results, e.g., the FID score on CelebA", + "type": "text" + }, + { + "bbox": [ + 306, + 106, + 350, + 117 + ], + "score": 0.89, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + }, + { + "bbox": [ + 350, + 105, + 506, + 119 + ], + "score": 1.0, + "content": "dataset is further improved from 10.26", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 116, + 139, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 116, + 139, + 128 + ], + "score": 1.0, + "content": "to 8.93.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "title", + "bbox": [ + 107, + 138, + 423, + 152 + ], + "lines": [ + { + "bbox": [ + 104, + 137, + 424, + 155 + ], + "spans": [ + { + "bbox": [ + 104, + 137, + 424, + 155 + ], + "score": 1.0, + "content": "5 Conclusions, Limitation, and Discussions of Broad Impact", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 158, + 505, + 235 + ], + "lines": [ + { + "bbox": [ + 106, + 159, + 505, + 171 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 505, + 171 + ], + "score": 1.0, + "content": "In this work, we provide the first pilot study of building GAN with pure transformers. We have", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 169, + 505, + 182 + ], + "spans": [ + { + "bbox": [ + 106, + 169, + 505, + 182 + ], + "score": 1.0, + "content": "carefully crafted the architectures and thoughtfully designed training techniques. As a result, the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 180, + 506, + 193 + ], + "spans": [ + { + "bbox": [ + 105, + 180, + 506, + 193 + ], + "score": 1.0, + "content": "proposed TransGAN has achieved state-of-the-art performance across multiple popular datasets, and", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 191, + 506, + 205 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 506, + 205 + ], + "score": 1.0, + "content": "easily scales up to higher-resolution generative tasks. Although TransGAN provides an encouraging", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 202, + 505, + 214 + ], + "spans": [ + { + "bbox": [ + 106, + 202, + 505, + 214 + ], + "score": 1.0, + "content": "starting point, there is still a large room to explore further, such as achieving state-of-the-art results", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 212, + 507, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 119, + 227 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 120, + 213, + 164, + 224 + ], + "score": 0.89, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 212, + 507, + 227 + ], + "score": 1.0, + "content": "generation tasks or going towards extremely high resolution generation tasks (e.g.,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 224, + 318, + 236 + ], + "spans": [ + { + "bbox": [ + 106, + 225, + 160, + 235 + ], + "score": 0.87, + "content": "1 0 2 4 \\times 1 0 2 4 ,", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 224, + 318, + 236 + ], + "score": 1.0, + "content": "), which would be our future directions.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 107, + 240, + 505, + 284 + ], + "lines": [ + { + "bbox": [ + 105, + 239, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 505, + 253 + ], + "score": 1.0, + "content": "Broader Impact. The proposed generative model can serve as a data engine to alleviate the challenge", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 251, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 251, + 506, + 264 + ], + "score": 1.0, + "content": "of data collection. More importantly, using synthesized image examples helps avoid privacy concerns.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 261, + 505, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 505, + 275 + ], + "score": 1.0, + "content": "However, the abuse of advanced generative models may create fake media materials, which demands", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 273, + 192, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 192, + 285 + ], + "score": 1.0, + "content": "caution in the future.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14.5 + }, + { + "type": "title", + "bbox": [ + 107, + 295, + 207, + 308 + ], + "lines": [ + { + "bbox": [ + 106, + 293, + 208, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 293, + 208, + 312 + ], + "score": 1.0, + "content": "Acknowledgements", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 107, + 315, + 505, + 359 + ], + "lines": [ + { + "bbox": [ + 105, + 315, + 505, + 328 + ], + "spans": [ + { + "bbox": [ + 105, + 315, + 505, + 328 + ], + "score": 1.0, + "content": "We would like to express our deepest gratitude to the MIT-IBM Watson AI Lab, in particular John", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 326, + 506, + 339 + ], + "spans": [ + { + "bbox": [ + 105, + 326, + 506, + 339 + ], + "score": 1.0, + "content": "Cohn for generously providing us with the computing resources necessary to conduct this research. Z", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 337, + 505, + 350 + ], + "spans": [ + { + "bbox": [ + 105, + 337, + 505, + 350 + ], + "score": 1.0, + "content": "Wang’s work is in part supported by an IBM Faculty Research Award, and the NSF AI Institute for", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 347, + 280, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 347, + 280, + 360 + ], + "score": 1.0, + "content": "Foundations of Machine Learning (IFML).", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19.5 + }, + { + "type": "title", + "bbox": [ + 107, + 375, + 163, + 387 + ], + "lines": [ + { + "bbox": [ + 106, + 373, + 165, + 389 + ], + "spans": [ + { + "bbox": [ + 106, + 373, + 165, + 389 + ], + "score": 1.0, + "content": "References", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 109, + 392, + 506, + 722 + ], + "lines": [ + { + "bbox": [ + 110, + 393, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 110, + 393, + 505, + 406 + ], + "score": 1.0, + "content": "[1] Ishaan Gulrajani, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, and Aaron C Courville. Improved", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 124, + 402, + 507, + 416 + ], + "spans": [ + { + "bbox": [ + 124, + 402, + 507, + 416 + ], + "score": 1.0, + "content": "training of wasserstein gans. In Advances in neural information processing systems, pages 5767–5777,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 125, + 411, + 150, + 425 + ], + "spans": [ + { + "bbox": [ + 125, + 411, + 150, + 425 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 110, + 431, + 506, + 445 + ], + "spans": [ + { + "bbox": [ + 110, + 431, + 506, + 445 + ], + "score": 1.0, + "content": "[2] Andrew Brock, Jeff Donahue, and Karen Simonyan. Large scale gan training for high fidelity natural", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 125, + 442, + 335, + 453 + ], + "spans": [ + { + "bbox": [ + 125, + 442, + 335, + 453 + ], + "score": 1.0, + "content": "image synthesis. arXiv preprint arXiv:1809.11096, 2018.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 110, + 460, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 110, + 460, + 505, + 473 + ], + "score": 1.0, + "content": "[3] Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A Efros. Image-to-image translation with conditional", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 124, + 469, + 506, + 483 + ], + "spans": [ + { + "bbox": [ + 124, + 469, + 506, + 483 + ], + "score": 1.0, + "content": "adversarial networks. In Proceedings of the IEEE conference on computer vision and pattern recognition,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 125, + 480, + 217, + 492 + ], + "spans": [ + { + "bbox": [ + 125, + 480, + 217, + 492 + ], + "score": 1.0, + "content": "pages 1125–1134, 2017.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 109, + 498, + 506, + 513 + ], + "spans": [ + { + "bbox": [ + 109, + 498, + 506, + 513 + ], + "score": 1.0, + "content": "[4] Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired image-to-image translation using", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 126, + 509, + 506, + 522 + ], + "spans": [ + { + "bbox": [ + 126, + 509, + 506, + 522 + ], + "score": 1.0, + "content": "cycle-consistent adversarial networks. In Proceedings of the IEEE international conference on computer", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 126, + 519, + 242, + 530 + ], + "spans": [ + { + "bbox": [ + 126, + 519, + 242, + 530 + ], + "score": 1.0, + "content": "vision, pages 2223–2232, 2017.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 110, + 538, + 505, + 549 + ], + "spans": [ + { + "bbox": [ + 110, + 538, + 505, + 549 + ], + "score": 1.0, + "content": "[5] Jun-Yan Zhu, Richard Zhang, Deepak Pathak, Trevor Darrell, Alexei A Efros, Oliver Wang, and Eli", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 125, + 546, + 505, + 561 + ], + "spans": [ + { + "bbox": [ + 125, + 546, + 505, + 561 + ], + "score": 1.0, + "content": "Shechtman. Toward multimodal image-to-image translation. In Advances in neural information processing", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 124, + 558, + 240, + 569 + ], + "spans": [ + { + "bbox": [ + 124, + 558, + 240, + 569 + ], + "score": 1.0, + "content": "systems, pages 465–476, 2017.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 110, + 576, + 505, + 589 + ], + "spans": [ + { + "bbox": [ + 110, + 576, + 505, + 589 + ], + "score": 1.0, + "content": "[6] Shuai Yang, Zhangyang Wang, Zhaowen Wang, Ning Xu, Jiaying Liu, and Zongming Guo. Controllable", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 124, + 585, + 506, + 600 + ], + "spans": [ + { + "bbox": [ + 124, + 585, + 506, + 600 + ], + "score": 1.0, + "content": "artistic text style transfer via shape-matching gan. In Proceedings of the IEEE International Conference on", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 126, + 596, + 281, + 609 + ], + "spans": [ + { + "bbox": [ + 126, + 596, + 281, + 609 + ], + "score": 1.0, + "content": "Computer Vision, pages 4442–4451, 2019.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 110, + 614, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 110, + 614, + 505, + 627 + ], + "score": 1.0, + "content": "[7] Yifan Jiang, Xinyu Gong, Ding Liu, Yu Cheng, Chen Fang, Xiaohui Shen, Jianchao Yang, Pan Zhou, and", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 124, + 624, + 506, + 638 + ], + "spans": [ + { + "bbox": [ + 124, + 624, + 506, + 638 + ], + "score": 1.0, + "content": "Zhangyang Wang. Enlightengan: Deep light enhancement without paired supervision. IEEE Transactions", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 125, + 635, + 285, + 646 + ], + "spans": [ + { + "bbox": [ + 125, + 635, + 285, + 646 + ], + "score": 1.0, + "content": "on Image Processing, 30:2340–2349, 2021.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 110, + 653, + 505, + 665 + ], + "spans": [ + { + "bbox": [ + 110, + 653, + 505, + 665 + ], + "score": 1.0, + "content": "[8] Karol Kurach, Mario Luciˇ c, Xiaohua Zhai, Marcin Michalski, and Sylvain Gelly. A large-scale study ´", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 125, + 663, + 505, + 676 + ], + "spans": [ + { + "bbox": [ + 125, + 663, + 505, + 676 + ], + "score": 1.0, + "content": "on regularization and normalization in gans. In International Conference on Machine Learning, pages", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 125, + 672, + 223, + 685 + ], + "spans": [ + { + "bbox": [ + 125, + 672, + 223, + 685 + ], + "score": 1.0, + "content": "3581–3590. PMLR, 2019.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 110, + 691, + 505, + 705 + ], + "spans": [ + { + "bbox": [ + 110, + 691, + 505, + 705 + ], + "score": 1.0, + "content": "[9] Kevin Roth, Aurelien Lucchi, Sebastian Nowozin, and Thomas Hofmann. Stabilizing training of generative", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 125, + 700, + 505, + 715 + ], + "spans": [ + { + "bbox": [ + 125, + 700, + 505, + 715 + ], + "score": 1.0, + "content": "adversarial networks through regularization. In Advances in neural information processing systems, pages", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 126, + 712, + 195, + 723 + ], + "spans": [ + { + "bbox": [ + 126, + 712, + 195, + 723 + ], + "score": 1.0, + "content": "2018–2028, 2017.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 35.5 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "10", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 72, + 505, + 127 + ], + "lines": [], + "index": 2, + "bbox_fs": [ + 105, + 73, + 506, + 128 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 107, + 138, + 423, + 152 + ], + "lines": [ + { + "bbox": [ + 104, + 137, + 424, + 155 + ], + "spans": [ + { + "bbox": [ + 104, + 137, + 424, + 155 + ], + "score": 1.0, + "content": "5 Conclusions, Limitation, and Discussions of Broad Impact", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 158, + 505, + 235 + ], + "lines": [ + { + "bbox": [ + 106, + 159, + 505, + 171 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 505, + 171 + ], + "score": 1.0, + "content": "In this work, we provide the first pilot study of building GAN with pure transformers. We have", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 169, + 505, + 182 + ], + "spans": [ + { + "bbox": [ + 106, + 169, + 505, + 182 + ], + "score": 1.0, + "content": "carefully crafted the architectures and thoughtfully designed training techniques. As a result, the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 180, + 506, + 193 + ], + "spans": [ + { + "bbox": [ + 105, + 180, + 506, + 193 + ], + "score": 1.0, + "content": "proposed TransGAN has achieved state-of-the-art performance across multiple popular datasets, and", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 191, + 506, + 205 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 506, + 205 + ], + "score": 1.0, + "content": "easily scales up to higher-resolution generative tasks. Although TransGAN provides an encouraging", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 202, + 505, + 214 + ], + "spans": [ + { + "bbox": [ + 106, + 202, + 505, + 214 + ], + "score": 1.0, + "content": "starting point, there is still a large room to explore further, such as achieving state-of-the-art results", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 212, + 507, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 119, + 227 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 120, + 213, + 164, + 224 + ], + "score": 0.89, + "content": "2 5 6 \\times 2 5 6", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 212, + 507, + 227 + ], + "score": 1.0, + "content": "generation tasks or going towards extremely high resolution generation tasks (e.g.,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 224, + 318, + 236 + ], + "spans": [ + { + "bbox": [ + 106, + 225, + 160, + 235 + ], + "score": 0.87, + "content": "1 0 2 4 \\times 1 0 2 4 ,", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 224, + 318, + 236 + ], + "score": 1.0, + "content": "), which would be our future directions.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 159, + 507, + 236 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 240, + 505, + 284 + ], + "lines": [ + { + "bbox": [ + 105, + 239, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 505, + 253 + ], + "score": 1.0, + "content": "Broader Impact. The proposed generative model can serve as a data engine to alleviate the challenge", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 251, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 251, + 506, + 264 + ], + "score": 1.0, + "content": "of data collection. More importantly, using synthesized image examples helps avoid privacy concerns.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 261, + 505, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 505, + 275 + ], + "score": 1.0, + "content": "However, the abuse of advanced generative models may create fake media materials, which demands", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 273, + 192, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 192, + 285 + ], + "score": 1.0, + "content": "caution in the future.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 239, + 506, + 285 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 295, + 207, + 308 + ], + "lines": [ + { + "bbox": [ + 106, + 293, + 208, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 293, + 208, + 312 + ], + "score": 1.0, + "content": "Acknowledgements", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 107, + 315, + 505, + 359 + ], + "lines": [ + { + "bbox": [ + 105, + 315, + 505, + 328 + ], + "spans": [ + { + "bbox": [ + 105, + 315, + 505, + 328 + ], + "score": 1.0, + "content": "We would like to express our deepest gratitude to the MIT-IBM Watson AI Lab, in particular John", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 326, + 506, + 339 + ], + "spans": [ + { + "bbox": [ + 105, + 326, + 506, + 339 + ], + "score": 1.0, + "content": "Cohn for generously providing us with the computing resources necessary to conduct this research. Z", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 337, + 505, + 350 + ], + "spans": [ + { + "bbox": [ + 105, + 337, + 505, + 350 + ], + "score": 1.0, + "content": "Wang’s work is in part supported by an IBM Faculty Research Award, and the NSF AI Institute for", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 347, + 280, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 347, + 280, + 360 + ], + "score": 1.0, + "content": "Foundations of Machine Learning (IFML).", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19.5, + "bbox_fs": [ + 105, + 315, + 506, + 360 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 375, + 163, + 387 + ], + "lines": [ + { + "bbox": [ + 106, + 373, + 165, + 389 + ], + "spans": [ + { + "bbox": [ + 106, + 373, + 165, + 389 + ], + "score": 1.0, + "content": "References", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "list", + "bbox": [ + 109, + 392, + 506, + 722 + ], + "lines": [ + { + "bbox": [ + 110, + 393, + 505, + 406 + ], + "spans": [ + { + "bbox": [ + 110, + 393, + 505, + 406 + ], + "score": 1.0, + "content": "[1] Ishaan Gulrajani, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, and Aaron C Courville. Improved", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 402, + 507, + 416 + ], + "spans": [ + { + "bbox": [ + 124, + 402, + 507, + 416 + ], + "score": 1.0, + "content": "training of wasserstein gans. In Advances in neural information processing systems, pages 5767–5777,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 125, + 411, + 150, + 425 + ], + "spans": [ + { + "bbox": [ + 125, + 411, + 150, + 425 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 25, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 431, + 506, + 445 + ], + "spans": [ + { + "bbox": [ + 110, + 431, + 506, + 445 + ], + "score": 1.0, + "content": "[2] Andrew Brock, Jeff Donahue, and Karen Simonyan. Large scale gan training for high fidelity natural", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 442, + 335, + 453 + ], + "spans": [ + { + "bbox": [ + 125, + 442, + 335, + 453 + ], + "score": 1.0, + "content": "image synthesis. arXiv preprint arXiv:1809.11096, 2018.", + "type": "text" + } + ], + "index": 27, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 460, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 110, + 460, + 505, + 473 + ], + "score": 1.0, + "content": "[3] Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A Efros. Image-to-image translation with conditional", + "type": "text" + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 469, + 506, + 483 + ], + "spans": [ + { + "bbox": [ + 124, + 469, + 506, + 483 + ], + "score": 1.0, + "content": "adversarial networks. In Proceedings of the IEEE conference on computer vision and pattern recognition,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 125, + 480, + 217, + 492 + ], + "spans": [ + { + "bbox": [ + 125, + 480, + 217, + 492 + ], + "score": 1.0, + "content": "pages 1125–1134, 2017.", + "type": "text" + } + ], + "index": 30, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 498, + 506, + 513 + ], + "spans": [ + { + "bbox": [ + 109, + 498, + 506, + 513 + ], + "score": 1.0, + "content": "[4] Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired image-to-image translation using", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 509, + 506, + 522 + ], + "spans": [ + { + "bbox": [ + 126, + 509, + 506, + 522 + ], + "score": 1.0, + "content": "cycle-consistent adversarial networks. In Proceedings of the IEEE international conference on computer", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 126, + 519, + 242, + 530 + ], + "spans": [ + { + "bbox": [ + 126, + 519, + 242, + 530 + ], + "score": 1.0, + "content": "vision, pages 2223–2232, 2017.", + "type": "text" + } + ], + "index": 33, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 538, + 505, + 549 + ], + "spans": [ + { + "bbox": [ + 110, + 538, + 505, + 549 + ], + "score": 1.0, + "content": "[5] Jun-Yan Zhu, Richard Zhang, Deepak Pathak, Trevor Darrell, Alexei A Efros, Oliver Wang, and Eli", + "type": "text" + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 546, + 505, + 561 + ], + "spans": [ + { + "bbox": [ + 125, + 546, + 505, + 561 + ], + "score": 1.0, + "content": "Shechtman. Toward multimodal image-to-image translation. In Advances in neural information processing", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 124, + 558, + 240, + 569 + ], + "spans": [ + { + "bbox": [ + 124, + 558, + 240, + 569 + ], + "score": 1.0, + "content": "systems, pages 465–476, 2017.", + "type": "text" + } + ], + "index": 36, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 576, + 505, + 589 + ], + "spans": [ + { + "bbox": [ + 110, + 576, + 505, + 589 + ], + "score": 1.0, + "content": "[6] Shuai Yang, Zhangyang Wang, Zhaowen Wang, Ning Xu, Jiaying Liu, and Zongming Guo. Controllable", + "type": "text" + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 585, + 506, + 600 + ], + "spans": [ + { + "bbox": [ + 124, + 585, + 506, + 600 + ], + "score": 1.0, + "content": "artistic text style transfer via shape-matching gan. In Proceedings of the IEEE International Conference on", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 126, + 596, + 281, + 609 + ], + "spans": [ + { + "bbox": [ + 126, + 596, + 281, + 609 + ], + "score": 1.0, + "content": "Computer Vision, pages 4442–4451, 2019.", + "type": "text" + } + ], + "index": 39, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 614, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 110, + 614, + 505, + 627 + ], + "score": 1.0, + "content": "[7] Yifan Jiang, Xinyu Gong, Ding Liu, Yu Cheng, Chen Fang, Xiaohui Shen, Jianchao Yang, Pan Zhou, and", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 624, + 506, + 638 + ], + "spans": [ + { + "bbox": [ + 124, + 624, + 506, + 638 + ], + "score": 1.0, + "content": "Zhangyang Wang. Enlightengan: Deep light enhancement without paired supervision. IEEE Transactions", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 125, + 635, + 285, + 646 + ], + "spans": [ + { + "bbox": [ + 125, + 635, + 285, + 646 + ], + "score": 1.0, + "content": "on Image Processing, 30:2340–2349, 2021.", + "type": "text" + } + ], + "index": 42, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 653, + 505, + 665 + ], + "spans": [ + { + "bbox": [ + 110, + 653, + 505, + 665 + ], + "score": 1.0, + "content": "[8] Karol Kurach, Mario Luciˇ c, Xiaohua Zhai, Marcin Michalski, and Sylvain Gelly. A large-scale study ´", + "type": "text" + } + ], + "index": 43, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 663, + 505, + 676 + ], + "spans": [ + { + "bbox": [ + 125, + 663, + 505, + 676 + ], + "score": 1.0, + "content": "on regularization and normalization in gans. In International Conference on Machine Learning, pages", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 125, + 672, + 223, + 685 + ], + "spans": [ + { + "bbox": [ + 125, + 672, + 223, + 685 + ], + "score": 1.0, + "content": "3581–3590. PMLR, 2019.", + "type": "text" + } + ], + "index": 45, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 691, + 505, + 705 + ], + "spans": [ + { + "bbox": [ + 110, + 691, + 505, + 705 + ], + "score": 1.0, + "content": "[9] Kevin Roth, Aurelien Lucchi, Sebastian Nowozin, and Thomas Hofmann. Stabilizing training of generative", + "type": "text" + } + ], + "index": 46, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 700, + 505, + 715 + ], + "spans": [ + { + "bbox": [ + 125, + 700, + 505, + 715 + ], + "score": 1.0, + "content": "adversarial networks through regularization. In Advances in neural information processing systems, pages", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 126, + 712, + 195, + 723 + ], + "spans": [ + { + "bbox": [ + 126, + 712, + 195, + 723 + ], + "score": 1.0, + "content": "2018–2028, 2017.", + "type": "text" + } + ], + "index": 48, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 74, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 74, + 506, + 86 + ], + "score": 1.0, + "content": "[10] Han Zhang, Zizhao Zhang, Augustus Odena, and Honglak Lee. Consistency regularization for generative", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 83, + 352, + 95 + ], + "spans": [ + { + "bbox": [ + 125, + 83, + 352, + 95 + ], + "score": 1.0, + "content": "adversarial networks. arXiv preprint arXiv:1910.12027, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 1, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 100, + 506, + 114 + ], + "spans": [ + { + "bbox": [ + 104, + 100, + 506, + 114 + ], + "score": 1.0, + "content": "[11] Lars Mescheder, Andreas Geiger, and Sebastian Nowozin. Which training methods for gans do actually", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 123, + 110, + 314, + 124 + ], + "spans": [ + { + "bbox": [ + 123, + 110, + 314, + 124 + ], + "score": 1.0, + "content": "converge? arXiv preprint arXiv:1801.04406, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 3, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 129, + 506, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 129, + 506, + 141 + ], + "score": 1.0, + "content": "[12] Xudong Mao, Qing Li, Haoran Xie, Raymond YK Lau, Zhen Wang, and Stephen Paul Smolley. Least", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 138, + 506, + 153 + ], + "spans": [ + { + "bbox": [ + 125, + 138, + 506, + 153 + ], + "score": 1.0, + "content": "squares generative adversarial networks. In Proceedings of the IEEE international conference on computer", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 124, + 148, + 244, + 162 + ], + "spans": [ + { + "bbox": [ + 124, + 148, + 244, + 162 + ], + "score": 1.0, + "content": "vision, pages 2794–2802, 2017.", + "type": "text", + "cross_page": true + } + ], + "index": 6, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 167, + 506, + 179 + ], + "spans": [ + { + "bbox": [ + 105, + 167, + 506, + 179 + ], + "score": 1.0, + "content": "[13] Alexia Jolicoeur-Martineau. The relativistic discriminator: a key element missing from standard gan. arXiv", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 123, + 177, + 251, + 190 + ], + "spans": [ + { + "bbox": [ + 123, + 177, + 251, + 190 + ], + "score": 1.0, + "content": "preprint arXiv:1807.00734, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 194, + 506, + 208 + ], + "spans": [ + { + "bbox": [ + 104, + 194, + 506, + 208 + ], + "score": 1.0, + "content": "[14] Chun-Liang Li, Wei-Cheng Chang, Yu Cheng, Yiming Yang, and Barnabás Póczos. Mmd gan: Towards", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 204, + 506, + 218 + ], + "spans": [ + { + "bbox": [ + 125, + 204, + 506, + 218 + ], + "score": 1.0, + "content": "deeper understanding of moment matching network. In Advances in neural information processing systems,", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 123, + 214, + 217, + 227 + ], + "spans": [ + { + "bbox": [ + 123, + 214, + 217, + 227 + ], + "score": 1.0, + "content": "pages 2203–2213, 2017.", + "type": "text", + "cross_page": true + } + ], + "index": 11, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 232, + 506, + 246 + ], + "spans": [ + { + "bbox": [ + 104, + 232, + 506, + 246 + ], + "score": 1.0, + "content": "[15] Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen. Improved", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 242, + 380, + 255 + ], + "spans": [ + { + "bbox": [ + 124, + 242, + 380, + 255 + ], + "score": 1.0, + "content": "techniques for training gans. arXiv preprint arXiv:1606.03498, 2016.", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 260, + 506, + 275 + ], + "spans": [ + { + "bbox": [ + 104, + 260, + 506, + 275 + ], + "score": 1.0, + "content": "[16] Tero Karras, Timo Aila, Samuli Laine, and Jaakko Lehtinen. Progressive growing of gans for improved", + "type": "text", + "cross_page": true + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 271, + 388, + 284 + ], + "spans": [ + { + "bbox": [ + 124, + 271, + 388, + 284 + ], + "score": 1.0, + "content": "quality, stability, and variation. arXiv preprint arXiv:1710.10196, 2017.", + "type": "text", + "cross_page": true + } + ], + "index": 15, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 289, + 506, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 506, + 301 + ], + "score": 1.0, + "content": "[17] Mario Lucic, Karol Kurach, Marcin Michalski, Olivier Bousquet, and Sylvain Gelly. Are gans created", + "type": "text", + "cross_page": true + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 299, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 126, + 299, + 505, + 311 + ], + "score": 1.0, + "content": "equal? a large-scale study. In Proceedings of the 32nd International Conference on Neural Information", + "type": "text", + "cross_page": true + } + ], + "index": 17 + }, + { + "bbox": [ + 123, + 308, + 283, + 321 + ], + "spans": [ + { + "bbox": [ + 123, + 308, + 283, + 321 + ], + "score": 1.0, + "content": "Processing Systems, pages 698–707, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 18, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 326, + 507, + 340 + ], + "spans": [ + { + "bbox": [ + 104, + 326, + 507, + 340 + ], + "score": 1.0, + "content": "[18] Xinyu Gong, Shiyu Chang, Yifan Jiang, and Zhangyang Wang. Autogan: Neural architecture search for", + "type": "text", + "cross_page": true + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 123, + 335, + 508, + 351 + ], + "spans": [ + { + "bbox": [ + 123, + 335, + 508, + 351 + ], + "score": 1.0, + "content": "generative adversarial networks. In Proceedings of the IEEE International Conference on Computer Vision,", + "type": "text", + "cross_page": true + } + ], + "index": 20 + }, + { + "bbox": [ + 123, + 346, + 217, + 358 + ], + "spans": [ + { + "bbox": [ + 123, + 346, + 217, + 358 + ], + "score": 1.0, + "content": "pages 3224–3234, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 21, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 363, + 506, + 377 + ], + "spans": [ + { + "bbox": [ + 104, + 363, + 506, + 377 + ], + "score": 1.0, + "content": "[19] Chen Gao, Yunpeng Chen, Si Liu, Zhenxiong Tan, and Shuicheng Yan. Adversarialnas: Adversarial neural", + "type": "text", + "cross_page": true + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 375, + 506, + 387 + ], + "spans": [ + { + "bbox": [ + 125, + 375, + 506, + 387 + ], + "score": 1.0, + "content": "architecture search for gans. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern", + "type": "text", + "cross_page": true + } + ], + "index": 23 + }, + { + "bbox": [ + 125, + 385, + 264, + 397 + ], + "spans": [ + { + "bbox": [ + 125, + 385, + 264, + 397 + ], + "score": 1.0, + "content": "Recognition, pages 5680–5689, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 24, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 401, + 507, + 415 + ], + "spans": [ + { + "bbox": [ + 104, + 401, + 507, + 415 + ], + "score": 1.0, + "content": "[20] Yuan Tian, Qin Wang, Zhiwu Huang, Wen Li, Dengxin Dai, Minghao Yang, Jun Wang, and Olga Fink. Off-", + "type": "text", + "cross_page": true + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 412, + 506, + 425 + ], + "spans": [ + { + "bbox": [ + 124, + 412, + 506, + 425 + ], + "score": 1.0, + "content": "policy reinforcement learning for efficient and effective gan architecture search. In European Conference", + "type": "text", + "cross_page": true + } + ], + "index": 26 + }, + { + "bbox": [ + 126, + 423, + 318, + 434 + ], + "spans": [ + { + "bbox": [ + 126, + 423, + 318, + 434 + ], + "score": 1.0, + "content": "on Computer Vision, pages 175–192. Springer, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 27, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 440, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 106, + 440, + 505, + 452 + ], + "score": 1.0, + "content": "[21] Han Zhang, Ian Goodfellow, Dimitris Metaxas, and Augustus Odena. Self-attention generative adversarial", + "type": "text", + "cross_page": true + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 450, + 464, + 464 + ], + "spans": [ + { + "bbox": [ + 124, + 450, + 464, + 464 + ], + "score": 1.0, + "content": "networks. In International conference on machine learning, pages 7354–7363. PMLR, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 29, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 469, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 106, + 469, + 506, + 480 + ], + "score": 1.0, + "content": "[22] Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative adversarial", + "type": "text", + "cross_page": true + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 477, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 124, + 477, + 506, + 492 + ], + "score": 1.0, + "content": "networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages", + "type": "text", + "cross_page": true + } + ], + "index": 31 + }, + { + "bbox": [ + 127, + 490, + 193, + 500 + ], + "spans": [ + { + "bbox": [ + 127, + 490, + 193, + 500 + ], + "score": 1.0, + "content": "4401–4410, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 32, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 506, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 506, + 518 + ], + "score": 1.0, + "content": "[23] Patrick Esser, Robin Rombach, and Björn Ommer. Taming transformers for high-resolution image synthesis.", + "type": "text", + "cross_page": true + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 516, + 272, + 528 + ], + "spans": [ + { + "bbox": [ + 125, + 516, + 272, + 528 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2012.09841, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 34, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 534, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 534, + 505, + 546 + ], + "score": 1.0, + "content": "[24] Ian J Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron", + "type": "text", + "cross_page": true + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 543, + 503, + 557 + ], + "spans": [ + { + "bbox": [ + 124, + 543, + 503, + 557 + ], + "score": 1.0, + "content": "Courville, and Yoshua Bengio. Generative adversarial networks. arXiv preprint arXiv:1406.2661, 2014.", + "type": "text", + "cross_page": true + } + ], + "index": 36 + }, + { + "bbox": [ + 104, + 561, + 506, + 576 + ], + "spans": [ + { + "bbox": [ + 104, + 561, + 506, + 576 + ], + "score": 1.0, + "content": "[25] Emily Denton, Soumith Chintala, Arthur Szlam, and Rob Fergus. Deep generative image models using a", + "type": "text", + "cross_page": true + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 571, + 430, + 585 + ], + "spans": [ + { + "bbox": [ + 124, + 571, + 430, + 585 + ], + "score": 1.0, + "content": "laplacian pyramid of adversarial networks. arXiv preprint arXiv:1506.05751, 2015.", + "type": "text", + "cross_page": true + } + ], + "index": 38, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 588, + 506, + 604 + ], + "spans": [ + { + "bbox": [ + 104, + 588, + 506, + 604 + ], + "score": 1.0, + "content": "[26] Alec Radford, Luke Metz, and Soumith Chintala. Unsupervised representation learning with deep", + "type": "text", + "cross_page": true + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 600, + 442, + 612 + ], + "spans": [ + { + "bbox": [ + 126, + 600, + 442, + 612 + ], + "score": 1.0, + "content": "convolutional generative adversarial networks. arXiv preprint arXiv:1511.06434, 2015.", + "type": "text", + "cross_page": true + } + ], + "index": 40, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 618, + 506, + 630 + ], + "spans": [ + { + "bbox": [ + 106, + 618, + 506, + 630 + ], + "score": 1.0, + "content": "[27] Xiaolong Wang, Ross Girshick, Abhinav Gupta, and Kaiming He. Non-local neural networks. In", + "type": "text", + "cross_page": true + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 627, + 506, + 641 + ], + "spans": [ + { + "bbox": [ + 124, + 627, + 506, + 641 + ], + "score": 1.0, + "content": "Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7794–7803, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 646, + 506, + 658 + ], + "spans": [ + { + "bbox": [ + 106, + 646, + 506, + 658 + ], + "score": 1.0, + "content": "[28] Han Hu, Zheng Zhang, Zhenda Xie, and Stephen Lin. Local relation networks for image recognition. In", + "type": "text", + "cross_page": true + } + ], + "index": 43, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 655, + 496, + 669 + ], + "spans": [ + { + "bbox": [ + 125, + 655, + 496, + 669 + ], + "score": 1.0, + "content": "Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3464–3473, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 674, + 506, + 686 + ], + "spans": [ + { + "bbox": [ + 106, + 674, + 506, + 686 + ], + "score": 1.0, + "content": "[29] Yu-Syuan Xu, Shou-Yao Roy Tseng, Yu Tseng, Hsien-Kai Kuo, and Yi-Min Tsai. Unified dynamic", + "type": "text", + "cross_page": true + } + ], + "index": 45, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 682, + 506, + 696 + ], + "spans": [ + { + "bbox": [ + 124, + 682, + 506, + 696 + ], + "score": 1.0, + "content": "convolutional network for super-resolution with variational degradations. In Proceedings of the IEEE/CVF", + "type": "text", + "cross_page": true + } + ], + "index": 46 + }, + { + "bbox": [ + 126, + 694, + 434, + 706 + ], + "spans": [ + { + "bbox": [ + 126, + 694, + 434, + 706 + ], + "score": 1.0, + "content": "Conference on Computer Vision and Pattern Recognition, pages 12496–12505, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 47, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 709, + 482, + 725 + ], + "spans": [ + { + "bbox": [ + 104, + 709, + 482, + 725 + ], + "score": 1.0, + "content": "[30] David Ha, Andrew Dai, and Quoc V Le. Hypernetworks. arXiv preprint arXiv:1609.09106, 2016.", + "type": "text", + "cross_page": true + } + ], + "index": 48, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 73, + 506, + 85 + ], + "spans": [ + { + "bbox": [ + 105, + 73, + 506, + 85 + ], + "score": 1.0, + "content": "[31] Ben Mildenhall, Jonathan T Barron, Jiawen Chen, Dillon Sharlet, Ren Ng, and Robert Carroll. Burst", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 82, + 507, + 97 + ], + "spans": [ + { + "bbox": [ + 125, + 82, + 507, + 97 + ], + "score": 1.0, + "content": "denoising with kernel prediction networks. In Proceedings of the IEEE Conference on Computer Vision", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 125, + 94, + 308, + 106 + ], + "spans": [ + { + "bbox": [ + 125, + 94, + 308, + 106 + ], + "score": 1.0, + "content": "and Pattern Recognition, pages 2502–2510, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 109, + 505, + 126 + ], + "spans": [ + { + "bbox": [ + 105, + 109, + 505, + 126 + ], + "score": 1.0, + "content": "[32] Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey", + "type": "text", + "cross_page": true + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 122, + 489, + 134 + ], + "spans": [ + { + "bbox": [ + 126, + 122, + 489, + 134 + ], + "score": 1.0, + "content": "Zagoruyko. End-to-end object detection with transformers. arXiv preprint arXiv:2005.12872, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 140, + 506, + 152 + ], + "spans": [ + { + "bbox": [ + 106, + 140, + 506, + 152 + ], + "score": 1.0, + "content": "[33] Yanhong Zeng, Jianlong Fu, and Hongyang Chao. Learning joint spatial-temporal transformations for", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 150, + 285, + 162 + ], + "spans": [ + { + "bbox": [ + 126, + 150, + 285, + 162 + ], + "score": 1.0, + "content": "video inpainting. In ECCV. Springer, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 6, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 168, + 505, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 168, + 505, + 180 + ], + "score": 1.0, + "content": "[34] Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 178, + 505, + 190 + ], + "spans": [ + { + "bbox": [ + 126, + 178, + 505, + 190 + ], + "score": 1.0, + "content": "Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 126, + 188, + 486, + 200 + ], + "spans": [ + { + "bbox": [ + 126, + 188, + 486, + 200 + ], + "score": 1.0, + "content": "16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 205, + 506, + 220 + ], + "spans": [ + { + "bbox": [ + 105, + 205, + 506, + 220 + ], + "score": 1.0, + "content": "[35] Drew A Hudson and C Lawrence Zitnick. Generative adversarial transformers. arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 217, + 220, + 229 + ], + "spans": [ + { + "bbox": [ + 125, + 217, + 220, + 229 + ], + "score": 1.0, + "content": "arXiv:2103.01209, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 11, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 235, + 506, + 247 + ], + "spans": [ + { + "bbox": [ + 106, + 235, + 506, + 247 + ], + "score": 1.0, + "content": "[36] Fuzhi Yang, Huan Yang, Jianlong Fu, Hongtao Lu, and Baining Guo. Learning texture transformer network", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 245, + 505, + 257 + ], + "spans": [ + { + "bbox": [ + 126, + 245, + 505, + 257 + ], + "score": 1.0, + "content": "for image super-resolution. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern", + "type": "text", + "cross_page": true + } + ], + "index": 13 + }, + { + "bbox": [ + 124, + 255, + 266, + 267 + ], + "spans": [ + { + "bbox": [ + 124, + 255, + 266, + 267 + ], + "score": 1.0, + "content": "Recognition, pages 5791–5800, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 14, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 271, + 506, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 506, + 286 + ], + "score": 1.0, + "content": "[37] Hanting Chen, Yunhe Wang, Tianyu Guo, Chang Xu, Yiping Deng, Zhenhua Liu, Siwei Ma, Chunjing", + "type": "text", + "cross_page": true + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 282, + 507, + 297 + ], + "spans": [ + { + "bbox": [ + 124, + 282, + 507, + 297 + ], + "score": 1.0, + "content": "Xu, Chao Xu, and Wen Gao. Pre-trained image processing transformer. arXiv preprint arXiv:2012.00364,", + "type": "text", + "cross_page": true + } + ], + "index": 16 + }, + { + "bbox": [ + 125, + 293, + 151, + 306 + ], + "spans": [ + { + "bbox": [ + 125, + 293, + 151, + 306 + ], + "score": 1.0, + "content": "2020.", + "type": "text", + "cross_page": true + } + ], + "index": 17, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 311, + 505, + 323 + ], + "spans": [ + { + "bbox": [ + 106, + 311, + 505, + 323 + ], + "score": 1.0, + "content": "[38] Jie Gui, Zhenan Sun, Yonggang Wen, Dacheng Tao, and Jieping Ye. A review on generative adversarial", + "type": "text", + "cross_page": true + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 321, + 447, + 333 + ], + "spans": [ + { + "bbox": [ + 125, + 321, + 447, + 333 + ], + "score": 1.0, + "content": "networks: Algorithms, theory, and applications. arXiv preprint arXiv:2001.06937, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 19, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 339, + 506, + 351 + ], + "spans": [ + { + "bbox": [ + 106, + 339, + 506, + 351 + ], + "score": 1.0, + "content": "[39] Edgar Schonfeld, Bernt Schiele, and Anna Khoreva. A u-net based discriminator for generative adversarial", + "type": "text", + "cross_page": true + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 349, + 506, + 362 + ], + "spans": [ + { + "bbox": [ + 125, + 349, + 506, + 362 + ], + "score": 1.0, + "content": "networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,", + "type": "text", + "cross_page": true + } + ], + "index": 21 + }, + { + "bbox": [ + 123, + 359, + 218, + 372 + ], + "spans": [ + { + "bbox": [ + 123, + 359, + 218, + 372 + ], + "score": 1.0, + "content": "pages 8207–8216, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 22, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 377, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 377, + 505, + 389 + ], + "score": 1.0, + "content": "[40] Tero Karras, Samuli Laine, Miika Aittala, Janne Hellsten, Jaakko Lehtinen, and Timo Aila. Analyzing and", + "type": "text", + "cross_page": true + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 387, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 124, + 387, + 506, + 401 + ], + "score": 1.0, + "content": "improving the image quality of stylegan. In Proceedings of the IEEE/CVF Conference on Computer Vision", + "type": "text", + "cross_page": true + } + ], + "index": 24 + }, + { + "bbox": [ + 123, + 397, + 309, + 410 + ], + "spans": [ + { + "bbox": [ + 123, + 397, + 309, + 410 + ], + "score": 1.0, + "content": "and Pattern Recognition, pages 8110–8119, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 25, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 416, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 416, + 505, + 428 + ], + "score": 1.0, + "content": "[41] Haotao Wang, Shupeng Gui, Haichuan Yang, Ji Liu, and Zhangyang Wang. Gan slimming: All-in-one gan", + "type": "text", + "cross_page": true + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 424, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 125, + 424, + 505, + 438 + ], + "score": 1.0, + "content": "compression by a unified optimization framework. In European Conference on Computer Vision, pages", + "type": "text", + "cross_page": true + } + ], + "index": 27 + }, + { + "bbox": [ + 126, + 436, + 213, + 448 + ], + "spans": [ + { + "bbox": [ + 126, + 436, + 213, + 448 + ], + "score": 1.0, + "content": "54–73. Springer, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 28, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 453, + 507, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 507, + 467 + ], + "score": 1.0, + "content": "[42] Christian Ledig, Lucas Theis, Ferenc Huszár, Jose Caballero, Andrew Cunningham, Alejandro Acosta,", + "type": "text", + "cross_page": true + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 464, + 506, + 477 + ], + "spans": [ + { + "bbox": [ + 125, + 464, + 506, + 477 + ], + "score": 1.0, + "content": "Andrew Aitken, Alykhan Tejani, Johannes Totz, Zehan Wang, et al. Photo-realistic single image super-", + "type": "text", + "cross_page": true + } + ], + "index": 30 + }, + { + "bbox": [ + 124, + 473, + 506, + 488 + ], + "spans": [ + { + "bbox": [ + 124, + 473, + 506, + 488 + ], + "score": 1.0, + "content": "resolution using a generative adversarial network. In Proceedings of the IEEE conference on computer", + "type": "text", + "cross_page": true + } + ], + "index": 31 + }, + { + "bbox": [ + 124, + 483, + 331, + 497 + ], + "spans": [ + { + "bbox": [ + 124, + 483, + 331, + 497 + ], + "score": 1.0, + "content": "vision and pattern recognition, pages 4681–4690, 2017.", + "type": "text", + "cross_page": true + } + ], + "index": 32, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 502, + 506, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 506, + 514 + ], + "score": 1.0, + "content": "[43] Orest Kupyn, Tetiana Martyniuk, Junru Wu, and Zhangyang Wang. Deblurgan-v2: Deblurring (orders-", + "type": "text", + "cross_page": true + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 512, + 505, + 524 + ], + "spans": [ + { + "bbox": [ + 126, + 512, + 505, + 524 + ], + "score": 1.0, + "content": "of-magnitude) faster and better. In Proceedings of the IEEE/CVF International Conference on Computer", + "type": "text", + "cross_page": true + } + ], + "index": 34 + }, + { + "bbox": [ + 124, + 521, + 244, + 535 + ], + "spans": [ + { + "bbox": [ + 124, + 521, + 244, + 535 + ], + "score": 1.0, + "content": "Vision, pages 8878–8887, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 35, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 539, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 505, + 554 + ], + "score": 1.0, + "content": "[44] Xi Ouyang, Yu Cheng, Yifan Jiang, Chun-Liang Li, and Pan Zhou. Pedestrian-synthesis-gan: Generating", + "type": "text", + "cross_page": true + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 123, + 549, + 424, + 563 + ], + "spans": [ + { + "bbox": [ + 123, + 549, + 424, + 563 + ], + "score": 1.0, + "content": "pedestrian data in real scene and beyond. arXiv preprint arXiv:1804.02047, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 37, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 567, + 506, + 582 + ], + "spans": [ + { + "bbox": [ + 104, + 567, + 506, + 582 + ], + "score": 1.0, + "content": "[45] Jiahui Yu, Zhe Lin, Jimei Yang, Xiaohui Shen, Xin Lu, and Thomas S Huang. Generative image", + "type": "text", + "cross_page": true + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 578, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 125, + 578, + 505, + 590 + ], + "score": 1.0, + "content": "inpainting with contextual attention. In Proceedings of the IEEE conference on computer vision and", + "type": "text", + "cross_page": true + } + ], + "index": 39 + }, + { + "bbox": [ + 123, + 588, + 290, + 601 + ], + "spans": [ + { + "bbox": [ + 123, + 588, + 290, + 601 + ], + "score": 1.0, + "content": "pattern recognition, pages 5505–5514, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 40, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 605, + 507, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 507, + 620 + ], + "score": 1.0, + "content": "[46] Martin Arjovsky, Soumith Chintala, and Léon Bottou. Wasserstein gan. arXiv preprint arXiv:1701.07875,", + "type": "text", + "cross_page": true + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 614, + 151, + 630 + ], + "spans": [ + { + "bbox": [ + 125, + 614, + 151, + 630 + ], + "score": 1.0, + "content": "2017.", + "type": "text", + "cross_page": true + } + ], + "index": 42, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 634, + 506, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 634, + 506, + 649 + ], + "score": 1.0, + "content": "[47] Tianlong Chen, Yu Cheng, Zhe Gan, Jingjing Liu, and Zhangyang Wang. Ultra-data-efficient gan training:", + "type": "text", + "cross_page": true + } + ], + "index": 43, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 644, + 467, + 659 + ], + "spans": [ + { + "bbox": [ + 125, + 644, + 467, + 659 + ], + "score": 1.0, + "content": "Drawing a lottery ticket first, then training it toughly. arXiv preprint arXiv:2103.00397, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 44, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 663, + 505, + 675 + ], + "spans": [ + { + "bbox": [ + 106, + 663, + 505, + 675 + ], + "score": 1.0, + "content": "[48] Takeru Miyato, Toshiki Kataoka, Masanori Koyama, and Yuichi Yoshida. Spectral normalization for", + "type": "text", + "cross_page": true + } + ], + "index": 45, + "is_list_start_line": true + }, + { + "bbox": [ + 123, + 673, + 392, + 686 + ], + "spans": [ + { + "bbox": [ + 123, + 673, + 392, + 686 + ], + "score": 1.0, + "content": "generative adversarial networks. arXiv preprint arXiv:1802.05957, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 46, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 691, + 505, + 703 + ], + "spans": [ + { + "bbox": [ + 106, + 691, + 505, + 703 + ], + "score": 1.0, + "content": "[49] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz", + "type": "text", + "cross_page": true + } + ], + "index": 47, + "is_list_start_line": true + }, + { + "bbox": [ + 123, + 700, + 506, + 715 + ], + "spans": [ + { + "bbox": [ + 123, + 700, + 506, + 715 + ], + "score": 1.0, + "content": "Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in neural information processing", + "type": "text", + "cross_page": true + } + ], + "index": 48 + }, + { + "bbox": [ + 123, + 711, + 249, + 724 + ], + "spans": [ + { + "bbox": [ + 123, + 711, + 249, + 724 + ], + "score": 1.0, + "content": "systems, pages 5998–6008, 2017.", + "type": "text", + "cross_page": true + } + ], + "index": 49, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 73, + 505, + 85 + ], + "spans": [ + { + "bbox": [ + 106, + 73, + 505, + 85 + ], + "score": 1.0, + "content": "[50] Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Hervé", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 123, + 82, + 506, + 97 + ], + "spans": [ + { + "bbox": [ + 123, + 82, + 506, + 97 + ], + "score": 1.0, + "content": "Jégou. Training data-efficient image transformers & distillation through attention. arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 126, + 93, + 220, + 106 + ], + "spans": [ + { + "bbox": [ + 126, + 93, + 220, + 106 + ], + "score": 1.0, + "content": "arXiv:2012.12877, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 109, + 506, + 125 + ], + "spans": [ + { + "bbox": [ + 104, + 109, + 506, + 125 + ], + "score": 1.0, + "content": "[51] Hugo Touvron, Matthieu Cord, Alexandre Sablayrolles, Gabriel Synnaeve, and Hervé Jégou. Going deeper", + "type": "text", + "cross_page": true + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 121, + 366, + 133 + ], + "spans": [ + { + "bbox": [ + 126, + 121, + 366, + 133 + ], + "score": 1.0, + "content": "with image transformers. arXiv preprint arXiv:2103.17239, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 138, + 506, + 151 + ], + "spans": [ + { + "bbox": [ + 106, + 138, + 506, + 151 + ], + "score": 1.0, + "content": "[52] Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 146, + 507, + 162 + ], + "spans": [ + { + "bbox": [ + 124, + 146, + 507, + 162 + ], + "score": 1.0, + "content": "transformer: Hierarchical vision transformer using shifted windows. arXiv preprint arXiv:2103.14030,", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 125, + 156, + 151, + 172 + ], + "spans": [ + { + "bbox": [ + 125, + 156, + 151, + 172 + ], + "score": 1.0, + "content": "2021.", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 174, + 506, + 188 + ], + "spans": [ + { + "bbox": [ + 104, + 174, + 506, + 188 + ], + "score": 1.0, + "content": "[53] Chun-Fu Chen, Quanfu Fan, and Rameswar Panda. Crossvit: Cross-attention multi-scale vision transformer", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 185, + 363, + 198 + ], + "spans": [ + { + "bbox": [ + 124, + 185, + 363, + 198 + ], + "score": 1.0, + "content": "for image classification. arXiv preprint arXiv:2103.14899, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 201, + 507, + 217 + ], + "spans": [ + { + "bbox": [ + 104, + 201, + 507, + 217 + ], + "score": 1.0, + "content": "[54] Pengchuan Zhang, Xiyang Dai, Jianwei Yang, Bin Xiao, Lu Yuan, Lei Zhang, and Jianfeng Gao. Multi-", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 213, + 505, + 225 + ], + "spans": [ + { + "bbox": [ + 126, + 213, + 505, + 225 + ], + "score": 1.0, + "content": "scale vision longformer: A new vision transformer for high-resolution image encoding. arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 126, + 223, + 220, + 235 + ], + "spans": [ + { + "bbox": [ + 126, + 223, + 220, + 235 + ], + "score": 1.0, + "content": "arXiv:2103.15358, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 239, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 506, + 254 + ], + "score": 1.0, + "content": "[55] Changlin Li, Tao Tang, Guangrun Wang, Jiefeng Peng, Bing Wang, Xiaodan Liang, and Xiaojun Chang.", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 249, + 506, + 262 + ], + "spans": [ + { + "bbox": [ + 125, + 249, + 506, + 262 + ], + "score": 1.0, + "content": "Bossnas: Exploring hybrid cnn-transformers with block-wisely self-supervised neural architecture search.", + "type": "text", + "cross_page": true + } + ], + "index": 14 + }, + { + "bbox": [ + 126, + 260, + 274, + 272 + ], + "spans": [ + { + "bbox": [ + 126, + 260, + 274, + 272 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2103.12424, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 15, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 278, + 505, + 290 + ], + "spans": [ + { + "bbox": [ + 106, + 278, + 505, + 290 + ], + "score": 1.0, + "content": "[56] Yutong Xie, Jianpeng Zhang, Chunhua Shen, and Yong Xia. Cotr: Efficiently bridging cnn and transformer", + "type": "text", + "cross_page": true + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 287, + 405, + 302 + ], + "spans": [ + { + "bbox": [ + 124, + 287, + 405, + 302 + ], + "score": 1.0, + "content": "for 3d medical image segmentation. arXiv preprint arXiv:2103.03024, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 17, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 306, + 506, + 318 + ], + "spans": [ + { + "bbox": [ + 106, + 306, + 506, + 318 + ], + "score": 1.0, + "content": "[57] Hengshuang Zhao, Li Jiang, Jiaya Jia, Philip Torr, and Vladlen Koltun. Point transformer. arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 315, + 220, + 327 + ], + "spans": [ + { + "bbox": [ + 125, + 315, + 220, + 327 + ], + "score": 1.0, + "content": "arXiv:2012.09164, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 19, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 331, + 506, + 347 + ], + "spans": [ + { + "bbox": [ + 104, + 331, + 506, + 347 + ], + "score": 1.0, + "content": "[58] Sixiao Zheng, Jiachen Lu, Hengshuang Zhao, Xiatian Zhu, Zekun Luo, Yabiao Wang, Yanwei Fu, Jianfeng", + "type": "text", + "cross_page": true + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 341, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 125, + 341, + 506, + 356 + ], + "score": 1.0, + "content": "Feng, Tao Xiang, Philip HS Torr, et al. Rethinking semantic segmentation from a sequence-to-sequence", + "type": "text", + "cross_page": true + } + ], + "index": 21 + }, + { + "bbox": [ + 123, + 352, + 385, + 365 + ], + "spans": [ + { + "bbox": [ + 123, + 352, + 385, + 365 + ], + "score": 1.0, + "content": "perspective with transformers. arXiv preprint arXiv:2012.15840, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 22, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 369, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 506, + 383 + ], + "score": 1.0, + "content": "[59] Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, and Jifeng Dai. Deformable detr: Deformable", + "type": "text", + "cross_page": true + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 380, + 437, + 393 + ], + "spans": [ + { + "bbox": [ + 126, + 380, + 437, + 393 + ], + "score": 1.0, + "content": "transformers for end-to-end object detection. arXiv preprint arXiv:2010.04159, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 24, + "is_list_end_line": true + }, + { + "bbox": [ + 103, + 397, + 507, + 410 + ], + "spans": [ + { + "bbox": [ + 103, + 397, + 507, + 410 + ], + "score": 1.0, + "content": "[60] Kai Han, Yunhe Wang, Hanting Chen, Xinghao Chen, Jianyuan Guo, Zhenhua Liu, Yehui Tang, An Xiao,", + "type": "text", + "cross_page": true + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 406, + 504, + 421 + ], + "spans": [ + { + "bbox": [ + 125, + 406, + 504, + 421 + ], + "score": 1.0, + "content": "Chunjing Xu, Yixing Xu, et al. A survey on visual transformer. arXiv preprint arXiv:2012.12556, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 425, + 506, + 437 + ], + "spans": [ + { + "bbox": [ + 106, + 425, + 506, + 437 + ], + "score": 1.0, + "content": "[61] Niki Parmar, Ashish Vaswani, Jakob Uszkoreit, Lukasz Kaiser, Noam Shazeer, Alexander Ku, and Dustin", + "type": "text", + "cross_page": true + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 434, + 506, + 448 + ], + "spans": [ + { + "bbox": [ + 125, + 434, + 506, + 448 + ], + "score": 1.0, + "content": "Tran. Image transformer. In International Conference on Machine Learning, pages 4055–4064. PMLR,", + "type": "text", + "cross_page": true + } + ], + "index": 28 + }, + { + "bbox": [ + 125, + 444, + 151, + 457 + ], + "spans": [ + { + "bbox": [ + 125, + 444, + 151, + 457 + ], + "score": 1.0, + "content": "2018.", + "type": "text", + "cross_page": true + } + ], + "index": 29, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 461, + 506, + 476 + ], + "spans": [ + { + "bbox": [ + 104, + 461, + 506, + 476 + ], + "score": 1.0, + "content": "[62] Rewon Child, Scott Gray, Alec Radford, and Ilya Sutskever. Generating long sequences with sparse", + "type": "text", + "cross_page": true + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 472, + 324, + 484 + ], + "spans": [ + { + "bbox": [ + 125, + 472, + 324, + 484 + ], + "score": 1.0, + "content": "transformers. arXiv preprint arXiv:1904.10509, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 31, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 490, + 506, + 502 + ], + "spans": [ + { + "bbox": [ + 106, + 490, + 506, + 502 + ], + "score": 1.0, + "content": "[63] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirec-", + "type": "text", + "cross_page": true + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 500, + 446, + 512 + ], + "spans": [ + { + "bbox": [ + 125, + 500, + 446, + 512 + ], + "score": 1.0, + "content": "tional transformers for language understanding. arXiv preprint arXiv:1810.04805, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 33, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 516, + 506, + 529 + ], + "spans": [ + { + "bbox": [ + 104, + 516, + 506, + 529 + ], + "score": 1.0, + "content": "[64] Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, and Vikas", + "type": "text", + "cross_page": true + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 527, + 506, + 541 + ], + "spans": [ + { + "bbox": [ + 125, + 527, + 506, + 541 + ], + "score": 1.0, + "content": "Singh. Nystromformer: A nystrom-based algorithm for approximating self-attention. arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 35 + }, + { + "bbox": [ + 126, + 538, + 219, + 548 + ], + "spans": [ + { + "bbox": [ + 126, + 538, + 219, + 548 + ], + "score": 1.0, + "content": "arXiv:2102.03902, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 36, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 554, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 106, + 554, + 506, + 567 + ], + "score": 1.0, + "content": "[65] Iz Beltagy, Matthew E Peters, and Arman Cohan. Longformer: The long-document transformer. arXiv", + "type": "text", + "cross_page": true + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 123, + 564, + 252, + 578 + ], + "spans": [ + { + "bbox": [ + 123, + 564, + 252, + 578 + ], + "score": 1.0, + "content": "preprint arXiv:2004.05150, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 38, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 581, + 506, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 506, + 595 + ], + "score": 1.0, + "content": "[66] Gedas Bertasius, Heng Wang, and Lorenzo Torresani. Is space-time attention all you need for video", + "type": "text", + "cross_page": true + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 591, + 334, + 605 + ], + "spans": [ + { + "bbox": [ + 124, + 591, + 334, + 605 + ], + "score": 1.0, + "content": "understanding? arXiv preprint arXiv:2102.05095, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 40, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 609, + 505, + 622 + ], + "spans": [ + { + "bbox": [ + 106, + 609, + 505, + 622 + ], + "score": 1.0, + "content": "[67] Manoj Kumar, Dirk Weissenborn, and Nal Kalchbrenner. Colorization transformer. arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 619, + 219, + 631 + ], + "spans": [ + { + "bbox": [ + 126, + 619, + 219, + 631 + ], + "score": 1.0, + "content": "arXiv:2102.04432, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 42, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 637, + 506, + 649 + ], + "spans": [ + { + "bbox": [ + 106, + 637, + 506, + 649 + ], + "score": 1.0, + "content": "[68] Chen Sun, Abhinav Shrivastava, Saurabh Singh, and Abhinav Gupta. Revisiting unreasonable effectiveness", + "type": "text", + "cross_page": true + } + ], + "index": 43, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 647, + 506, + 659 + ], + "spans": [ + { + "bbox": [ + 125, + 647, + 506, + 659 + ], + "score": 1.0, + "content": "of data in deep learning era. In Proceedings of the IEEE international conference on computer vision,", + "type": "text", + "cross_page": true + } + ], + "index": 44 + }, + { + "bbox": [ + 124, + 656, + 209, + 669 + ], + "spans": [ + { + "bbox": [ + 124, + 656, + 209, + 669 + ], + "score": 1.0, + "content": "pages 843–852, 2017.", + "type": "text", + "cross_page": true + } + ], + "index": 45, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 674, + 506, + 686 + ], + "spans": [ + { + "bbox": [ + 106, + 674, + 506, + 686 + ], + "score": 1.0, + "content": "[69] Shengyu Zhao, Zhijian Liu, Ji Lin, Jun-Yan Zhu, and Song Han. Differentiable augmentation for data-", + "type": "text", + "cross_page": true + } + ], + "index": 46, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 684, + 353, + 696 + ], + "spans": [ + { + "bbox": [ + 126, + 684, + 353, + 696 + ], + "score": 1.0, + "content": "efficient gan training. arXiv preprint arXiv:2006.10738, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 47, + "is_list_end_line": true + }, + { + "bbox": [ + 103, + 699, + 508, + 716 + ], + "spans": [ + { + "bbox": [ + 103, + 699, + 508, + 716 + ], + "score": 1.0, + "content": "[70] Tero Karras, Miika Aittala, Janne Hellsten, Samuli Laine, Jaakko Lehtinen, and Timo Aila. Training", + "type": "text", + "cross_page": true + } + ], + "index": 48, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 711, + 456, + 725 + ], + "spans": [ + { + "bbox": [ + 124, + 711, + 456, + 725 + ], + "score": 1.0, + "content": "generative adversarial networks with limited data. arXiv preprint arXiv:2006.06676, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 49, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 72, + 507, + 87 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 507, + 87 + ], + "score": 1.0, + "content": "[71] Peter Shaw, Jakob Uszkoreit, and Ashish Vaswani. Self-attention with relative position representations.", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 83, + 274, + 96 + ], + "spans": [ + { + "bbox": [ + 124, + 83, + 274, + 96 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1803.02155, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 1, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 101, + 505, + 114 + ], + "spans": [ + { + "bbox": [ + 106, + 101, + 505, + 114 + ], + "score": 1.0, + "content": "[72] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou,", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 110, + 507, + 125 + ], + "spans": [ + { + "bbox": [ + 125, + 110, + 507, + 125 + ], + "score": 1.0, + "content": "Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer.", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 125, + 121, + 274, + 134 + ], + "spans": [ + { + "bbox": [ + 125, + 121, + 274, + 134 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1910.10683, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 139, + 506, + 152 + ], + "spans": [ + { + "bbox": [ + 106, + 139, + 506, + 152 + ], + "score": 1.0, + "content": "[73] Cheng-Zhi Anna Huang, Ashish Vaswani, Jakob Uszkoreit, Noam Shazeer, Curtis Hawthorne, AM Dai,", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 149, + 505, + 162 + ], + "spans": [ + { + "bbox": [ + 126, + 149, + 505, + 162 + ], + "score": 1.0, + "content": "MD Hoffman, and D Eck. Music transformer: Generating music with long-term structure (2018). arXiv", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 124, + 160, + 251, + 171 + ], + "spans": [ + { + "bbox": [ + 124, + 160, + 251, + 171 + ], + "score": 1.0, + "content": "preprint arXiv:1809.04281, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 177, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 177, + 506, + 190 + ], + "score": 1.0, + "content": "[74] Chieh Hubert Lin, Chia-Che Chang, Yu-Sheng Chen, Da-Cheng Juan, Wei Wei, and Hwann-Tzong Chen.", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 186, + 507, + 201 + ], + "spans": [ + { + "bbox": [ + 126, + 186, + 507, + 201 + ], + "score": 1.0, + "content": "Coco-gan: Generation by parts via conditional coordinating. In Proceedings of the IEEE/CVF International", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 126, + 197, + 336, + 210 + ], + "spans": [ + { + "bbox": [ + 126, + 197, + 336, + 210 + ], + "score": 1.0, + "content": "Conference on Computer Vision, pages 4512–4521, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 215, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 506, + 228 + ], + "score": 1.0, + "content": "[75] Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional", + "type": "text", + "cross_page": true + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 224, + 457, + 238 + ], + "spans": [ + { + "bbox": [ + 124, + 224, + 457, + 238 + ], + "score": 1.0, + "content": "neural networks. Advances in neural information processing systems, 25:1097–1105, 2012.", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 241, + 507, + 257 + ], + "spans": [ + { + "bbox": [ + 104, + 241, + 507, + 257 + ], + "score": 1.0, + "content": "[76] Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. Layer normalization. arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 253, + 220, + 265 + ], + "spans": [ + { + "bbox": [ + 125, + 253, + 220, + 265 + ], + "score": 1.0, + "content": "arXiv:1607.06450, 2016.", + "type": "text", + "cross_page": true + } + ], + "index": 14, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 271, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 505, + 284 + ], + "score": 1.0, + "content": "[77] Sergey Ioffe and Christian Szegedy. Batch normalization: Accelerating deep network training by reducing", + "type": "text", + "cross_page": true + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 280, + 504, + 295 + ], + "spans": [ + { + "bbox": [ + 124, + 280, + 504, + 295 + ], + "score": 1.0, + "content": "internal covariate shift. In International conference on machine learning, pages 448–456. PMLR, 2015.", + "type": "text", + "cross_page": true + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 299, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 299, + 505, + 312 + ], + "score": 1.0, + "content": "[78] Dmitry Ulyanov, Andrea Vedaldi, and Victor Lempitsky. Instance normalization: The missing ingredient", + "type": "text", + "cross_page": true + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 308, + 344, + 322 + ], + "spans": [ + { + "bbox": [ + 125, + 308, + 344, + 322 + ], + "score": 1.0, + "content": "for fast stylization. arXiv preprint arXiv:1607.08022, 2016.", + "type": "text", + "cross_page": true + } + ], + "index": 18, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 326, + 495, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 326, + 495, + 341 + ], + "score": 1.0, + "content": "[79] Alex Krizhevsky, Geoffrey Hinton, et al. Learning multiple layers of features from tiny images. 2009.", + "type": "text", + "cross_page": true + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 106, + 344, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 344, + 505, + 357 + ], + "score": 1.0, + "content": "[80] Adam Coates, Andrew Ng, and Honglak Lee. An analysis of single-layer networks in unsupervised feature", + "type": "text", + "cross_page": true + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 353, + 507, + 369 + ], + "spans": [ + { + "bbox": [ + 124, + 353, + 507, + 369 + ], + "score": 1.0, + "content": "learning. In Proceedings of the fourteenth international conference on artificial intelligence and statistics,", + "type": "text", + "cross_page": true + } + ], + "index": 21 + }, + { + "bbox": [ + 123, + 363, + 380, + 379 + ], + "spans": [ + { + "bbox": [ + 123, + 363, + 380, + 379 + ], + "score": 1.0, + "content": "pages 215–223. JMLR Workshop and Conference Proceedings, 2011.", + "type": "text", + "cross_page": true + } + ], + "index": 22, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 383, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 506, + 396 + ], + "score": 1.0, + "content": "[81] Ziwei Liu, Ping Luo, Xiaogang Wang, and Xiaoou Tang. Deep learning face attributes in the wild. In", + "type": "text", + "cross_page": true + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 392, + 402, + 405 + ], + "spans": [ + { + "bbox": [ + 126, + 392, + 402, + 405 + ], + "score": 1.0, + "content": "Proceedings of International Conference on Computer Vision (ICCV), 2015.", + "type": "text", + "cross_page": true + } + ], + "index": 24, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 410, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 106, + 410, + 505, + 423 + ], + "score": 1.0, + "content": "[82] Fisher Yu, Ari Seff, Yinda Zhang, Shuran Song, Thomas Funkhouser, and Jianxiong Xiao. Lsun: Con-", + "type": "text", + "cross_page": true + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 421, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 126, + 421, + 506, + 434 + ], + "score": 1.0, + "content": "struction of a large-scale image dataset using deep learning with humans in the loop. arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 26 + }, + { + "bbox": [ + 126, + 430, + 219, + 443 + ], + "spans": [ + { + "bbox": [ + 126, + 430, + 219, + 443 + ], + "score": 1.0, + "content": "arXiv:1506.03365, 2015.", + "type": "text", + "cross_page": true + } + ], + "index": 27, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 448, + 505, + 461 + ], + "spans": [ + { + "bbox": [ + 106, + 448, + 505, + 461 + ], + "score": 1.0, + "content": "[83] Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans", + "type": "text", + "cross_page": true + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 457, + 506, + 471 + ], + "spans": [ + { + "bbox": [ + 124, + 457, + 506, + 471 + ], + "score": 1.0, + "content": "trained by a two time-scale update rule converge to a local nash equilibrium. arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 29 + }, + { + "bbox": [ + 125, + 468, + 219, + 480 + ], + "spans": [ + { + "bbox": [ + 125, + 468, + 219, + 480 + ], + "score": 1.0, + "content": "arXiv:1706.08500, 2017.", + "type": "text", + "cross_page": true + } + ], + "index": 30, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 486, + 506, + 499 + ], + "spans": [ + { + "bbox": [ + 106, + 486, + 506, + 499 + ], + "score": 1.0, + "content": "[84] Animesh Karnewar and Oliver Wang. Msg-gan: Multi-scale gradients for generative adversarial networks.", + "type": "text", + "cross_page": true + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 496, + 506, + 509 + ], + "spans": [ + { + "bbox": [ + 126, + 496, + 506, + 509 + ], + "score": 1.0, + "content": "In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7799–", + "type": "text", + "cross_page": true + } + ], + "index": 32 + }, + { + "bbox": [ + 126, + 506, + 171, + 518 + ], + "spans": [ + { + "bbox": [ + 126, + 506, + 171, + 518 + ], + "score": 1.0, + "content": "7808, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 33, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 523, + 506, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 506, + 538 + ], + "score": 1.0, + "content": "[85] Ting Chen, Mario Lucic, Neil Houlsby, and Sylvain Gelly. On self modulation for generative adversarial", + "type": "text", + "cross_page": true + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 534, + 311, + 547 + ], + "spans": [ + { + "bbox": [ + 126, + 534, + 311, + 547 + ], + "score": 1.0, + "content": "networks. arXiv preprint arXiv:1810.01365, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 35, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 552, + 506, + 565 + ], + "spans": [ + { + "bbox": [ + 106, + 552, + 506, + 565 + ], + "score": 1.0, + "content": "[86] Long Zhao, Zizhao Zhang, Ting Chen, Dimitris N Metaxas, and Han Zhang. Improved transformer for", + "type": "text", + "cross_page": true + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 562, + 352, + 575 + ], + "spans": [ + { + "bbox": [ + 126, + 562, + 352, + 575 + ], + "score": 1.0, + "content": "high-resolution gans. arXiv preprint arXiv:2106.07631, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 37, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 581, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 106, + 581, + 506, + 592 + ], + "score": 1.0, + "content": "[87] Tero Karras, Miika Aittala, Samuli Laine, Erik Härkönen, Janne Hellsten, Jaakko Lehtinen, and Timo Aila.", + "type": "text", + "cross_page": true + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 589, + 429, + 602 + ], + "spans": [ + { + "bbox": [ + 126, + 589, + 429, + 602 + ], + "score": 1.0, + "content": "Alias-free generative adversarial networks. arXiv preprint arXiv:2106.12423, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 39, + "is_list_end_line": true + } + ], + "index": 35.5, + "bbox_fs": [ + 109, + 393, + 507, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 102, + 41, + 507, + 730 + ], + "lines": [ + { + "bbox": [ + 105, + 74, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 74, + 506, + 86 + ], + "score": 1.0, + "content": "[10] Han Zhang, Zizhao Zhang, Augustus Odena, and Honglak Lee. Consistency regularization for generative", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 125, + 83, + 352, + 95 + ], + "spans": [ + { + "bbox": [ + 125, + 83, + 352, + 95 + ], + "score": 1.0, + "content": "adversarial networks. arXiv preprint arXiv:1910.12027, 2019.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 104, + 100, + 506, + 114 + ], + "spans": [ + { + "bbox": [ + 104, + 100, + 506, + 114 + ], + "score": 1.0, + "content": "[11] Lars Mescheder, Andreas Geiger, and Sebastian Nowozin. Which training methods for gans do actually", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 123, + 110, + 314, + 124 + ], + "spans": [ + { + "bbox": [ + 123, + 110, + 314, + 124 + ], + "score": 1.0, + "content": "converge? arXiv preprint arXiv:1801.04406, 2018.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 129, + 506, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 129, + 506, + 141 + ], + "score": 1.0, + "content": "[12] Xudong Mao, Qing Li, Haoran Xie, Raymond YK Lau, Zhen Wang, and Stephen Paul Smolley. Least", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 125, + 138, + 506, + 153 + ], + "spans": [ + { + "bbox": [ + 125, + 138, + 506, + 153 + ], + "score": 1.0, + "content": "squares generative adversarial networks. In Proceedings of the IEEE international conference on computer", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 124, + 148, + 244, + 162 + ], + "spans": [ + { + "bbox": [ + 124, + 148, + 244, + 162 + ], + "score": 1.0, + "content": "vision, pages 2794–2802, 2017.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 167, + 506, + 179 + ], + "spans": [ + { + "bbox": [ + 105, + 167, + 506, + 179 + ], + "score": 1.0, + "content": "[13] Alexia Jolicoeur-Martineau. The relativistic discriminator: a key element missing from standard gan. arXiv", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 123, + 177, + 251, + 190 + ], + "spans": [ + { + "bbox": [ + 123, + 177, + 251, + 190 + ], + "score": 1.0, + "content": "preprint arXiv:1807.00734, 2018.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 104, + 194, + 506, + 208 + ], + "spans": [ + { + "bbox": [ + 104, + 194, + 506, + 208 + ], + "score": 1.0, + "content": "[14] Chun-Liang Li, Wei-Cheng Chang, Yu Cheng, Yiming Yang, and Barnabás Póczos. Mmd gan: Towards", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 125, + 204, + 506, + 218 + ], + "spans": [ + { + "bbox": [ + 125, + 204, + 506, + 218 + ], + "score": 1.0, + "content": "deeper understanding of moment matching network. In Advances in neural information processing systems,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 123, + 214, + 217, + 227 + ], + "spans": [ + { + "bbox": [ + 123, + 214, + 217, + 227 + ], + "score": 1.0, + "content": "pages 2203–2213, 2017.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 232, + 506, + 246 + ], + "spans": [ + { + "bbox": [ + 104, + 232, + 506, + 246 + ], + "score": 1.0, + "content": "[15] Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen. Improved", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 124, + 242, + 380, + 255 + ], + "spans": [ + { + "bbox": [ + 124, + 242, + 380, + 255 + ], + "score": 1.0, + "content": "techniques for training gans. arXiv preprint arXiv:1606.03498, 2016.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 104, + 260, + 506, + 275 + ], + "spans": [ + { + "bbox": [ + 104, + 260, + 506, + 275 + ], + "score": 1.0, + "content": "[16] Tero Karras, Timo Aila, Samuli Laine, and Jaakko Lehtinen. Progressive growing of gans for improved", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 124, + 271, + 388, + 284 + ], + "spans": [ + { + "bbox": [ + 124, + 271, + 388, + 284 + ], + "score": 1.0, + "content": "quality, stability, and variation. arXiv preprint arXiv:1710.10196, 2017.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 289, + 506, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 506, + 301 + ], + "score": 1.0, + "content": "[17] Mario Lucic, Karol Kurach, Marcin Michalski, Olivier Bousquet, and Sylvain Gelly. Are gans created", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 126, + 299, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 126, + 299, + 505, + 311 + ], + "score": 1.0, + "content": "equal? a large-scale study. In Proceedings of the 32nd International Conference on Neural Information", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 123, + 308, + 283, + 321 + ], + "spans": [ + { + "bbox": [ + 123, + 308, + 283, + 321 + ], + "score": 1.0, + "content": "Processing Systems, pages 698–707, 2018.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 104, + 326, + 507, + 340 + ], + "spans": [ + { + "bbox": [ + 104, + 326, + 507, + 340 + ], + "score": 1.0, + "content": "[18] Xinyu Gong, Shiyu Chang, Yifan Jiang, and Zhangyang Wang. Autogan: Neural architecture search for", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 123, + 335, + 508, + 351 + ], + "spans": [ + { + "bbox": [ + 123, + 335, + 508, + 351 + ], + "score": 1.0, + "content": "generative adversarial networks. In Proceedings of the IEEE International Conference on Computer Vision,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 123, + 346, + 217, + 358 + ], + "spans": [ + { + "bbox": [ + 123, + 346, + 217, + 358 + ], + "score": 1.0, + "content": "pages 3224–3234, 2019.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 104, + 363, + 506, + 377 + ], + "spans": [ + { + "bbox": [ + 104, + 363, + 506, + 377 + ], + "score": 1.0, + "content": "[19] Chen Gao, Yunpeng Chen, Si Liu, Zhenxiong Tan, and Shuicheng Yan. Adversarialnas: Adversarial neural", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 125, + 375, + 506, + 387 + ], + "spans": [ + { + "bbox": [ + 125, + 375, + 506, + 387 + ], + "score": 1.0, + "content": "architecture search for gans. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 125, + 385, + 264, + 397 + ], + "spans": [ + { + "bbox": [ + 125, + 385, + 264, + 397 + ], + "score": 1.0, + "content": "Recognition, pages 5680–5689, 2020.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 401, + 507, + 415 + ], + "spans": [ + { + "bbox": [ + 104, + 401, + 507, + 415 + ], + "score": 1.0, + "content": "[20] Yuan Tian, Qin Wang, Zhiwu Huang, Wen Li, Dengxin Dai, Minghao Yang, Jun Wang, and Olga Fink. Off-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 124, + 412, + 506, + 425 + ], + "spans": [ + { + "bbox": [ + 124, + 412, + 506, + 425 + ], + "score": 1.0, + "content": "policy reinforcement learning for efficient and effective gan architecture search. In European Conference", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 126, + 423, + 318, + 434 + ], + "spans": [ + { + "bbox": [ + 126, + 423, + 318, + 434 + ], + "score": 1.0, + "content": "on Computer Vision, pages 175–192. Springer, 2020.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 440, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 106, + 440, + 505, + 452 + ], + "score": 1.0, + "content": "[21] Han Zhang, Ian Goodfellow, Dimitris Metaxas, and Augustus Odena. Self-attention generative adversarial", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 124, + 450, + 464, + 464 + ], + "spans": [ + { + "bbox": [ + 124, + 450, + 464, + 464 + ], + "score": 1.0, + "content": "networks. In International conference on machine learning, pages 7354–7363. PMLR, 2019.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 469, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 106, + 469, + 506, + 480 + ], + "score": 1.0, + "content": "[22] Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative adversarial", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 124, + 477, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 124, + 477, + 506, + 492 + ], + "score": 1.0, + "content": "networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 127, + 490, + 193, + 500 + ], + "spans": [ + { + "bbox": [ + 127, + 490, + 193, + 500 + ], + "score": 1.0, + "content": "4401–4410, 2019.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 506, + 506, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 506, + 518 + ], + "score": 1.0, + "content": "[23] Patrick Esser, Robin Rombach, and Björn Ommer. Taming transformers for high-resolution image synthesis.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 125, + 516, + 272, + 528 + ], + "spans": [ + { + "bbox": [ + 125, + 516, + 272, + 528 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2012.09841, 2020.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 534, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 534, + 505, + 546 + ], + "score": 1.0, + "content": "[24] Ian J Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 124, + 543, + 503, + 557 + ], + "spans": [ + { + "bbox": [ + 124, + 543, + 503, + 557 + ], + "score": 1.0, + "content": "Courville, and Yoshua Bengio. Generative adversarial networks. arXiv preprint arXiv:1406.2661, 2014.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 104, + 561, + 506, + 576 + ], + "spans": [ + { + "bbox": [ + 104, + 561, + 506, + 576 + ], + "score": 1.0, + "content": "[25] Emily Denton, Soumith Chintala, Arthur Szlam, and Rob Fergus. Deep generative image models using a", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 124, + 571, + 430, + 585 + ], + "spans": [ + { + "bbox": [ + 124, + 571, + 430, + 585 + ], + "score": 1.0, + "content": "laplacian pyramid of adversarial networks. arXiv preprint arXiv:1506.05751, 2015.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 104, + 588, + 506, + 604 + ], + "spans": [ + { + "bbox": [ + 104, + 588, + 506, + 604 + ], + "score": 1.0, + "content": "[26] Alec Radford, Luke Metz, and Soumith Chintala. Unsupervised representation learning with deep", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 126, + 600, + 442, + 612 + ], + "spans": [ + { + "bbox": [ + 126, + 600, + 442, + 612 + ], + "score": 1.0, + "content": "convolutional generative adversarial networks. arXiv preprint arXiv:1511.06434, 2015.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 618, + 506, + 630 + ], + "spans": [ + { + "bbox": [ + 106, + 618, + 506, + 630 + ], + "score": 1.0, + "content": "[27] Xiaolong Wang, Ross Girshick, Abhinav Gupta, and Kaiming He. Non-local neural networks. In", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 124, + 627, + 506, + 641 + ], + "spans": [ + { + "bbox": [ + 124, + 627, + 506, + 641 + ], + "score": 1.0, + "content": "Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7794–7803, 2018.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 646, + 506, + 658 + ], + "spans": [ + { + "bbox": [ + 106, + 646, + 506, + 658 + ], + "score": 1.0, + "content": "[28] Han Hu, Zheng Zhang, Zhenda Xie, and Stephen Lin. Local relation networks for image recognition. In", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 125, + 655, + 496, + 669 + ], + "spans": [ + { + "bbox": [ + 125, + 655, + 496, + 669 + ], + "score": 1.0, + "content": "Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3464–3473, 2019.", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 674, + 506, + 686 + ], + "spans": [ + { + "bbox": [ + 106, + 674, + 506, + 686 + ], + "score": 1.0, + "content": "[29] Yu-Syuan Xu, Shou-Yao Roy Tseng, Yu Tseng, Hsien-Kai Kuo, and Yi-Min Tsai. Unified dynamic", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 124, + 682, + 506, + 696 + ], + "spans": [ + { + "bbox": [ + 124, + 682, + 506, + 696 + ], + "score": 1.0, + "content": "convolutional network for super-resolution with variational degradations. In Proceedings of the IEEE/CVF", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 126, + 694, + 434, + 706 + ], + "spans": [ + { + "bbox": [ + 126, + 694, + 434, + 706 + ], + "score": 1.0, + "content": "Conference on Computer Vision and Pattern Recognition, pages 12496–12505, 2020.", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 104, + 709, + 482, + 725 + ], + "spans": [ + { + "bbox": [ + 104, + 709, + 482, + 725 + ], + "score": 1.0, + "content": "[30] David Ha, Andrew Dai, and Quoc V Le. Hypernetworks. arXiv preprint arXiv:1609.09106, 2016.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 24 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 310, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 312, + 755 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 312, + 755 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 102, + 41, + 507, + 730 + ], + "lines": [], + "index": 24, + "bbox_fs": [ + 104, + 74, + 508, + 725 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 38, + 507, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 73, + 506, + 85 + ], + "spans": [ + { + "bbox": [ + 105, + 73, + 506, + 85 + ], + "score": 1.0, + "content": "[31] Ben Mildenhall, Jonathan T Barron, Jiawen Chen, Dillon Sharlet, Ren Ng, and Robert Carroll. Burst", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 125, + 82, + 507, + 97 + ], + "spans": [ + { + "bbox": [ + 125, + 82, + 507, + 97 + ], + "score": 1.0, + "content": "denoising with kernel prediction networks. In Proceedings of the IEEE Conference on Computer Vision", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 125, + 94, + 308, + 106 + ], + "spans": [ + { + "bbox": [ + 125, + 94, + 308, + 106 + ], + "score": 1.0, + "content": "and Pattern Recognition, pages 2502–2510, 2018.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 109, + 505, + 126 + ], + "spans": [ + { + "bbox": [ + 105, + 109, + 505, + 126 + ], + "score": 1.0, + "content": "[32] Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 126, + 122, + 489, + 134 + ], + "spans": [ + { + "bbox": [ + 126, + 122, + 489, + 134 + ], + "score": 1.0, + "content": "Zagoruyko. End-to-end object detection with transformers. arXiv preprint arXiv:2005.12872, 2020.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 140, + 506, + 152 + ], + "spans": [ + { + "bbox": [ + 106, + 140, + 506, + 152 + ], + "score": 1.0, + "content": "[33] Yanhong Zeng, Jianlong Fu, and Hongyang Chao. Learning joint spatial-temporal transformations for", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 126, + 150, + 285, + 162 + ], + "spans": [ + { + "bbox": [ + 126, + 150, + 285, + 162 + ], + "score": 1.0, + "content": "video inpainting. In ECCV. Springer, 2020.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 168, + 505, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 168, + 505, + 180 + ], + "score": 1.0, + "content": "[34] Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 126, + 178, + 505, + 190 + ], + "spans": [ + { + "bbox": [ + 126, + 178, + 505, + 190 + ], + "score": 1.0, + "content": "Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 126, + 188, + 486, + 200 + ], + "spans": [ + { + "bbox": [ + 126, + 188, + 486, + 200 + ], + "score": 1.0, + "content": "16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 205, + 506, + 220 + ], + "spans": [ + { + "bbox": [ + 105, + 205, + 506, + 220 + ], + "score": 1.0, + "content": "[35] Drew A Hudson and C Lawrence Zitnick. Generative adversarial transformers. arXiv preprint", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 125, + 217, + 220, + 229 + ], + "spans": [ + { + "bbox": [ + 125, + 217, + 220, + 229 + ], + "score": 1.0, + "content": "arXiv:2103.01209, 2021.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 235, + 506, + 247 + ], + "spans": [ + { + "bbox": [ + 106, + 235, + 506, + 247 + ], + "score": 1.0, + "content": "[36] Fuzhi Yang, Huan Yang, Jianlong Fu, Hongtao Lu, and Baining Guo. Learning texture transformer network", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 126, + 245, + 505, + 257 + ], + "spans": [ + { + "bbox": [ + 126, + 245, + 505, + 257 + ], + "score": 1.0, + "content": "for image super-resolution. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 124, + 255, + 266, + 267 + ], + "spans": [ + { + "bbox": [ + 124, + 255, + 266, + 267 + ], + "score": 1.0, + "content": "Recognition, pages 5791–5800, 2020.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 271, + 506, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 506, + 286 + ], + "score": 1.0, + "content": "[37] Hanting Chen, Yunhe Wang, Tianyu Guo, Chang Xu, Yiping Deng, Zhenhua Liu, Siwei Ma, Chunjing", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 124, + 282, + 507, + 297 + ], + "spans": [ + { + "bbox": [ + 124, + 282, + 507, + 297 + ], + "score": 1.0, + "content": "Xu, Chao Xu, and Wen Gao. Pre-trained image processing transformer. arXiv preprint arXiv:2012.00364,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 125, + 293, + 151, + 306 + ], + "spans": [ + { + "bbox": [ + 125, + 293, + 151, + 306 + ], + "score": 1.0, + "content": "2020.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 311, + 505, + 323 + ], + "spans": [ + { + "bbox": [ + 106, + 311, + 505, + 323 + ], + "score": 1.0, + "content": "[38] Jie Gui, Zhenan Sun, Yonggang Wen, Dacheng Tao, and Jieping Ye. A review on generative adversarial", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 125, + 321, + 447, + 333 + ], + "spans": [ + { + "bbox": [ + 125, + 321, + 447, + 333 + ], + "score": 1.0, + "content": "networks: Algorithms, theory, and applications. arXiv preprint arXiv:2001.06937, 2020.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 339, + 506, + 351 + ], + "spans": [ + { + "bbox": [ + 106, + 339, + 506, + 351 + ], + "score": 1.0, + "content": "[39] Edgar Schonfeld, Bernt Schiele, and Anna Khoreva. A u-net based discriminator for generative adversarial", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 125, + 349, + 506, + 362 + ], + "spans": [ + { + "bbox": [ + 125, + 349, + 506, + 362 + ], + "score": 1.0, + "content": "networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 123, + 359, + 218, + 372 + ], + "spans": [ + { + "bbox": [ + 123, + 359, + 218, + 372 + ], + "score": 1.0, + "content": "pages 8207–8216, 2020.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 377, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 377, + 505, + 389 + ], + "score": 1.0, + "content": "[40] Tero Karras, Samuli Laine, Miika Aittala, Janne Hellsten, Jaakko Lehtinen, and Timo Aila. Analyzing and", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 124, + 387, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 124, + 387, + 506, + 401 + ], + "score": 1.0, + "content": "improving the image quality of stylegan. In Proceedings of the IEEE/CVF Conference on Computer Vision", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 123, + 397, + 309, + 410 + ], + "spans": [ + { + "bbox": [ + 123, + 397, + 309, + 410 + ], + "score": 1.0, + "content": "and Pattern Recognition, pages 8110–8119, 2020.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 416, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 416, + 505, + 428 + ], + "score": 1.0, + "content": "[41] Haotao Wang, Shupeng Gui, Haichuan Yang, Ji Liu, and Zhangyang Wang. Gan slimming: All-in-one gan", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 125, + 424, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 125, + 424, + 505, + 438 + ], + "score": 1.0, + "content": "compression by a unified optimization framework. In European Conference on Computer Vision, pages", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 126, + 436, + 213, + 448 + ], + "spans": [ + { + "bbox": [ + 126, + 436, + 213, + 448 + ], + "score": 1.0, + "content": "54–73. Springer, 2020.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 453, + 507, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 507, + 467 + ], + "score": 1.0, + "content": "[42] Christian Ledig, Lucas Theis, Ferenc Huszár, Jose Caballero, Andrew Cunningham, Alejandro Acosta,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 125, + 464, + 506, + 477 + ], + "spans": [ + { + "bbox": [ + 125, + 464, + 506, + 477 + ], + "score": 1.0, + "content": "Andrew Aitken, Alykhan Tejani, Johannes Totz, Zehan Wang, et al. Photo-realistic single image super-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 124, + 473, + 506, + 488 + ], + "spans": [ + { + "bbox": [ + 124, + 473, + 506, + 488 + ], + "score": 1.0, + "content": "resolution using a generative adversarial network. In Proceedings of the IEEE conference on computer", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 124, + 483, + 331, + 497 + ], + "spans": [ + { + "bbox": [ + 124, + 483, + 331, + 497 + ], + "score": 1.0, + "content": "vision and pattern recognition, pages 4681–4690, 2017.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 502, + 506, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 506, + 514 + ], + "score": 1.0, + "content": "[43] Orest Kupyn, Tetiana Martyniuk, Junru Wu, and Zhangyang Wang. Deblurgan-v2: Deblurring (orders-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 126, + 512, + 505, + 524 + ], + "spans": [ + { + "bbox": [ + 126, + 512, + 505, + 524 + ], + "score": 1.0, + "content": "of-magnitude) faster and better. In Proceedings of the IEEE/CVF International Conference on Computer", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 124, + 521, + 244, + 535 + ], + "spans": [ + { + "bbox": [ + 124, + 521, + 244, + 535 + ], + "score": 1.0, + "content": "Vision, pages 8878–8887, 2019.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 539, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 505, + 554 + ], + "score": 1.0, + "content": "[44] Xi Ouyang, Yu Cheng, Yifan Jiang, Chun-Liang Li, and Pan Zhou. Pedestrian-synthesis-gan: Generating", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 123, + 549, + 424, + 563 + ], + "spans": [ + { + "bbox": [ + 123, + 549, + 424, + 563 + ], + "score": 1.0, + "content": "pedestrian data in real scene and beyond. arXiv preprint arXiv:1804.02047, 2018.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 567, + 506, + 582 + ], + "spans": [ + { + "bbox": [ + 104, + 567, + 506, + 582 + ], + "score": 1.0, + "content": "[45] Jiahui Yu, Zhe Lin, Jimei Yang, Xiaohui Shen, Xin Lu, and Thomas S Huang. Generative image", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 125, + 578, + 505, + 590 + ], + "spans": [ + { + "bbox": [ + 125, + 578, + 505, + 590 + ], + "score": 1.0, + "content": "inpainting with contextual attention. In Proceedings of the IEEE conference on computer vision and", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 123, + 588, + 290, + 601 + ], + "spans": [ + { + "bbox": [ + 123, + 588, + 290, + 601 + ], + "score": 1.0, + "content": "pattern recognition, pages 5505–5514, 2018.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 605, + 507, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 507, + 620 + ], + "score": 1.0, + "content": "[46] Martin Arjovsky, Soumith Chintala, and Léon Bottou. Wasserstein gan. arXiv preprint arXiv:1701.07875,", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 125, + 614, + 151, + 630 + ], + "spans": [ + { + "bbox": [ + 125, + 614, + 151, + 630 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 634, + 506, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 634, + 506, + 649 + ], + "score": 1.0, + "content": "[47] Tianlong Chen, Yu Cheng, Zhe Gan, Jingjing Liu, and Zhangyang Wang. Ultra-data-efficient gan training:", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 125, + 644, + 467, + 659 + ], + "spans": [ + { + "bbox": [ + 125, + 644, + 467, + 659 + ], + "score": 1.0, + "content": "Drawing a lottery ticket first, then training it toughly. arXiv preprint arXiv:2103.00397, 2021.", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 663, + 505, + 675 + ], + "spans": [ + { + "bbox": [ + 106, + 663, + 505, + 675 + ], + "score": 1.0, + "content": "[48] Takeru Miyato, Toshiki Kataoka, Masanori Koyama, and Yuichi Yoshida. Spectral normalization for", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 123, + 673, + 392, + 686 + ], + "spans": [ + { + "bbox": [ + 123, + 673, + 392, + 686 + ], + "score": 1.0, + "content": "generative adversarial networks. arXiv preprint arXiv:1802.05957, 2018.", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 691, + 505, + 703 + ], + "spans": [ + { + "bbox": [ + 106, + 691, + 505, + 703 + ], + "score": 1.0, + "content": "[49] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 123, + 700, + 506, + 715 + ], + "spans": [ + { + "bbox": [ + 123, + 700, + 506, + 715 + ], + "score": 1.0, + "content": "Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in neural information processing", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 123, + 711, + 249, + 724 + ], + "spans": [ + { + "bbox": [ + 123, + 711, + 249, + 724 + ], + "score": 1.0, + "content": "systems, pages 5998–6008, 2017.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 24.5 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 105, + 38, + 507, + 731 + ], + "lines": [], + "index": 24.5, + "bbox_fs": [ + 104, + 73, + 507, + 724 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 29, + 507, + 727 + ], + "lines": [ + { + "bbox": [ + 106, + 73, + 505, + 85 + ], + "spans": [ + { + "bbox": [ + 106, + 73, + 505, + 85 + ], + "score": 1.0, + "content": "[50] Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Hervé", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 123, + 82, + 506, + 97 + ], + "spans": [ + { + "bbox": [ + 123, + 82, + 506, + 97 + ], + "score": 1.0, + "content": "Jégou. Training data-efficient image transformers & distillation through attention. arXiv preprint", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 126, + 93, + 220, + 106 + ], + "spans": [ + { + "bbox": [ + 126, + 93, + 220, + 106 + ], + "score": 1.0, + "content": "arXiv:2012.12877, 2020.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 104, + 109, + 506, + 125 + ], + "spans": [ + { + "bbox": [ + 104, + 109, + 506, + 125 + ], + "score": 1.0, + "content": "[51] Hugo Touvron, Matthieu Cord, Alexandre Sablayrolles, Gabriel Synnaeve, and Hervé Jégou. Going deeper", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 126, + 121, + 366, + 133 + ], + "spans": [ + { + "bbox": [ + 126, + 121, + 366, + 133 + ], + "score": 1.0, + "content": "with image transformers. arXiv preprint arXiv:2103.17239, 2021.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 138, + 506, + 151 + ], + "spans": [ + { + "bbox": [ + 106, + 138, + 506, + 151 + ], + "score": 1.0, + "content": "[52] Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 124, + 146, + 507, + 162 + ], + "spans": [ + { + "bbox": [ + 124, + 146, + 507, + 162 + ], + "score": 1.0, + "content": "transformer: Hierarchical vision transformer using shifted windows. arXiv preprint arXiv:2103.14030,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 125, + 156, + 151, + 172 + ], + "spans": [ + { + "bbox": [ + 125, + 156, + 151, + 172 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 174, + 506, + 188 + ], + "spans": [ + { + "bbox": [ + 104, + 174, + 506, + 188 + ], + "score": 1.0, + "content": "[53] Chun-Fu Chen, Quanfu Fan, and Rameswar Panda. Crossvit: Cross-attention multi-scale vision transformer", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 124, + 185, + 363, + 198 + ], + "spans": [ + { + "bbox": [ + 124, + 185, + 363, + 198 + ], + "score": 1.0, + "content": "for image classification. arXiv preprint arXiv:2103.14899, 2021.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 104, + 201, + 507, + 217 + ], + "spans": [ + { + "bbox": [ + 104, + 201, + 507, + 217 + ], + "score": 1.0, + "content": "[54] Pengchuan Zhang, Xiyang Dai, Jianwei Yang, Bin Xiao, Lu Yuan, Lei Zhang, and Jianfeng Gao. Multi-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 126, + 213, + 505, + 225 + ], + "spans": [ + { + "bbox": [ + 126, + 213, + 505, + 225 + ], + "score": 1.0, + "content": "scale vision longformer: A new vision transformer for high-resolution image encoding. arXiv preprint", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 126, + 223, + 220, + 235 + ], + "spans": [ + { + "bbox": [ + 126, + 223, + 220, + 235 + ], + "score": 1.0, + "content": "arXiv:2103.15358, 2021.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 239, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 506, + 254 + ], + "score": 1.0, + "content": "[55] Changlin Li, Tao Tang, Guangrun Wang, Jiefeng Peng, Bing Wang, Xiaodan Liang, and Xiaojun Chang.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 125, + 249, + 506, + 262 + ], + "spans": [ + { + "bbox": [ + 125, + 249, + 506, + 262 + ], + "score": 1.0, + "content": "Bossnas: Exploring hybrid cnn-transformers with block-wisely self-supervised neural architecture search.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 126, + 260, + 274, + 272 + ], + "spans": [ + { + "bbox": [ + 126, + 260, + 274, + 272 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2103.12424, 2021.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 278, + 505, + 290 + ], + "spans": [ + { + "bbox": [ + 106, + 278, + 505, + 290 + ], + "score": 1.0, + "content": "[56] Yutong Xie, Jianpeng Zhang, Chunhua Shen, and Yong Xia. Cotr: Efficiently bridging cnn and transformer", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 124, + 287, + 405, + 302 + ], + "spans": [ + { + "bbox": [ + 124, + 287, + 405, + 302 + ], + "score": 1.0, + "content": "for 3d medical image segmentation. arXiv preprint arXiv:2103.03024, 2021.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 306, + 506, + 318 + ], + "spans": [ + { + "bbox": [ + 106, + 306, + 506, + 318 + ], + "score": 1.0, + "content": "[57] Hengshuang Zhao, Li Jiang, Jiaya Jia, Philip Torr, and Vladlen Koltun. Point transformer. arXiv preprint", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 125, + 315, + 220, + 327 + ], + "spans": [ + { + "bbox": [ + 125, + 315, + 220, + 327 + ], + "score": 1.0, + "content": "arXiv:2012.09164, 2020.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 104, + 331, + 506, + 347 + ], + "spans": [ + { + "bbox": [ + 104, + 331, + 506, + 347 + ], + "score": 1.0, + "content": "[58] Sixiao Zheng, Jiachen Lu, Hengshuang Zhao, Xiatian Zhu, Zekun Luo, Yabiao Wang, Yanwei Fu, Jianfeng", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 125, + 341, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 125, + 341, + 506, + 356 + ], + "score": 1.0, + "content": "Feng, Tao Xiang, Philip HS Torr, et al. Rethinking semantic segmentation from a sequence-to-sequence", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 123, + 352, + 385, + 365 + ], + "spans": [ + { + "bbox": [ + 123, + 352, + 385, + 365 + ], + "score": 1.0, + "content": "perspective with transformers. arXiv preprint arXiv:2012.15840, 2020.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 369, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 506, + 383 + ], + "score": 1.0, + "content": "[59] Xizhou Zhu, Weijie Su, Lewei Lu, Bin Li, Xiaogang Wang, and Jifeng Dai. Deformable detr: Deformable", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 126, + 380, + 437, + 393 + ], + "spans": [ + { + "bbox": [ + 126, + 380, + 437, + 393 + ], + "score": 1.0, + "content": "transformers for end-to-end object detection. arXiv preprint arXiv:2010.04159, 2020.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 103, + 397, + 507, + 410 + ], + "spans": [ + { + "bbox": [ + 103, + 397, + 507, + 410 + ], + "score": 1.0, + "content": "[60] Kai Han, Yunhe Wang, Hanting Chen, Xinghao Chen, Jianyuan Guo, Zhenhua Liu, Yehui Tang, An Xiao,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 125, + 406, + 504, + 421 + ], + "spans": [ + { + "bbox": [ + 125, + 406, + 504, + 421 + ], + "score": 1.0, + "content": "Chunjing Xu, Yixing Xu, et al. A survey on visual transformer. arXiv preprint arXiv:2012.12556, 2020.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 425, + 506, + 437 + ], + "spans": [ + { + "bbox": [ + 106, + 425, + 506, + 437 + ], + "score": 1.0, + "content": "[61] Niki Parmar, Ashish Vaswani, Jakob Uszkoreit, Lukasz Kaiser, Noam Shazeer, Alexander Ku, and Dustin", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 125, + 434, + 506, + 448 + ], + "spans": [ + { + "bbox": [ + 125, + 434, + 506, + 448 + ], + "score": 1.0, + "content": "Tran. Image transformer. In International Conference on Machine Learning, pages 4055–4064. PMLR,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 125, + 444, + 151, + 457 + ], + "spans": [ + { + "bbox": [ + 125, + 444, + 151, + 457 + ], + "score": 1.0, + "content": "2018.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 461, + 506, + 476 + ], + "spans": [ + { + "bbox": [ + 104, + 461, + 506, + 476 + ], + "score": 1.0, + "content": "[62] Rewon Child, Scott Gray, Alec Radford, and Ilya Sutskever. Generating long sequences with sparse", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 125, + 472, + 324, + 484 + ], + "spans": [ + { + "bbox": [ + 125, + 472, + 324, + 484 + ], + "score": 1.0, + "content": "transformers. arXiv preprint arXiv:1904.10509, 2019.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 490, + 506, + 502 + ], + "spans": [ + { + "bbox": [ + 106, + 490, + 506, + 502 + ], + "score": 1.0, + "content": "[63] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirec-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 125, + 500, + 446, + 512 + ], + "spans": [ + { + "bbox": [ + 125, + 500, + 446, + 512 + ], + "score": 1.0, + "content": "tional transformers for language understanding. arXiv preprint arXiv:1810.04805, 2018.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 104, + 516, + 506, + 529 + ], + "spans": [ + { + "bbox": [ + 104, + 516, + 506, + 529 + ], + "score": 1.0, + "content": "[64] Yunyang Xiong, Zhanpeng Zeng, Rudrasis Chakraborty, Mingxing Tan, Glenn Fung, Yin Li, and Vikas", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 125, + 527, + 506, + 541 + ], + "spans": [ + { + "bbox": [ + 125, + 527, + 506, + 541 + ], + "score": 1.0, + "content": "Singh. Nystromformer: A nystrom-based algorithm for approximating self-attention. arXiv preprint", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 126, + 538, + 219, + 548 + ], + "spans": [ + { + "bbox": [ + 126, + 538, + 219, + 548 + ], + "score": 1.0, + "content": "arXiv:2102.03902, 2021.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 554, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 106, + 554, + 506, + 567 + ], + "score": 1.0, + "content": "[65] Iz Beltagy, Matthew E Peters, and Arman Cohan. Longformer: The long-document transformer. arXiv", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 123, + 564, + 252, + 578 + ], + "spans": [ + { + "bbox": [ + 123, + 564, + 252, + 578 + ], + "score": 1.0, + "content": "preprint arXiv:2004.05150, 2020.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 581, + 506, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 506, + 595 + ], + "score": 1.0, + "content": "[66] Gedas Bertasius, Heng Wang, and Lorenzo Torresani. Is space-time attention all you need for video", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 124, + 591, + 334, + 605 + ], + "spans": [ + { + "bbox": [ + 124, + 591, + 334, + 605 + ], + "score": 1.0, + "content": "understanding? arXiv preprint arXiv:2102.05095, 2021.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 609, + 505, + 622 + ], + "spans": [ + { + "bbox": [ + 106, + 609, + 505, + 622 + ], + "score": 1.0, + "content": "[67] Manoj Kumar, Dirk Weissenborn, and Nal Kalchbrenner. Colorization transformer. arXiv preprint", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 126, + 619, + 219, + 631 + ], + "spans": [ + { + "bbox": [ + 126, + 619, + 219, + 631 + ], + "score": 1.0, + "content": "arXiv:2102.04432, 2021.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 637, + 506, + 649 + ], + "spans": [ + { + "bbox": [ + 106, + 637, + 506, + 649 + ], + "score": 1.0, + "content": "[68] Chen Sun, Abhinav Shrivastava, Saurabh Singh, and Abhinav Gupta. Revisiting unreasonable effectiveness", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 125, + 647, + 506, + 659 + ], + "spans": [ + { + "bbox": [ + 125, + 647, + 506, + 659 + ], + "score": 1.0, + "content": "of data in deep learning era. In Proceedings of the IEEE international conference on computer vision,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 124, + 656, + 209, + 669 + ], + "spans": [ + { + "bbox": [ + 124, + 656, + 209, + 669 + ], + "score": 1.0, + "content": "pages 843–852, 2017.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 674, + 506, + 686 + ], + "spans": [ + { + "bbox": [ + 106, + 674, + 506, + 686 + ], + "score": 1.0, + "content": "[69] Shengyu Zhao, Zhijian Liu, Ji Lin, Jun-Yan Zhu, and Song Han. Differentiable augmentation for data-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 126, + 684, + 353, + 696 + ], + "spans": [ + { + "bbox": [ + 126, + 684, + 353, + 696 + ], + "score": 1.0, + "content": "efficient gan training. arXiv preprint arXiv:2006.10738, 2020.", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 103, + 699, + 508, + 716 + ], + "spans": [ + { + "bbox": [ + 103, + 699, + 508, + 716 + ], + "score": 1.0, + "content": "[70] Tero Karras, Miika Aittala, Janne Hellsten, Samuli Laine, Jaakko Lehtinen, and Timo Aila. Training", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 124, + 711, + 456, + 725 + ], + "spans": [ + { + "bbox": [ + 124, + 711, + 456, + 725 + ], + "score": 1.0, + "content": "generative adversarial networks with limited data. arXiv preprint arXiv:2006.06676, 2020.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 24.5 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "13", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 105, + 29, + 507, + 727 + ], + "lines": [], + "index": 24.5, + "bbox_fs": [ + 103, + 73, + 508, + 725 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 71, + 507, + 603 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 507, + 87 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 507, + 87 + ], + "score": 1.0, + "content": "[71] Peter Shaw, Jakob Uszkoreit, and Ashish Vaswani. Self-attention with relative position representations.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 124, + 83, + 274, + 96 + ], + "spans": [ + { + "bbox": [ + 124, + 83, + 274, + 96 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1803.02155, 2018.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 101, + 505, + 114 + ], + "spans": [ + { + "bbox": [ + 106, + 101, + 505, + 114 + ], + "score": 1.0, + "content": "[72] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 125, + 110, + 507, + 125 + ], + "spans": [ + { + "bbox": [ + 125, + 110, + 507, + 125 + ], + "score": 1.0, + "content": "Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 125, + 121, + 274, + 134 + ], + "spans": [ + { + "bbox": [ + 125, + 121, + 274, + 134 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1910.10683, 2019.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 139, + 506, + 152 + ], + "spans": [ + { + "bbox": [ + 106, + 139, + 506, + 152 + ], + "score": 1.0, + "content": "[73] Cheng-Zhi Anna Huang, Ashish Vaswani, Jakob Uszkoreit, Noam Shazeer, Curtis Hawthorne, AM Dai,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 126, + 149, + 505, + 162 + ], + "spans": [ + { + "bbox": [ + 126, + 149, + 505, + 162 + ], + "score": 1.0, + "content": "MD Hoffman, and D Eck. Music transformer: Generating music with long-term structure (2018). arXiv", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 124, + 160, + 251, + 171 + ], + "spans": [ + { + "bbox": [ + 124, + 160, + 251, + 171 + ], + "score": 1.0, + "content": "preprint arXiv:1809.04281, 2018.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 177, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 177, + 506, + 190 + ], + "score": 1.0, + "content": "[74] Chieh Hubert Lin, Chia-Che Chang, Yu-Sheng Chen, Da-Cheng Juan, Wei Wei, and Hwann-Tzong Chen.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 126, + 186, + 507, + 201 + ], + "spans": [ + { + "bbox": [ + 126, + 186, + 507, + 201 + ], + "score": 1.0, + "content": "Coco-gan: Generation by parts via conditional coordinating. In Proceedings of the IEEE/CVF International", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 126, + 197, + 336, + 210 + ], + "spans": [ + { + "bbox": [ + 126, + 197, + 336, + 210 + ], + "score": 1.0, + "content": "Conference on Computer Vision, pages 4512–4521, 2019.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 215, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 506, + 228 + ], + "score": 1.0, + "content": "[75] Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 124, + 224, + 457, + 238 + ], + "spans": [ + { + "bbox": [ + 124, + 224, + 457, + 238 + ], + "score": 1.0, + "content": "neural networks. Advances in neural information processing systems, 25:1097–1105, 2012.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 104, + 241, + 507, + 257 + ], + "spans": [ + { + "bbox": [ + 104, + 241, + 507, + 257 + ], + "score": 1.0, + "content": "[76] Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. Layer normalization. arXiv preprint", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 125, + 253, + 220, + 265 + ], + "spans": [ + { + "bbox": [ + 125, + 253, + 220, + 265 + ], + "score": 1.0, + "content": "arXiv:1607.06450, 2016.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 271, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 505, + 284 + ], + "score": 1.0, + "content": "[77] Sergey Ioffe and Christian Szegedy. Batch normalization: Accelerating deep network training by reducing", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 124, + 280, + 504, + 295 + ], + "spans": [ + { + "bbox": [ + 124, + 280, + 504, + 295 + ], + "score": 1.0, + "content": "internal covariate shift. In International conference on machine learning, pages 448–456. PMLR, 2015.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 299, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 299, + 505, + 312 + ], + "score": 1.0, + "content": "[78] Dmitry Ulyanov, Andrea Vedaldi, and Victor Lempitsky. Instance normalization: The missing ingredient", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 125, + 308, + 344, + 322 + ], + "spans": [ + { + "bbox": [ + 125, + 308, + 344, + 322 + ], + "score": 1.0, + "content": "for fast stylization. arXiv preprint arXiv:1607.08022, 2016.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 326, + 495, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 326, + 495, + 341 + ], + "score": 1.0, + "content": "[79] Alex Krizhevsky, Geoffrey Hinton, et al. Learning multiple layers of features from tiny images. 2009.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 344, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 344, + 505, + 357 + ], + "score": 1.0, + "content": "[80] Adam Coates, Andrew Ng, and Honglak Lee. An analysis of single-layer networks in unsupervised feature", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 124, + 353, + 507, + 369 + ], + "spans": [ + { + "bbox": [ + 124, + 353, + 507, + 369 + ], + "score": 1.0, + "content": "learning. In Proceedings of the fourteenth international conference on artificial intelligence and statistics,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 123, + 363, + 380, + 379 + ], + "spans": [ + { + "bbox": [ + 123, + 363, + 380, + 379 + ], + "score": 1.0, + "content": "pages 215–223. JMLR Workshop and Conference Proceedings, 2011.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 383, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 506, + 396 + ], + "score": 1.0, + "content": "[81] Ziwei Liu, Ping Luo, Xiaogang Wang, and Xiaoou Tang. Deep learning face attributes in the wild. In", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 126, + 392, + 402, + 405 + ], + "spans": [ + { + "bbox": [ + 126, + 392, + 402, + 405 + ], + "score": 1.0, + "content": "Proceedings of International Conference on Computer Vision (ICCV), 2015.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 410, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 106, + 410, + 505, + 423 + ], + "score": 1.0, + "content": "[82] Fisher Yu, Ari Seff, Yinda Zhang, Shuran Song, Thomas Funkhouser, and Jianxiong Xiao. Lsun: Con-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 126, + 421, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 126, + 421, + 506, + 434 + ], + "score": 1.0, + "content": "struction of a large-scale image dataset using deep learning with humans in the loop. arXiv preprint", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 126, + 430, + 219, + 443 + ], + "spans": [ + { + "bbox": [ + 126, + 430, + 219, + 443 + ], + "score": 1.0, + "content": "arXiv:1506.03365, 2015.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 448, + 505, + 461 + ], + "spans": [ + { + "bbox": [ + 106, + 448, + 505, + 461 + ], + "score": 1.0, + "content": "[83] Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 124, + 457, + 506, + 471 + ], + "spans": [ + { + "bbox": [ + 124, + 457, + 506, + 471 + ], + "score": 1.0, + "content": "trained by a two time-scale update rule converge to a local nash equilibrium. arXiv preprint", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 125, + 468, + 219, + 480 + ], + "spans": [ + { + "bbox": [ + 125, + 468, + 219, + 480 + ], + "score": 1.0, + "content": "arXiv:1706.08500, 2017.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 486, + 506, + 499 + ], + "spans": [ + { + "bbox": [ + 106, + 486, + 506, + 499 + ], + "score": 1.0, + "content": "[84] Animesh Karnewar and Oliver Wang. Msg-gan: Multi-scale gradients for generative adversarial networks.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 126, + 496, + 506, + 509 + ], + "spans": [ + { + "bbox": [ + 126, + 496, + 506, + 509 + ], + "score": 1.0, + "content": "In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7799–", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 126, + 506, + 171, + 518 + ], + "spans": [ + { + "bbox": [ + 126, + 506, + 171, + 518 + ], + "score": 1.0, + "content": "7808, 2020.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 523, + 506, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 506, + 538 + ], + "score": 1.0, + "content": "[85] Ting Chen, Mario Lucic, Neil Houlsby, and Sylvain Gelly. On self modulation for generative adversarial", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 126, + 534, + 311, + 547 + ], + "spans": [ + { + "bbox": [ + 126, + 534, + 311, + 547 + ], + "score": 1.0, + "content": "networks. arXiv preprint arXiv:1810.01365, 2018.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 552, + 506, + 565 + ], + "spans": [ + { + "bbox": [ + 106, + 552, + 506, + 565 + ], + "score": 1.0, + "content": "[86] Long Zhao, Zizhao Zhang, Ting Chen, Dimitris N Metaxas, and Han Zhang. Improved transformer for", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 126, + 562, + 352, + 575 + ], + "spans": [ + { + "bbox": [ + 126, + 562, + 352, + 575 + ], + "score": 1.0, + "content": "high-resolution gans. arXiv preprint arXiv:2106.07631, 2021.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 581, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 106, + 581, + 506, + 592 + ], + "score": 1.0, + "content": "[87] Tero Karras, Miika Aittala, Samuli Laine, Erik Härkönen, Janne Hellsten, Jaakko Lehtinen, and Timo Aila.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 126, + 589, + 429, + 602 + ], + "spans": [ + { + "bbox": [ + 126, + 589, + 429, + 602 + ], + "score": 1.0, + "content": "Alias-free generative adversarial networks. arXiv preprint arXiv:2106.12423, 2021.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 19.5 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "14", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 105, + 71, + 507, + 603 + ], + "lines": [], + "index": 19.5, + "bbox_fs": [ + 104, + 72, + 507, + 602 + ], + "lines_deleted": true + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/train/1GTpBZvNUrk/1GTpBZvNUrk_model.json b/parse/train/1GTpBZvNUrk/1GTpBZvNUrk_model.json new file mode 100644 index 0000000000000000000000000000000000000000..a53182730c32ac1917b393799ba92819d4b6b6c7 --- /dev/null +++ b/parse/train/1GTpBZvNUrk/1GTpBZvNUrk_model.json @@ -0,0 +1,16905 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 397, + 751, + 1303, + 751, + 1303, + 1538, + 397, + 1538 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1649, + 1405, + 1649, + 1405, + 1983, + 298, + 1983 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 465, + 495, + 1237, + 495, + 1237, + 625, + 465, + 625 + ], + "score": 0.97 + }, + { + "category_id": 0, + "poly": [ + 317, + 271, + 1387, + 271, + 1387, + 383, + 317, + 383 + ], + "score": 0.964 + }, + { + "category_id": 0, + "poly": [ + 299, + 1593, + 530, + 1593, + 530, + 1630, + 299, + 1630 + ], + "score": 0.905 + }, + { + "category_id": 0, + "poly": [ + 788, + 677, + 912, + 677, + 912, + 714, + 788, + 714 + ], + "score": 0.882 + }, + { + "category_id": 2, + "poly": [ + 297, + 2033, + 1070, + 2033, + 1070, + 2061, + 297, + 2061 + ], + "score": 0.825 + }, + { + "category_id": 13, + "poly": [ + 486, + 1325, + 614, + 1325, + 614, + 1355, + 486, + 1355 + ], + "score": 0.89, + "latex": "\\mathbf { 1 2 8 \\times 1 2 8 }" + }, + { + "category_id": 13, + "poly": [ + 754, + 1356, + 890, + 1356, + 890, + 1385, + 754, + 1385 + ], + "score": 0.89, + "latex": "\\mathbf { 2 5 6 \\times 2 5 6 } )" + }, + { + "category_id": 15, + "poly": [ + 315.0, + 270.0, + 1386.0, + 270.0, + 1386.0, + 328.0, + 315.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 466.0, + 324.0, + 1235.0, + 324.0, + 1235.0, + 389.0, + 466.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1589.0, + 536.0, + 1589.0, + 536.0, + 1636.0, + 292.0, + 1636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 677.0, + 917.0, + 677.0, + 917.0, + 716.0, + 784.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2028.0, + 1075.0, + 2028.0, + 1075.0, + 2067.0, + 293.0, + 2067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 751.0, + 1303.0, + 751.0, + 1303.0, + 781.0, + 396.0, + 781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 778.0, + 1308.0, + 778.0, + 1308.0, + 814.0, + 393.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 809.0, + 1305.0, + 809.0, + 1305.0, + 844.0, + 394.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 839.0, + 1308.0, + 839.0, + 1308.0, + 875.0, + 394.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 869.0, + 1306.0, + 869.0, + 1306.0, + 906.0, + 394.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 900.0, + 1308.0, + 900.0, + 1308.0, + 936.0, + 393.0, + 936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 930.0, + 1309.0, + 930.0, + 1309.0, + 966.0, + 393.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 960.0, + 1306.0, + 960.0, + 1306.0, + 995.0, + 394.0, + 995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 990.0, + 1305.0, + 990.0, + 1305.0, + 1026.0, + 392.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1021.0, + 1308.0, + 1021.0, + 1308.0, + 1056.0, + 393.0, + 1056.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1050.0, + 1306.0, + 1050.0, + 1306.0, + 1088.0, + 393.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1083.0, + 1305.0, + 1083.0, + 1305.0, + 1116.0, + 394.0, + 1116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1114.0, + 1307.0, + 1114.0, + 1307.0, + 1146.0, + 393.0, + 1146.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1142.0, + 1306.0, + 1142.0, + 1306.0, + 1178.0, + 394.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1174.0, + 1306.0, + 1174.0, + 1306.0, + 1206.0, + 393.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1205.0, + 1306.0, + 1205.0, + 1306.0, + 1238.0, + 395.0, + 1238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 1234.0, + 1305.0, + 1234.0, + 1305.0, + 1267.0, + 396.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1264.0, + 1304.0, + 1264.0, + 1304.0, + 1297.0, + 393.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1295.0, + 1305.0, + 1295.0, + 1305.0, + 1327.0, + 393.0, + 1327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1324.0, + 485.0, + 1324.0, + 485.0, + 1357.0, + 395.0, + 1357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 615.0, + 1324.0, + 1306.0, + 1324.0, + 1306.0, + 1357.0, + 615.0, + 1357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1355.0, + 753.0, + 1355.0, + 753.0, + 1388.0, + 393.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 1355.0, + 1308.0, + 1355.0, + 1308.0, + 1388.0, + 891.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1384.0, + 1305.0, + 1384.0, + 1305.0, + 1417.0, + 394.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1416.0, + 1304.0, + 1416.0, + 1304.0, + 1449.0, + 395.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1447.0, + 1305.0, + 1447.0, + 1305.0, + 1477.0, + 395.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1477.0, + 1305.0, + 1477.0, + 1305.0, + 1510.0, + 394.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1508.0, + 1000.0, + 1508.0, + 1000.0, + 1540.0, + 393.0, + 1540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1646.0, + 1410.0, + 1646.0, + 1410.0, + 1687.0, + 293.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1678.0, + 1406.0, + 1678.0, + 1406.0, + 1712.0, + 295.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1709.0, + 1408.0, + 1709.0, + 1408.0, + 1747.0, + 293.0, + 1747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1738.0, + 1406.0, + 1738.0, + 1406.0, + 1775.0, + 293.0, + 1775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1769.0, + 1406.0, + 1769.0, + 1406.0, + 1807.0, + 293.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1800.0, + 1406.0, + 1800.0, + 1406.0, + 1838.0, + 293.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1830.0, + 1406.0, + 1830.0, + 1406.0, + 1865.0, + 292.0, + 1865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1862.0, + 1405.0, + 1862.0, + 1405.0, + 1896.0, + 295.0, + 1896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1892.0, + 1405.0, + 1892.0, + 1405.0, + 1927.0, + 295.0, + 1927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1921.0, + 1405.0, + 1921.0, + 1405.0, + 1958.0, + 293.0, + 1958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1949.0, + 1405.0, + 1949.0, + 1405.0, + 1991.0, + 292.0, + 1991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 492.0, + 1141.0, + 492.0, + 1141.0, + 537.0, + 554.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 676.0, + 527.0, + 1024.0, + 527.0, + 1024.0, + 566.0, + 676.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 556.0, + 1126.0, + 556.0, + 1126.0, + 599.0, + 575.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 594.0, + 1237.0, + 594.0, + 1237.0, + 628.0, + 462.0, + 628.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1414, + 1405, + 1414, + 1405, + 1810, + 298, + 1810 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1032, + 1405, + 1032, + 1405, + 1337, + 298, + 1337 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 805, + 1404, + 805, + 1404, + 1019, + 298, + 1019 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 299, + 1824, + 1401, + 1824, + 1401, + 2007, + 299, + 2007 + ], + "score": 0.976 + }, + { + "category_id": 3, + "poly": [ + 305, + 197, + 1395, + 197, + 1395, + 584, + 305, + 584 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 295, + 729, + 1401, + 729, + 1401, + 792, + 295, + 792 + ], + "score": 0.951 + }, + { + "category_id": 4, + "poly": [ + 298, + 599, + 1405, + 599, + 1405, + 691, + 298, + 691 + ], + "score": 0.949 + }, + { + "category_id": 0, + "poly": [ + 300, + 1368, + 586, + 1368, + 586, + 1401, + 300, + 1401 + ], + "score": 0.912 + }, + { + "category_id": 2, + "poly": [ + 841, + 2062, + 858, + 2062, + 858, + 2085, + 841, + 2085 + ], + "score": 0.733 + }, + { + "category_id": 13, + "poly": [ + 1002, + 630, + 1132, + 630, + 1132, + 659, + 1002, + 659 + ], + "score": 0.86, + "latex": "2 5 6 \\times 2 5 6 )" + }, + { + "category_id": 13, + "poly": [ + 1018, + 661, + 1142, + 661, + 1142, + 690, + 1018, + 690 + ], + "score": 0.84, + "latex": "2 5 6 \\times 2 5 6 )" + }, + { + "category_id": 15, + "poly": [ + 855.0, + 196.0, + 1388.0, + 196.0, + 1388.0, + 294.0, + 855.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 321.0, + 768.0, + 321.0, + 768.0, + 347.0, + 721.0, + 347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1135.0, + 332.0, + 1150.0, + 332.0, + 1150.0, + 342.0, + 1135.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1232.0, + 314.0, + 1346.0, + 314.0, + 1346.0, + 368.0, + 1232.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 421.0, + 747.0, + 421.0, + 747.0, + 442.0, + 724.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 813.0, + 427.0, + 828.0, + 427.0, + 828.0, + 437.0, + 813.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 404.0, + 943.0, + 404.0, + 943.0, + 453.0, + 846.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 947.0, + 394.0, + 1381.0, + 394.0, + 1381.0, + 467.0, + 947.0, + 467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 463.0, + 1395.0, + 463.0, + 1395.0, + 559.0, + 695.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 553.0, + 612.0, + 553.0, + 612.0, + 590.0, + 372.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 862.0, + 553.0, + 1213.0, + 553.0, + 1213.0, + 588.0, + 862.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 597.0, + 1407.0, + 597.0, + 1407.0, + 638.0, + 292.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 631.0, + 1001.0, + 631.0, + 1001.0, + 664.0, + 295.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1133.0, + 631.0, + 1405.0, + 631.0, + 1405.0, + 664.0, + 1133.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 661.0, + 1017.0, + 661.0, + 1017.0, + 694.0, + 295.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1143.0, + 661.0, + 1241.0, + 661.0, + 1241.0, + 694.0, + 1143.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1363.0, + 590.0, + 1363.0, + 590.0, + 1407.0, + 292.0, + 1407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2059.0, + 862.0, + 2059.0, + 862.0, + 2094.0, + 839.0, + 2094.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1410.0, + 1406.0, + 1410.0, + 1406.0, + 1453.0, + 292.0, + 1453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1445.0, + 1407.0, + 1445.0, + 1407.0, + 1480.0, + 293.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1478.0, + 1403.0, + 1478.0, + 1403.0, + 1510.0, + 296.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1505.0, + 1405.0, + 1505.0, + 1405.0, + 1541.0, + 296.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1537.0, + 1407.0, + 1537.0, + 1407.0, + 1573.0, + 295.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1568.0, + 1403.0, + 1568.0, + 1403.0, + 1600.0, + 296.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1597.0, + 1406.0, + 1597.0, + 1406.0, + 1631.0, + 293.0, + 1631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1625.0, + 1410.0, + 1625.0, + 1410.0, + 1665.0, + 292.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1657.0, + 1403.0, + 1657.0, + 1403.0, + 1693.0, + 293.0, + 1693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1684.0, + 1408.0, + 1684.0, + 1408.0, + 1726.0, + 292.0, + 1726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1718.0, + 1405.0, + 1718.0, + 1405.0, + 1751.0, + 293.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1749.0, + 1406.0, + 1749.0, + 1406.0, + 1784.0, + 293.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1776.0, + 1302.0, + 1776.0, + 1302.0, + 1816.0, + 293.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1028.0, + 1405.0, + 1028.0, + 1405.0, + 1071.0, + 293.0, + 1071.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1062.0, + 1406.0, + 1062.0, + 1406.0, + 1100.0, + 293.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1088.0, + 1406.0, + 1088.0, + 1406.0, + 1133.0, + 292.0, + 1133.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1122.0, + 1403.0, + 1122.0, + 1403.0, + 1158.0, + 295.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1155.0, + 1405.0, + 1155.0, + 1405.0, + 1187.0, + 295.0, + 1187.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1184.0, + 1406.0, + 1184.0, + 1406.0, + 1220.0, + 295.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1215.0, + 1408.0, + 1215.0, + 1408.0, + 1250.0, + 295.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1240.0, + 1406.0, + 1240.0, + 1406.0, + 1285.0, + 292.0, + 1285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1276.0, + 1406.0, + 1276.0, + 1406.0, + 1311.0, + 295.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1306.0, + 1121.0, + 1306.0, + 1121.0, + 1338.0, + 295.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 805.0, + 1404.0, + 805.0, + 1404.0, + 840.0, + 296.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 835.0, + 1406.0, + 835.0, + 1406.0, + 872.0, + 292.0, + 872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 863.0, + 1406.0, + 863.0, + 1406.0, + 906.0, + 291.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 897.0, + 1406.0, + 897.0, + 1406.0, + 932.0, + 296.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 925.0, + 1406.0, + 925.0, + 1406.0, + 964.0, + 293.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 957.0, + 1410.0, + 957.0, + 1410.0, + 994.0, + 292.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 987.0, + 1370.0, + 987.0, + 1370.0, + 1024.0, + 292.0, + 1024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1819.0, + 1405.0, + 1819.0, + 1405.0, + 1860.0, + 293.0, + 1860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1850.0, + 1406.0, + 1850.0, + 1406.0, + 1893.0, + 293.0, + 1893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1883.0, + 1406.0, + 1883.0, + 1406.0, + 1919.0, + 294.0, + 1919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1914.0, + 1405.0, + 1914.0, + 1405.0, + 1951.0, + 293.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1946.0, + 1406.0, + 1946.0, + 1406.0, + 1981.0, + 293.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1974.0, + 1403.0, + 1974.0, + 1403.0, + 2010.0, + 294.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 729.0, + 1403.0, + 729.0, + 1403.0, + 764.0, + 297.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 759.0, + 1143.0, + 759.0, + 1143.0, + 797.0, + 293.0, + 797.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1262, + 1405, + 1262, + 1405, + 1627, + 297, + 1627 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1642, + 1404, + 1642, + 1404, + 2007, + 298, + 2007 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 1066, + 1405, + 1066, + 1405, + 1250, + 298, + 1250 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 299, + 202, + 1403, + 202, + 1403, + 326, + 299, + 326 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 301, + 339, + 1404, + 339, + 1404, + 491, + 301, + 491 + ], + "score": 0.935 + }, + { + "category_id": 0, + "poly": [ + 299, + 1008, + 562, + 1008, + 562, + 1046, + 299, + 1046 + ], + "score": 0.923 + }, + { + "category_id": 1, + "poly": [ + 371, + 507, + 1405, + 507, + 1405, + 977, + 371, + 977 + ], + "score": 0.653 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 858, + 2061, + 858, + 2085, + 841, + 2085 + ], + "score": 0.635 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.367 + }, + { + "category_id": 13, + "poly": [ + 859, + 1355, + 949, + 1355, + 949, + 1384, + 859, + 1384 + ], + "score": 0.9, + "latex": "1 6 \\times 1 6" + }, + { + "category_id": 13, + "poly": [ + 1133, + 886, + 1252, + 886, + 1252, + 915, + 1133, + 915 + ], + "score": 0.89, + "latex": "1 2 8 \\times 1 2 8" + }, + { + "category_id": 13, + "poly": [ + 683, + 295, + 806, + 295, + 806, + 325, + 683, + 325 + ], + "score": 0.87, + "latex": "2 5 6 \\times 2 5 6" + }, + { + "category_id": 13, + "poly": [ + 1147, + 918, + 1272, + 918, + 1272, + 947, + 1147, + 947 + ], + "score": 0.87, + "latex": "2 5 6 \\times 2 5 6 )" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1007.0, + 565.0, + 1007.0, + 565.0, + 1051.0, + 291.0, + 1051.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2058.0, + 860.0, + 2058.0, + 860.0, + 2089.0, + 839.0, + 2089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2091.0, + 838.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1262.0, + 1406.0, + 1262.0, + 1406.0, + 1299.0, + 294.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1294.0, + 1405.0, + 1294.0, + 1405.0, + 1328.0, + 294.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1321.0, + 1405.0, + 1321.0, + 1405.0, + 1361.0, + 294.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1354.0, + 858.0, + 1354.0, + 858.0, + 1391.0, + 294.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 950.0, + 1354.0, + 1406.0, + 1354.0, + 1406.0, + 1391.0, + 950.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1385.0, + 1405.0, + 1385.0, + 1405.0, + 1419.0, + 296.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1415.0, + 1408.0, + 1415.0, + 1408.0, + 1449.0, + 295.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1444.0, + 1405.0, + 1444.0, + 1405.0, + 1481.0, + 294.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1472.0, + 1408.0, + 1472.0, + 1408.0, + 1512.0, + 292.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1507.0, + 1406.0, + 1507.0, + 1406.0, + 1538.0, + 297.0, + 1538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1537.0, + 1405.0, + 1537.0, + 1405.0, + 1570.0, + 295.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1568.0, + 1405.0, + 1568.0, + 1405.0, + 1602.0, + 295.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1597.0, + 1110.0, + 1597.0, + 1110.0, + 1631.0, + 295.0, + 1631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1643.0, + 1404.0, + 1643.0, + 1404.0, + 1677.0, + 296.0, + 1677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1673.0, + 1405.0, + 1673.0, + 1405.0, + 1707.0, + 293.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1702.0, + 1405.0, + 1702.0, + 1405.0, + 1740.0, + 293.0, + 1740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1734.0, + 1405.0, + 1734.0, + 1405.0, + 1766.0, + 292.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1764.0, + 1404.0, + 1764.0, + 1404.0, + 1798.0, + 293.0, + 1798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1794.0, + 1406.0, + 1794.0, + 1406.0, + 1828.0, + 294.0, + 1828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1824.0, + 1406.0, + 1824.0, + 1406.0, + 1858.0, + 296.0, + 1858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1854.0, + 1406.0, + 1854.0, + 1406.0, + 1889.0, + 292.0, + 1889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1884.0, + 1406.0, + 1884.0, + 1406.0, + 1920.0, + 292.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1915.0, + 1406.0, + 1915.0, + 1406.0, + 1949.0, + 294.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1946.0, + 1405.0, + 1946.0, + 1405.0, + 1979.0, + 294.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1976.0, + 1404.0, + 1976.0, + 1404.0, + 2010.0, + 296.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1065.0, + 1405.0, + 1065.0, + 1405.0, + 1101.0, + 295.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1095.0, + 1408.0, + 1095.0, + 1408.0, + 1131.0, + 293.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1125.0, + 1407.0, + 1125.0, + 1407.0, + 1162.0, + 291.0, + 1162.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1157.0, + 1406.0, + 1157.0, + 1406.0, + 1192.0, + 295.0, + 1192.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1188.0, + 1407.0, + 1188.0, + 1407.0, + 1220.0, + 296.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1216.0, + 1392.0, + 1216.0, + 1392.0, + 1252.0, + 295.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 199.0, + 1407.0, + 199.0, + 1407.0, + 242.0, + 292.0, + 242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 232.0, + 1405.0, + 232.0, + 1405.0, + 268.0, + 294.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 263.0, + 1404.0, + 263.0, + 1404.0, + 298.0, + 294.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 296.0, + 682.0, + 296.0, + 682.0, + 328.0, + 295.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 296.0, + 822.0, + 296.0, + 822.0, + 328.0, + 807.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 336.0, + 1404.0, + 336.0, + 1404.0, + 374.0, + 296.0, + 374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 366.0, + 1405.0, + 366.0, + 1405.0, + 405.0, + 295.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 398.0, + 1405.0, + 398.0, + 1405.0, + 436.0, + 295.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 429.0, + 1404.0, + 429.0, + 1404.0, + 466.0, + 294.0, + 466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 461.0, + 933.0, + 461.0, + 933.0, + 496.0, + 295.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 504.0, + 1407.0, + 504.0, + 1407.0, + 544.0, + 369.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 537.0, + 1405.0, + 537.0, + 1405.0, + 573.0, + 393.0, + 573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 570.0, + 1405.0, + 570.0, + 1405.0, + 603.0, + 395.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 599.0, + 1404.0, + 599.0, + 1404.0, + 633.0, + 395.0, + 633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 628.0, + 1078.0, + 628.0, + 1078.0, + 664.0, + 393.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 380.0, + 667.0, + 1407.0, + 667.0, + 1407.0, + 700.0, + 380.0, + 700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 697.0, + 1407.0, + 697.0, + 1407.0, + 734.0, + 393.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 728.0, + 1409.0, + 728.0, + 1409.0, + 765.0, + 393.0, + 765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 759.0, + 767.0, + 759.0, + 767.0, + 792.0, + 392.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 384.0, + 793.0, + 1405.0, + 793.0, + 1405.0, + 829.0, + 384.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 825.0, + 1405.0, + 825.0, + 1405.0, + 862.0, + 391.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 854.0, + 1405.0, + 854.0, + 1405.0, + 892.0, + 391.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 882.0, + 1132.0, + 882.0, + 1132.0, + 923.0, + 392.0, + 923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1253.0, + 882.0, + 1408.0, + 882.0, + 1408.0, + 923.0, + 1253.0, + 923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 914.0, + 1146.0, + 914.0, + 1146.0, + 953.0, + 392.0, + 953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 914.0, + 1405.0, + 914.0, + 1405.0, + 953.0, + 1273.0, + 953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 945.0, + 1322.0, + 945.0, + 1322.0, + 982.0, + 394.0, + 982.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1626, + 1405, + 1626, + 1405, + 1932, + 296, + 1932 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 299, + 1303, + 1404, + 1303, + 1404, + 1424, + 299, + 1424 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 299, + 1439, + 1403, + 1439, + 1403, + 1563, + 299, + 1563 + ], + "score": 0.976 + }, + { + "category_id": 4, + "poly": [ + 297, + 956, + 1405, + 956, + 1405, + 1140, + 297, + 1140 + ], + "score": 0.964 + }, + { + "category_id": 1, + "poly": [ + 298, + 1156, + 1396, + 1156, + 1396, + 1219, + 298, + 1219 + ], + "score": 0.953 + }, + { + "category_id": 0, + "poly": [ + 296, + 1246, + 1368, + 1246, + 1368, + 1285, + 296, + 1285 + ], + "score": 0.939 + }, + { + "category_id": 3, + "poly": [ + 317, + 200, + 1389, + 200, + 1389, + 945, + 317, + 945 + ], + "score": 0.934 + }, + { + "category_id": 1, + "poly": [ + 299, + 1945, + 1397, + 1945, + 1397, + 2008, + 299, + 2008 + ], + "score": 0.933 + }, + { + "category_id": 0, + "poly": [ + 299, + 1585, + 697, + 1585, + 697, + 1617, + 299, + 1617 + ], + "score": 0.924 + }, + { + "category_id": 2, + "poly": [ + 841, + 2062, + 858, + 2062, + 858, + 2084, + 841, + 2084 + ], + "score": 0.763 + }, + { + "category_id": 3, + "poly": [ + 317, + 203, + 1389, + 203, + 1389, + 944, + 317, + 944 + ], + "score": 0.605 + }, + { + "category_id": 13, + "poly": [ + 721, + 1049, + 783, + 1049, + 783, + 1077, + 721, + 1077 + ], + "score": 0.9, + "latex": "8 \\times 8" + }, + { + "category_id": 13, + "poly": [ + 296, + 988, + 411, + 988, + 411, + 1017, + 296, + 1017 + ], + "score": 0.9, + "latex": "2 5 6 \\times 2 5 6" + }, + { + "category_id": 13, + "poly": [ + 1122, + 1749, + 1217, + 1749, + 1217, + 1779, + 1122, + 1779 + ], + "score": 0.88, + "latex": "3 2 \\times 3 2" + }, + { + "category_id": 13, + "poly": [ + 412, + 1023, + 431, + 1023, + 431, + 1050, + 412, + 1050 + ], + "score": 0.8, + "latex": "p" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 955.0, + 1405.0, + 955.0, + 1405.0, + 991.0, + 295.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 412.0, + 987.0, + 1405.0, + 987.0, + 1405.0, + 1023.0, + 412.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1018.0, + 411.0, + 1018.0, + 411.0, + 1052.0, + 292.0, + 1052.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 1018.0, + 1405.0, + 1018.0, + 1405.0, + 1052.0, + 432.0, + 1052.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1049.0, + 720.0, + 1049.0, + 720.0, + 1081.0, + 295.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1049.0, + 1403.0, + 1049.0, + 1403.0, + 1081.0, + 784.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1080.0, + 1405.0, + 1080.0, + 1405.0, + 1111.0, + 295.0, + 1111.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1109.0, + 783.0, + 1109.0, + 783.0, + 1142.0, + 293.0, + 1142.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1245.0, + 1372.0, + 1245.0, + 1372.0, + 1290.0, + 290.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 212.0, + 973.0, + 212.0, + 973.0, + 251.0, + 873.0, + 251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1025.0, + 224.0, + 1039.0, + 224.0, + 1039.0, + 240.0, + 1025.0, + 240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 220.0, + 1206.0, + 220.0, + 1206.0, + 246.0, + 1183.0, + 246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 247.0, + 576.0, + 247.0, + 576.0, + 275.0, + 462.0, + 275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 256.0, + 1211.0, + 256.0, + 1211.0, + 286.0, + 982.0, + 286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 246.0, + 1282.0, + 246.0, + 1282.0, + 302.0, + 1261.0, + 302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 292.0, + 1146.0, + 292.0, + 1146.0, + 323.0, + 1045.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1352.0, + 307.0, + 1378.0, + 307.0, + 1378.0, + 320.0, + 1352.0, + 320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 328.0, + 1126.0, + 328.0, + 1126.0, + 341.0, + 1103.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1321.0, + 325.0, + 1378.0, + 325.0, + 1378.0, + 366.0, + 1321.0, + 366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 337.0, + 701.0, + 337.0, + 701.0, + 371.0, + 545.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1104.0, + 354.0, + 1117.0, + 354.0, + 1117.0, + 362.0, + 1104.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 373.0, + 582.0, + 373.0, + 582.0, + 382.0, + 570.0, + 382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 586.0, + 371.0, + 604.0, + 371.0, + 604.0, + 384.0, + 586.0, + 384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 370.0, + 731.0, + 370.0, + 731.0, + 386.0, + 703.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 371.0, + 1150.0, + 371.0, + 1150.0, + 395.0, + 1046.0, + 395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 390.0, + 895.0, + 390.0, + 895.0, + 409.0, + 871.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 949.0, + 388.0, + 975.0, + 388.0, + 975.0, + 412.0, + 949.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 389.0, + 1185.0, + 389.0, + 1185.0, + 431.0, + 1040.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 383.0, + 1283.0, + 383.0, + 1283.0, + 444.0, + 1258.0, + 444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 868.0, + 401.0, + 937.0, + 401.0, + 937.0, + 428.0, + 868.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 952.0, + 408.0, + 973.0, + 408.0, + 973.0, + 428.0, + 952.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1026.0, + 406.0, + 1036.0, + 406.0, + 1036.0, + 416.0, + 1026.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 416.0, + 470.0, + 416.0, + 470.0, + 460.0, + 332.0, + 460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 422.0, + 737.0, + 422.0, + 737.0, + 452.0, + 510.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1006.0, + 443.0, + 1188.0, + 443.0, + 1188.0, + 474.0, + 1006.0, + 474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 571.0, + 456.0, + 676.0, + 456.0, + 676.0, + 491.0, + 571.0, + 491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1317.0, + 462.0, + 1382.0, + 462.0, + 1382.0, + 494.0, + 1317.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 488.0, + 707.0, + 488.0, + 707.0, + 526.0, + 540.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 484.0, + 1148.0, + 484.0, + 1148.0, + 515.0, + 1046.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 445.0, + 529.0, + 465.0, + 529.0, + 465.0, + 550.0, + 445.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 586.0, + 518.0, + 661.0, + 518.0, + 661.0, + 551.0, + 586.0, + 551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1262.0, + 524.0, + 1283.0, + 524.0, + 1283.0, + 582.0, + 1262.0, + 582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 558.0, + 400.0, + 558.0, + 400.0, + 572.0, + 385.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 557.0, + 632.0, + 557.0, + 632.0, + 574.0, + 616.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 570.0, + 401.0, + 570.0, + 401.0, + 587.0, + 385.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 571.0, + 633.0, + 571.0, + 633.0, + 587.0, + 616.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 566.0, + 1150.0, + 566.0, + 1150.0, + 592.0, + 1045.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 587.0, + 401.0, + 587.0, + 401.0, + 601.0, + 385.0, + 601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 587.0, + 685.0, + 587.0, + 685.0, + 647.0, + 563.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 592.0, + 898.0, + 592.0, + 898.0, + 631.0, + 871.0, + 631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 593.0, + 975.0, + 593.0, + 975.0, + 633.0, + 908.0, + 633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 627.0, + 463.0, + 627.0, + 463.0, + 659.0, + 345.0, + 659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 637.0, + 662.0, + 637.0, + 662.0, + 673.0, + 585.0, + 673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1039.0, + 636.0, + 1156.0, + 636.0, + 1156.0, + 666.0, + 1039.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 571.0, + 671.0, + 675.0, + 671.0, + 675.0, + 702.0, + 571.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 657.0, + 1135.0, + 657.0, + 1135.0, + 693.0, + 1059.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 718.0, + 447.0, + 718.0, + 447.0, + 747.0, + 350.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 707.0, + 674.0, + 707.0, + 674.0, + 741.0, + 573.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 692.0, + 1128.0, + 692.0, + 1128.0, + 743.0, + 1038.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1184.0, + 705.0, + 1231.0, + 705.0, + 1231.0, + 732.0, + 1184.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 592.0, + 732.0, + 656.0, + 732.0, + 656.0, + 761.0, + 592.0, + 761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 742.0, + 1157.0, + 742.0, + 1157.0, + 773.0, + 1037.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 771.0, + 693.0, + 771.0, + 693.0, + 799.0, + 556.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 767.0, + 1135.0, + 767.0, + 1135.0, + 799.0, + 1059.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 592.0, + 815.0, + 653.0, + 815.0, + 653.0, + 850.0, + 592.0, + 850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 810.0, + 1159.0, + 810.0, + 1159.0, + 847.0, + 1094.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 564.0, + 859.0, + 682.0, + 859.0, + 682.0, + 898.0, + 564.0, + 898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1061.0, + 855.0, + 1188.0, + 855.0, + 1188.0, + 893.0, + 1061.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 902.0, + 695.0, + 902.0, + 695.0, + 945.0, + 552.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1003.0, + 906.0, + 1192.0, + 906.0, + 1192.0, + 945.0, + 1003.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 903.75, + 385.5, + 958.75, + 385.5, + 958.75, + 420.0, + 903.75, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.25, + 506.5, + 474.25, + 506.5, + 474.25, + 549.5, + 326.25, + 549.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 900.25, + 602.5, + 906.25, + 602.5, + 906.25, + 616.5, + 900.25, + 616.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1579.0, + 703.0, + 1579.0, + 703.0, + 1627.0, + 291.0, + 1627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2059.0, + 862.0, + 2059.0, + 862.0, + 2091.0, + 838.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 213.0, + 892.0, + 213.0, + 892.0, + 231.0, + 871.0, + 231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 226.0, + 905.0, + 226.0, + 905.0, + 238.0, + 895.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 907.0, + 212.0, + 973.0, + 212.0, + 973.0, + 243.0, + 907.0, + 243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 223.0, + 1013.0, + 223.0, + 1013.0, + 239.0, + 998.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 221.0, + 1122.0, + 221.0, + 1122.0, + 240.0, + 1067.0, + 240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1184.0, + 221.0, + 1205.0, + 221.0, + 1205.0, + 244.0, + 1184.0, + 244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 231.0, + 892.0, + 231.0, + 892.0, + 250.0, + 873.0, + 250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 909.0, + 231.0, + 927.0, + 231.0, + 927.0, + 251.0, + 909.0, + 251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 953.0, + 231.0, + 973.0, + 231.0, + 973.0, + 251.0, + 953.0, + 251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 247.0, + 575.0, + 247.0, + 575.0, + 274.0, + 462.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 255.0, + 1208.0, + 255.0, + 1208.0, + 284.0, + 982.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 246.0, + 1282.0, + 246.0, + 1282.0, + 304.0, + 1261.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 615.0, + 307.0, + 631.0, + 307.0, + 631.0, + 320.0, + 615.0, + 320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 292.0, + 1147.0, + 292.0, + 1147.0, + 323.0, + 1046.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1295.0, + 309.0, + 1306.0, + 309.0, + 1306.0, + 318.0, + 1295.0, + 318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 306.0, + 1380.0, + 306.0, + 1380.0, + 321.0, + 1325.0, + 321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 318.0, + 628.0, + 318.0, + 628.0, + 329.0, + 616.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 327.0, + 1126.0, + 327.0, + 1126.0, + 341.0, + 1103.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1321.0, + 325.0, + 1378.0, + 325.0, + 1378.0, + 364.0, + 1321.0, + 364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 339.0, + 699.0, + 339.0, + 699.0, + 367.0, + 547.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 708.0, + 373.0, + 725.0, + 373.0, + 725.0, + 386.0, + 708.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 370.0, + 1150.0, + 370.0, + 1150.0, + 394.0, + 1046.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 388.0, + 934.0, + 388.0, + 934.0, + 429.0, + 869.0, + 429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 390.0, + 973.0, + 390.0, + 973.0, + 428.0, + 936.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1027.0, + 406.0, + 1037.0, + 406.0, + 1037.0, + 419.0, + 1027.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 391.0, + 1185.0, + 391.0, + 1185.0, + 430.0, + 1042.0, + 430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 385.0, + 1282.0, + 385.0, + 1282.0, + 437.0, + 1261.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 333.0, + 416.0, + 470.0, + 416.0, + 470.0, + 460.0, + 333.0, + 460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.0, + 419.0, + 737.0, + 419.0, + 737.0, + 453.0, + 509.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1007.0, + 443.0, + 1189.0, + 443.0, + 1189.0, + 473.0, + 1007.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 458.0, + 674.0, + 458.0, + 674.0, + 489.0, + 572.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1317.0, + 462.0, + 1382.0, + 462.0, + 1382.0, + 495.0, + 1317.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 497.0, + 481.0, + 497.0, + 481.0, + 572.0, + 321.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 491.0, + 705.0, + 491.0, + 705.0, + 573.0, + 541.0, + 573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 484.0, + 1147.0, + 484.0, + 1147.0, + 515.0, + 1046.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1262.0, + 523.0, + 1283.0, + 523.0, + 1283.0, + 583.0, + 1262.0, + 583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 572.0, + 401.0, + 572.0, + 401.0, + 587.0, + 385.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 572.0, + 632.0, + 572.0, + 632.0, + 587.0, + 616.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 566.0, + 1150.0, + 566.0, + 1150.0, + 593.0, + 1045.0, + 593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 387.0, + 589.0, + 399.0, + 589.0, + 399.0, + 600.0, + 387.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 588.0, + 685.0, + 588.0, + 685.0, + 645.0, + 563.0, + 645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 593.0, + 975.0, + 593.0, + 975.0, + 632.0, + 873.0, + 632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 628.0, + 462.0, + 628.0, + 462.0, + 657.0, + 346.0, + 657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 639.0, + 661.0, + 639.0, + 661.0, + 671.0, + 587.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 637.0, + 1156.0, + 637.0, + 1156.0, + 664.0, + 1040.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 668.0, + 675.0, + 668.0, + 675.0, + 703.0, + 570.0, + 703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 658.0, + 1135.0, + 658.0, + 1135.0, + 693.0, + 1059.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 718.0, + 447.0, + 718.0, + 447.0, + 749.0, + 350.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 707.0, + 674.0, + 707.0, + 674.0, + 741.0, + 572.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 692.0, + 1128.0, + 692.0, + 1128.0, + 743.0, + 1040.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 703.0, + 1231.0, + 703.0, + 1231.0, + 732.0, + 1183.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 592.0, + 732.0, + 655.0, + 732.0, + 655.0, + 761.0, + 592.0, + 761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 742.0, + 1157.0, + 742.0, + 1157.0, + 773.0, + 1037.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 771.0, + 693.0, + 771.0, + 693.0, + 799.0, + 556.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 767.0, + 1135.0, + 767.0, + 1135.0, + 799.0, + 1059.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 593.0, + 816.0, + 653.0, + 816.0, + 653.0, + 849.0, + 593.0, + 849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 810.0, + 1159.0, + 810.0, + 1159.0, + 847.0, + 1094.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 565.0, + 859.0, + 683.0, + 859.0, + 683.0, + 899.0, + 565.0, + 899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1063.0, + 861.0, + 1188.0, + 861.0, + 1188.0, + 892.0, + 1063.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 903.0, + 695.0, + 903.0, + 695.0, + 945.0, + 552.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 905.0, + 1193.0, + 905.0, + 1193.0, + 946.0, + 1002.0, + 946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1625.0, + 1406.0, + 1625.0, + 1406.0, + 1661.0, + 294.0, + 1661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1656.0, + 1407.0, + 1656.0, + 1407.0, + 1691.0, + 291.0, + 1691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1685.0, + 1405.0, + 1685.0, + 1405.0, + 1723.0, + 293.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1716.0, + 1407.0, + 1716.0, + 1407.0, + 1754.0, + 291.0, + 1754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1746.0, + 1121.0, + 1746.0, + 1121.0, + 1784.0, + 294.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 1746.0, + 1405.0, + 1746.0, + 1405.0, + 1784.0, + 1218.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1779.0, + 1405.0, + 1779.0, + 1405.0, + 1815.0, + 294.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1807.0, + 1410.0, + 1807.0, + 1410.0, + 1846.0, + 293.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1840.0, + 1405.0, + 1840.0, + 1405.0, + 1875.0, + 294.0, + 1875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1872.0, + 1403.0, + 1872.0, + 1403.0, + 1903.0, + 295.0, + 1903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1900.0, + 713.0, + 1900.0, + 713.0, + 1935.0, + 295.0, + 1935.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1301.0, + 1408.0, + 1301.0, + 1408.0, + 1338.0, + 294.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1334.0, + 1405.0, + 1334.0, + 1405.0, + 1368.0, + 293.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1363.0, + 1404.0, + 1363.0, + 1404.0, + 1399.0, + 293.0, + 1399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1396.0, + 688.0, + 1396.0, + 688.0, + 1428.0, + 294.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1440.0, + 1404.0, + 1440.0, + 1404.0, + 1473.0, + 294.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1469.0, + 1407.0, + 1469.0, + 1407.0, + 1506.0, + 294.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1502.0, + 1403.0, + 1502.0, + 1403.0, + 1535.0, + 295.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1531.0, + 1398.0, + 1531.0, + 1398.0, + 1566.0, + 293.0, + 1566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1154.0, + 1401.0, + 1154.0, + 1401.0, + 1193.0, + 293.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1188.0, + 1029.0, + 1188.0, + 1029.0, + 1219.0, + 295.0, + 1219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1945.0, + 1402.0, + 1945.0, + 1402.0, + 1980.0, + 295.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1974.0, + 1403.0, + 1974.0, + 1403.0, + 2012.0, + 293.0, + 2012.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 368, + 1406, + 368, + 1406, + 715, + 297, + 715 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 297, + 1070, + 1405, + 1070, + 1405, + 1483, + 297, + 1483 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 297, + 782, + 1405, + 782, + 1405, + 1056, + 297, + 1056 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1717, + 1404, + 1717, + 1404, + 1931, + 298, + 1931 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 1552, + 1405, + 1552, + 1405, + 1705, + 297, + 1705 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 299, + 202, + 1403, + 202, + 1403, + 355, + 299, + 355 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 299, + 1945, + 1399, + 1945, + 1399, + 2008, + 299, + 2008 + ], + "score": 0.942 + }, + { + "category_id": 0, + "poly": [ + 298, + 1507, + 1299, + 1507, + 1299, + 1540, + 298, + 1540 + ], + "score": 0.928 + }, + { + "category_id": 0, + "poly": [ + 299, + 739, + 669, + 739, + 669, + 771, + 299, + 771 + ], + "score": 0.924 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 858, + 2061, + 858, + 2085, + 841, + 2085 + ], + "score": 0.714 + }, + { + "category_id": 13, + "poly": [ + 1212, + 1127, + 1374, + 1127, + 1374, + 1168, + 1212, + 1168 + ], + "score": 0.93, + "latex": "\\textstyle \\left( { \\frac { H } { P } } \\times { \\frac { W } { P } } \\right) \\times 3" + }, + { + "category_id": 13, + "poly": [ + 564, + 1162, + 735, + 1162, + 735, + 1203, + 564, + 1203 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\left( \\frac { H } { P } \\times \\frac { W } { P } \\right) \\times \\frac { C } { 4 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1106, + 491, + 1328, + 491, + 1328, + 529, + 1106, + 529 + ], + "score": 0.93, + "latex": "\\boldsymbol { X } _ { i } ^ { ' } \\in \\mathbb { R } ^ { 2 H _ { i } \\times 2 W _ { i } \\times C }" + }, + { + "category_id": 13, + "poly": [ + 1015, + 1098, + 1186, + 1098, + 1186, + 1128, + 1015, + 1128 + ], + "score": 0.92, + "latex": "Y \\in R ^ { H \\times W \\times 3 ^ { \\bullet } }" + }, + { + "category_id": 13, + "poly": [ + 850, + 1235, + 1038, + 1235, + 1038, + 1276, + 850, + 1276 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\big ( \\frac { H } { 2 P } \\times \\frac { W } { 2 P } \\big ) \\times \\frac { C } { 4 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 873, + 264, + 983, + 264, + 983, + 295, + 873, + 295 + ], + "score": 0.91, + "latex": "H _ { 0 } \\times W _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 485, + 458, + 689, + 458, + 689, + 491, + 485, + 491 + ], + "score": 0.91, + "latex": "X _ { i } ~ \\overset { \\cdot } { \\in } ~ \\mathbb { R } ^ { \\cup _ { i } \\times W _ { i } \\times C }" + }, + { + "category_id": 13, + "poly": [ + 1096, + 1236, + 1285, + 1236, + 1285, + 1277, + 1096, + 1277 + ], + "score": 0.91, + "latex": "\\bigl ( { \\frac { H } { 4 P } } \\times { \\frac { W } { 4 P } } \\bigr ) \\times { \\frac { C } { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 373, + 264, + 539, + 264, + 539, + 294, + 373, + 294 + ], + "score": 0.91, + "latex": "H _ { 0 } \\times W _ { 0 } \\times C" + }, + { + "category_id": 13, + "poly": [ + 590, + 527, + 626, + 527, + 626, + 565, + 590, + 565 + ], + "score": 0.9, + "latex": "X _ { i } ^ { ' }" + }, + { + "category_id": 13, + "poly": [ + 1215, + 682, + 1385, + 682, + 1385, + 712, + 1215, + 712 + ], + "score": 0.89, + "latex": "Y \\in \\mathbb { R } ^ { H \\times W \\times 3 }" + }, + { + "category_id": 13, + "poly": [ + 342, + 296, + 509, + 296, + 509, + 325, + 342, + 325 + ], + "score": 0.89, + "latex": "H _ { 0 } = W _ { 0 } = 8" + }, + { + "category_id": 13, + "poly": [ + 765, + 1200, + 791, + 1200, + 791, + 1240, + 765, + 1240 + ], + "score": 0.88, + "latex": "\\frac { C } { 4 }" + }, + { + "category_id": 13, + "poly": [ + 806, + 1809, + 900, + 1809, + 900, + 1838, + 806, + 1838 + ], + "score": 0.88, + "latex": "3 2 \\times 3 2" + }, + { + "category_id": 13, + "poly": [ + 876, + 204, + 970, + 204, + 970, + 233, + 876, + 233 + ], + "score": 0.88, + "latex": "H \\times W" + }, + { + "category_id": 13, + "poly": [ + 1308, + 652, + 1396, + 652, + 1396, + 684, + 1308, + 684 + ], + "score": 0.87, + "latex": "( H , W )" + }, + { + "category_id": 13, + "poly": [ + 297, + 431, + 395, + 431, + 395, + 461, + 297, + 461 + ], + "score": 0.87, + "latex": "6 4 \\times 6 4 )" + }, + { + "category_id": 13, + "poly": [ + 838, + 594, + 876, + 594, + 876, + 622, + 838, + 622 + ], + "score": 0.87, + "latex": "2 \\times" + }, + { + "category_id": 13, + "poly": [ + 549, + 326, + 573, + 326, + 573, + 352, + 549, + 352 + ], + "score": 0.83, + "latex": "C" + }, + { + "category_id": 13, + "poly": [ + 666, + 296, + 690, + 296, + 690, + 323, + 666, + 323 + ], + "score": 0.82, + "latex": "C" + }, + { + "category_id": 13, + "poly": [ + 802, + 1131, + 943, + 1131, + 943, + 1162, + 802, + 1162 + ], + "score": 0.79, + "latex": "( P , 2 P , 4 P )" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1501.0, + 1304.0, + 1501.0, + 1304.0, + 1550.0, + 289.0, + 1550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 736.0, + 672.0, + 736.0, + 672.0, + 776.0, + 294.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2058.0, + 861.0, + 2058.0, + 861.0, + 2093.0, + 839.0, + 2093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 367.0, + 1406.0, + 367.0, + 1406.0, + 406.0, + 292.0, + 406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 401.0, + 1404.0, + 401.0, + 1404.0, + 436.0, + 294.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 428.0, + 1406.0, + 428.0, + 1406.0, + 467.0, + 396.0, + 467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 453.0, + 484.0, + 453.0, + 484.0, + 498.0, + 291.0, + 498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 453.0, + 1407.0, + 453.0, + 1407.0, + 498.0, + 690.0, + 498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 490.0, + 1105.0, + 490.0, + 1105.0, + 531.0, + 288.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1329.0, + 490.0, + 1409.0, + 490.0, + 1409.0, + 531.0, + 1329.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 529.0, + 589.0, + 529.0, + 589.0, + 564.0, + 294.0, + 564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 627.0, + 529.0, + 1404.0, + 529.0, + 1404.0, + 564.0, + 627.0, + 564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 561.0, + 1407.0, + 561.0, + 1407.0, + 597.0, + 292.0, + 597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 591.0, + 837.0, + 591.0, + 837.0, + 628.0, + 292.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 591.0, + 1406.0, + 591.0, + 1406.0, + 628.0, + 877.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 620.0, + 1407.0, + 620.0, + 1407.0, + 661.0, + 291.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 652.0, + 1307.0, + 652.0, + 1307.0, + 688.0, + 295.0, + 688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1397.0, + 652.0, + 1408.0, + 652.0, + 1408.0, + 688.0, + 1397.0, + 688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 680.0, + 1214.0, + 680.0, + 1214.0, + 719.0, + 290.0, + 719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1386.0, + 680.0, + 1402.0, + 680.0, + 1402.0, + 719.0, + 1386.0, + 719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1069.0, + 1405.0, + 1069.0, + 1405.0, + 1104.0, + 295.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1093.0, + 1014.0, + 1093.0, + 1014.0, + 1137.0, + 290.0, + 1137.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1187.0, + 1093.0, + 1408.0, + 1093.0, + 1408.0, + 1137.0, + 1187.0, + 1137.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1120.0, + 801.0, + 1120.0, + 801.0, + 1174.0, + 289.0, + 1174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 1120.0, + 1211.0, + 1120.0, + 1211.0, + 1174.0, + 944.0, + 1174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1375.0, + 1120.0, + 1411.0, + 1120.0, + 1411.0, + 1174.0, + 1375.0, + 1174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1156.0, + 563.0, + 1156.0, + 563.0, + 1214.0, + 294.0, + 1214.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 1156.0, + 1412.0, + 1156.0, + 1412.0, + 1214.0, + 736.0, + 1214.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1199.0, + 764.0, + 1199.0, + 764.0, + 1239.0, + 291.0, + 1239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 792.0, + 1199.0, + 1407.0, + 1199.0, + 1407.0, + 1239.0, + 792.0, + 1239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 1221.0, + 849.0, + 1221.0, + 849.0, + 1290.0, + 285.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1039.0, + 1221.0, + 1095.0, + 1221.0, + 1095.0, + 1290.0, + 1039.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1286.0, + 1221.0, + 1414.0, + 1221.0, + 1414.0, + 1290.0, + 1286.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1268.0, + 1406.0, + 1268.0, + 1406.0, + 1306.0, + 291.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1299.0, + 1405.0, + 1299.0, + 1405.0, + 1336.0, + 294.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1328.0, + 1405.0, + 1328.0, + 1405.0, + 1365.0, + 294.0, + 1365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1362.0, + 1405.0, + 1362.0, + 1405.0, + 1396.0, + 294.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1391.0, + 1407.0, + 1391.0, + 1407.0, + 1426.0, + 295.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1418.0, + 1406.0, + 1418.0, + 1406.0, + 1458.0, + 291.0, + 1458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1454.0, + 731.0, + 1454.0, + 731.0, + 1485.0, + 295.0, + 1485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 781.0, + 1405.0, + 781.0, + 1405.0, + 818.0, + 296.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 813.0, + 1402.0, + 813.0, + 1402.0, + 845.0, + 296.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 842.0, + 1408.0, + 842.0, + 1408.0, + 879.0, + 295.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 874.0, + 1408.0, + 874.0, + 1408.0, + 907.0, + 294.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 903.0, + 1407.0, + 903.0, + 1407.0, + 940.0, + 295.0, + 940.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 931.0, + 1408.0, + 931.0, + 1408.0, + 970.0, + 292.0, + 970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 963.0, + 1405.0, + 963.0, + 1405.0, + 1000.0, + 295.0, + 1000.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 994.0, + 1403.0, + 994.0, + 1403.0, + 1031.0, + 295.0, + 1031.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1024.0, + 1259.0, + 1024.0, + 1259.0, + 1061.0, + 295.0, + 1061.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1717.0, + 1406.0, + 1717.0, + 1406.0, + 1752.0, + 294.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1749.0, + 1408.0, + 1749.0, + 1408.0, + 1784.0, + 294.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1777.0, + 1406.0, + 1777.0, + 1406.0, + 1813.0, + 293.0, + 1813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1811.0, + 805.0, + 1811.0, + 805.0, + 1841.0, + 296.0, + 1841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 1811.0, + 1402.0, + 1811.0, + 1402.0, + 1841.0, + 901.0, + 1841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1840.0, + 1404.0, + 1840.0, + 1404.0, + 1874.0, + 294.0, + 1874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1869.0, + 1405.0, + 1869.0, + 1405.0, + 1904.0, + 294.0, + 1904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1900.0, + 1053.0, + 1900.0, + 1053.0, + 1934.0, + 296.0, + 1934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1551.0, + 1405.0, + 1551.0, + 1405.0, + 1588.0, + 295.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1580.0, + 1407.0, + 1580.0, + 1407.0, + 1615.0, + 294.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1610.0, + 1406.0, + 1610.0, + 1406.0, + 1646.0, + 292.0, + 1646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1642.0, + 1408.0, + 1642.0, + 1408.0, + 1679.0, + 294.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1671.0, + 923.0, + 1671.0, + 923.0, + 1709.0, + 290.0, + 1709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 202.0, + 875.0, + 202.0, + 875.0, + 239.0, + 294.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 971.0, + 202.0, + 1405.0, + 202.0, + 1405.0, + 239.0, + 971.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 235.0, + 1404.0, + 235.0, + 1404.0, + 268.0, + 295.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 261.0, + 372.0, + 261.0, + 372.0, + 300.0, + 293.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 261.0, + 872.0, + 261.0, + 872.0, + 300.0, + 540.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 261.0, + 1405.0, + 261.0, + 1405.0, + 300.0, + 984.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 293.0, + 341.0, + 293.0, + 341.0, + 330.0, + 295.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 293.0, + 665.0, + 293.0, + 665.0, + 330.0, + 510.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 293.0, + 1405.0, + 293.0, + 1405.0, + 330.0, + 691.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 322.0, + 548.0, + 322.0, + 548.0, + 363.0, + 294.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 322.0, + 1359.0, + 322.0, + 1359.0, + 363.0, + 574.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1945.0, + 1405.0, + 1945.0, + 1405.0, + 1979.0, + 295.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1973.0, + 1404.0, + 1973.0, + 1404.0, + 2012.0, + 294.0, + 2012.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1316, + 1405, + 1316, + 1405, + 1508, + 297, + 1508 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 296, + 856, + 1405, + 856, + 1405, + 1103, + 296, + 1103 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 296, + 1771, + 1405, + 1771, + 1405, + 2009, + 296, + 2009 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 691, + 1404, + 691, + 1404, + 844, + 297, + 844 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 1605, + 1404, + 1605, + 1404, + 1759, + 297, + 1759 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 297, + 1114, + 1405, + 1114, + 1405, + 1237, + 297, + 1237 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 504, + 1404, + 504, + 1404, + 627, + 298, + 627 + ], + "score": 0.977 + }, + { + "category_id": 3, + "poly": [ + 309, + 171, + 1396, + 171, + 1396, + 396, + 309, + 396 + ], + "score": 0.954 + }, + { + "category_id": 4, + "poly": [ + 295, + 400, + 1403, + 400, + 1403, + 490, + 295, + 490 + ], + "score": 0.954 + }, + { + "category_id": 8, + "poly": [ + 556, + 1521, + 1144, + 1521, + 1144, + 1595, + 556, + 1595 + ], + "score": 0.941 + }, + { + "category_id": 0, + "poly": [ + 300, + 648, + 723, + 648, + 723, + 682, + 300, + 682 + ], + "score": 0.929 + }, + { + "category_id": 8, + "poly": [ + 596, + 1238, + 1104, + 1238, + 1104, + 1310, + 596, + 1310 + ], + "score": 0.929 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1543, + 1400, + 1543, + 1400, + 1573, + 1366, + 1573 + ], + "score": 0.881 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1259, + 1400, + 1259, + 1400, + 1288, + 1366, + 1288 + ], + "score": 0.88 + }, + { + "category_id": 2, + "poly": [ + 840, + 2062, + 859, + 2062, + 859, + 2085, + 840, + 2085 + ], + "score": 0.795 + }, + { + "category_id": 13, + "poly": [ + 607, + 1895, + 952, + 1895, + 952, + 1950, + 607, + 1950 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { Y = X / \\sqrt { \\frac { 1 } { C } \\sum _ { i = 0 } ^ { C - 1 } ( X ^ { i } ) ^ { 2 } + \\epsilon } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 554, + 1518, + 1146, + 1518, + 1146, + 1596, + 554, + 1596 + ], + "score": 0.93, + "latex": "A t t e n t i o n ( Q , K , V ) = s o f t m a x ( ( ( \\frac { Q K ^ { T } } { \\sqrt { d _ { k } } } + E ) V )" + }, + { + "category_id": 14, + "poly": [ + 593, + 1233, + 1107, + 1233, + 1107, + 1311, + 593, + 1311 + ], + "score": 0.92, + "latex": "A t t e n t i o n ( Q , K , V ) = s o f t m a x ( ( \\frac { Q K ^ { T } } { \\sqrt { d _ { k } } } V )" + }, + { + "category_id": 13, + "poly": [ + 598, + 1473, + 665, + 1473, + 665, + 1506, + 598, + 1506 + ], + "score": 0.92, + "latex": "Q K ^ { T }" + }, + { + "category_id": 13, + "poly": [ + 369, + 1317, + 620, + 1317, + 620, + 1352, + 369, + 1352 + ], + "score": 0.91, + "latex": "Q , K , V \\in \\mathbb { R } ^ { ( H \\times W ) \\times C }" + }, + { + "category_id": 13, + "poly": [ + 684, + 1380, + 903, + 1380, + 903, + 1414, + 684, + 1414 + ], + "score": 0.91, + "latex": "[ - ( H - \\bar { 1 } ) , H - 1 ]" + }, + { + "category_id": 13, + "poly": [ + 298, + 1440, + 556, + 1440, + 556, + 1473, + 298, + 1473 + ], + "score": 0.9, + "latex": "M \\in \\mathbb { R } ^ { ( 2 \\hat { H _ { - 1 } } ) \\times ( 2 W - 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 979, + 430, + 1075, + 430, + 1075, + 459, + 979, + 459 + ], + "score": 0.9, + "latex": "3 2 \\times 3 2" + }, + { + "category_id": 13, + "poly": [ + 1038, + 1909, + 1165, + 1909, + 1165, + 1938, + 1038, + 1938 + ], + "score": 0.89, + "latex": "\\epsilon = 1 e - 8" + }, + { + "category_id": 13, + "poly": [ + 296, + 460, + 389, + 460, + 389, + 490, + 296, + 490 + ], + "score": 0.88, + "latex": "1 6 \\times 1 6" + }, + { + "category_id": 13, + "poly": [ + 296, + 980, + 365, + 980, + 365, + 1009, + 296, + 1009 + ], + "score": 0.87, + "latex": "100 \\%" + }, + { + "category_id": 13, + "poly": [ + 1025, + 1320, + 1119, + 1320, + 1119, + 1351, + 1025, + 1351 + ], + "score": 0.84, + "latex": "H , W , C" + }, + { + "category_id": 13, + "poly": [ + 491, + 1412, + 519, + 1412, + 519, + 1438, + 491, + 1438 + ], + "score": 0.84, + "latex": "H" + }, + { + "category_id": 13, + "poly": [ + 1067, + 1446, + 1092, + 1446, + 1092, + 1472, + 1067, + 1472 + ], + "score": 0.83, + "latex": "E" + }, + { + "category_id": 13, + "poly": [ + 380, + 1382, + 409, + 1382, + 409, + 1408, + 380, + 1408 + ], + "score": 0.8, + "latex": "H" + }, + { + "category_id": 13, + "poly": [ + 1295, + 1909, + 1324, + 1909, + 1324, + 1937, + 1295, + 1937 + ], + "score": 0.8, + "latex": "X" + }, + { + "category_id": 13, + "poly": [ + 1323, + 1446, + 1355, + 1446, + 1355, + 1472, + 1323, + 1472 + ], + "score": 0.77, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 840, + 1948, + 864, + 1948, + 864, + 1973, + 840, + 1973 + ], + "score": 0.75, + "latex": "C" + }, + { + "category_id": 13, + "poly": [ + 570, + 1411, + 603, + 1411, + 603, + 1438, + 570, + 1438 + ], + "score": 0.71, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 1375, + 1909, + 1402, + 1909, + 1402, + 1937, + 1375, + 1937 + ], + "score": 0.68, + "latex": "Y" + }, + { + "category_id": 13, + "poly": [ + 1088, + 1381, + 1121, + 1381, + 1121, + 1409, + 1088, + 1409 + ], + "score": 0.58, + "latex": "W" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 173.0, + 390.0, + 173.0, + 390.0, + 198.0, + 308.0, + 198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 458.0, + 171.0, + 519.0, + 171.0, + 519.0, + 201.0, + 458.0, + 201.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 172.0, + 716.0, + 172.0, + 716.0, + 198.0, + 628.0, + 198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.0, + 173.0, + 986.0, + 173.0, + 986.0, + 198.0, + 906.0, + 198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 171.0, + 1115.0, + 171.0, + 1115.0, + 201.0, + 1054.0, + 201.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1226.0, + 175.0, + 1311.0, + 175.0, + 1311.0, + 197.0, + 1226.0, + 197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 315.0, + 195.0, + 385.0, + 195.0, + 385.0, + 220.0, + 315.0, + 220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 446.0, + 195.0, + 531.0, + 195.0, + 531.0, + 220.0, + 446.0, + 220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 631.0, + 193.0, + 716.0, + 193.0, + 716.0, + 219.0, + 631.0, + 219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 911.0, + 193.0, + 984.0, + 193.0, + 984.0, + 223.0, + 911.0, + 223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 195.0, + 1127.0, + 195.0, + 1127.0, + 219.0, + 1042.0, + 219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1227.0, + 195.0, + 1311.0, + 195.0, + 1311.0, + 219.0, + 1227.0, + 219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 219.0, + 1265.0, + 219.0, + 1265.0, + 228.0, + 1238.0, + 228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1268.0, + 219.0, + 1301.0, + 219.0, + 1301.0, + 228.0, + 1268.0, + 228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 238.0, + 1076.0, + 238.0, + 1076.0, + 247.0, + 1029.0, + 247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 923.0, + 280.0, + 939.0, + 280.0, + 939.0, + 290.0, + 923.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 375.0, + 630.0, + 375.0, + 630.0, + 399.0, + 423.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1071.0, + 372.0, + 1244.0, + 372.0, + 1244.0, + 400.0, + 1071.0, + 400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 400.0, + 1401.0, + 400.0, + 1401.0, + 433.0, + 295.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 430.0, + 978.0, + 430.0, + 978.0, + 463.0, + 295.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1076.0, + 430.0, + 1404.0, + 430.0, + 1404.0, + 463.0, + 1076.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 460.0, + 514.0, + 460.0, + 514.0, + 491.0, + 390.0, + 491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 644.0, + 724.0, + 644.0, + 724.0, + 689.0, + 293.0, + 689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2060.0, + 862.0, + 2060.0, + 862.0, + 2091.0, + 840.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1309.0, + 368.0, + 1309.0, + 368.0, + 1361.0, + 289.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 621.0, + 1309.0, + 1024.0, + 1309.0, + 1024.0, + 1361.0, + 621.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1120.0, + 1309.0, + 1411.0, + 1309.0, + 1411.0, + 1361.0, + 1120.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1349.0, + 1403.0, + 1349.0, + 1403.0, + 1382.0, + 296.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1378.0, + 379.0, + 1378.0, + 379.0, + 1415.0, + 294.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 1378.0, + 683.0, + 1378.0, + 683.0, + 1415.0, + 410.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 1378.0, + 1087.0, + 1378.0, + 1087.0, + 1415.0, + 904.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1122.0, + 1378.0, + 1403.0, + 1378.0, + 1403.0, + 1415.0, + 1122.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1412.0, + 490.0, + 1412.0, + 490.0, + 1445.0, + 296.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 520.0, + 1412.0, + 569.0, + 1412.0, + 569.0, + 1445.0, + 520.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 604.0, + 1412.0, + 1405.0, + 1412.0, + 1405.0, + 1445.0, + 604.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1434.0, + 297.0, + 1434.0, + 297.0, + 1484.0, + 292.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 557.0, + 1434.0, + 1066.0, + 1434.0, + 1066.0, + 1484.0, + 557.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 1434.0, + 1322.0, + 1434.0, + 1322.0, + 1484.0, + 1093.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1356.0, + 1434.0, + 1409.0, + 1434.0, + 1409.0, + 1484.0, + 1356.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1469.0, + 597.0, + 1469.0, + 597.0, + 1512.0, + 292.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 666.0, + 1469.0, + 1066.0, + 1469.0, + 1066.0, + 1512.0, + 666.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 857.0, + 1407.0, + 857.0, + 1407.0, + 893.0, + 294.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 888.0, + 1403.0, + 888.0, + 1403.0, + 922.0, + 294.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 917.0, + 1403.0, + 917.0, + 1403.0, + 953.0, + 294.0, + 953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 949.0, + 1405.0, + 949.0, + 1405.0, + 983.0, + 293.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 977.0, + 295.0, + 977.0, + 295.0, + 1018.0, + 291.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 366.0, + 977.0, + 1406.0, + 977.0, + 1406.0, + 1018.0, + 366.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1007.0, + 1406.0, + 1007.0, + 1406.0, + 1046.0, + 293.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1039.0, + 1405.0, + 1039.0, + 1405.0, + 1073.0, + 293.0, + 1073.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1071.0, + 1281.0, + 1071.0, + 1281.0, + 1105.0, + 295.0, + 1105.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1766.0, + 1406.0, + 1766.0, + 1406.0, + 1812.0, + 291.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1802.0, + 1403.0, + 1802.0, + 1403.0, + 1837.0, + 295.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1831.0, + 1405.0, + 1831.0, + 1405.0, + 1869.0, + 293.0, + 1869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1861.0, + 1406.0, + 1861.0, + 1406.0, + 1901.0, + 291.0, + 1901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 283.0, + 1886.0, + 606.0, + 1886.0, + 606.0, + 1960.0, + 283.0, + 1960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 953.0, + 1886.0, + 1037.0, + 1886.0, + 1037.0, + 1960.0, + 953.0, + 1960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1166.0, + 1886.0, + 1294.0, + 1886.0, + 1294.0, + 1960.0, + 1166.0, + 1960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 1886.0, + 1374.0, + 1886.0, + 1374.0, + 1960.0, + 1325.0, + 1960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1403.0, + 1886.0, + 1416.0, + 1886.0, + 1416.0, + 1960.0, + 1403.0, + 1960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1944.0, + 839.0, + 1944.0, + 839.0, + 1981.0, + 295.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 865.0, + 1944.0, + 1406.0, + 1944.0, + 1406.0, + 1981.0, + 865.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1976.0, + 1405.0, + 1976.0, + 1405.0, + 2009.0, + 295.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 690.0, + 1404.0, + 690.0, + 1404.0, + 727.0, + 295.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 719.0, + 1405.0, + 719.0, + 1405.0, + 756.0, + 295.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 753.0, + 1405.0, + 753.0, + 1405.0, + 786.0, + 297.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 782.0, + 1405.0, + 782.0, + 1405.0, + 818.0, + 294.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 811.0, + 1148.0, + 811.0, + 1148.0, + 849.0, + 294.0, + 849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1605.0, + 1402.0, + 1605.0, + 1402.0, + 1642.0, + 295.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1630.0, + 1405.0, + 1630.0, + 1405.0, + 1677.0, + 292.0, + 1677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1663.0, + 1405.0, + 1663.0, + 1405.0, + 1703.0, + 292.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1694.0, + 1405.0, + 1694.0, + 1405.0, + 1735.0, + 292.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1724.0, + 712.0, + 1724.0, + 712.0, + 1764.0, + 294.0, + 1764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1114.0, + 1403.0, + 1114.0, + 1403.0, + 1150.0, + 294.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1146.0, + 1408.0, + 1146.0, + 1408.0, + 1182.0, + 294.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1176.0, + 1406.0, + 1176.0, + 1406.0, + 1212.0, + 294.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1204.0, + 518.0, + 1204.0, + 518.0, + 1242.0, + 294.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 502.0, + 1406.0, + 502.0, + 1406.0, + 541.0, + 292.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 535.0, + 1402.0, + 535.0, + 1402.0, + 568.0, + 294.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 564.0, + 1405.0, + 564.0, + 1405.0, + 600.0, + 293.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 598.0, + 624.0, + 598.0, + 624.0, + 627.0, + 294.0, + 627.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1081, + 1405, + 1081, + 1405, + 1418, + 297, + 1418 + ], + "score": 0.984 + }, + { + "category_id": 5, + "poly": [ + 312, + 264, + 1382, + 264, + 1382, + 618, + 312, + 618 + ], + "score": 0.982, + "html": "
MethodsCIFAR-10STL-10CelebA
IS↑FID↓IS↑FID↓FID↓
WGAN-GP [1]6.49 ± 0.0939.681=
SN-GAN [48]8.22 ± 0.059.16 ± 0.1240.1
AutoGAN[18]8.55 ±0.1012.429.16 ± 0.1231.01
AdversarialNAS-GAN[18]8.74 ±0.0710.879.63 ± 0.1926.98=
Progressive-GAN[16]8.80 ± 0.0515.52=7.30
COCO-GAN [74]-=-5.74
StyleGAN-V2 [69]9.1811.0710.21* ± 0.1420.84*5.59*
StyleGAN-V2 + DiffAug. [69]9.409.8910.31*± 0.1219.15*5.40*
TransGAN9.02 ± 0.129.2610.43 ± 0.1618.285.28
" + }, + { + "category_id": 1, + "poly": [ + 297, + 794, + 1405, + 794, + 1405, + 1068, + 297, + 1068 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 1482, + 1406, + 1482, + 1406, + 1697, + 297, + 1697 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 1708, + 1404, + 1708, + 1404, + 1922, + 297, + 1922 + ], + "score": 0.972 + }, + { + "category_id": 6, + "poly": [ + 297, + 166, + 1407, + 166, + 1407, + 258, + 297, + 258 + ], + "score": 0.959 + }, + { + "category_id": 2, + "poly": [ + 330, + 1948, + 1030, + 1948, + 1030, + 2007, + 330, + 2007 + ], + "score": 0.952 + }, + { + "category_id": 0, + "poly": [ + 299, + 1440, + 838, + 1440, + 838, + 1472, + 299, + 1472 + ], + "score": 0.909 + }, + { + "category_id": 0, + "poly": [ + 297, + 747, + 534, + 747, + 534, + 784, + 297, + 784 + ], + "score": 0.906 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 858, + 2061, + 858, + 2084, + 841, + 2084 + ], + "score": 0.742 + }, + { + "category_id": 1, + "poly": [ + 302, + 639, + 1399, + 639, + 1399, + 732, + 302, + 732 + ], + "score": 0.697 + }, + { + "category_id": 7, + "poly": [ + 302, + 639, + 1399, + 639, + 1399, + 732, + 302, + 732 + ], + "score": 0.336 + }, + { + "category_id": 13, + "poly": [ + 1268, + 1113, + 1353, + 1113, + 1353, + 1144, + 1268, + 1144 + ], + "score": 0.92, + "latex": "\\beta _ { 1 } = 0" + }, + { + "category_id": 13, + "poly": [ + 710, + 1038, + 831, + 1038, + 831, + 1066, + 710, + 1066 + ], + "score": 0.9, + "latex": "2 5 6 \\times 2 5 6" + }, + { + "category_id": 13, + "poly": [ + 298, + 1144, + 414, + 1144, + 414, + 1175, + 298, + 1175 + ], + "score": 0.9, + "latex": "\\beta _ { 2 } = 0 . 9 9" + }, + { + "category_id": 13, + "poly": [ + 815, + 916, + 910, + 916, + 910, + 945, + 815, + 945 + ], + "score": 0.9, + "latex": "4 8 \\times 4 8" + }, + { + "category_id": 13, + "poly": [ + 1163, + 946, + 1283, + 946, + 1283, + 975, + 1163, + 975 + ], + "score": 0.89, + "latex": "1 2 8 \\times 1 2 8" + }, + { + "category_id": 13, + "poly": [ + 478, + 1113, + 557, + 1113, + 557, + 1142, + 478, + 1142 + ], + "score": 0.87, + "latex": "1 e - 4" + }, + { + "category_id": 13, + "poly": [ + 1100, + 825, + 1246, + 825, + 1246, + 854, + 1100, + 854 + ], + "score": 0.86, + "latex": "6 0 \\mathbf { k } \\ 3 2 \\times 3 2" + }, + { + "category_id": 13, + "poly": [ + 1361, + 1831, + 1402, + 1831, + 1402, + 1860, + 1361, + 1860 + ], + "score": 0.85, + "latex": "2 \\times" + }, + { + "category_id": 13, + "poly": [ + 1266, + 167, + 1398, + 167, + 1398, + 197, + 1266, + 197 + ], + "score": 0.81, + "latex": "( 1 2 8 \\times 1 2 8 )" + }, + { + "category_id": 13, + "poly": [ + 297, + 856, + 345, + 856, + 345, + 885, + 297, + 885 + ], + "score": 0.54, + "latex": "5 0 \\mathrm { k }" + }, + { + "category_id": 13, + "poly": [ + 1355, + 856, + 1403, + 856, + 1403, + 885, + 1355, + 885 + ], + "score": 0.46, + "latex": "5 0 \\mathrm { k }" + }, + { + "category_id": 13, + "poly": [ + 765, + 1007, + 811, + 1007, + 811, + 1035, + 765, + 1035 + ], + "score": 0.43, + "latex": "3 0 \\mathrm { k }" + }, + { + "category_id": 13, + "poly": [ + 1259, + 1007, + 1317, + 1007, + 1317, + 1036, + 1259, + 1036 + ], + "score": 0.36, + "latex": "1 2 5 \\mathrm { k }" + }, + { + "category_id": 13, + "poly": [ + 481, + 554, + 502, + 554, + 502, + 577, + 481, + 577 + ], + "score": 0.35, + "latex": "^ +" + }, + { + "category_id": 13, + "poly": [ + 495, + 856, + 541, + 856, + 541, + 885, + 495, + 885 + ], + "score": 0.3, + "latex": "1 0 \\mathrm { k }" + }, + { + "category_id": 13, + "poly": [ + 297, + 946, + 356, + 946, + 356, + 975, + 297, + 975 + ], + "score": 0.29, + "latex": "2 0 0 \\mathrm { k }" + }, + { + "category_id": 13, + "poly": [ + 296, + 916, + 356, + 916, + 356, + 944, + 296, + 944 + ], + "score": 0.27, + "latex": "1 0 0 \\mathrm { k }" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 164.0, + 1265.0, + 164.0, + 1265.0, + 201.0, + 292.0, + 201.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 164.0, + 1407.0, + 164.0, + 1407.0, + 201.0, + 1399.0, + 201.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 193.0, + 1408.0, + 193.0, + 1408.0, + 230.0, + 291.0, + 230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 226.0, + 736.0, + 226.0, + 736.0, + 262.0, + 295.0, + 262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 333.0, + 1943.0, + 1032.0, + 1943.0, + 1032.0, + 1981.0, + 333.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 1973.0, + 709.0, + 1973.0, + 709.0, + 2010.0, + 330.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1435.0, + 843.0, + 1435.0, + 843.0, + 1478.0, + 291.0, + 1478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 740.0, + 537.0, + 740.0, + 537.0, + 797.0, + 289.0, + 797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2059.0, + 860.0, + 2059.0, + 860.0, + 2091.0, + 839.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 636.0, + 1404.0, + 636.0, + 1404.0, + 675.0, + 293.0, + 675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 670.0, + 1404.0, + 670.0, + 1404.0, + 704.0, + 296.0, + 704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 701.0, + 1233.0, + 701.0, + 1233.0, + 735.0, + 298.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1082.0, + 1406.0, + 1082.0, + 1406.0, + 1117.0, + 296.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1112.0, + 477.0, + 1112.0, + 477.0, + 1146.0, + 294.0, + 1146.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 558.0, + 1112.0, + 1267.0, + 1112.0, + 1267.0, + 1146.0, + 558.0, + 1146.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 1112.0, + 1406.0, + 1112.0, + 1406.0, + 1146.0, + 1354.0, + 1146.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 1144.0, + 1405.0, + 1144.0, + 1405.0, + 1179.0, + 415.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1170.0, + 1406.0, + 1170.0, + 1406.0, + 1212.0, + 292.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1202.0, + 1407.0, + 1202.0, + 1407.0, + 1241.0, + 294.0, + 1241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1233.0, + 1406.0, + 1233.0, + 1406.0, + 1268.0, + 294.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1264.0, + 1406.0, + 1264.0, + 1406.0, + 1299.0, + 295.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1294.0, + 1403.0, + 1294.0, + 1403.0, + 1329.0, + 295.0, + 1329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1319.0, + 1407.0, + 1319.0, + 1407.0, + 1361.0, + 292.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1354.0, + 1405.0, + 1354.0, + 1405.0, + 1391.0, + 294.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1385.0, + 540.0, + 1385.0, + 540.0, + 1425.0, + 294.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 795.0, + 1407.0, + 795.0, + 1407.0, + 827.0, + 295.0, + 827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 823.0, + 1099.0, + 823.0, + 1099.0, + 860.0, + 294.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1247.0, + 823.0, + 1406.0, + 823.0, + 1406.0, + 860.0, + 1247.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 854.0, + 494.0, + 854.0, + 494.0, + 891.0, + 346.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 542.0, + 854.0, + 1354.0, + 854.0, + 1354.0, + 891.0, + 542.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 884.0, + 1406.0, + 884.0, + 1406.0, + 921.0, + 294.0, + 921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 913.0, + 295.0, + 913.0, + 295.0, + 950.0, + 292.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 913.0, + 814.0, + 913.0, + 814.0, + 950.0, + 357.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 911.0, + 913.0, + 1406.0, + 913.0, + 1406.0, + 950.0, + 911.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 943.0, + 296.0, + 943.0, + 296.0, + 980.0, + 292.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 943.0, + 1162.0, + 943.0, + 1162.0, + 980.0, + 357.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1284.0, + 943.0, + 1408.0, + 943.0, + 1408.0, + 980.0, + 1284.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 975.0, + 1405.0, + 975.0, + 1405.0, + 1012.0, + 295.0, + 1012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1002.0, + 764.0, + 1002.0, + 764.0, + 1045.0, + 291.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 1002.0, + 1258.0, + 1002.0, + 1258.0, + 1045.0, + 812.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1318.0, + 1002.0, + 1405.0, + 1002.0, + 1405.0, + 1045.0, + 1318.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1036.0, + 709.0, + 1036.0, + 709.0, + 1069.0, + 295.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 1036.0, + 955.0, + 1036.0, + 955.0, + 1069.0, + 832.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1482.0, + 1404.0, + 1482.0, + 1404.0, + 1517.0, + 296.0, + 1517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1511.0, + 1406.0, + 1511.0, + 1406.0, + 1548.0, + 294.0, + 1548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1541.0, + 1406.0, + 1541.0, + 1406.0, + 1577.0, + 294.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1574.0, + 1408.0, + 1574.0, + 1408.0, + 1608.0, + 294.0, + 1608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1603.0, + 1407.0, + 1603.0, + 1407.0, + 1637.0, + 294.0, + 1637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1634.0, + 1407.0, + 1634.0, + 1407.0, + 1668.0, + 295.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1665.0, + 1375.0, + 1665.0, + 1375.0, + 1699.0, + 295.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1708.0, + 1405.0, + 1708.0, + 1405.0, + 1743.0, + 295.0, + 1743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1739.0, + 1405.0, + 1739.0, + 1405.0, + 1773.0, + 294.0, + 1773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1770.0, + 1406.0, + 1770.0, + 1406.0, + 1804.0, + 294.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1802.0, + 1405.0, + 1802.0, + 1405.0, + 1832.0, + 296.0, + 1832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1828.0, + 1360.0, + 1828.0, + 1360.0, + 1864.0, + 292.0, + 1864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1862.0, + 1404.0, + 1862.0, + 1404.0, + 1896.0, + 294.0, + 1896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1890.0, + 685.0, + 1890.0, + 685.0, + 1925.0, + 294.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 636.0, + 1404.0, + 636.0, + 1404.0, + 675.0, + 293.0, + 675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 670.0, + 1404.0, + 670.0, + 1404.0, + 704.0, + 296.0, + 704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 701.0, + 1233.0, + 701.0, + 1233.0, + 735.0, + 298.0, + 735.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 362, + 963, + 1332, + 963, + 1332, + 1278, + 362, + 1278 + ], + "score": 0.983, + "html": "
Training ConfigurationCelebA (64x64)CelebA (128x128)LSUN Church (256x256)
(A). Standard Self-Attention8.9200M0OM
(B). Nystrom Self-Attention [64]13.4717.4239.92
(C). Axis Self-Attention [67]12.3913.9529.30
(D). Grid Self-Attention9.8910.5820.39
+Multi-scale Discriminator9.288.0315.29
+Modified Normalization7.057.1313.27
+Relative Position Encoding6.146.3211.93
(E). Converge5.015.288.94
" + }, + { + "category_id": 1, + "poly": [ + 297, + 1534, + 1406, + 1534, + 1406, + 1750, + 297, + 1750 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 1822, + 1405, + 1822, + 1405, + 2008, + 298, + 2008 + ], + "score": 0.979 + }, + { + "category_id": 5, + "poly": [ + 331, + 702, + 1367, + 702, + 1367, + 863, + 331, + 863 + ], + "score": 0.977, + "html": "
MethodsWGAN-GPAutoGANStyleGAN-V2TransGAN
IS个FID←IS个FID←IS个FID←IS个FID↓
Original6.4939.688.5512.429.1811.078.3622.53
+ DiffAug [69]6.2937.148.6012.729.409.899.029.26
" + }, + { + "category_id": 3, + "poly": [ + 303, + 181, + 1392, + 181, + 1392, + 533, + 303, + 533 + ], + "score": 0.968 + }, + { + "category_id": 6, + "poly": [ + 299, + 636, + 1396, + 636, + 1396, + 699, + 299, + 699 + ], + "score": 0.936 + }, + { + "category_id": 0, + "poly": [ + 298, + 1778, + 897, + 1778, + 897, + 1812, + 298, + 1812 + ], + "score": 0.934 + }, + { + "category_id": 4, + "poly": [ + 298, + 541, + 1400, + 541, + 1400, + 604, + 298, + 604 + ], + "score": 0.934 + }, + { + "category_id": 0, + "poly": [ + 299, + 1489, + 748, + 1489, + 748, + 1524, + 299, + 1524 + ], + "score": 0.915 + }, + { + "category_id": 6, + "poly": [ + 303, + 896, + 1405, + 896, + 1405, + 959, + 303, + 959 + ], + "score": 0.908 + }, + { + "category_id": 1, + "poly": [ + 297, + 1307, + 1407, + 1307, + 1407, + 1461, + 297, + 1461 + ], + "score": 0.885 + }, + { + "category_id": 2, + "poly": [ + 840, + 2062, + 858, + 2062, + 858, + 2085, + 840, + 2085 + ], + "score": 0.805 + }, + { + "category_id": 7, + "poly": [ + 297, + 1307, + 1407, + 1307, + 1407, + 1461, + 297, + 1461 + ], + "score": 0.108 + }, + { + "category_id": 13, + "poly": [ + 1010, + 1657, + 1131, + 1657, + 1131, + 1687, + 1010, + 1687 + ], + "score": 0.92, + "latex": "2 5 6 \\times 2 5 6" + }, + { + "category_id": 13, + "poly": [ + 558, + 572, + 680, + 572, + 680, + 601, + 558, + 601 + ], + "score": 0.89, + "latex": "2 5 6 \\times 2 5 6" + }, + { + "category_id": 13, + "poly": [ + 432, + 572, + 527, + 572, + 527, + 601, + 432, + 601 + ], + "score": 0.89, + "latex": "3 2 \\times 3 2" + }, + { + "category_id": 13, + "poly": [ + 1151, + 1400, + 1281, + 1400, + 1281, + 1429, + 1151, + 1429 + ], + "score": 0.87, + "latex": "( 1 2 8 \\times 1 2 8 )" + }, + { + "category_id": 13, + "poly": [ + 1285, + 1308, + 1403, + 1308, + 1403, + 1337, + 1285, + 1337 + ], + "score": 0.87, + "latex": "( 1 2 8 \\times 1 2 8" + }, + { + "category_id": 13, + "poly": [ + 888, + 1536, + 1019, + 1536, + 1019, + 1566, + 888, + 1566 + ], + "score": 0.86, + "latex": "( 2 5 6 \\times 2 5 6 )" + }, + { + "category_id": 13, + "poly": [ + 1289, + 897, + 1394, + 897, + 1394, + 928, + 1289, + 928 + ], + "score": 0.84, + "latex": "( 6 4 \\times 6 4 )" + }, + { + "category_id": 13, + "poly": [ + 388, + 927, + 512, + 927, + 512, + 956, + 388, + 956 + ], + "score": 0.83, + "latex": "( 1 2 8 \\times 1 2 8" + }, + { + "category_id": 13, + "poly": [ + 736, + 927, + 869, + 927, + 869, + 958, + 736, + 958 + ], + "score": 0.83, + "latex": "( 2 5 6 \\times 2 5 6 )" + }, + { + "category_id": 13, + "poly": [ + 403, + 1189, + 423, + 1189, + 423, + 1210, + 403, + 1210 + ], + "score": 0.59, + "latex": "^ +" + }, + { + "category_id": 13, + "poly": [ + 403, + 1160, + 423, + 1160, + 423, + 1182, + 403, + 1182 + ], + "score": 0.54, + "latex": "^ +" + }, + { + "category_id": 13, + "poly": [ + 403, + 1216, + 423, + 1216, + 423, + 1238, + 403, + 1238 + ], + "score": 0.45, + "latex": "^ +" + }, + { + "category_id": 13, + "poly": [ + 1181, + 1657, + 1287, + 1657, + 1287, + 1687, + 1181, + 1687 + ], + "score": 0.26, + "latex": "8 . 9 4 \\mathrm { F I D }" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 182.0, + 476.0, + 182.0, + 476.0, + 216.0, + 362.0, + 216.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 617.0, + 182.0, + 705.0, + 182.0, + 705.0, + 217.0, + 617.0, + 217.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 900.0, + 181.0, + 986.0, + 181.0, + 986.0, + 216.0, + 900.0, + 216.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1135.0, + 182.0, + 1363.0, + 182.0, + 1363.0, + 216.0, + 1135.0, + 216.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 209.0, + 462.0, + 209.0, + 462.0, + 244.0, + 376.0, + 244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 208.0, + 705.0, + 208.0, + 705.0, + 247.0, + 616.0, + 247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 208.0, + 997.0, + 208.0, + 997.0, + 245.0, + 890.0, + 245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1192.0, + 207.0, + 1305.0, + 207.0, + 1305.0, + 244.0, + 1192.0, + 244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 474.0, + 1380.0, + 474.0, + 1380.0, + 529.0, + 1238.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 482.0, + 1239.0, + 482.0, + 1239.0, + 520.0, + 1183.0, + 520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 633.0, + 1402.0, + 633.0, + 1402.0, + 671.0, + 294.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 666.0, + 843.0, + 666.0, + 843.0, + 702.0, + 295.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1774.0, + 900.0, + 1774.0, + 900.0, + 1816.0, + 292.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 539.0, + 1404.0, + 539.0, + 1404.0, + 575.0, + 296.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 569.0, + 431.0, + 569.0, + 431.0, + 607.0, + 292.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 528.0, + 569.0, + 557.0, + 569.0, + 557.0, + 607.0, + 528.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 569.0, + 1258.0, + 569.0, + 1258.0, + 607.0, + 681.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1488.0, + 749.0, + 1488.0, + 749.0, + 1528.0, + 294.0, + 1528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 896.0, + 1288.0, + 896.0, + 1288.0, + 931.0, + 297.0, + 931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1395.0, + 896.0, + 1407.0, + 896.0, + 1407.0, + 931.0, + 1395.0, + 931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 926.0, + 387.0, + 926.0, + 387.0, + 962.0, + 298.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 926.0, + 735.0, + 926.0, + 735.0, + 962.0, + 513.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 926.0, + 1349.0, + 926.0, + 1349.0, + 962.0, + 870.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2090.0, + 838.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1306.0, + 1284.0, + 1306.0, + 1284.0, + 1343.0, + 296.0, + 1343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 1306.0, + 1407.0, + 1306.0, + 1407.0, + 1343.0, + 1404.0, + 1343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1336.0, + 1407.0, + 1336.0, + 1407.0, + 1373.0, + 294.0, + 1373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1365.0, + 1403.0, + 1365.0, + 1403.0, + 1403.0, + 296.0, + 1403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1397.0, + 1150.0, + 1397.0, + 1150.0, + 1434.0, + 294.0, + 1434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1282.0, + 1397.0, + 1405.0, + 1397.0, + 1405.0, + 1434.0, + 1282.0, + 1434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1426.0, + 1104.0, + 1426.0, + 1104.0, + 1466.0, + 294.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1533.0, + 887.0, + 1533.0, + 887.0, + 1570.0, + 294.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1020.0, + 1533.0, + 1409.0, + 1533.0, + 1409.0, + 1570.0, + 1020.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1560.0, + 1407.0, + 1560.0, + 1407.0, + 1605.0, + 291.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1594.0, + 1406.0, + 1594.0, + 1406.0, + 1630.0, + 294.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1625.0, + 1407.0, + 1625.0, + 1407.0, + 1664.0, + 292.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1657.0, + 1009.0, + 1657.0, + 1009.0, + 1689.0, + 292.0, + 1689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 1657.0, + 1180.0, + 1657.0, + 1180.0, + 1689.0, + 1132.0, + 1689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1288.0, + 1657.0, + 1404.0, + 1657.0, + 1404.0, + 1689.0, + 1288.0, + 1689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1686.0, + 1406.0, + 1686.0, + 1406.0, + 1721.0, + 296.0, + 1721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1716.0, + 1409.0, + 1716.0, + 1409.0, + 1753.0, + 292.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1822.0, + 1408.0, + 1822.0, + 1408.0, + 1858.0, + 295.0, + 1858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1854.0, + 1405.0, + 1854.0, + 1405.0, + 1889.0, + 295.0, + 1889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1882.0, + 1405.0, + 1882.0, + 1405.0, + 1918.0, + 295.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1912.0, + 1408.0, + 1912.0, + 1408.0, + 1951.0, + 292.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1944.0, + 1405.0, + 1944.0, + 1405.0, + 1980.0, + 293.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1975.0, + 1405.0, + 1975.0, + 1405.0, + 2009.0, + 293.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1306.0, + 1284.0, + 1306.0, + 1284.0, + 1343.0, + 296.0, + 1343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 1306.0, + 1407.0, + 1306.0, + 1407.0, + 1343.0, + 1404.0, + 1343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1336.0, + 1407.0, + 1336.0, + 1407.0, + 1373.0, + 294.0, + 1373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1365.0, + 1403.0, + 1365.0, + 1403.0, + 1403.0, + 296.0, + 1403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1397.0, + 1150.0, + 1397.0, + 1150.0, + 1434.0, + 294.0, + 1434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1282.0, + 1397.0, + 1405.0, + 1397.0, + 1405.0, + 1434.0, + 1282.0, + 1434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1426.0, + 1104.0, + 1426.0, + 1104.0, + 1466.0, + 294.0, + 1466.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1702, + 1404, + 1702, + 1404, + 2007, + 298, + 2007 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 297, + 1300, + 1405, + 1300, + 1405, + 1606, + 297, + 1606 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 928, + 1405, + 928, + 1405, + 1233, + 298, + 1233 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 610, + 1404, + 610, + 1404, + 915, + 298, + 915 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 297, + 479, + 1400, + 479, + 1400, + 542, + 297, + 542 + ], + "score": 0.952 + }, + { + "category_id": 3, + "poly": [ + 307, + 171, + 1390, + 171, + 1390, + 385, + 307, + 385 + ], + "score": 0.951 + }, + { + "category_id": 0, + "poly": [ + 301, + 1644, + 1143, + 1644, + 1143, + 1678, + 301, + 1678 + ], + "score": 0.927 + }, + { + "category_id": 0, + "poly": [ + 297, + 1257, + 995, + 1257, + 995, + 1290, + 297, + 1290 + ], + "score": 0.926 + }, + { + "category_id": 4, + "poly": [ + 295, + 399, + 1402, + 399, + 1402, + 462, + 295, + 462 + ], + "score": 0.922 + }, + { + "category_id": 0, + "poly": [ + 298, + 566, + 541, + 566, + 541, + 599, + 298, + 599 + ], + "score": 0.904 + }, + { + "category_id": 2, + "poly": [ + 840, + 2062, + 858, + 2062, + 858, + 2084, + 840, + 2084 + ], + "score": 0.792 + }, + { + "category_id": 13, + "poly": [ + 423, + 1202, + 517, + 1202, + 517, + 1231, + 423, + 1231 + ], + "score": 0.92, + "latex": "6 4 \\times 6 4" + }, + { + "category_id": 13, + "poly": [ + 1120, + 990, + 1204, + 990, + 1204, + 1019, + 1120, + 1019 + ], + "score": 0.9, + "latex": "6 4 \\times 6 4" + }, + { + "category_id": 13, + "poly": [ + 387, + 1081, + 507, + 1081, + 507, + 1110, + 387, + 1110 + ], + "score": 0.89, + "latex": "2 5 6 \\times 2 5 6" + }, + { + "category_id": 13, + "poly": [ + 692, + 1202, + 813, + 1202, + 813, + 1231, + 692, + 1231 + ], + "score": 0.89, + "latex": "1 2 8 \\times 1 2 8" + }, + { + "category_id": 13, + "poly": [ + 1111, + 1202, + 1231, + 1202, + 1231, + 1231, + 1111, + 1231 + ], + "score": 0.88, + "latex": "2 5 6 \\times 2 5 6" + }, + { + "category_id": 13, + "poly": [ + 1036, + 1051, + 1132, + 1051, + 1132, + 1080, + 1036, + 1080 + ], + "score": 0.87, + "latex": "6 4 \\times 6 4" + }, + { + "category_id": 13, + "poly": [ + 296, + 960, + 391, + 960, + 391, + 990, + 296, + 990 + ], + "score": 0.87, + "latex": "3 2 \\times 3 2 )" + }, + { + "category_id": 13, + "poly": [ + 1110, + 793, + 1211, + 793, + 1211, + 823, + 1110, + 823 + ], + "score": 0.87, + "latex": "( 6 4 \\times 6 4 )" + }, + { + "category_id": 13, + "poly": [ + 1145, + 1051, + 1269, + 1051, + 1269, + 1080, + 1145, + 1080 + ], + "score": 0.87, + "latex": "1 2 8 \\times 1 2 8" + }, + { + "category_id": 13, + "poly": [ + 620, + 1886, + 659, + 1886, + 659, + 1914, + 620, + 1914 + ], + "score": 0.86, + "latex": "4 \\times" + }, + { + "category_id": 13, + "poly": [ + 410, + 1916, + 449, + 1916, + 449, + 1944, + 410, + 1944 + ], + "score": 0.86, + "latex": "2 \\times" + }, + { + "category_id": 13, + "poly": [ + 444, + 430, + 573, + 430, + 573, + 460, + 444, + 460 + ], + "score": 0.86, + "latex": "( 2 5 6 \\times 2 5 6 )" + }, + { + "category_id": 13, + "poly": [ + 1099, + 1855, + 1139, + 1855, + 1139, + 1884, + 1099, + 1884 + ], + "score": 0.86, + "latex": "2 \\times" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 179.0, + 332.0, + 179.0, + 332.0, + 260.0, + 308.0, + 260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 209.0, + 707.0, + 209.0, + 707.0, + 230.0, + 685.0, + 230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 172.0, + 1390.0, + 172.0, + 1390.0, + 273.0, + 812.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 287.0, + 331.0, + 287.0, + 331.0, + 327.0, + 310.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 816.0, + 257.0, + 1387.0, + 257.0, + 1387.0, + 350.0, + 816.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 362.0, + 412.0, + 362.0, + 412.0, + 388.0, + 346.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 459.0, + 363.0, + 485.0, + 363.0, + 485.0, + 387.0, + 459.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 549.0, + 363.0, + 576.0, + 363.0, + 576.0, + 386.0, + 549.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 359.0, + 740.0, + 359.0, + 740.0, + 389.0, + 653.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 357.0, + 1207.0, + 357.0, + 1207.0, + 388.0, + 996.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1643.0, + 1146.0, + 1643.0, + 1146.0, + 1683.0, + 295.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1256.0, + 997.0, + 1256.0, + 997.0, + 1293.0, + 293.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 399.0, + 1404.0, + 399.0, + 1404.0, + 434.0, + 295.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 428.0, + 443.0, + 428.0, + 443.0, + 464.0, + 295.0, + 464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 428.0, + 1267.0, + 428.0, + 1267.0, + 464.0, + 574.0, + 464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 559.0, + 545.0, + 559.0, + 545.0, + 608.0, + 293.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2060.0, + 860.0, + 2060.0, + 860.0, + 2090.0, + 839.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1700.0, + 1408.0, + 1700.0, + 1408.0, + 1737.0, + 293.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1733.0, + 1405.0, + 1733.0, + 1405.0, + 1770.0, + 293.0, + 1770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1763.0, + 1406.0, + 1763.0, + 1406.0, + 1799.0, + 293.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1794.0, + 1408.0, + 1794.0, + 1408.0, + 1829.0, + 294.0, + 1829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1824.0, + 1406.0, + 1824.0, + 1406.0, + 1860.0, + 293.0, + 1860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1854.0, + 1098.0, + 1854.0, + 1098.0, + 1890.0, + 294.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 1854.0, + 1406.0, + 1854.0, + 1406.0, + 1890.0, + 1140.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1885.0, + 619.0, + 1885.0, + 619.0, + 1920.0, + 293.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 1885.0, + 1404.0, + 1885.0, + 1404.0, + 1920.0, + 660.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1913.0, + 409.0, + 1913.0, + 409.0, + 1949.0, + 292.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1913.0, + 1405.0, + 1913.0, + 1405.0, + 1949.0, + 450.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1946.0, + 1405.0, + 1946.0, + 1405.0, + 1981.0, + 294.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1976.0, + 1405.0, + 1976.0, + 1405.0, + 2011.0, + 296.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1299.0, + 1406.0, + 1299.0, + 1406.0, + 1336.0, + 294.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1332.0, + 1406.0, + 1332.0, + 1406.0, + 1367.0, + 294.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1360.0, + 1405.0, + 1360.0, + 1405.0, + 1398.0, + 294.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1390.0, + 1406.0, + 1390.0, + 1406.0, + 1431.0, + 292.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1420.0, + 1408.0, + 1420.0, + 1408.0, + 1464.0, + 290.0, + 1464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1454.0, + 1406.0, + 1454.0, + 1406.0, + 1490.0, + 295.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1483.0, + 1408.0, + 1483.0, + 1408.0, + 1519.0, + 294.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1514.0, + 1406.0, + 1514.0, + 1406.0, + 1549.0, + 295.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1544.0, + 1406.0, + 1544.0, + 1406.0, + 1577.0, + 292.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1574.0, + 1172.0, + 1574.0, + 1172.0, + 1609.0, + 292.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 927.0, + 1405.0, + 927.0, + 1405.0, + 963.0, + 295.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 959.0, + 1406.0, + 959.0, + 1406.0, + 991.0, + 392.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 988.0, + 1119.0, + 988.0, + 1119.0, + 1025.0, + 292.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1205.0, + 988.0, + 1408.0, + 988.0, + 1408.0, + 1025.0, + 1205.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1020.0, + 1406.0, + 1020.0, + 1406.0, + 1055.0, + 295.0, + 1055.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1051.0, + 1035.0, + 1051.0, + 1035.0, + 1083.0, + 295.0, + 1083.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1133.0, + 1051.0, + 1144.0, + 1051.0, + 1144.0, + 1083.0, + 1133.0, + 1083.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1270.0, + 1051.0, + 1403.0, + 1051.0, + 1403.0, + 1083.0, + 1270.0, + 1083.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1077.0, + 386.0, + 1077.0, + 386.0, + 1117.0, + 292.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 1077.0, + 1407.0, + 1077.0, + 1407.0, + 1117.0, + 508.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1112.0, + 1403.0, + 1112.0, + 1403.0, + 1144.0, + 296.0, + 1144.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1140.0, + 1405.0, + 1140.0, + 1405.0, + 1174.0, + 293.0, + 1174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1169.0, + 1406.0, + 1169.0, + 1406.0, + 1207.0, + 292.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1202.0, + 422.0, + 1202.0, + 422.0, + 1234.0, + 296.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 1202.0, + 691.0, + 1202.0, + 691.0, + 1234.0, + 518.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 814.0, + 1202.0, + 1110.0, + 1202.0, + 1110.0, + 1234.0, + 814.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1232.0, + 1202.0, + 1314.0, + 1202.0, + 1314.0, + 1234.0, + 1232.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 606.0, + 1406.0, + 606.0, + 1406.0, + 649.0, + 292.0, + 649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 641.0, + 1406.0, + 641.0, + 1406.0, + 674.0, + 293.0, + 674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 673.0, + 1405.0, + 673.0, + 1405.0, + 704.0, + 296.0, + 704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 700.0, + 1406.0, + 700.0, + 1406.0, + 736.0, + 294.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 732.0, + 1404.0, + 732.0, + 1404.0, + 768.0, + 293.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 764.0, + 1405.0, + 764.0, + 1405.0, + 795.0, + 296.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 786.0, + 1109.0, + 786.0, + 1109.0, + 831.0, + 292.0, + 831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1212.0, + 786.0, + 1410.0, + 786.0, + 1410.0, + 831.0, + 1212.0, + 831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 821.0, + 1406.0, + 821.0, + 1406.0, + 857.0, + 292.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 852.0, + 1405.0, + 852.0, + 1405.0, + 888.0, + 293.0, + 888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 883.0, + 724.0, + 883.0, + 724.0, + 918.0, + 294.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 479.0, + 1404.0, + 479.0, + 1404.0, + 514.0, + 296.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 511.0, + 1242.0, + 511.0, + 1242.0, + 546.0, + 296.0, + 546.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 440, + 1405, + 440, + 1405, + 654, + 297, + 654 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 202, + 1404, + 202, + 1404, + 355, + 298, + 355 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 299, + 668, + 1404, + 668, + 1404, + 790, + 299, + 790 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 298, + 877, + 1404, + 877, + 1404, + 999, + 298, + 999 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 303, + 1090, + 1408, + 1090, + 1408, + 2008, + 303, + 2008 + ], + "score": 0.945 + }, + { + "category_id": 0, + "poly": [ + 298, + 384, + 1175, + 384, + 1175, + 423, + 298, + 423 + ], + "score": 0.923 + }, + { + "category_id": 0, + "poly": [ + 299, + 822, + 576, + 822, + 576, + 858, + 299, + 858 + ], + "score": 0.917 + }, + { + "category_id": 0, + "poly": [ + 299, + 1042, + 455, + 1042, + 455, + 1076, + 299, + 1076 + ], + "score": 0.905 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2085, + 836, + 2085 + ], + "score": 0.839 + }, + { + "category_id": 13, + "poly": [ + 852, + 295, + 973, + 295, + 973, + 325, + 852, + 325 + ], + "score": 0.89, + "latex": "2 5 6 \\times 2 5 6" + }, + { + "category_id": 13, + "poly": [ + 334, + 594, + 457, + 594, + 457, + 623, + 334, + 623 + ], + "score": 0.89, + "latex": "2 5 6 \\times 2 5 6" + }, + { + "category_id": 13, + "poly": [ + 296, + 625, + 447, + 625, + 447, + 654, + 296, + 654 + ], + "score": 0.87, + "latex": "1 0 2 4 \\times 1 0 2 4 ," + }, + { + "category_id": 15, + "poly": [ + 290.0, + 381.0, + 1178.0, + 381.0, + 1178.0, + 431.0, + 290.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 816.0, + 579.0, + 816.0, + 579.0, + 867.0, + 295.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1038.0, + 460.0, + 1038.0, + 460.0, + 1082.0, + 295.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2057.0, + 871.0, + 2057.0, + 871.0, + 2096.0, + 832.0, + 2096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 442.0, + 1405.0, + 442.0, + 1405.0, + 476.0, + 295.0, + 476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 472.0, + 1405.0, + 472.0, + 1405.0, + 507.0, + 295.0, + 507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 502.0, + 1406.0, + 502.0, + 1406.0, + 537.0, + 292.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 531.0, + 1406.0, + 531.0, + 1406.0, + 570.0, + 292.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 562.0, + 1403.0, + 562.0, + 1403.0, + 596.0, + 295.0, + 596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 590.0, + 333.0, + 590.0, + 333.0, + 633.0, + 292.0, + 633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 458.0, + 590.0, + 1410.0, + 590.0, + 1410.0, + 633.0, + 458.0, + 633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 623.0, + 885.0, + 623.0, + 885.0, + 656.0, + 448.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 203.0, + 1406.0, + 203.0, + 1406.0, + 238.0, + 293.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 235.0, + 1405.0, + 235.0, + 1405.0, + 268.0, + 296.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 263.0, + 1405.0, + 263.0, + 1405.0, + 298.0, + 293.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 293.0, + 851.0, + 293.0, + 851.0, + 331.0, + 292.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 293.0, + 1407.0, + 293.0, + 1407.0, + 331.0, + 974.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 324.0, + 388.0, + 324.0, + 388.0, + 356.0, + 292.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 665.0, + 1404.0, + 665.0, + 1404.0, + 704.0, + 293.0, + 704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 698.0, + 1408.0, + 698.0, + 1408.0, + 734.0, + 294.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 727.0, + 1405.0, + 727.0, + 1405.0, + 764.0, + 293.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 759.0, + 535.0, + 759.0, + 535.0, + 792.0, + 296.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 877.0, + 1405.0, + 877.0, + 1405.0, + 913.0, + 294.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 907.0, + 1408.0, + 907.0, + 1408.0, + 943.0, + 294.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 937.0, + 1405.0, + 937.0, + 1405.0, + 973.0, + 294.0, + 973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 966.0, + 779.0, + 966.0, + 779.0, + 1002.0, + 293.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 1092.0, + 1405.0, + 1092.0, + 1405.0, + 1128.0, + 308.0, + 1128.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1119.0, + 1409.0, + 1119.0, + 1409.0, + 1157.0, + 347.0, + 1157.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1144.0, + 419.0, + 1144.0, + 419.0, + 1182.0, + 348.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 1198.0, + 1407.0, + 1198.0, + 1407.0, + 1237.0, + 306.0, + 1237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1228.0, + 933.0, + 1228.0, + 933.0, + 1261.0, + 348.0, + 1261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 1279.0, + 1405.0, + 1279.0, + 1405.0, + 1315.0, + 307.0, + 1315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1305.0, + 1407.0, + 1305.0, + 1407.0, + 1344.0, + 347.0, + 1344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1335.0, + 604.0, + 1335.0, + 604.0, + 1368.0, + 348.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 1384.0, + 1407.0, + 1384.0, + 1407.0, + 1426.0, + 305.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1414.0, + 1407.0, + 1414.0, + 1407.0, + 1450.0, + 351.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1442.0, + 673.0, + 1442.0, + 673.0, + 1474.0, + 351.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 1496.0, + 1404.0, + 1496.0, + 1404.0, + 1527.0, + 308.0, + 1527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1518.0, + 1405.0, + 1518.0, + 1405.0, + 1559.0, + 348.0, + 1559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1550.0, + 667.0, + 1550.0, + 667.0, + 1582.0, + 347.0, + 1582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 1601.0, + 1405.0, + 1601.0, + 1405.0, + 1637.0, + 307.0, + 1637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1625.0, + 1407.0, + 1625.0, + 1407.0, + 1667.0, + 347.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1656.0, + 783.0, + 1656.0, + 783.0, + 1692.0, + 351.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 1708.0, + 1405.0, + 1708.0, + 1405.0, + 1744.0, + 307.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1735.0, + 1407.0, + 1735.0, + 1407.0, + 1774.0, + 347.0, + 1774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1765.0, + 793.0, + 1765.0, + 793.0, + 1797.0, + 348.0, + 1797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 1814.0, + 1403.0, + 1814.0, + 1403.0, + 1849.0, + 307.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1842.0, + 1405.0, + 1842.0, + 1405.0, + 1880.0, + 348.0, + 1880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1868.0, + 620.0, + 1868.0, + 620.0, + 1904.0, + 349.0, + 1904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 1921.0, + 1405.0, + 1921.0, + 1405.0, + 1959.0, + 306.0, + 1959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1947.0, + 1405.0, + 1947.0, + 1405.0, + 1988.0, + 348.0, + 1988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1978.0, + 542.0, + 1978.0, + 542.0, + 2011.0, + 350.0, + 2011.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 835, + 2061, + 863, + 2061, + 863, + 2086, + 835, + 2086 + ], + "score": 0.817 + }, + { + "category_id": 1, + "poly": [ + 286, + 115, + 1409, + 115, + 1409, + 2029, + 286, + 2029 + ], + "score": 0.604 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 868.0, + 2058.0, + 868.0, + 2098.0, + 832.0, + 2098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 206.0, + 1406.0, + 206.0, + 1406.0, + 239.0, + 293.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 233.0, + 979.0, + 233.0, + 979.0, + 266.0, + 349.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 279.0, + 1408.0, + 279.0, + 1408.0, + 319.0, + 291.0, + 319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 308.0, + 873.0, + 308.0, + 873.0, + 346.0, + 344.0, + 346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 361.0, + 1406.0, + 361.0, + 1406.0, + 394.0, + 293.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 386.0, + 1408.0, + 386.0, + 1408.0, + 426.0, + 349.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 413.0, + 680.0, + 413.0, + 680.0, + 451.0, + 346.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 466.0, + 1406.0, + 466.0, + 1406.0, + 499.0, + 293.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 493.0, + 699.0, + 493.0, + 699.0, + 528.0, + 344.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 541.0, + 1408.0, + 541.0, + 1408.0, + 579.0, + 289.0, + 579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 568.0, + 1408.0, + 568.0, + 1408.0, + 606.0, + 349.0, + 606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 596.0, + 605.0, + 596.0, + 605.0, + 631.0, + 344.0, + 631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 646.0, + 1406.0, + 646.0, + 1406.0, + 684.0, + 289.0, + 684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 673.0, + 1056.0, + 673.0, + 1056.0, + 711.0, + 346.0, + 711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 724.0, + 1408.0, + 724.0, + 1408.0, + 764.0, + 291.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 753.0, + 1079.0, + 753.0, + 1079.0, + 789.0, + 346.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 803.0, + 1406.0, + 803.0, + 1406.0, + 837.0, + 293.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 831.0, + 1404.0, + 831.0, + 1404.0, + 864.0, + 351.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 856.0, + 788.0, + 856.0, + 788.0, + 894.0, + 344.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 906.0, + 1410.0, + 906.0, + 1410.0, + 946.0, + 291.0, + 946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 931.0, + 1412.0, + 931.0, + 1412.0, + 975.0, + 342.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 963.0, + 605.0, + 963.0, + 605.0, + 996.0, + 344.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1011.0, + 1408.0, + 1011.0, + 1408.0, + 1049.0, + 289.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1043.0, + 1406.0, + 1043.0, + 1406.0, + 1076.0, + 349.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1070.0, + 735.0, + 1070.0, + 735.0, + 1103.0, + 349.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1116.0, + 1410.0, + 1116.0, + 1410.0, + 1154.0, + 289.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 1145.0, + 1408.0, + 1145.0, + 1408.0, + 1183.0, + 346.0, + 1183.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1175.0, + 886.0, + 1175.0, + 886.0, + 1208.0, + 351.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1223.0, + 1404.0, + 1223.0, + 1404.0, + 1257.0, + 295.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 1250.0, + 1289.0, + 1250.0, + 1289.0, + 1290.0, + 346.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1303.0, + 1406.0, + 1303.0, + 1406.0, + 1336.0, + 295.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 1326.0, + 1408.0, + 1326.0, + 1408.0, + 1368.0, + 346.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1362.0, + 538.0, + 1362.0, + 538.0, + 1389.0, + 353.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1408.0, + 1408.0, + 1408.0, + 1408.0, + 1441.0, + 295.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1435.0, + 758.0, + 1435.0, + 758.0, + 1469.0, + 349.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1485.0, + 1404.0, + 1485.0, + 1404.0, + 1519.0, + 293.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 1510.0, + 1399.0, + 1510.0, + 1399.0, + 1548.0, + 346.0, + 1548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1561.0, + 1408.0, + 1561.0, + 1408.0, + 1601.0, + 291.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 1588.0, + 1196.0, + 1588.0, + 1196.0, + 1626.0, + 346.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1634.0, + 1408.0, + 1634.0, + 1408.0, + 1678.0, + 291.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1668.0, + 1230.0, + 1668.0, + 1230.0, + 1701.0, + 351.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1718.0, + 1406.0, + 1718.0, + 1406.0, + 1752.0, + 295.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 1743.0, + 1408.0, + 1743.0, + 1408.0, + 1783.0, + 346.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1796.0, + 1406.0, + 1796.0, + 1406.0, + 1829.0, + 295.0, + 1829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1821.0, + 1378.0, + 1821.0, + 1378.0, + 1861.0, + 349.0, + 1861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1873.0, + 1406.0, + 1873.0, + 1406.0, + 1907.0, + 295.0, + 1907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 1897.0, + 1408.0, + 1897.0, + 1408.0, + 1934.0, + 346.0, + 1934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1928.0, + 1208.0, + 1928.0, + 1208.0, + 1962.0, + 351.0, + 1962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1972.0, + 1340.0, + 1972.0, + 1340.0, + 2014.0, + 289.0, + 2014.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2086, + 836, + 2086 + ], + "score": 0.798 + }, + { + "category_id": 1, + "poly": [ + 293, + 107, + 1410, + 107, + 1410, + 2032, + 293, + 2032 + ], + "score": 0.586 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2086, + 836, + 2086 + ], + "score": 0.136 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 205.0, + 1407.0, + 205.0, + 1407.0, + 238.0, + 294.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 230.0, + 1409.0, + 230.0, + 1409.0, + 270.0, + 349.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 262.0, + 858.0, + 262.0, + 858.0, + 295.0, + 349.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 304.0, + 1405.0, + 304.0, + 1405.0, + 350.0, + 292.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 340.0, + 1359.0, + 340.0, + 1359.0, + 373.0, + 351.0, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 390.0, + 1407.0, + 390.0, + 1407.0, + 424.0, + 296.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 418.0, + 794.0, + 418.0, + 794.0, + 451.0, + 351.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 468.0, + 1405.0, + 468.0, + 1405.0, + 502.0, + 294.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 496.0, + 1405.0, + 496.0, + 1405.0, + 530.0, + 351.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 523.0, + 1350.0, + 523.0, + 1350.0, + 557.0, + 351.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 572.0, + 1407.0, + 572.0, + 1407.0, + 612.0, + 292.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 603.0, + 613.0, + 603.0, + 613.0, + 637.0, + 349.0, + 637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 654.0, + 1407.0, + 654.0, + 1407.0, + 688.0, + 296.0, + 688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 681.0, + 1405.0, + 681.0, + 1405.0, + 715.0, + 351.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 709.0, + 739.0, + 709.0, + 739.0, + 742.0, + 346.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 755.0, + 1407.0, + 755.0, + 1407.0, + 797.0, + 294.0, + 797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 785.0, + 1409.0, + 785.0, + 1409.0, + 825.0, + 347.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 814.0, + 420.0, + 814.0, + 420.0, + 850.0, + 349.0, + 850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 865.0, + 1405.0, + 865.0, + 1405.0, + 899.0, + 296.0, + 899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 892.0, + 1242.0, + 892.0, + 1242.0, + 926.0, + 349.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 943.0, + 1407.0, + 943.0, + 1407.0, + 977.0, + 296.0, + 977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 972.0, + 1407.0, + 972.0, + 1407.0, + 1006.0, + 349.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 998.0, + 608.0, + 998.0, + 608.0, + 1036.0, + 344.0, + 1036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1048.0, + 1405.0, + 1048.0, + 1405.0, + 1082.0, + 294.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1076.0, + 1407.0, + 1076.0, + 1407.0, + 1116.0, + 347.0, + 1116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1103.0, + 860.0, + 1103.0, + 860.0, + 1141.0, + 344.0, + 1141.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1156.0, + 1405.0, + 1156.0, + 1405.0, + 1190.0, + 296.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1179.0, + 1403.0, + 1179.0, + 1403.0, + 1219.0, + 349.0, + 1219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1213.0, + 592.0, + 1213.0, + 592.0, + 1247.0, + 351.0, + 1247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1259.0, + 1409.0, + 1259.0, + 1409.0, + 1299.0, + 292.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1291.0, + 1407.0, + 1291.0, + 1407.0, + 1325.0, + 349.0, + 1325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1316.0, + 1407.0, + 1316.0, + 1407.0, + 1356.0, + 347.0, + 1356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 1344.0, + 921.0, + 1344.0, + 921.0, + 1382.0, + 346.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1396.0, + 1407.0, + 1396.0, + 1407.0, + 1430.0, + 294.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1424.0, + 1405.0, + 1424.0, + 1405.0, + 1458.0, + 351.0, + 1458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 1449.0, + 680.0, + 1449.0, + 680.0, + 1487.0, + 346.0, + 1487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1498.0, + 1405.0, + 1498.0, + 1405.0, + 1540.0, + 294.0, + 1540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1527.0, + 1179.0, + 1527.0, + 1179.0, + 1565.0, + 344.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1576.0, + 1407.0, + 1576.0, + 1407.0, + 1618.0, + 290.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1607.0, + 1405.0, + 1607.0, + 1405.0, + 1641.0, + 349.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 1635.0, + 807.0, + 1635.0, + 807.0, + 1671.0, + 342.0, + 1671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1683.0, + 1409.0, + 1683.0, + 1409.0, + 1723.0, + 292.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1708.0, + 421.0, + 1708.0, + 421.0, + 1752.0, + 349.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1763.0, + 1407.0, + 1763.0, + 1407.0, + 1804.0, + 292.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1791.0, + 1299.0, + 1791.0, + 1299.0, + 1831.0, + 349.0, + 1831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1844.0, + 1405.0, + 1844.0, + 1405.0, + 1877.0, + 296.0, + 1877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1871.0, + 1090.0, + 1871.0, + 1090.0, + 1907.0, + 344.0, + 1907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1922.0, + 1405.0, + 1922.0, + 1405.0, + 1955.0, + 296.0, + 1955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1945.0, + 1407.0, + 1945.0, + 1407.0, + 1987.0, + 344.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1975.0, + 693.0, + 1975.0, + 693.0, + 2012.0, + 344.0, + 2012.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2086, + 836, + 2086 + ], + "score": 0.809 + }, + { + "category_id": 1, + "poly": [ + 294, + 81, + 1411, + 81, + 1411, + 2022, + 294, + 2022 + ], + "score": 0.788 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2086, + 836, + 2086 + ], + "score": 0.099 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2096.0, + 832.0, + 2096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2096.0, + 832.0, + 2096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 203.0, + 1404.0, + 203.0, + 1404.0, + 237.0, + 295.0, + 237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 229.0, + 1408.0, + 229.0, + 1408.0, + 271.0, + 343.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 261.0, + 612.0, + 261.0, + 612.0, + 295.0, + 350.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 305.0, + 1406.0, + 305.0, + 1406.0, + 348.0, + 291.0, + 348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 337.0, + 1017.0, + 337.0, + 1017.0, + 371.0, + 350.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 386.0, + 1406.0, + 386.0, + 1406.0, + 420.0, + 295.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 407.0, + 1410.0, + 407.0, + 1410.0, + 450.0, + 345.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 434.0, + 420.0, + 434.0, + 420.0, + 478.0, + 348.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 486.0, + 1406.0, + 486.0, + 1406.0, + 524.0, + 290.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 514.0, + 1011.0, + 514.0, + 1011.0, + 552.0, + 345.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 560.0, + 1410.0, + 560.0, + 1410.0, + 603.0, + 291.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 592.0, + 1404.0, + 592.0, + 1404.0, + 626.0, + 350.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 620.0, + 612.0, + 620.0, + 612.0, + 654.0, + 350.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 665.0, + 1406.0, + 665.0, + 1406.0, + 707.0, + 293.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 694.0, + 1408.0, + 694.0, + 1408.0, + 730.0, + 348.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 724.0, + 762.0, + 724.0, + 762.0, + 758.0, + 350.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 773.0, + 1404.0, + 773.0, + 1404.0, + 807.0, + 295.0, + 807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 799.0, + 1127.0, + 799.0, + 1127.0, + 839.0, + 345.0, + 839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 850.0, + 1406.0, + 850.0, + 1406.0, + 884.0, + 295.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 875.0, + 612.0, + 875.0, + 612.0, + 909.0, + 348.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 920.0, + 1408.0, + 920.0, + 1408.0, + 964.0, + 291.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 949.0, + 1408.0, + 949.0, + 1408.0, + 990.0, + 348.0, + 990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 979.0, + 1070.0, + 979.0, + 1070.0, + 1015.0, + 343.0, + 1015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1026.0, + 1408.0, + 1026.0, + 1408.0, + 1066.0, + 293.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1058.0, + 1214.0, + 1058.0, + 1214.0, + 1092.0, + 350.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1103.0, + 1410.0, + 1103.0, + 1410.0, + 1141.0, + 288.0, + 1141.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1130.0, + 1400.0, + 1130.0, + 1400.0, + 1171.0, + 348.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1181.0, + 1406.0, + 1181.0, + 1406.0, + 1215.0, + 295.0, + 1215.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1207.0, + 1408.0, + 1207.0, + 1408.0, + 1247.0, + 348.0, + 1247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1234.0, + 421.0, + 1234.0, + 421.0, + 1270.0, + 348.0, + 1270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1281.0, + 1408.0, + 1281.0, + 1408.0, + 1324.0, + 291.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1313.0, + 901.0, + 1313.0, + 901.0, + 1347.0, + 348.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1362.0, + 1406.0, + 1362.0, + 1406.0, + 1396.0, + 295.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1390.0, + 1241.0, + 1390.0, + 1241.0, + 1424.0, + 348.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1434.0, + 1408.0, + 1434.0, + 1408.0, + 1472.0, + 290.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1464.0, + 1408.0, + 1464.0, + 1408.0, + 1504.0, + 348.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1496.0, + 610.0, + 1496.0, + 610.0, + 1523.0, + 352.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1540.0, + 1406.0, + 1540.0, + 1406.0, + 1575.0, + 295.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1568.0, + 702.0, + 1568.0, + 702.0, + 1606.0, + 343.0, + 1606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1615.0, + 1408.0, + 1615.0, + 1408.0, + 1655.0, + 293.0, + 1655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1643.0, + 928.0, + 1643.0, + 928.0, + 1681.0, + 345.0, + 1681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1694.0, + 1404.0, + 1694.0, + 1404.0, + 1728.0, + 295.0, + 1728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1721.0, + 610.0, + 1721.0, + 610.0, + 1755.0, + 350.0, + 1755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1770.0, + 1406.0, + 1770.0, + 1406.0, + 1804.0, + 295.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1798.0, + 1408.0, + 1798.0, + 1408.0, + 1832.0, + 348.0, + 1832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1824.0, + 582.0, + 1824.0, + 582.0, + 1859.0, + 345.0, + 1859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1874.0, + 1408.0, + 1874.0, + 1408.0, + 1908.0, + 295.0, + 1908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1902.0, + 983.0, + 1902.0, + 983.0, + 1936.0, + 350.0, + 1936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1942.0, + 1412.0, + 1942.0, + 1412.0, + 1991.0, + 288.0, + 1991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1976.0, + 1267.0, + 1976.0, + 1267.0, + 2015.0, + 345.0, + 2015.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2085, + 836, + 2085 + ], + "score": 0.839 + }, + { + "category_id": 1, + "poly": [ + 292, + 199, + 1410, + 199, + 1410, + 1677, + 292, + 1677 + ], + "score": 0.756 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2057.0, + 871.0, + 2057.0, + 871.0, + 2095.0, + 832.0, + 2095.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 202.0, + 1410.0, + 202.0, + 1410.0, + 244.0, + 292.0, + 244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 233.0, + 762.0, + 233.0, + 762.0, + 269.0, + 346.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 281.0, + 1405.0, + 281.0, + 1405.0, + 317.0, + 295.0, + 317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 308.0, + 1409.0, + 308.0, + 1409.0, + 348.0, + 348.0, + 348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 338.0, + 762.0, + 338.0, + 762.0, + 374.0, + 348.0, + 374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 387.0, + 1407.0, + 387.0, + 1407.0, + 424.0, + 295.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 415.0, + 1405.0, + 415.0, + 1405.0, + 451.0, + 350.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 445.0, + 698.0, + 445.0, + 698.0, + 476.0, + 346.0, + 476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 493.0, + 1407.0, + 493.0, + 1407.0, + 529.0, + 293.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 519.0, + 1409.0, + 519.0, + 1409.0, + 560.0, + 350.0, + 560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 548.0, + 935.0, + 548.0, + 935.0, + 585.0, + 350.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 598.0, + 1407.0, + 598.0, + 1407.0, + 634.0, + 293.0, + 634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 624.0, + 1271.0, + 624.0, + 1271.0, + 663.0, + 346.0, + 663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 672.0, + 1409.0, + 672.0, + 1409.0, + 714.0, + 290.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 703.0, + 612.0, + 703.0, + 612.0, + 737.0, + 348.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 754.0, + 1405.0, + 754.0, + 1405.0, + 790.0, + 293.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 779.0, + 1400.0, + 779.0, + 1400.0, + 820.0, + 346.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 831.0, + 1405.0, + 831.0, + 1405.0, + 867.0, + 295.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 856.0, + 957.0, + 856.0, + 957.0, + 897.0, + 348.0, + 897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 907.0, + 1377.0, + 907.0, + 1377.0, + 948.0, + 293.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 958.0, + 1405.0, + 958.0, + 1405.0, + 994.0, + 295.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 982.0, + 1409.0, + 982.0, + 1409.0, + 1025.0, + 346.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 1010.0, + 1056.0, + 1010.0, + 1056.0, + 1053.0, + 343.0, + 1053.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1065.0, + 1407.0, + 1065.0, + 1407.0, + 1101.0, + 295.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1091.0, + 1119.0, + 1091.0, + 1119.0, + 1127.0, + 351.0, + 1127.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1139.0, + 1405.0, + 1139.0, + 1405.0, + 1175.0, + 295.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1170.0, + 1407.0, + 1170.0, + 1407.0, + 1206.0, + 350.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1196.0, + 611.0, + 1196.0, + 611.0, + 1232.0, + 350.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1245.0, + 1405.0, + 1245.0, + 1405.0, + 1282.0, + 295.0, + 1282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 1272.0, + 1407.0, + 1272.0, + 1407.0, + 1311.0, + 346.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1301.0, + 611.0, + 1301.0, + 611.0, + 1334.0, + 348.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1351.0, + 1407.0, + 1351.0, + 1407.0, + 1387.0, + 295.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1379.0, + 1407.0, + 1379.0, + 1407.0, + 1415.0, + 350.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1406.0, + 475.0, + 1406.0, + 475.0, + 1439.0, + 350.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1454.0, + 1407.0, + 1454.0, + 1407.0, + 1495.0, + 293.0, + 1495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1484.0, + 866.0, + 1484.0, + 866.0, + 1520.0, + 350.0, + 1520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1535.0, + 1407.0, + 1535.0, + 1407.0, + 1571.0, + 295.0, + 1571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1563.0, + 978.0, + 1563.0, + 978.0, + 1599.0, + 350.0, + 1599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1614.0, + 1407.0, + 1614.0, + 1407.0, + 1645.0, + 297.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1638.0, + 1194.0, + 1638.0, + 1194.0, + 1674.0, + 350.0, + 1674.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 13, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/train/3ZeGLibhFo0/3ZeGLibhFo0.md b/parse/train/3ZeGLibhFo0/3ZeGLibhFo0.md new file mode 100644 index 0000000000000000000000000000000000000000..e22430a93ba0c119db36f7ed634cf1e9e36e3d38 --- /dev/null +++ b/parse/train/3ZeGLibhFo0/3ZeGLibhFo0.md @@ -0,0 +1,285 @@ +# ENABLING COUNTERFACTUAL SURVIVAL ANALYSIS WITH BALANCED REPRESENTATIONS + +Anonymous authors Paper under double-blind review + +# ABSTRACT + +Balanced representation learning methods have been applied successfully to counterfactual inference from observational data. However, approaches that account for survival outcomes are relatively limited. Survival data are frequently encountered across diverse medical applications, i.e., drug development, risk profiling, and clinical trials, and such data are also relevant in fields like manufacturing (for equipment monitoring). When the outcome of interest is time-to-event, special precautions for handling censored events need to be taken, as ignoring censored outcomes may lead to biased estimates. We propose a theoretically grounded unified framework for counterfactual inference applicable to survival outcomes. Further, we formulate a nonparametric hazard ratio metric for evaluating average and individualized treatment effects. Experimental results on real-world and semi-synthetic datasets, the latter which we introduce, demonstrate that the proposed approach significantly outperforms competitive alternatives in both survival-outcome predictions and treatment-effect estimation. + +# 1 INTRODUCTION + +Survival analysis or time-to-event studies focus on modeling the time of a future event, such as death or failure, and investigate its relationship with covariates or predictors of interest. Specifically, we may be interested in the causal effect of a given intervention or treatment on survival time. A typical question may be: will a given therapy increase the chances of survival of an individual or population? Such causal inquiries on survival outcomes are common in the fields of epidemiology and medicine (Robins, 1986; Hammer et al., 1996; Yusuf et al., 2016). As an important current example, the COVID-19 pandemic is creating a demand for methodological development to address such questions, specifically, when evaluating the effectiveness of a potential vaccine or therapeutic outside randomized controlled trial settings. + +Traditional causal survival analysis is typically carried out in the context of a randomized controlled trial (RCT), where the treatment assignment is controlled by researchers. Though they are the gold standard for causal inference, RCTs are usually long-term engagements, expensive and limited in sample size. Alternatively, the availability of observational data with comprehensive information about patients, such as electronic health records (EHRs), constitutes a more accessible but also more challenging source for estimating causal effects (Häyrinen et al., 2008; Jha et al., 2009). Such observational data may be used to augment and verify an RCT, after a particular treatment is approved and in use (Gombar et al., 2019; Frankovich et al., 2011; Longhurst et al., 2014). Moreover, the wealth of information from observational data also allows for the estimation of the individualized treatment effect (ITE), namely, the causal effect of an intervention at the individual level. In this work, we develop a novel framework for counterfactual time-to-event prediction to estimate the ITE for survival or time-to-event outcomes from observational data. + +Estimating the causal effect for survival outcomes in observational data manifests two principal challenges. First, the treatment assignment mechanism is not known a priori. Therefore, there may be variables, known as confounders, affecting both the treatment and survival time, which lead to selection bias (Bareinboim & Pearl, 2012), i.e., that the distributions across treatment groups are not the same. In this work, we focus on selection biases due to confounding, but other sources may also be considered. For instance, patients who are severely ill are likely to receive more aggressive therapy, however, their health status may also inevitably influence survival. Traditional survival analysis neglects such bias, leading to incorrect causal estimation. Second, the exact time-to-event is not always observed, i.e., sometimes we only know that an event has not occurred up to a certain point in time. This is known as the censoring problem. Moreover, censoring might be informative depending on the characteristics of the individuals and their treatment assignments, thus proper adjustment is required for accurate causal estimation (Cole & Hernán, 2004; Díaz, 2019). + +Traditional causal survival-analysis approaches typically model the effect of the treatment or covariates (not time or survival) in a parametric manner. Two commonly used models are the Cox proportional hazards (CoxPH) model (Cox, 1972) and the accelerated failure time (AFT) model (Wei, 1992), which presume a linear relationship between the covariates and survival probability. Further, proper weighting for each individual has been employed to account for confounding bias from these models (Austin, 2007; 2014; Hernán et al., 2005). For instance, probability weighting schemes that account for both selection bias and covariate dependent censoring have been considered for adjusted survival curves (Cole & Hernán, 2004; Díaz, 2019). Moreover, such probability weighting schemes have been applied to causal survival-analysis under time-varying treatment and confounding (Robins, 1986; Hernán et al., 2000). See van der Laan & Robins (2003); Tsiatis (2007); Van der Laan & Rose (2011); Hernán & Robins (2020) for an overview. Such linear specification makes these models interpretable but compromises their flexibility, and makes it difficult to adapt them for high-dimensional data or to capture complex interactions among covariates. Importantly, these methods lack a counterfactual prediction mechanism, which is key for ITE estimation (see Section 2). + +Fortunately, recent advances in machine learning, such as representation learning or generative modeling, have enabled causal inference methods to handle high-dimensional data and to characterize complex interactions effectively. For instance, there has been recent interest in tree-based (Chipman et al., 2010; Wager & Athey, 2018) and neural-network-based (Shalit et al., 2017; Zhang et al., 2020) approaches. For pre-specified time-horizons, the nonparametric Random Survival Forest (RSF) (Ishwaran et al., 2008) and Bayesian Additive regression trees (BART) (Chipman et al., 2010) have been extended to causal survival analysis. RSF has been applied to causal survival forests with weighted bootstrap inference (Shen et al., 2018; Cui et al., 2020) while a BART is extended to account for survival outcomes in Surv-BART (Sparapani et al., 2016), and AFT-BART (Henderson et al., 2020). See Hu et al. (2020) for an extensive investigation of the causal survival tree-based methods. Alternatively, when estimating the ITE, neural-network-based methods propose to regularize the transformed covariates or representations for an individual to have balanced distributions across treatment groups, thus accounting for the confounding bias and improving ITE prediction. However, most approaches employing representation learning techniques for counterfactual inference deal with continuous or binary outcomes, instead of time-to-event outcomes with censoring (informative or non-informative). Hence, a principled generalization to the context of counterfactual survival analysis is needed. + +In this work we leverage balanced (latent) representation learning to estimate ITE via counterfactual prediction of survival outcomes in observational studies. We develop a framework to predict event times from a low-dimensional transformation of the original covariate space. To address the specific challenges associated with counterfactual survival analysis, we make the following contributions: + +• We develop an optimization objective incorporating adjustments for informative censoring, as well as a balanced regularization term bounding the generalization error for ITE prediction. For the latter, we repurpose a recently proposed bound (Shalit et al., 2017) for our time-to-event scenario. +• We propose a generative model for event times to relax restrictive survival linear and parametric assumptions, thus allowing for more flexible modeling. Our approach can also provide nonparametric uncertainty quantification for ITE predictions. +• We provide survival-specific evaluation metrics, including a new nonparametric hazard ratio estimator, and discuss how to perform model selection for survival outcomes. The proposed model demonstrates superior performance relative to the commonly used baselines in real-world and semi-synthetic datasets. +• We introduce a survival-specific semi-synthetic dataset and demonstrate an approach for leveraging prior randomized experiments in longitudinal studies for model validation. + +![](images/482e9df6a7b5d031a444403a4be41d381d10c980808568d8cbf42f012f3882e6.jpg) +Figure 1: (a) Illustration of the proposed counterfactual survival analysis (CSA). Covariates $X =$ $x$ are mapped into latent representation $r$ via deterministic mapping $r \ = \ \Phi ( x )$ . The potential outcomes are sampled from $\bar { t } _ { a } \sim p ( T _ { A } | X = x )$ for $A = a$ via stochastic mapping $h _ { A } ( r , \tilde { \epsilon } )$ , where randomness is induced with a flow-based transformation, $\tilde { \epsilon }$ , of a simple distribution $p ( \epsilon )$ , i.e., uniform or Gaussian. (b) and (c) show the proposed causal graphs for non-informative and informative censoring, respectively. + +# 2 PROBLEM FORMULATION + +We first introduce the basic setup for performing causal survival analysis in observational studies. Suppose we have $N$ units, with $N _ { 1 }$ units being treated and $N _ { 0 }$ in the control group $( N = N _ { 1 } + N _ { 0 } )$ . For each unit (individual), we have covariates $X$ , which can be heterogeneous, e.g., a mixture of categorical and continuous covariates which, in the context of medicine, may include labs, vitals, procedure codes, etc. We also have a treatment indicator $A$ , where $A = 0$ for the controls and $A = 1$ for the treated, as well as the outcome (event) of interest $T$ . Under the potential-outcomes framework (Rubin, 2005), let $T _ { 0 }$ and $T _ { 1 }$ be the potential event times for a given subject under control and treatment, respectively. In practice we only observe one realization of the potential outcomes, i.e., the factual outcome $T = T _ { A }$ , while the counterfactual outcome $T _ { 1 - A }$ is unobserved. + +In survival analysis, the problem becomes more difficult because we do not always observe the exact event time for each individual, but rather the time up to which we are certain that the event has not occurred; specifically, we have a (right) censoring problem, most likely due to the loss of follow-up. We denote the censoring time as $C$ and censoring indicator as $\delta \in \{ 0 , 1 \}$ . The actual observed time is $Y = \operatorname* { m i n } ( T _ { A } , C )$ , i.e., the outcome is observed (non-censored) if $T _ { A } < C$ and $\delta = 1$ . + +In this work, we are interested in the expected difference between the $T _ { 1 }$ and $T _ { 0 }$ conditioned on $X$ for a given unit (individual), which is commonly known as the individualized treatment effect (ITE). Specifically, we wish to perform inference on the conditional distributions of $T _ { 1 }$ and $T _ { 0 }$ , i.e., $p ( T _ { 1 } | X )$ and $p ( T _ { 0 } | X )$ , respectively, as shown in Figure 1a. In practice, we observe $N$ realizations of $( Y , \delta , X , A )$ for observed time, censoring indicator, covariates and treatment indicator, respectively; hence, from an observational study the dataset takes the form $\mathcal { D } = \{ ( y _ { i } , \delta _ { i } , x _ { i } , a _ { i } ) \} _ { i = 1 } ^ { N }$ . Below, we discuss several common choices of estimands in survival analysis. + +Estimands of Interest We begin by considering survival analysis in the absence of an intervening treatment choice, $A$ . Let $F ( t | x ) \triangleq P ( T \leq t | X = x )$ be the cumulative distribution function of the event (failure) time, $t$ , given a realization of the covariates, $x$ . Survival analysis is primarily concerned with characterization of the survival function conditioned on covariates $S ( t | x ) \triangleq 1 - F ( t | x )$ , and the hazard function or risk score, $\lambda ( t | x )$ , defined below. $S ( t | x )$ is a monotonically decreasing function indicating the probability of survival up to time $t$ . The hazard function measures the instantaneous probability of the event occurring between $\{ t , t + \Delta t \}$ given $T > t$ and $\Delta t 0$ . From standard definitions (Kleinbaum & Klein, 2010), the relationship between cumulative and hazard function is formulated as + +$$ +\lambda ( t | x ) = \operatorname* { l i m } _ { d t 0 } { \frac { P ( t < T < t + d t | X = x ) } { P ( T > t | X = x ) d t } } = - { \frac { d \log S ( t | x ) } { d t } } = { \frac { f ( t | x ) } { S ( t | x ) } } . +$$ + +From (1) we see that $f ( t | x ) \triangleq P ( T = t | X = x ) = \lambda ( t | x ) S ( t | x )$ , is the conditional event time density function (Kleinbaum $\&$ Klein, 2010). + +Given the binary treatment $A$ , we are interested in its impact on the survival time. For ITE estimation, we are also interested in the difference between the two potential outcomes $T _ { 1 } , T _ { 0 }$ . Let $S _ { A } ( t | x )$ and $\lambda _ { A } ( t | x )$ denote the survival and hazard functions for the potential outcomes $T _ { A }$ , i.e., $T _ { 1 }$ and $T _ { 0 }$ . Several common estimands of interest include (Zhao et al., 2012; Trinquart et al., 2016): difference in expected lifetime: $\begin{array} { r } { \mathrm { { I T E } } ( t , x ) = \int _ { 0 } ^ { t _ { \operatorname* { m a x } } } \{ S _ { 1 } ( t | x ) - S _ { 0 } ( t | x ) \} \mathrm { d } t = \mathbb { E } \{ T _ { 1 } - T _ { 0 } | X = x \} . } \end{array}$ , difference in survival function: $\mathrm { I T E } ( t , x ) = S _ { 1 } ( t | x ) - S _ { 0 } ( t | x )$ , and hazard ratio: $\mathrm { I T E } ( t , x ) = \lambda _ { 1 } ( t | x ) / \lambda _ { 0 } ( t | x )$ . The inference difficulties associated with the above estimands from observational data are two-fold. First, there are confounders affecting both the treatment assignment and outcomes, which stem from selection bias, i.e., the treatment and control covariate distributions are not necessarily the same. Also, we do not have direct knowledge of the conditional treatment assignment mechanism, i.e., $P ( A = a | X = x )$ , also known as the propensity score. Let ⊥⊥ denote statistical independence. For estimands to be identifiable from observational data, we make two assumptions: (i) $\{ T _ { 1 } , T _ { 0 } \} \perp \perp A | X$ , i.e., no unobserved confounders or ignorability, and $( i i )$ overlap in the covariate support $0 < P ( A =$ $1 | X = x ) < 1$ almost surely if $p ( X = x ) > 0$ . Second, the censoring mechanism is also unknown and may lead to bias without proper adjustment. We consider two censoring mechanisms in our work, $( i )$ conditionally independent or informative censoring: $T \perp \perp C | X , A$ , and $( i i )$ random or non-informative censoring: $T \perp \perp C$ . Note that for informative censoring, we also have to consider potential censoring times $C _ { 1 }$ and $C _ { 0 }$ and their conditionals $p ( C _ { 1 } | X )$ and $p ( C _ { 0 } | X )$ , respectively. Figure 1 shows causal graphs illustrating these modeling assumptions. + +# 3 MODELING + +To overcome the above challenges and adjust for observational biases, we propose a unified framework for counterfactual survival analysis (CSA). Specifically, we repurpose the counterfactual bound in Shalit et al. (2017) for our time-to-event scenario and introduce a nonparametric approach for stochastic survival outcome predictions. Below we formulate a theoretically grounded and unified approach for estimating $( i )$ the encoder function $r = \Phi ( x )$ , which deterministically maps covariates $x$ to their corresponding latent representation $r \in { \mathbb { R } } ^ { d }$ , and $( i i )$ two stochastic time-to-event generative functions, $h _ { A } ( \cdot )$ , to implicitly draw samples from both potential outcome conditionals $\bar { t } _ { a } \sim p _ { h , \Phi } ( T _ { A } | X = x )$ , for $A = \{ 1 , 0 \}$ , and where $t _ { a }$ indicates the sample from $p _ { h , \Phi } ( T _ { A } | X = x )$ is for $A = a$ . Further, we formulate a general extension that accounts for informative censoring by introducing two stochastic censoring generative functions, $\nu _ { A } ( \cdot )$ , to draw samples for potential censoring times $c _ { a } \sim p _ { \nu , \Phi } ( C _ { A } | X = x )$ . The model-specifying functions, $\{ h _ { A } ( \cdot ) , \nu _ { A } ( \cdot ) , \Phi ( \cdot ) \}$ , are parameterized via neural networks. See the Supplementary Material (SM) for details. Figure 1a summarizes our modeling approach. + +Accounting for selection bias We wish to estimate the potential outcomes, i.e., event times, which are sampled by distributions parameterized by functions $\{ h _ { A } ( \cdot ) , \Phi ( \cdot ) \}$ , i.e., + +$$ +\begin{array} { c } { { t \sim p _ { h , \Phi } ( T | X = x , A = a ) } } \\ { { { \it t } _ { a } \sim p _ { h , \Phi } ( T _ { a } | X = x ) } } \end{array} +$$ + +We obtain (3) from (2) via the strong ignorability assumption, i.e., $\{ T _ { 0 } , T _ { 1 } \}$ ⊥⊥ $A | X$ (consistent with the causal graphs in Figure 1b and 1c) and $0 < P ( A = \mathsf { \bar { a } } | X \doteq x ) < 1 .$ , and the consistency assumption, i.e., $T = T _ { A } | A = a$ . A similar argument can be made for informative censoring based on Figure 1c, so we can also write $c _ { a } \sim p _ { \nu , \Phi } ( C _ { A } | X = x )$ . Given (3), model functions $\{ h _ { A } ( \cdot ) , \Phi ( \cdot ) \bar \}$ and $\nu _ { A } ( \cdot )$ for informative censoring can be learned by leveraging standard statistical optimization approaches, that minimize a loss hypothesis $\mathcal { L }$ given samples from the empirical distribution $( y , \delta , x , a ) \sim p ( Y , \delta , X , A )$ , i.e., from dataset $\mathcal { D }$ . Specifically, we write $\mathcal { L } = \mathbb { E } _ { ( y , \delta , x , a ) \sim p ( Y , \delta , X , A ) } \left[ \ell _ { h , \Phi } ( t _ { a } , y , \delta ) \right]$ , where $\ell _ { h , \Phi } ( t _ { a } , y , \delta )$ is a loss function that measures the agreement of $t _ { a } \sim p _ { h , \Phi } ( T _ { A } | X = x )$ (and $c _ { a } \sim p _ { \nu , \Phi } ( C _ { A } | X = x )$ for informative censoring) with ground truth $\{ y , \delta \}$ , the observed time and censoring indicator, respectively. + +For some parametric formulations of event time distribution $p _ { h , \Phi } ( T _ { A } | X = x )$ , e.g., exponential, Weibull, log-Normal, etc., and provided the censoring mechanism is non-informative, $- \ell _ { h , \Phi } ( t _ { a } , y , \delta )$ is the closed form log likelihood. Specifically, $- \ell _ { h , \Phi } ( t _ { a } , y , \delta ) \triangleq \log p _ { h , \Phi } ( T _ { a } | X = x ) = \delta$ · $\log f _ { h , \Phi } ( t _ { a } | x ) + ( 1 - \bar { \delta } ) \cdot \log S _ { h , \Phi } ( t _ { a } | x )$ , which implies that the conditional event time density and survival functions can be calculated in closed form from transformations $\{ h _ { A } ( \cdot ) , \Phi ( \cdot ) \}$ of $x$ . See the SM for parametric examples of $\mathcal { L }$ accounting for informative censoring. + +We further define the expected loss for a given realization of covariates $x$ and treatment assignment $a$ over observed times $y$ (censored and non-censored), and the censoring indicator $\delta$ as $\zeta _ { h , \Phi } ( x , a ) \triangleq$ $\mathbb { E } _ { ( y , \delta , x ) \sim p ( Y , \delta | X ) } \ell _ { h , \Phi } ( t _ { a } , y , \delta )$ as in Shalit et al. (2017). For a given subject with covariates $x$ and treatment assignment $a$ , we wish to minimize both the factual and counterfactual losses, $\mathcal { L } _ { \mathrm { F } }$ and $\mathcal { L } _ { \mathrm { C F } }$ respectively, by decomposing $\mathcal { L } = \mathcal { L } _ { \mathrm { F } } + \mathcal { L } _ { \mathrm { C F } }$ as follows + +$$ +\begin{array} { r } { \mathcal { L } _ { \mathrm { F } } = \mathbb { E } _ { ( x , a ) \sim p ( A , X ) } \zeta _ { h , \Phi } ( x , a ) , \quad \mathcal { L } _ { \mathrm { C F } } = \mathbb { E } _ { ( x , a ) \sim p ( 1 - A , X ) } \zeta _ { h , \Phi } ( x , a ) . } \end{array} +$$ + +Let $u \triangleq P ( A = 1 )$ denote the marginal probability of treatment assignment. We can readily decompose the losses in (4) according to treatment assignments. The decomposed factual $\mathcal { L } _ { \mathrm { F } } =$ $u \cdot \mathcal { L } _ { \mathrm { F } } ^ { A = 1 } + \left( 1 - u \right) \cdot \mathcal { L } _ { \mathrm { F } } ^ { A = 0 }$ , and similarly, the decomposed counterfactual $\mathcal { L } _ { \mathrm { C F } } = \dot { ( } 1 - u ) \cdot \mathcal { L } _ { \mathrm { C F } } ^ { A = 1 } + u$ $\mathcal { L } _ { \mathrm { C F } } ^ { A = 0 }$ F CF . In practice, only factual outcomes are observed, hence, for a non-randomized non-controlled experiment, we cannot obtain an unbiased estimate of $\mathcal { L } _ { \mathrm { C F } }$ from data due to selection bias (or confounding). Therefore, we bound $\mathcal { L } _ { \mathrm { C F } }$ and $\mathcal { L }$ below following Shalit et al. (2017). + +Corollary 1 Assume $\Phi ( \cdot )$ is an invertible map, and $\alpha ^ { - 1 } \zeta _ { h , \Phi } ( x , a ) \in G$ , where $G$ is a family of functions, $p _ { \Phi } ^ { A = a } \triangleq p _ { \Phi } ( R | A = a )$ is the latent distribution for group $A = a$ , and $\alpha > 0$ is a constant. Then, we have: + +$$ +\begin{array} { r l } & { \mathcal { L } _ { \mathrm { C F } } \leq ( 1 - u ) \cdot \mathcal { L } _ { \mathrm { F } } ^ { A = 1 } + u \cdot \mathcal { L } _ { \mathrm { F } } ^ { A = 0 } + \alpha \cdot \mathrm { I P M } _ { G } ( p _ { \Phi } ^ { A = 1 } , p _ { \Phi } ^ { A = 0 } ) } \\ & { \quad \mathcal { L } \leq \mathcal { L } _ { \mathrm { F } } ^ { A = 1 } + \mathcal { L } _ { \mathrm { F } } ^ { A = 0 } + \alpha \cdot \mathrm { I P M } _ { G } ( p _ { \Phi } ^ { A = 1 } , p _ { \Phi } ^ { A = 0 } ) . } \end{array} +$$ + +The integral probability metric (IPM) (Müller, 1997; Sriperumbudur et al., 2012) measures the distance between two probability distributions $p$ and $q$ defined over $M$ , i.e., the latent space of $R$ . Formally, $\begin{array} { r } { \mathrm { I P M } _ { G } ( p , q ) \triangleq \operatorname* { s u p } _ { g \in G } \vert \int _ { M } g ( m ) \left( p ( m ) - q ( m ) \right) d m \vert } \end{array}$ , where $g : m \mathbb { R }$ , represents a class of real-valued bounded measurable functions on $M$ (Shalit et al., 2017). Therefore, model functions $\{ h _ { a } ( \cdot ) , \Phi ( \cdot ) \}$ can be learned by minimizing the upper bound in (5) consisting of $( i )$ only factual losses under both treatment assignments and $( i i )$ an IPM regularizer enforcing latent distributional equivalence between the treatment groups. Note that if the data originates from a RCT it follows (by construction) that $\mathrm { I P M } _ { G } ( p _ { \Phi } ^ { A = 1 } , p _ { \Phi } ^ { \tilde { A = 0 } } ) \stackrel { . } { = } 0$ . + +Accounting for censoring bias Below we formulate an approach for estimating functions $h _ { A } ( \cdot )$ and $\nu _ { A } ( \cdot )$ for synthesizing (sampling) non-censored $t _ { a } \sim \stackrel { \sim } { p } _ { h , \Phi } ( T _ { A } | X = x )$ and censored $c _ { a } \sim$ $p _ { \nu , \Phi } ( C _ { A } | X = x )$ times, respectively. While some parametric assumptions for $p _ { h , \Phi } ( T _ { A } | X = x )$ yield easy-to-evaluate closed forms for $S _ { h , \Phi } ( t _ { a } | x )$ that can be used as likelihood for censored observations, they are restrictive, and have been shown to generate unrealistic high variance samples (Chapfuwa et al., 2018). So motivated, we seek a nonparametric likelihood-based approach that can model a flexible family of distributions, with an easy-to-sample approach for event times $t _ { a } \sim$ $p _ { h , \Phi } ( T _ { a } | X = x )$ . We model the event time generation process with a source of randomness, $p ( \epsilon )$ , e.g. Gaussian or uniform, which is obtained from a neural-network-based nonlinear transformation. In the experiments we use a planar flow formulation parameterized by $\{ U _ { h } , W _ { h } , b _ { h } \}$ (Rezende & Mohamed, 2015), however, other specifications can also be used. Note that Miscouridou et al. (2018) has previously leveraged normalizing flows for survival analysis, however, our approach is very different in that it focuses on formulating $\romannumeral 1$ ) a counterfactual survival analysis framework that accounts for informative or non-informative censoring mechanisms and confounding, and $_ { i i }$ ) model event times as a continuous variable instead of discretizing them. Specifically, we transform the source of randomness, $\epsilon$ , using a single layer specification as follows + +$$ +{ \widetilde { \epsilon } } _ { h } = \epsilon + U _ { h } \operatorname { t a n h } ( W _ { h } \epsilon + b _ { h } ) , \quad \epsilon \sim \operatorname { U n i f o r m } ( 0 , 1 ) , t _ { a } = h _ { A } ( r , { \widetilde { \epsilon } } _ { h } ) , \quad r = \Phi ( x ) +$$ + +where $\{ U _ { h } , W _ { h } \} \in \mathbb { R } ^ { d \times d }$ , $\{ b _ { h } , \epsilon \} \in \mathbb { R } ^ { d }$ , $d$ is the dimensionality of the normalizing flow; each component of $\epsilon$ is drawn independently from Uniform $( 0 , 1 )$ , and $\tilde { \epsilon } _ { h }$ may be viewed as a skip connection with stochasticity in $\epsilon$ . Further, $h _ { A } ( r , \tilde { \epsilon } _ { h } )$ and $\Phi ( x )$ are time-to-event generative and encoding functions, respectively, parameterized as neural networks. For simplicity, the dimensions of $r$ and $\epsilon$ are set to $d$ , however, they can be set independently if desired. In practice, we are interested in generating realistic event-time samples; therefore, we account for both censored and non-censored observations by adopting the objective from Chapfuwa et al. (2018), formulated as + +$$ +\mathcal { L } _ { \mathrm { F } } ^ { \mathrm { C S A } } \triangleq \mathbb { E } _ { ( y , \delta , x , a ) \sim p ( Y , \delta , X , A ) , \epsilon \sim p ( \epsilon ) } \left[ \delta \cdot \left( \left| y - t _ { a } \right| \right) + \left( 1 - \delta \right) \cdot \left( \operatorname* { m a x } ( 0 , y - t _ { a } ) \right) \right] , +$$ + +where the first term encourages sampled event times $t _ { a }$ to be close to $y$ , the ground truth for observed events, i.e., $\delta = 1$ , while penalizing $t _ { a }$ for being smaller than the censoring time when $\delta = 0$ . Further, the expectation is taken over samples (a minibatch) from empirical distribution $p ( Y , \delta , X , A )$ . + +Informative censoring We model informative censoring similar to (7) but mirroring the censoring indicators to encourage accurate censoring time samples $c _ { a }$ for $\delta = 0$ , while penalizing $c _ { a }$ for being smaller than $y$ for $\delta \ : = \ : 1$ (observed events). Specifically, we set an independent source of randomness like in (6) but parameterized by $\{ U _ { \nu } , W _ { \nu } , b _ { \nu } \}$ and censoring generative functions $\nu _ { A } ( r , \tilde { \epsilon } _ { \nu } )$ , parameterized as neural networks, where $c _ { a } \sim p _ { \nu , \Phi } ( C _ { A } | X = x )$ formulated as + +$$ +\ell _ { c } ( \nu , \Phi ) = \mathbb { E } _ { ( y , \delta , x , a ) \sim p ( y , \delta , X , A ) , \epsilon \sim p ( \epsilon ) } \left[ ( 1 - \delta ) \cdot ( | y - c _ { a } | ) + \delta \cdot ( \operatorname* { m a x } ( 0 , y - c _ { a } ) ) \right] . +$$ + +Further, we introduce an additional time-order-consistency loss that enforces the correct order of the observed time relative to the censoring indicator, i.e., $c _ { a } < t _ { a }$ if $\delta = 0$ and $t \sb a < c \sb a$ if $\delta = 1$ , thus + +$$ +\ell _ { \mathrm { T C } } ( h , \nu , \Phi ) = \mathbb { E } _ { ( \delta , x , a ) \sim p ( \delta , X , A ) , \epsilon \sim p ( \epsilon ) } \left[ \delta \cdot ( \operatorname* { m a x } ( 0 , t _ { a } - c _ { a } ) ) + ( 1 - \delta ) \cdot ( \operatorname* { m a x } ( 0 , c _ { a } - t _ { a } ) ) \right] . +$$ + +Note that $\ell _ { \mathrm { T C } } ( h , \nu , \Phi )$ does not depend on the observed event times but only on the censoring indicators. Finally, we write the consolidated CSA loss for informative censoring (CSA-INFO) by aggregating (7), (8) and (9) as $\mathcal { L } _ { \mathrm { F } } ^ { \mathrm { C S A - I N F O } } \triangleq \mathcal { L } _ { \mathrm { F } } ^ { \mathrm { C S A } } + \ell _ { \mathrm { c } } + \ell _ { \mathrm { T C } }$ . + +Learning Model functions $\{ h _ { A } ( \cdot ) , \Phi ( \cdot ) , \nu _ { A } ( \cdot ) \}$ are learned by minimizing the bound (5), via stochastic gradient descent on minibatches from $\mathcal { D }$ , with $\mathcal { L } _ { \mathrm { F } } ^ { \mathrm { C S A } }$ for non-informative censoring and $\mathcal { L } _ { \mathrm { F } } ^ { \mathrm { C S A - I N F O } }$ for informative censoring. Further, for the IPM regularization loss in (5), we optimize the dual formulation of the Wasserstein distance, via the regularized optimal transport (Villani, 2008; Cuturi, 2013). Consequently, we only require $\alpha ^ { - 1 } \zeta _ { h , \Phi } ( \bar { x , } a )$ to be 1-Lipschitz (Shalit et al., 2017) and $\alpha$ is selected by grid search on the validation set using only factual data (details below). + +# 4 METRICS + +We propose a comprehensive evaluation approach that accounts for both factual and causal metrics. Factual survival outcome predictions are evaluated according to standard survival metrics that measure diverse performance characteristics, such as concordance index (C-Index) (Harrell Jr et al., 1984), mean coefficient of variation (COV) and calibration slope (C-slope) (Chapfuwa et al., 2020). See the SM for more details on these metrics. For causal metrics, defined below, we introduce a nonparametric hazard ratio (HR) between treatment outcomes, and adopt the conventional precision in estimation of heterogeneous effect (PEHE) and average treatment effect (ATE) performance metrics (Hill, 2011). Note that PEHE and ATE require ground truth counterfactual event times, which is only possible in (semi-)synthetic data. For HR, we compare our findings with those independently reported in the literature from gold-standard RCT data. + +Nonparametric Hazard Ratio In a medical setting, the population hazard ratio $\mathrm { H R } ( t )$ between treatment groups is considered informative thus has been widely used in drug development and RCT (Yusuf et al., 2016; Mihaylova et al., 2012). For example, $\mathrm { H R } ( t ) < 1 , > 1$ , or $\approx 1$ indicate population positive, negative and neutral treatment effects at time $t$ , respectively. Moreover, $\mathrm { H R } ( t )$ naturally accounts for both censored and non-censored outcomes. Standard approaches for computing $\mathrm { H R } ( { \dot { t } } )$ rely on the restrictive proportional hazard assumption from CoxPH (Cox, 1972), which is constituted as a semi-parametric linear model $\lambda ( t | a ) = \lambda _ { \mathrm { b } } \bar { ( } t ) \exp ( a \beta )$ . However, the constant covariate (time independent) effect is often violated in practice (see Figure 2b). For CoxPH, the marginal HR between treatment and control can be obtained from regression coefficient $\beta$ learned via maximum likelihood without the need for specifying the baseline hazard $\begin{array} { r } { \mathrm { ~ \cdot ~ } _ { \mathrm { b } } ( t ) \colon \mathrm { H R } _ { \mathrm { C o x P H } } ( t ) = \frac { \lambda ( t | a = 1 ) } { \lambda ( t | a = 0 ) } = \exp ( \beta ) } \end{array}$ . So motivated, we propose a nonparametric, model-free approach for computing $\mathrm { H R } ( t )$ , in which we do not assume a parametric form for the event time distribution or the proportional hazard assumption from CoxPH. This approach only relies on samples from the conditional event time density functions, $f ( t _ { 1 } | x )$ and $f ( t _ { 0 } | x )$ , via $t _ { a } = h _ { A } ( \cdot )$ from (6). + +Definition 1 We define the nonparametric marginal Hazard Ratio and its approximation, $\mathrm { \hat { H R } } ( t )$ , as + +$$ +\mathrm { H R } ( t ) = \frac { \lambda _ { 1 } ( t ) } { \lambda _ { 0 } ( t ) } = \frac { S _ { 0 } ( t ) } { S _ { 1 } ( t ) } \cdot \frac { S _ { 1 } ^ { \prime } ( t ) } { S _ { 0 } ^ { \prime } ( t ) } , \quad \mathrm { H R } ( t ) = \frac { \hat { S } _ { 0 } ^ { \mathrm { P K M } } ( t ) } { \hat { S } _ { 1 } ^ { \mathrm { P K M } } ( t ) } \cdot \frac { m _ { 1 } ( t ) } { m _ { 0 } ( t ) } , +$$ + +where for $\mathrm { H R } ( t )$ we leveraged (1) to obtain (10) and $S ^ { \prime } ( t ) \triangleq d S ( t ) / d t$ . The nonparametric assumption for $S ( t )$ makes the computation of $S ^ { \prime } ( t )$ challenging. Provided that $S ( t )$ is a monotonically decreasing function, for simplicity, we fit a linear function $S ( t ) = m \cdot t + c .$ , and set $S ^ { \prime } ( t ) \approx m$ . Note that the linear model is only used for estimating $S ^ { \prime } ( t )$ from the nonparametric estimation of $S ( t )$ . + +Table 1: Performance comparisons on ACTG-SYNTHETIC data, with $9 5 \%$ $\mathrm { H R } ( t )$ confidence interval. The ground truth, test set, hazard ratio is $\mathrm { H R ( t ) } = 0 . 5 2 _ { ( 0 . 3 9 , 0 . 7 1 ) }$ . + +
MethodCausalFactualC-Slope (A=0,A=1)
EPEHEEATEHR(t)C-Index (A=0,A=1)Mean COV
CoxPH-UniformNANA0.97(0.86,1.09)NANANA
CoxPH-IPWNANA0.48(0.03,7.21)NANANA
CoxPH-OWNANA0.60(0.53,0.68)NANANA
Surv-BART352.0777.890.0(0.0,0.0)(0.706,0.686)0.001(0.398,0)
AFT-Weibull367.92133.930.47(0.47,0.47)(0.21,0.267)6.209(0.707,0.729)
AFT-log-Normal377.76157.640.47(0.47,0.47)(0.675, 0.556)6.971(0.707,0.729)
SR369.4788.550.38(0.330.65)(0.791,0.744)0(0.985,1.027)
CSA (proposed)358.720.80.45(0.39,0.65)(0.787,0.767)0.131(0.985,1.026)
CSA-INFO (proposed)344.331.190.53(0.41,0.67)(0.78,0.764)0.13(0.999,1.029)
+ +Bias from $S ^ { \prime } ( t )$ can be reduced by considering more complex function approximations for $S ( t )$ , e.g., polynomial or spline. For the nonparametric estimation of $S ( t )$ we leverage the model-free population point-estimate-based nonparametric Kaplan-Meier (Kaplan & Meier, 1958) estimator of the survival function $\hat { S } ^ { \mathrm { P K M } } ( t )$ in Chapfuwa et al. (2020) to marginalize both factual and counterfactual predictions given covariates $x$ . The approximated hazard ratio, $\mathrm { \hat { H R } } ( t )$ , is thus obtained by combining the approximations $\hat { S } _ { a } ^ { \mathrm { P K M } } ( t )$ and $m _ { a }$ . A similar formulation for the conditional, $\mathrm { H } \mathrm { \hat { R } } ( t | x )$ , can also be derived. See the SM for full details on the evaluation or $\mathrm { \hat { H R } } ( t )$ and $\mathrm { H } \mathrm { \hat { R } } ( t | x )$ . Note that for some AFT- or CoxPH-based parametric formulations, $\mathrm { H R } ( t | x )$ , can be readily evaluated because $f ( t _ { a } | x )$ and $S ( t _ { a } | x )$ are available in closed form. + +In the experiments, we will use $\mathrm { H R } ( t )$ to compare different approaches against results reported in RCTs (see Tables 1 and 3). Further, we will use $\mathrm { H R } ( t | x )$ to illustrate stratified treatment effects (see Figure 2). Note that though a neural-based survival recommender system (Katzman et al., 2018) has been previously used to estimate $\operatorname { H R } ( t | x )$ , their approach does not account for confounding or informative censoring thus it is susceptible to bias. + +Precision in Estimation of Heterogeneous Effect (PEHE) A general individualized estimation error is formulated as $\epsilon _ { \mathrm { P E H E } } = \sqrt { \mathbb { E } _ { X } [ ( \mathrm { I T E } ( x ) - \mathrm { I } \hat { \mathrm { T E } } ( x ) ) ^ { 2 } ] }$ , where $\operatorname { I T E } ( x )$ is the ground truth, $\mathrm { I } \hat { \mathrm { T E } } ( x ) = \mathbb { E } _ { T } \left[ \gamma \left( T _ { 1 } \right) - \gamma \left( T _ { 0 } \right) \vert X = x \right]$ and $\gamma ( \cdot )$ is a deterministic transformation. In our experiments, $\gamma ( \cdot )$ is the average over samples from $\dot { t _ { a } } \sim p _ { h , \Phi } ( T _ { A } | X = x )$ . Alternative estimands, e.g., thresholding survival times $\gamma ( T _ { A } ) = I \{ T _ { A } > \tau \}$ , can also be considered as described above. + +Average Treatment Effect (ATE) The population treatment effect estimation error is defined as $\epsilon _ { \mathrm { A T E } } = | \mathrm { A T E } - \mathrm { A \hat { T } E } |$ , where $\mathrm { A T E } = \mathbb { E } _ { X } [ \mathrm { I T E } ( x ) ]$ (ground truth) and $\mathrm { A } \hat { \mathrm { T E } } = \mathbb { E } _ { X } [ \hat { \mathrm { I T E } } ( x ) ]$ . + +# 5 EXPERIMENTS + +We describe the baselines and datasets that will be used to evaluate the proposed counterfactual survival analysis methods (CSA and CSA-INFO). Pytorch code including the new semi-synthetic dataset (see below) will be made publicly available. Throughout the experiments, we use the standard $\mathrm { H R } ( t )$ for CoxPH based methods and (10) for all others. The bound in (5) is sensitive to $\alpha$ , thus we propose approximating proxy counterfactual outcomes $\{ Y _ { \mathrm { C F } } , \delta _ { \mathrm { C F } } \}$ for the validation set, according to the covariate Euclidean nearest-neighbour (NN) from the training set. We select the $\alpha$ that minimizes the validation loss $\mathcal { L } = \mathcal { L } _ { \mathrm { F } } + \mathcal { L } _ { \mathrm { C F } }$ from the set $( 0 , 0 . 1 , 1 , 1 0 , 1 0 0 )$ . + +Baselines We consider the following competitive baseline approaches: $( i )$ propensity weighted CoxPH (Schemper et al., 2009; Buchanan et al., 2014; Rosenbaum & Rubin, 1983); $( i i )$ IPM (5) regularized AFT (log-Normal and Weibull) models; $( i i i )$ an IPM (5) regularized deterministic semisupervised regression (SR) model with accuracy objective from (Chapfuwa et al., 2018), as a contrast for the proposed stochastic predictors (CSA and CSA-INFO); and $( i v )$ survival Bayesian additive regression trees (Surv-BART) (Sparapani et al., 2016). For CoxPH, we consider three normalized weigh2009), $\begin{array} { r } { \mathrm { I P W } _ { i } = \frac { a _ { i } } { \hat { e } _ { i } } + \frac { 1 - a _ { i } } { 1 - \hat { e } _ { i } } ; i i ) } \end{array}$ $( i )$ weilappi ng (IPW) (Horvitz & Thompson, 1952; Cao et al.,weights (OW) (Crump et al., 2006; Li et al., 2018), $\mathrm { O W } _ { i } = a _ { i } \cdot \left( 1 - \hat { e } _ { i } \right) + \left( 1 - a _ { i } \right) \cdot \hat { e } _ { i }$ $i i i$ +linear logistic model $\hat { e } _ { i } = \sigma ( x _ { i } ; w )$ , is used as an approximation, $\boldsymbol { \hat { e } } _ { i }$ , to the unknown propensity score $P ( A = { \bar { 1 } } | X = x )$ . See the SM for a details of the baselines. + +Table 3: Performance comparisons on FRAMINGHAM data, with $9 5 \%$ $\mathrm { H R } ( t )$ confidence interval. Test set NN assignment of $y _ { \mathrm { C F } }$ and $\delta _ { \mathrm { C F } }$ yields biased $\mathrm { H R ( t ) } = 1 . 2 3 _ { ( 1 . 1 7 , 1 . 2 5 ) }$ , while previous large scale longitudinal RCT studies estimated $\mathrm { H R } ( \mathrm { t } ) = 0 . 7 5$ (0.64,0.88) (Yusuf et al., 2016). + +
MethodCausal HR(t)C-Index (A=0, A=1)Factual Mean COVC-Slope (A=0,A=1)
CoxPH-Uniform1.69(1.38,2.07)NANANA
CoxPH-IPW1.09(0.76,1.57)NANANA
CoxPH-OW0.88(0.73,1.08)NANANA
Surv-BART14.99(14.9,14.9e8)(0.629,0.630)0.003(0.232,0.084)
AFT-Weibull1.09(1.09,1.09)(0.734,0.395)8.609(0.857,0.89)
AFT-log-Normal1.55(1.46,1.55)(0.68,0.56)10.415(0.979,0.732)
SR0.58(0.53,0.71)(0.601, 0.57)0(0.491,0.63)
CSA(proposed)1.04(1.00,1.09)(0.763,0.728)0.161(0.891,0.81)
CSA-INFO (proposed)0.81(0.77,0.83)(0.752,0.651)0.156(0.907, 0.881)
+ +Datasets We consider the following datasets: (i) FRAMINGHAM, is an EHR-based longitudinal cardiovascular cohort study that we use to evaluate the effect of statins on future coronary heart disease outcomes (Benjamin et al., 1994); $( i i )$ ACTG, is a longitudinal RCT study comparing monotherapy with Zidovudine or Didanosine with combination therapy in HIV patients (Hammer et al., 1996); and $( i i i )$ ACTG + +Table 2: Summary statistics of the datasets. + +
FRAMINGHAMACTGACTG-SYNTHETIC
Events (%)26.026.948.9
Treatment (%)10.449.555.9
N3,4351,0542,139
p322323
Missing (%)0.231.411.38
tmax (days)7,2791,2311,313
+ +SYNTHETIC, is a semi-synthetic dataset based on ACTG covariates. We simulate potential outcomes according to a Gompertz-Cox distribution (Bender et al., 2005) with selection bias from a simple logistic model for $\bar { P ( A = 1 | X = x ) }$ and AFT-based censoring mechanism. The generative process is detailed in the SM. Table 2 summarizes the datasets according to $( i )$ covariates of size $p$ ; $( i i )$ proportion of non-censored events, treated units, and missing entries in the $N \times p$ covariate matrix; and $( i i i )$ time range $t _ { \mathrm { m a x } }$ for both censored and non-censored events. Missing entries are imputed with median or mode if continuous or categorical, respectively. + +Quantitative Results Experimental results for two data-sets in Tables 1 and 3, illustrate that AFTbased methods are high variance, inferior in calibration and C-Index than accuracy-based methods (SR, CSA, CSA-INFO). Surv-BART is the least calibrated but low variance method. CSA-INFO and CSA outperform all methods across all factual metrics, whereas CSA-INFO is better calibrated, low variance but slightly lower C-Index than CSA. Note that we fit CoxPH using the entire dataset; since it does not support counterfactual inference, we do not present factual metrics. By properly adjusting for both informative censoring and selection bias, CSA-INFO significantly outperforms all methods in treatment effect estimation according to $\mathrm { H R } ( t )$ and $\epsilon _ { \mathrm { P E H E } }$ , across non-RCT datasets, while remaining comparable to AFT-Weibull on the RCT dataset (see the SM). Further, RCT-based results on ACTG data in the SM illustrate comparable $\mathrm { H R } ( t )$ across all models except for AFT-log-Normal and Surv-BART, which overestimate, and SR, which underestimates risk. For non-RCT datasets (ACTG-SYNTHENTIC and FRAMINGHAM), CoxPH-OW has a clear advantage over all CoxPH based methods, mostly credited to the well-behaved bounded propensity weights $\in [ 0 , 1 ]$ . Interestingly, the FRAMINGHAM observational data exhibits a common paradox, where without proper adjustment of selection and censoring bias, naive approaches would result in a counter-intuitive treatment effect from statins. However, there is severe confounding from covariates such as age, BMI, diabetes, CAD, PAD, MI, stroke, etc., that influence both treatment likelihood and survival time. Table 3, demonstrates that CSA-INFO is clearly the best performing approach. Specifically, its $\mathrm { H R } ( t )$ , reverses the biased observational treatment effect, to demonstrate positive treatment from statins, which is consistent with prior large RCT longitudinal findings (Yusuf et al., 2016). + +Qualitative Results Figure 2a demonstrates that CSA-INFO matches the ground truth population hazard, $\mathrm { H R } ( t )$ , better than alternative methods on ACTG-SYNTHETIC data. See the SM for ACTG and FRAMINGHAM. Figure 2b shows sub-population log hazard ratios for four patient clusters obtained via hierarchical clustering on the individual log hazard ratios, $\log \mathrm { H R } ( t | x )$ , of the test set of FRAMINGHAM data. Interestingly, these clusters stratify treatment effects into: positive (2), negative (1 and 3), and neutral (4) sub-populations. Moreover, the estimated density of median $\log \mathrm { H } \bar { \mathrm { R } } ( t | x )$ values in Figure $_ { 2 \mathrm { c } }$ illustrates that nearly $7 0 \%$ of the testing set individuals have $\log \mathrm { H R } ( t | x ) < 0$ , thus may benefit from taking statins. Further, we isolated the extreme top and bottom quantiles, $\mathrm { H R } ( t | x ) < 0 . 0 2 4$ and $\mathrm { H R } ( \bar { t } | x ) > 1 . 9 1 6$ , respectively, of the median $\log \mathrm { \bar { H R } } ( t | x )$ values for the test set of FRAMINGHAM, as shown in Figure 2c. After comparing their covariates, we found that individuals with the following characteristics may benefit from taking statins: young, male, diabetic, without prior history (CAD, PAD, stroke or MI), high BMI, cholesterol, triglycerides, fasting glucose, and low high-density lipoprotein. There seem to be consensus that diabetics and high-cholesterol patients benefit from statins (Cheung et al., 2004; Wilt et al., 2004). See SM for additional results. + +![](images/4c89c16905f48a1bdcc35b1ade4f0f866e4e151f0c05e3f1b2a559591a62f777.jpg) +Figure 2: (a) Inferred population $\mathrm { H R } ( t )$ compared against ground truth (EMP) on ACTG-SYNTHETIC data. CSA-INFO-based (b) cluster-specific average l $\operatorname { \Pi } ^ { } \operatorname { \Pi } ^ { } \operatorname { \mathrm { ) g } } \operatorname { \mathrm { { H R } } } ( t | x )$ curves and (c) estimated density of median $\log \mathrm { H R } ( t | x )$ values on the test set of the FRAMINGHAM dataset. Clusters assignment were obtained via hierarchical clustering of individualized $\log \mathrm { H R } ( t | x )$ traces. + +# 6 CONCLUSIONS + +We have proposed a unified counterfactual inference framework for survival analysis. Our approach adjusts for bias from two unknown sources, namely, confounding due to covariate dependent selection bias and censoring (informative or non-informative). Relative to competitive alternatives, we demonstrate superior performance for both survival-outcome prediction and treatment-effect estimation, across three diverse datasets, including a semi-synthetic dataset which we introduce. Moreover, we formulate a model-free nonparametric hazard ratio metric for comparing treatment effects or leveraging prior randomized real-world experiments in longitudinal studies. + +# REFERENCES + +Peter C Austin. Propensity-score matching in the cardiovascular surgery literature from 2004 to 2006: a systematic review and suggestions for improvement. The Journal of Thoracic and Cardiovascular Surgery, 2007. + +Peter C Austin. The use of propensity score methods with survival or time-to-event outcomes: reporting measures of effect similar to those used in randomized experiments. Statistics in Medicine, 2014. + +Elias Bareinboim and Judea Pearl. Controlling selection bias in causal inference. In AISTATS, 2012. + +Ralf Bender, Thomas Augustin, and Maria Blettner. Generating survival times to simulate cox proportional hazards models. Statistics in medicine, 2005. + +Emelia J Benjamin, Daniel Levy, Sonya M Vaziri, Ralph B D’Agostino, Albert J Belanger, and Philip A Wolf. Independent risk factors for atrial fibrillation in a population-based cohort: the framingham heart study. Jama, 1994. + +Ashley L Buchanan, Michael G Hudgens, Stephen R Cole, Bryan Lau, Adaora A Adimora, and Women’s Interagency HIV Study. Worth the weight: using inverse probability weighted cox models in aids research. AIDS research and human retroviruses, 2014. + +Weihua Cao, Anastasios A Tsiatis, and Marie Davidian. Improving efficiency and robustness of the doubly robust estimator for a population mean with incomplete data. Biometrika, 2009. + +P. Chapfuwa, C. Tao, C. Li, I. Khan, K. J. Chandross, M. J. Pencina, L. Carin, and R. Henao. Calibration and uncertainty in neural time-to-event modeling. IEEE Transactions on Neural Networks and Learning Systems, 2020. + +Paidamoyo Chapfuwa, Chenyang Tao, Chunyuan Li, Courtney Page, Benjamin Goldstein, Lawrence Carin, and Ricardo Henao. Adversarial time-to-event modeling. In ICML, 2018. + +Bernard MY Cheung, Ian J Lauder, Chu-Pak Lau, and Cyrus R Kumana. Meta-analysis of large randomized controlled trials to evaluate the impact of statins on cardiovascular outcomes. British journal of clinical pharmacology, 2004. + +Hugh A Chipman, Edward I George, Robert E McCulloch, et al. Bart: Bayesian additive regression trees. The Annals of Applied Statistics, 2010. + +Stephen R Cole and Miguel A Hernán. Adjusted survival curves with inverse probability weights. Computer methods and programs in biomedicine, 2004. + +David R Cox. Regression models and life-tables. Journal of the Royal Statistical Society: Series B (Methodological), 1972. + +Richard K Crump, V Joseph Hotz, Guido W Imbens, and Oscar A Mitnik. Moving the goalposts: Addressing limited overlap in the estimation of average treatment effects by changing the estimand. Technical report, National Bureau of Economic Research, 2006. + +Yifan Cui, Michael R Kosorok, Stefan Wager, and Ruoqing Zhu. Estimating heterogeneous treatment effects with right-censored data via causal survival forests. arXiv, 2020. + +Marco Cuturi. Sinkhorn distances: Lightspeed computation of optimal transport. In NeurIPS, 2013. + +Iván Díaz. Statistical inference for data-adaptive doubly robust estimators with survival outcomes. Statistics in Medicine, 2019. + +Jennifer Frankovich, Christopher A Longhurst, and Scott M Sutherland. Evidence-based medicine in the emr era. N Engl J Med, 2011. + +Saurabh Gombar, Alison Callahan, Robert Califf, Robert Harrington, and Nigam H Shah. It is time to learn from patients like mine. NPJ digital medicine, 2019. + +Scott M Hammer, David A Katzenstein, Michael D Hughes, Holly Gundacker, Robert T Schooley, Richard H Haubrich, W Keith Henry, Michael M Lederman, John P Phair, Manette Niu, et al. A trial comparing nucleoside monotherapy with combination therapy in hiv-infected adults with cd4 cell counts from 200 to 500 per cubic millimeter. New England Journal of Medicine, 1996. + +Frank E Harrell Jr, Kerry L Lee, Robert M Califf, David B Pryor, and Robert A Rosati. Regression modelling strategies for improved prognostic prediction. Statistics in medicine, 1984. + +Kristiina Häyrinen, Kaija Saranto, and Pirkko Nykänen. Definition, structure, content, use and impacts of electronic health records: a review of the research literature. International Journal of Medical Informatics, 2008. + +Nicholas C Henderson, Thomas A Louis, Gary L Rosner, and Ravi Varadhan. Individualized treatment effects with censored data via fully nonparametric bayesian accelerated failure time models. Biostatistics, 2020. + +Miguel A Hernán and James M Robins. Causal inference: what if. Boca Raton: Chapman & Hill/CRC, 2020. + +Miguel A Hernán, Stephen R Cole, Joseph Margolick, Mardge Cohen, and James M Robins. Structural accelerated failure time models for survival analysis in studies with time-varying treatments. Pharmacoepidemiology and Drug Safety, 2005. + +Miguel Ángel Hernán, Babette Brumback, and James M Robins. Marginal structural models to estimate the causal effect of zidovudine on the survival of hiv-positive men. Epidemiology, 2000. + +Jennifer L Hill. Bayesian nonparametric modeling for causal inference. Journal of Computational and Graphical Statistics, 2011. + +Daniel G Horvitz and Donovan J Thompson. A generalization of sampling without replacement from a finite universe. Journal of the American statistical Association, 1952. + +Liangyuan Hu, Jiayi Ji, and Fan Li. Estimating heterogeneous survival treatment effect in observational data using machine learning. arXiv, 2020. + +Hemant Ishwaran, Udaya B Kogalur, Eugene H Blackstone, Michael S Lauer, et al. Random survival forests. The annals of applied statistics, 2008. + +Ashish K Jha, Catherine M DesRoches, Eric G Campbell, Karen Donelan, Sowmya R Rao, Timothy G Ferris, Alexandra Shields, Sara Rosenbaum, and David Blumenthal. Use of electronic health records in us hospitals. New England Journal of Medicine, 2009. + +Edward L Kaplan and Paul Meier. Nonparametric estimation from incomplete observations. Journal of the American statistical association, 1958. + +Jared L Katzman, Uri Shaham, Alexander Cloninger, Jonathan Bates, Tingting Jiang, and Yuval Kluger. Deepsurv: personalized treatment recommender system using a cox proportional hazards deep neural network. BMC medical research methodology, 2018. + +David G Kleinbaum and Mitchel Klein. Survival analysis. Springer, 2010. + +Fan Li, Kari Lock Morgan, and Alan M Zaslavsky. Balancing covariates via propensity score weighting. Journal of the American Statistical Association, 2018. + +Christopher A Longhurst, Robert A Harrington, and Nigam H Shah. A ‘green button’for using aggregate patient data at the point of care. Health affairs, 2014. + +B Mihaylova, J Emberson, L Blackwell, A Keech, J Simes, EH Barnes, M Voysey, 3A Gray, R Collins, and C Baigent. The effects of lowering ldl cholesterol with statin therapy in people at low risk of vascular disease: meta-analysis of individual data from 27 randomised trials., 2012. + +Xenia Miscouridou, Adler Perotte, Noémie Elhadad, and Rajesh Ranganath. Deep survival analysis: Nonparametrics and missingness. In Machine Learning for Healthcare Conference, 2018. + +Alfred Müller. Integral probability metrics and their generating classes of functions. Advances in Applied Probability, 1997. + +Judea Pearl and Elias Bareinboim. External validity: From do-calculus to transportability across populations. Statistical Science, 2014. + +Danilo Jimenez Rezende and Shakir Mohamed. Variational inference with normalizing flows. In ICML, 2015. + +James Robins. A new approach to causal inference in mortality studies with a sustained exposure period—application to control of the healthy worker survivor effect. Mathematical modelling, 1986. + +Paul R Rosenbaum and Donald B Rubin. The central role of the propensity score in observational studies for causal effects. Biometrika, 1983. + +Donald B Rubin. Causal inference using potential outcomes. Journal of the American Statistical Association, 2005. + +Michael Schemper, Samo Wakounig, and Georg Heinze. The estimation of average hazard ratios by weighted cox regression. Statistics in medicine, 2009. + +Uri Shalit, Fredrik D Johansson, and David Sontag. Estimating individual treatment effect: generalization bounds and algorithms. In ICML, 2017. + +Jincheng Shen, Lu Wang, Stephanie Daignault, Daniel E Spratt, Todd M Morgan, and Jeremy MG Taylor. Estimating the optimal personalized treatment strategy based on selected variables to prolong survival via random survival forest with weighted bootstrap. Journal of biopharmaceutical statistics, 2018. + +Rodney A Sparapani, Brent R Logan, Robert E McCulloch, and Purushottam W Laud. Nonparametric survival analysis using bayesian additive regression trees (bart). Statistics in medicine, 2016. + +Bharath K Sriperumbudur, Kenji Fukumizu, Arthur Gretton, Bernhard Schölkopf, Gert RG Lanckriet, et al. On the empirical estimation of integral probability metrics. Electronic Journal of Statistics, 2012. + +Ludovic Trinquart, Justine Jacot, Sarah C Conner, and Raphaël Porcher. Comparison of treatment effects measured by the hazard ratio and by the ratio of restricted mean survival times in oncology randomized controlled trials. Journal of Clinical Oncology, 2016. + +Anastasios Tsiatis. Semiparametric theory and missing data. Springer Science & Business Media, 2007. + +Mark J van der Laan and James M Robins. Unified approach for causal inference and censored data. In Unified Methods for Censored Longitudinal Data and Causality. Springer, 2003. + +Mark J Van der Laan and Sherri Rose. Targeted learning: causal inference for observational and experimental data. Springer Science & Business Media, 2011. + +Cédric Villani. Optimal transport: old and new. Springer Science & Business Media, 2008. + +Stefan Wager and Susan Athey. Estimation and inference of heterogeneous treatment effects using random forests. Journal of the American Statistical Association, 2018. + +Lee-Jen Wei. The accelerated failure time model: a useful alternative to the cox regression model in survival analysis. Statistics in medicine, 1992. + +Timothy J Wilt, Hanna E Bloomfield, Roderick MacDonald, David Nelson, Indulis Rutks, Michael Ho, Gregory Larsen, Anthony McCall, Sandra Pineros, and Anne Sales. Effectiveness of statin therapy in adults with coronary heart disease. Archives of internal medicine, 2004. + +Salim Yusuf, Jackie Bosch, Gilles Dagenais, Jun Zhu, Denis Xavier, Lisheng Liu, Prem Pais, Patricio López-Jaramillo, Lawrence A Leiter, Antonio Dans, et al. Cholesterol lowering in intermediate-risk persons without cardiovascular disease. New England Journal of Medicine, 2016. + +Yao Zhang, Alexis Bellot, and Mihaela van der Schaar. Learning overlapping representations for the estimation of individualized treatment effects. In AISTATS, 2020. + +Lihui Zhao, Lu Tian, Hajime Uno, Scott D Solomon, Marc A Pfeffer, Jerald S Schindler, and Lee Jen Wei. Utilizing the integrated difference of two survival functions to quantify the treatment contrast for designing, monitoring, and analyzing a comparative clinical study. Clinical trials, 2012. \ No newline at end of file diff --git a/parse/train/3ZeGLibhFo0/3ZeGLibhFo0_content_list.json b/parse/train/3ZeGLibhFo0/3ZeGLibhFo0_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..24aaf2bd1546da8e687f8e0bd0e21b9b6215ab29 --- /dev/null +++ b/parse/train/3ZeGLibhFo0/3ZeGLibhFo0_content_list.json @@ -0,0 +1,1504 @@ +[ + { + "type": "text", + "text": "ENABLING COUNTERFACTUAL SURVIVAL ANALYSIS WITH BALANCED REPRESENTATIONS ", + "text_level": 1, + "bbox": [ + 176, + 98, + 823, + 146 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Anonymous authors Paper under double-blind review ", + "bbox": [ + 183, + 170, + 398, + 198 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "ABSTRACT ", + "text_level": 1, + "bbox": [ + 454, + 234, + 544, + 251 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Balanced representation learning methods have been applied successfully to counterfactual inference from observational data. However, approaches that account for survival outcomes are relatively limited. Survival data are frequently encountered across diverse medical applications, i.e., drug development, risk profiling, and clinical trials, and such data are also relevant in fields like manufacturing (for equipment monitoring). When the outcome of interest is time-to-event, special precautions for handling censored events need to be taken, as ignoring censored outcomes may lead to biased estimates. We propose a theoretically grounded unified framework for counterfactual inference applicable to survival outcomes. Further, we formulate a nonparametric hazard ratio metric for evaluating average and individualized treatment effects. Experimental results on real-world and semi-synthetic datasets, the latter which we introduce, demonstrate that the proposed approach significantly outperforms competitive alternatives in both survival-outcome predictions and treatment-effect estimation. ", + "bbox": [ + 233, + 267, + 766, + 460 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 INTRODUCTION ", + "text_level": 1, + "bbox": [ + 178, + 489, + 336, + 505 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Survival analysis or time-to-event studies focus on modeling the time of a future event, such as death or failure, and investigate its relationship with covariates or predictors of interest. Specifically, we may be interested in the causal effect of a given intervention or treatment on survival time. A typical question may be: will a given therapy increase the chances of survival of an individual or population? Such causal inquiries on survival outcomes are common in the fields of epidemiology and medicine (Robins, 1986; Hammer et al., 1996; Yusuf et al., 2016). As an important current example, the COVID-19 pandemic is creating a demand for methodological development to address such questions, specifically, when evaluating the effectiveness of a potential vaccine or therapeutic outside randomized controlled trial settings. ", + "bbox": [ + 174, + 520, + 825, + 646 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Traditional causal survival analysis is typically carried out in the context of a randomized controlled trial (RCT), where the treatment assignment is controlled by researchers. Though they are the gold standard for causal inference, RCTs are usually long-term engagements, expensive and limited in sample size. Alternatively, the availability of observational data with comprehensive information about patients, such as electronic health records (EHRs), constitutes a more accessible but also more challenging source for estimating causal effects (Häyrinen et al., 2008; Jha et al., 2009). Such observational data may be used to augment and verify an RCT, after a particular treatment is approved and in use (Gombar et al., 2019; Frankovich et al., 2011; Longhurst et al., 2014). Moreover, the wealth of information from observational data also allows for the estimation of the individualized treatment effect (ITE), namely, the causal effect of an intervention at the individual level. In this work, we develop a novel framework for counterfactual time-to-event prediction to estimate the ITE for survival or time-to-event outcomes from observational data. ", + "bbox": [ + 174, + 652, + 825, + 819 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Estimating the causal effect for survival outcomes in observational data manifests two principal challenges. First, the treatment assignment mechanism is not known a priori. Therefore, there may be variables, known as confounders, affecting both the treatment and survival time, which lead to selection bias (Bareinboim & Pearl, 2012), i.e., that the distributions across treatment groups are not the same. In this work, we focus on selection biases due to confounding, but other sources may also be considered. For instance, patients who are severely ill are likely to receive more aggressive therapy, however, their health status may also inevitably influence survival. Traditional survival analysis neglects such bias, leading to incorrect causal estimation. Second, the exact time-to-event is not always observed, i.e., sometimes we only know that an event has not occurred up to a certain point in time. This is known as the censoring problem. Moreover, censoring might be informative depending on the characteristics of the individuals and their treatment assignments, thus proper adjustment is required for accurate causal estimation (Cole & Hernán, 2004; Díaz, 2019). ", + "bbox": [ + 174, + 827, + 823, + 924 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 103, + 825, + 172 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Traditional causal survival-analysis approaches typically model the effect of the treatment or covariates (not time or survival) in a parametric manner. Two commonly used models are the Cox proportional hazards (CoxPH) model (Cox, 1972) and the accelerated failure time (AFT) model (Wei, 1992), which presume a linear relationship between the covariates and survival probability. Further, proper weighting for each individual has been employed to account for confounding bias from these models (Austin, 2007; 2014; Hernán et al., 2005). For instance, probability weighting schemes that account for both selection bias and covariate dependent censoring have been considered for adjusted survival curves (Cole & Hernán, 2004; Díaz, 2019). Moreover, such probability weighting schemes have been applied to causal survival-analysis under time-varying treatment and confounding (Robins, 1986; Hernán et al., 2000). See van der Laan & Robins (2003); Tsiatis (2007); Van der Laan & Rose (2011); Hernán & Robins (2020) for an overview. Such linear specification makes these models interpretable but compromises their flexibility, and makes it difficult to adapt them for high-dimensional data or to capture complex interactions among covariates. Importantly, these methods lack a counterfactual prediction mechanism, which is key for ITE estimation (see Section 2). ", + "bbox": [ + 174, + 181, + 825, + 375 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Fortunately, recent advances in machine learning, such as representation learning or generative modeling, have enabled causal inference methods to handle high-dimensional data and to characterize complex interactions effectively. For instance, there has been recent interest in tree-based (Chipman et al., 2010; Wager & Athey, 2018) and neural-network-based (Shalit et al., 2017; Zhang et al., 2020) approaches. For pre-specified time-horizons, the nonparametric Random Survival Forest (RSF) (Ishwaran et al., 2008) and Bayesian Additive regression trees (BART) (Chipman et al., 2010) have been extended to causal survival analysis. RSF has been applied to causal survival forests with weighted bootstrap inference (Shen et al., 2018; Cui et al., 2020) while a BART is extended to account for survival outcomes in Surv-BART (Sparapani et al., 2016), and AFT-BART (Henderson et al., 2020). See Hu et al. (2020) for an extensive investigation of the causal survival tree-based methods. Alternatively, when estimating the ITE, neural-network-based methods propose to regularize the transformed covariates or representations for an individual to have balanced distributions across treatment groups, thus accounting for the confounding bias and improving ITE prediction. However, most approaches employing representation learning techniques for counterfactual inference deal with continuous or binary outcomes, instead of time-to-event outcomes with censoring (informative or non-informative). Hence, a principled generalization to the context of counterfactual survival analysis is needed. ", + "bbox": [ + 174, + 382, + 825, + 617 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "In this work we leverage balanced (latent) representation learning to estimate ITE via counterfactual prediction of survival outcomes in observational studies. We develop a framework to predict event times from a low-dimensional transformation of the original covariate space. To address the specific challenges associated with counterfactual survival analysis, we make the following contributions: ", + "bbox": [ + 174, + 625, + 825, + 680 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "• We develop an optimization objective incorporating adjustments for informative censoring, as well as a balanced regularization term bounding the generalization error for ITE prediction. For the latter, we repurpose a recently proposed bound (Shalit et al., 2017) for our time-to-event scenario. \n• We propose a generative model for event times to relax restrictive survival linear and parametric assumptions, thus allowing for more flexible modeling. Our approach can also provide nonparametric uncertainty quantification for ITE predictions. \n• We provide survival-specific evaluation metrics, including a new nonparametric hazard ratio estimator, and discuss how to perform model selection for survival outcomes. The proposed model demonstrates superior performance relative to the commonly used baselines in real-world and semi-synthetic datasets. \n• We introduce a survival-specific semi-synthetic dataset and demonstrate an approach for leveraging prior randomized experiments in longitudinal studies for model validation. ", + "bbox": [ + 176, + 688, + 826, + 854 + ], + "page_idx": 1 + }, + { + "type": "image", + "img_path": "images/482e9df6a7b5d031a444403a4be41d381d10c980808568d8cbf42f012f3882e6.jpg", + "image_caption": [ + "Figure 1: (a) Illustration of the proposed counterfactual survival analysis (CSA). Covariates $X =$ $x$ are mapped into latent representation $r$ via deterministic mapping $r \\ = \\ \\Phi ( x )$ . The potential outcomes are sampled from $\\bar { t } _ { a } \\sim p ( T _ { A } | X = x )$ for $A = a$ via stochastic mapping $h _ { A } ( r , \\tilde { \\epsilon } )$ , where randomness is induced with a flow-based transformation, $\\tilde { \\epsilon }$ , of a simple distribution $p ( \\epsilon )$ , i.e., uniform or Gaussian. (b) and (c) show the proposed causal graphs for non-informative and informative censoring, respectively. " + ], + "image_footnote": [], + "bbox": [ + 200, + 92, + 789, + 207 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "2 PROBLEM FORMULATION ", + "text_level": 1, + "bbox": [ + 176, + 308, + 418, + 324 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "We first introduce the basic setup for performing causal survival analysis in observational studies. Suppose we have $N$ units, with $N _ { 1 }$ units being treated and $N _ { 0 }$ in the control group $( N = N _ { 1 } + N _ { 0 } )$ . For each unit (individual), we have covariates $X$ , which can be heterogeneous, e.g., a mixture of categorical and continuous covariates which, in the context of medicine, may include labs, vitals, procedure codes, etc. We also have a treatment indicator $A$ , where $A = 0$ for the controls and $A = 1$ for the treated, as well as the outcome (event) of interest $T$ . Under the potential-outcomes framework (Rubin, 2005), let $T _ { 0 }$ and $T _ { 1 }$ be the potential event times for a given subject under control and treatment, respectively. In practice we only observe one realization of the potential outcomes, i.e., the factual outcome $T = T _ { A }$ , while the counterfactual outcome $T _ { 1 - A }$ is unobserved. ", + "bbox": [ + 173, + 338, + 826, + 465 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "In survival analysis, the problem becomes more difficult because we do not always observe the exact event time for each individual, but rather the time up to which we are certain that the event has not occurred; specifically, we have a (right) censoring problem, most likely due to the loss of follow-up. We denote the censoring time as $C$ and censoring indicator as $\\delta \\in \\{ 0 , 1 \\}$ . The actual observed time is $Y = \\operatorname* { m i n } ( T _ { A } , C )$ , i.e., the outcome is observed (non-censored) if $T _ { A } < C$ and $\\delta = 1$ . ", + "bbox": [ + 173, + 470, + 825, + 541 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "In this work, we are interested in the expected difference between the $T _ { 1 }$ and $T _ { 0 }$ conditioned on $X$ for a given unit (individual), which is commonly known as the individualized treatment effect (ITE). Specifically, we wish to perform inference on the conditional distributions of $T _ { 1 }$ and $T _ { 0 }$ , i.e., $p ( T _ { 1 } | X )$ and $p ( T _ { 0 } | X )$ , respectively, as shown in Figure 1a. In practice, we observe $N$ realizations of $( Y , \\delta , X , A )$ for observed time, censoring indicator, covariates and treatment indicator, respectively; hence, from an observational study the dataset takes the form $\\mathcal { D } = \\{ ( y _ { i } , \\delta _ { i } , x _ { i } , a _ { i } ) \\} _ { i = 1 } ^ { N }$ . Below, we discuss several common choices of estimands in survival analysis. ", + "bbox": [ + 173, + 547, + 825, + 646 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Estimands of Interest We begin by considering survival analysis in the absence of an intervening treatment choice, $A$ . Let $F ( t | x ) \\triangleq P ( T \\leq t | X = x )$ be the cumulative distribution function of the event (failure) time, $t$ , given a realization of the covariates, $x$ . Survival analysis is primarily concerned with characterization of the survival function conditioned on covariates $S ( t | x ) \\triangleq 1 - F ( t | x )$ , and the hazard function or risk score, $\\lambda ( t | x )$ , defined below. $S ( t | x )$ is a monotonically decreasing function indicating the probability of survival up to time $t$ . The hazard function measures the instantaneous probability of the event occurring between $\\{ t , t + \\Delta t \\}$ given $T > t$ and $\\Delta t 0$ . From standard definitions (Kleinbaum & Klein, 2010), the relationship between cumulative and hazard function is formulated as ", + "bbox": [ + 173, + 654, + 826, + 784 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/89816b6b4ab5614eea0ade34840cd19c6b9c6faa995393a64ef37e851560bce4.jpg", + "text": "$$\n\\lambda ( t | x ) = \\operatorname* { l i m } _ { d t 0 } { \\frac { P ( t < T < t + d t | X = x ) } { P ( T > t | X = x ) d t } } = - { \\frac { d \\log S ( t | x ) } { d t } } = { \\frac { f ( t | x ) } { S ( t | x ) } } .\n$$", + "text_format": "latex", + "bbox": [ + 263, + 787, + 736, + 823 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "From (1) we see that $f ( t | x ) \\triangleq P ( T = t | X = x ) = \\lambda ( t | x ) S ( t | x )$ , is the conditional event time density function (Kleinbaum $\\&$ Klein, 2010). ", + "bbox": [ + 176, + 832, + 823, + 861 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Given the binary treatment $A$ , we are interested in its impact on the survival time. For ITE estimation, we are also interested in the difference between the two potential outcomes $T _ { 1 } , T _ { 0 }$ . Let $S _ { A } ( t | x )$ and $\\lambda _ { A } ( t | x )$ denote the survival and hazard functions for the potential outcomes $T _ { A }$ , i.e., $T _ { 1 }$ and $T _ { 0 }$ . Several common estimands of interest include (Zhao et al., 2012; Trinquart et al., 2016): difference in expected lifetime: $\\begin{array} { r } { \\mathrm { { I T E } } ( t , x ) = \\int _ { 0 } ^ { t _ { \\operatorname* { m a x } } } \\{ S _ { 1 } ( t | x ) - S _ { 0 } ( t | x ) \\} \\mathrm { d } t = \\mathbb { E } \\{ T _ { 1 } - T _ { 0 } | X = x \\} . } \\end{array}$ , difference in survival function: $\\mathrm { I T E } ( t , x ) = S _ { 1 } ( t | x ) - S _ { 0 } ( t | x )$ , and hazard ratio: $\\mathrm { I T E } ( t , x ) = \\lambda _ { 1 } ( t | x ) / \\lambda _ { 0 } ( t | x )$ . The inference difficulties associated with the above estimands from observational data are two-fold. First, there are confounders affecting both the treatment assignment and outcomes, which stem from selection bias, i.e., the treatment and control covariate distributions are not necessarily the same. Also, we do not have direct knowledge of the conditional treatment assignment mechanism, i.e., $P ( A = a | X = x )$ , also known as the propensity score. Let ⊥⊥ denote statistical independence. For estimands to be identifiable from observational data, we make two assumptions: (i) $\\{ T _ { 1 } , T _ { 0 } \\} \\perp \\perp A | X$ , i.e., no unobserved confounders or ignorability, and $( i i )$ overlap in the covariate support $0 < P ( A =$ $1 | X = x ) < 1$ almost surely if $p ( X = x ) > 0$ . Second, the censoring mechanism is also unknown and may lead to bias without proper adjustment. We consider two censoring mechanisms in our work, $( i )$ conditionally independent or informative censoring: $T \\perp \\perp C | X , A$ , and $( i i )$ random or non-informative censoring: $T \\perp \\perp C$ . Note that for informative censoring, we also have to consider potential censoring times $C _ { 1 }$ and $C _ { 0 }$ and their conditionals $p ( C _ { 1 } | X )$ and $p ( C _ { 0 } | X )$ , respectively. Figure 1 shows causal graphs illustrating these modeling assumptions. ", + "bbox": [ + 174, + 867, + 825, + 924 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 101, + 826, + 314 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3 MODELING ", + "text_level": 1, + "bbox": [ + 174, + 330, + 300, + 347 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "To overcome the above challenges and adjust for observational biases, we propose a unified framework for counterfactual survival analysis (CSA). Specifically, we repurpose the counterfactual bound in Shalit et al. (2017) for our time-to-event scenario and introduce a nonparametric approach for stochastic survival outcome predictions. Below we formulate a theoretically grounded and unified approach for estimating $( i )$ the encoder function $r = \\Phi ( x )$ , which deterministically maps covariates $x$ to their corresponding latent representation $r \\in { \\mathbb { R } } ^ { d }$ , and $( i i )$ two stochastic time-to-event generative functions, $h _ { A } ( \\cdot )$ , to implicitly draw samples from both potential outcome conditionals $\\bar { t } _ { a } \\sim p _ { h , \\Phi } ( T _ { A } | X = x )$ , for $A = \\{ 1 , 0 \\}$ , and where $t _ { a }$ indicates the sample from $p _ { h , \\Phi } ( T _ { A } | X = x )$ is for $A = a$ . Further, we formulate a general extension that accounts for informative censoring by introducing two stochastic censoring generative functions, $\\nu _ { A } ( \\cdot )$ , to draw samples for potential censoring times $c _ { a } \\sim p _ { \\nu , \\Phi } ( C _ { A } | X = x )$ . The model-specifying functions, $\\{ h _ { A } ( \\cdot ) , \\nu _ { A } ( \\cdot ) , \\Phi ( \\cdot ) \\}$ , are parameterized via neural networks. See the Supplementary Material (SM) for details. Figure 1a summarizes our modeling approach. ", + "bbox": [ + 174, + 358, + 825, + 540 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Accounting for selection bias We wish to estimate the potential outcomes, i.e., event times, which are sampled by distributions parameterized by functions $\\{ h _ { A } ( \\cdot ) , \\Phi ( \\cdot ) \\}$ , i.e., ", + "bbox": [ + 173, + 547, + 821, + 577 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/9e497767f6b296dac4c8cf83d111947941f15982093bbf8fbe2814a7e006faf0.jpg", + "text": "$$\n\\begin{array} { c } { { t \\sim p _ { h , \\Phi } ( T | X = x , A = a ) } } \\\\ { { { \\it t } _ { a } \\sim p _ { h , \\Phi } ( T _ { a } | X = x ) } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 401, + 583, + 596, + 622 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "We obtain (3) from (2) via the strong ignorability assumption, i.e., $\\{ T _ { 0 } , T _ { 1 } \\}$ ⊥⊥ $A | X$ (consistent with the causal graphs in Figure 1b and 1c) and $0 < P ( A = \\mathsf { \\bar { a } } | X \\doteq x ) < 1 .$ , and the consistency assumption, i.e., $T = T _ { A } | A = a$ . A similar argument can be made for informative censoring based on Figure 1c, so we can also write $c _ { a } \\sim p _ { \\nu , \\Phi } ( C _ { A } | X = x )$ . Given (3), model functions $\\{ h _ { A } ( \\cdot ) , \\Phi ( \\cdot ) \\bar \\}$ and $\\nu _ { A } ( \\cdot )$ for informative censoring can be learned by leveraging standard statistical optimization approaches, that minimize a loss hypothesis $\\mathcal { L }$ given samples from the empirical distribution $( y , \\delta , x , a ) \\sim p ( Y , \\delta , X , A )$ , i.e., from dataset $\\mathcal { D }$ . Specifically, we write $\\mathcal { L } = \\mathbb { E } _ { ( y , \\delta , x , a ) \\sim p ( Y , \\delta , X , A ) } \\left[ \\ell _ { h , \\Phi } ( t _ { a } , y , \\delta ) \\right]$ , where $\\ell _ { h , \\Phi } ( t _ { a } , y , \\delta )$ is a loss function that measures the agreement of $t _ { a } \\sim p _ { h , \\Phi } ( T _ { A } | X = x )$ (and $c _ { a } \\sim p _ { \\nu , \\Phi } ( C _ { A } | X = x )$ for informative censoring) with ground truth $\\{ y , \\delta \\}$ , the observed time and censoring indicator, respectively. ", + "bbox": [ + 173, + 625, + 825, + 767 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "For some parametric formulations of event time distribution $p _ { h , \\Phi } ( T _ { A } | X = x )$ , e.g., exponential, Weibull, log-Normal, etc., and provided the censoring mechanism is non-informative, $- \\ell _ { h , \\Phi } ( t _ { a } , y , \\delta )$ is the closed form log likelihood. Specifically, $- \\ell _ { h , \\Phi } ( t _ { a } , y , \\delta ) \\triangleq \\log p _ { h , \\Phi } ( T _ { a } | X = x ) = \\delta$ · $\\log f _ { h , \\Phi } ( t _ { a } | x ) + ( 1 - \\bar { \\delta } ) \\cdot \\log S _ { h , \\Phi } ( t _ { a } | x )$ , which implies that the conditional event time density and survival functions can be calculated in closed form from transformations $\\{ h _ { A } ( \\cdot ) , \\Phi ( \\cdot ) \\}$ of $x$ . See the SM for parametric examples of $\\mathcal { L }$ accounting for informative censoring. ", + "bbox": [ + 173, + 772, + 825, + 859 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "We further define the expected loss for a given realization of covariates $x$ and treatment assignment $a$ over observed times $y$ (censored and non-censored), and the censoring indicator $\\delta$ as $\\zeta _ { h , \\Phi } ( x , a ) \\triangleq$ $\\mathbb { E } _ { ( y , \\delta , x ) \\sim p ( Y , \\delta | X ) } \\ell _ { h , \\Phi } ( t _ { a } , y , \\delta )$ as in Shalit et al. (2017). For a given subject with covariates $x$ and treatment assignment $a$ , we wish to minimize both the factual and counterfactual losses, $\\mathcal { L } _ { \\mathrm { F } }$ and $\\mathcal { L } _ { \\mathrm { C F } }$ respectively, by decomposing $\\mathcal { L } = \\mathcal { L } _ { \\mathrm { F } } + \\mathcal { L } _ { \\mathrm { C F } }$ as follows ", + "bbox": [ + 174, + 866, + 825, + 924 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 103, + 547, + 118 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/141da49929ef287744dea443052c2291ced1fb2cbc977e821917e72659db41f1.jpg", + "text": "$$\n\\begin{array} { r } { \\mathcal { L } _ { \\mathrm { F } } = \\mathbb { E } _ { ( x , a ) \\sim p ( A , X ) } \\zeta _ { h , \\Phi } ( x , a ) , \\quad \\mathcal { L } _ { \\mathrm { C F } } = \\mathbb { E } _ { ( x , a ) \\sim p ( 1 - A , X ) } \\zeta _ { h , \\Phi } ( x , a ) . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 264, + 128, + 732, + 147 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Let $u \\triangleq P ( A = 1 )$ denote the marginal probability of treatment assignment. We can readily decompose the losses in (4) according to treatment assignments. The decomposed factual $\\mathcal { L } _ { \\mathrm { F } } =$ $u \\cdot \\mathcal { L } _ { \\mathrm { F } } ^ { A = 1 } + \\left( 1 - u \\right) \\cdot \\mathcal { L } _ { \\mathrm { F } } ^ { A = 0 }$ , and similarly, the decomposed counterfactual $\\mathcal { L } _ { \\mathrm { C F } } = \\dot { ( } 1 - u ) \\cdot \\mathcal { L } _ { \\mathrm { C F } } ^ { A = 1 } + u$ $\\mathcal { L } _ { \\mathrm { C F } } ^ { A = 0 }$ F CF . In practice, only factual outcomes are observed, hence, for a non-randomized non-controlled experiment, we cannot obtain an unbiased estimate of $\\mathcal { L } _ { \\mathrm { C F } }$ from data due to selection bias (or confounding). Therefore, we bound $\\mathcal { L } _ { \\mathrm { C F } }$ and $\\mathcal { L }$ below following Shalit et al. (2017). ", + "bbox": [ + 173, + 159, + 825, + 246 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Corollary 1 Assume $\\Phi ( \\cdot )$ is an invertible map, and $\\alpha ^ { - 1 } \\zeta _ { h , \\Phi } ( x , a ) \\in G$ , where $G$ is a family of functions, $p _ { \\Phi } ^ { A = a } \\triangleq p _ { \\Phi } ( R | A = a )$ is the latent distribution for group $A = a$ , and $\\alpha > 0$ is a constant. Then, we have: ", + "bbox": [ + 173, + 257, + 826, + 303 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/370f28aa2b443822fd375b3828179499d1816978110177146821204aef6fa36a.jpg", + "text": "$$\n\\begin{array} { r l } & { \\mathcal { L } _ { \\mathrm { C F } } \\leq ( 1 - u ) \\cdot \\mathcal { L } _ { \\mathrm { F } } ^ { A = 1 } + u \\cdot \\mathcal { L } _ { \\mathrm { F } } ^ { A = 0 } + \\alpha \\cdot \\mathrm { I P M } _ { G } ( p _ { \\Phi } ^ { A = 1 } , p _ { \\Phi } ^ { A = 0 } ) } \\\\ & { \\quad \\mathcal { L } \\leq \\mathcal { L } _ { \\mathrm { F } } ^ { A = 1 } + \\mathcal { L } _ { \\mathrm { F } } ^ { A = 0 } + \\alpha \\cdot \\mathrm { I P M } _ { G } ( p _ { \\Phi } ^ { A = 1 } , p _ { \\Phi } ^ { A = 0 } ) . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 294, + 303, + 702, + 347 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "The integral probability metric (IPM) (Müller, 1997; Sriperumbudur et al., 2012) measures the distance between two probability distributions $p$ and $q$ defined over $M$ , i.e., the latent space of $R$ . Formally, $\\begin{array} { r } { \\mathrm { I P M } _ { G } ( p , q ) \\triangleq \\operatorname* { s u p } _ { g \\in G } \\vert \\int _ { M } g ( m ) \\left( p ( m ) - q ( m ) \\right) d m \\vert } \\end{array}$ , where $g : m \\mathbb { R }$ , represents a class of real-valued bounded measurable functions on $M$ (Shalit et al., 2017). Therefore, model functions $\\{ h _ { a } ( \\cdot ) , \\Phi ( \\cdot ) \\}$ can be learned by minimizing the upper bound in (5) consisting of $( i )$ only factual losses under both treatment assignments and $( i i )$ an IPM regularizer enforcing latent distributional equivalence between the treatment groups. Note that if the data originates from a RCT it follows (by construction) that $\\mathrm { I P M } _ { G } ( p _ { \\Phi } ^ { A = 1 } , p _ { \\Phi } ^ { \\tilde { A = 0 } } ) \\stackrel { . } { = } 0$ . ", + "bbox": [ + 173, + 359, + 826, + 477 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Accounting for censoring bias Below we formulate an approach for estimating functions $h _ { A } ( \\cdot )$ and $\\nu _ { A } ( \\cdot )$ for synthesizing (sampling) non-censored $t _ { a } \\sim \\stackrel { \\sim } { p } _ { h , \\Phi } ( T _ { A } | X = x )$ and censored $c _ { a } \\sim$ $p _ { \\nu , \\Phi } ( C _ { A } | X = x )$ times, respectively. While some parametric assumptions for $p _ { h , \\Phi } ( T _ { A } | X = x )$ yield easy-to-evaluate closed forms for $S _ { h , \\Phi } ( t _ { a } | x )$ that can be used as likelihood for censored observations, they are restrictive, and have been shown to generate unrealistic high variance samples (Chapfuwa et al., 2018). So motivated, we seek a nonparametric likelihood-based approach that can model a flexible family of distributions, with an easy-to-sample approach for event times $t _ { a } \\sim$ $p _ { h , \\Phi } ( T _ { a } | X = x )$ . We model the event time generation process with a source of randomness, $p ( \\epsilon )$ , e.g. Gaussian or uniform, which is obtained from a neural-network-based nonlinear transformation. In the experiments we use a planar flow formulation parameterized by $\\{ U _ { h } , W _ { h } , b _ { h } \\}$ (Rezende & Mohamed, 2015), however, other specifications can also be used. Note that Miscouridou et al. (2018) has previously leveraged normalizing flows for survival analysis, however, our approach is very different in that it focuses on formulating $\\romannumeral 1$ ) a counterfactual survival analysis framework that accounts for informative or non-informative censoring mechanisms and confounding, and $_ { i i }$ ) model event times as a continuous variable instead of discretizing them. Specifically, we transform the source of randomness, $\\epsilon$ , using a single layer specification as follows ", + "bbox": [ + 173, + 486, + 826, + 709 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/5d0fe636d034b8d80f53f9ef96d49f1ce924a440335a5cca19c9ccfdb29ad614.jpg", + "text": "$$\n{ \\widetilde { \\epsilon } } _ { h } = \\epsilon + U _ { h } \\operatorname { t a n h } ( W _ { h } \\epsilon + b _ { h } ) , \\quad \\epsilon \\sim \\operatorname { U n i f o r m } ( 0 , 1 ) , t _ { a } = h _ { A } ( r , { \\widetilde { \\epsilon } } _ { h } ) , \\quad r = \\Phi ( x )\n$$", + "text_format": "latex", + "bbox": [ + 225, + 719, + 772, + 737 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where $\\{ U _ { h } , W _ { h } \\} \\in \\mathbb { R } ^ { d \\times d }$ , $\\{ b _ { h } , \\epsilon \\} \\in \\mathbb { R } ^ { d }$ , $d$ is the dimensionality of the normalizing flow; each component of $\\epsilon$ is drawn independently from Uniform $( 0 , 1 )$ , and $\\tilde { \\epsilon } _ { h }$ may be viewed as a skip connection with stochasticity in $\\epsilon$ . Further, $h _ { A } ( r , \\tilde { \\epsilon } _ { h } )$ and $\\Phi ( x )$ are time-to-event generative and encoding functions, respectively, parameterized as neural networks. For simplicity, the dimensions of $r$ and $\\epsilon$ are set to $d$ , however, they can be set independently if desired. In practice, we are interested in generating realistic event-time samples; therefore, we account for both censored and non-censored observations by adopting the objective from Chapfuwa et al. (2018), formulated as ", + "bbox": [ + 173, + 747, + 825, + 847 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/7fde54abcbf1fd3fd611de6768c944eaf23863c85ea57f28571b691d6e808338.jpg", + "text": "$$\n\\mathcal { L } _ { \\mathrm { F } } ^ { \\mathrm { C S A } } \\triangleq \\mathbb { E } _ { ( y , \\delta , x , a ) \\sim p ( Y , \\delta , X , A ) , \\epsilon \\sim p ( \\epsilon ) } \\left[ \\delta \\cdot \\left( \\left| y - t _ { a } \\right| \\right) + \\left( 1 - \\delta \\right) \\cdot \\left( \\operatorname* { m a x } ( 0 , y - t _ { a } ) \\right) \\right] ,\n$$", + "text_format": "latex", + "bbox": [ + 230, + 856, + 766, + 877 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where the first term encourages sampled event times $t _ { a }$ to be close to $y$ , the ground truth for observed events, i.e., $\\delta = 1$ , while penalizing $t _ { a }$ for being smaller than the censoring time when $\\delta = 0$ . Further, the expectation is taken over samples (a minibatch) from empirical distribution $p ( Y , \\delta , X , A )$ . ", + "bbox": [ + 173, + 886, + 826, + 928 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Informative censoring We model informative censoring similar to (7) but mirroring the censoring indicators to encourage accurate censoring time samples $c _ { a }$ for $\\delta = 0$ , while penalizing $c _ { a }$ for being smaller than $y$ for $\\delta \\ : = \\ : 1$ (observed events). Specifically, we set an independent source of randomness like in (6) but parameterized by $\\{ U _ { \\nu } , W _ { \\nu } , b _ { \\nu } \\}$ and censoring generative functions $\\nu _ { A } ( r , \\tilde { \\epsilon } _ { \\nu } )$ , parameterized as neural networks, where $c _ { a } \\sim p _ { \\nu , \\Phi } ( C _ { A } | X = x )$ formulated as ", + "bbox": [ + 173, + 102, + 825, + 174 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/30f94ae655fc0f98b5a994dfdb2c983552b9b03f08d4eae0320432c33861540e.jpg", + "text": "$$\n\\ell _ { c } ( \\nu , \\Phi ) = \\mathbb { E } _ { ( y , \\delta , x , a ) \\sim p ( y , \\delta , X , A ) , \\epsilon \\sim p ( \\epsilon ) } \\left[ ( 1 - \\delta ) \\cdot ( | y - c _ { a } | ) + \\delta \\cdot ( \\operatorname* { m a x } ( 0 , y - c _ { a } ) ) \\right] .\n$$", + "text_format": "latex", + "bbox": [ + 220, + 180, + 777, + 199 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Further, we introduce an additional time-order-consistency loss that enforces the correct order of the observed time relative to the censoring indicator, i.e., $c _ { a } < t _ { a }$ if $\\delta = 0$ and $t \\sb a < c \\sb a$ if $\\delta = 1$ , thus ", + "bbox": [ + 171, + 204, + 823, + 233 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/5a8c95490dd905c65beac4f71a501b50581c5ec62609efa122fed7c912a3ba02.jpg", + "text": "$$\n\\ell _ { \\mathrm { T C } } ( h , \\nu , \\Phi ) = \\mathbb { E } _ { ( \\delta , x , a ) \\sim p ( \\delta , X , A ) , \\epsilon \\sim p ( \\epsilon ) } \\left[ \\delta \\cdot ( \\operatorname* { m a x } ( 0 , t _ { a } - c _ { a } ) ) + ( 1 - \\delta ) \\cdot ( \\operatorname* { m a x } ( 0 , c _ { a } - t _ { a } ) ) \\right] .\n$$", + "text_format": "latex", + "bbox": [ + 178, + 238, + 789, + 257 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Note that $\\ell _ { \\mathrm { T C } } ( h , \\nu , \\Phi )$ does not depend on the observed event times but only on the censoring indicators. Finally, we write the consolidated CSA loss for informative censoring (CSA-INFO) by aggregating (7), (8) and (9) as $\\mathcal { L } _ { \\mathrm { F } } ^ { \\mathrm { C S A - I N F O } } \\triangleq \\mathcal { L } _ { \\mathrm { F } } ^ { \\mathrm { C S A } } + \\ell _ { \\mathrm { c } } + \\ell _ { \\mathrm { T C } }$ . ", + "bbox": [ + 174, + 263, + 825, + 309 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Learning Model functions $\\{ h _ { A } ( \\cdot ) , \\Phi ( \\cdot ) , \\nu _ { A } ( \\cdot ) \\}$ are learned by minimizing the bound (5), via stochastic gradient descent on minibatches from $\\mathcal { D }$ , with $\\mathcal { L } _ { \\mathrm { F } } ^ { \\mathrm { C S A } }$ for non-informative censoring and $\\mathcal { L } _ { \\mathrm { F } } ^ { \\mathrm { C S A - I N F O } }$ for informative censoring. Further, for the IPM regularization loss in (5), we optimize the dual formulation of the Wasserstein distance, via the regularized optimal transport (Villani, 2008; Cuturi, 2013). Consequently, we only require $\\alpha ^ { - 1 } \\zeta _ { h , \\Phi } ( \\bar { x , } a )$ to be 1-Lipschitz (Shalit et al., 2017) and $\\alpha$ is selected by grid search on the validation set using only factual data (details below). ", + "bbox": [ + 173, + 315, + 826, + 402 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4 METRICS ", + "text_level": 1, + "bbox": [ + 174, + 422, + 284, + 438 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "We propose a comprehensive evaluation approach that accounts for both factual and causal metrics. Factual survival outcome predictions are evaluated according to standard survival metrics that measure diverse performance characteristics, such as concordance index (C-Index) (Harrell Jr et al., 1984), mean coefficient of variation (COV) and calibration slope (C-slope) (Chapfuwa et al., 2020). See the SM for more details on these metrics. For causal metrics, defined below, we introduce a nonparametric hazard ratio (HR) between treatment outcomes, and adopt the conventional precision in estimation of heterogeneous effect (PEHE) and average treatment effect (ATE) performance metrics (Hill, 2011). Note that PEHE and ATE require ground truth counterfactual event times, which is only possible in (semi-)synthetic data. For HR, we compare our findings with those independently reported in the literature from gold-standard RCT data. ", + "bbox": [ + 173, + 443, + 826, + 582 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Nonparametric Hazard Ratio In a medical setting, the population hazard ratio $\\mathrm { H R } ( t )$ between treatment groups is considered informative thus has been widely used in drug development and RCT (Yusuf et al., 2016; Mihaylova et al., 2012). For example, $\\mathrm { H R } ( t ) < 1 , > 1$ , or $\\approx 1$ indicate population positive, negative and neutral treatment effects at time $t$ , respectively. Moreover, $\\mathrm { H R } ( t )$ naturally accounts for both censored and non-censored outcomes. Standard approaches for computing $\\mathrm { H R } ( { \\dot { t } } )$ rely on the restrictive proportional hazard assumption from CoxPH (Cox, 1972), which is constituted as a semi-parametric linear model $\\lambda ( t | a ) = \\lambda _ { \\mathrm { b } } \\bar { ( } t ) \\exp ( a \\beta )$ . However, the constant covariate (time independent) effect is often violated in practice (see Figure 2b). For CoxPH, the marginal HR between treatment and control can be obtained from regression coefficient $\\beta$ learned via maximum likelihood without the need for specifying the baseline hazard $\\begin{array} { r } { \\mathrm { ~ \\cdot ~ } _ { \\mathrm { b } } ( t ) \\colon \\mathrm { H R } _ { \\mathrm { C o x P H } } ( t ) = \\frac { \\lambda ( t | a = 1 ) } { \\lambda ( t | a = 0 ) } = \\exp ( \\beta ) } \\end{array}$ . So motivated, we propose a nonparametric, model-free approach for computing $\\mathrm { H R } ( t )$ , in which we do not assume a parametric form for the event time distribution or the proportional hazard assumption from CoxPH. This approach only relies on samples from the conditional event time density functions, $f ( t _ { 1 } | x )$ and $f ( t _ { 0 } | x )$ , via $t _ { a } = h _ { A } ( \\cdot )$ from (6). ", + "bbox": [ + 173, + 587, + 826, + 789 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Definition 1 We define the nonparametric marginal Hazard Ratio and its approximation, $\\mathrm { \\hat { H R } } ( t )$ , as ", + "bbox": [ + 171, + 794, + 823, + 810 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/d9ca3513b198afb6432afe6930b65739952ee5cac99ad41dbc5cf652e4ac4216.jpg", + "text": "$$\n\\mathrm { H R } ( t ) = \\frac { \\lambda _ { 1 } ( t ) } { \\lambda _ { 0 } ( t ) } = \\frac { S _ { 0 } ( t ) } { S _ { 1 } ( t ) } \\cdot \\frac { S _ { 1 } ^ { \\prime } ( t ) } { S _ { 0 } ^ { \\prime } ( t ) } , \\quad \\mathrm { H R } ( t ) = \\frac { \\hat { S } _ { 0 } ^ { \\mathrm { P K M } } ( t ) } { \\hat { S } _ { 1 } ^ { \\mathrm { P K M } } ( t ) } \\cdot \\frac { m _ { 1 } ( t ) } { m _ { 0 } ( t ) } ,\n$$", + "text_format": "latex", + "bbox": [ + 279, + 816, + 717, + 854 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "where for $\\mathrm { H R } ( t )$ we leveraged (1) to obtain (10) and $S ^ { \\prime } ( t ) \\triangleq d S ( t ) / d t$ . The nonparametric assumption for $S ( t )$ makes the computation of $S ^ { \\prime } ( t )$ challenging. Provided that $S ( t )$ is a monotonically decreasing function, for simplicity, we fit a linear function $S ( t ) = m \\cdot t + c .$ , and set $S ^ { \\prime } ( t ) \\approx m$ . Note that the linear model is only used for estimating $S ^ { \\prime } ( t )$ from the nonparametric estimation of $S ( t )$ . ", + "bbox": [ + 174, + 866, + 825, + 924 + ], + "page_idx": 5 + }, + { + "type": "table", + "img_path": "images/5d3d824a74b40f44971099aa53bc084369e33021afa3cae740ca983e4448a9de.jpg", + "table_caption": [ + "Table 1: Performance comparisons on ACTG-SYNTHETIC data, with $9 5 \\%$ $\\mathrm { H R } ( t )$ confidence interval. The ground truth, test set, hazard ratio is $\\mathrm { H R ( t ) } = 0 . 5 2 _ { ( 0 . 3 9 , 0 . 7 1 ) }$ . " + ], + "table_footnote": [], + "table_body": "
MethodCausalFactualC-Slope (A=0,A=1)
EPEHEEATEHR(t)C-Index (A=0,A=1)Mean COV
CoxPH-UniformNANA0.97(0.86,1.09)NANANA
CoxPH-IPWNANA0.48(0.03,7.21)NANANA
CoxPH-OWNANA0.60(0.53,0.68)NANANA
Surv-BART352.0777.890.0(0.0,0.0)(0.706,0.686)0.001(0.398,0)
AFT-Weibull367.92133.930.47(0.47,0.47)(0.21,0.267)6.209(0.707,0.729)
AFT-log-Normal377.76157.640.47(0.47,0.47)(0.675, 0.556)6.971(0.707,0.729)
SR369.4788.550.38(0.330.65)(0.791,0.744)0(0.985,1.027)
CSA (proposed)358.720.80.45(0.39,0.65)(0.787,0.767)0.131(0.985,1.026)
CSA-INFO (proposed)344.331.190.53(0.41,0.67)(0.78,0.764)0.13(0.999,1.029)
", + "bbox": [ + 176, + 133, + 821, + 265 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Bias from $S ^ { \\prime } ( t )$ can be reduced by considering more complex function approximations for $S ( t )$ , e.g., polynomial or spline. For the nonparametric estimation of $S ( t )$ we leverage the model-free population point-estimate-based nonparametric Kaplan-Meier (Kaplan & Meier, 1958) estimator of the survival function $\\hat { S } ^ { \\mathrm { P K M } } ( t )$ in Chapfuwa et al. (2020) to marginalize both factual and counterfactual predictions given covariates $x$ . The approximated hazard ratio, $\\mathrm { \\hat { H R } } ( t )$ , is thus obtained by combining the approximations $\\hat { S } _ { a } ^ { \\mathrm { P K M } } ( t )$ and $m _ { a }$ . A similar formulation for the conditional, $\\mathrm { H } \\mathrm { \\hat { R } } ( t | x )$ , can also be derived. See the SM for full details on the evaluation or $\\mathrm { \\hat { H R } } ( t )$ and $\\mathrm { H } \\mathrm { \\hat { R } } ( t | x )$ . Note that for some AFT- or CoxPH-based parametric formulations, $\\mathrm { H R } ( t | x )$ , can be readily evaluated because $f ( t _ { a } | x )$ and $S ( t _ { a } | x )$ are available in closed form. ", + "bbox": [ + 173, + 276, + 826, + 412 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "In the experiments, we will use $\\mathrm { H R } ( t )$ to compare different approaches against results reported in RCTs (see Tables 1 and 3). Further, we will use $\\mathrm { H R } ( t | x )$ to illustrate stratified treatment effects (see Figure 2). Note that though a neural-based survival recommender system (Katzman et al., 2018) has been previously used to estimate $\\operatorname { H R } ( t | x )$ , their approach does not account for confounding or informative censoring thus it is susceptible to bias. ", + "bbox": [ + 174, + 417, + 825, + 489 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Precision in Estimation of Heterogeneous Effect (PEHE) A general individualized estimation error is formulated as $\\epsilon _ { \\mathrm { P E H E } } = \\sqrt { \\mathbb { E } _ { X } [ ( \\mathrm { I T E } ( x ) - \\mathrm { I } \\hat { \\mathrm { T E } } ( x ) ) ^ { 2 } ] }$ , where $\\operatorname { I T E } ( x )$ is the ground truth, $\\mathrm { I } \\hat { \\mathrm { T E } } ( x ) = \\mathbb { E } _ { T } \\left[ \\gamma \\left( T _ { 1 } \\right) - \\gamma \\left( T _ { 0 } \\right) \\vert X = x \\right]$ and $\\gamma ( \\cdot )$ is a deterministic transformation. In our experiments, $\\gamma ( \\cdot )$ is the average over samples from $\\dot { t _ { a } } \\sim p _ { h , \\Phi } ( T _ { A } | X = x )$ . Alternative estimands, e.g., thresholding survival times $\\gamma ( T _ { A } ) = I \\{ T _ { A } > \\tau \\}$ , can also be considered as described above. ", + "bbox": [ + 173, + 497, + 826, + 582 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Average Treatment Effect (ATE) The population treatment effect estimation error is defined as $\\epsilon _ { \\mathrm { A T E } } = | \\mathrm { A T E } - \\mathrm { A \\hat { T } E } |$ , where $\\mathrm { A T E } = \\mathbb { E } _ { X } [ \\mathrm { I T E } ( x ) ]$ (ground truth) and $\\mathrm { A } \\hat { \\mathrm { T E } } = \\mathbb { E } _ { X } [ \\hat { \\mathrm { I T E } } ( x ) ]$ . ", + "bbox": [ + 173, + 590, + 825, + 622 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "5 EXPERIMENTS ", + "text_level": 1, + "bbox": [ + 176, + 636, + 328, + 651 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "We describe the baselines and datasets that will be used to evaluate the proposed counterfactual survival analysis methods (CSA and CSA-INFO). Pytorch code including the new semi-synthetic dataset (see below) will be made publicly available. Throughout the experiments, we use the standard $\\mathrm { H R } ( t )$ for CoxPH based methods and (10) for all others. The bound in (5) is sensitive to $\\alpha$ , thus we propose approximating proxy counterfactual outcomes $\\{ Y _ { \\mathrm { C F } } , \\delta _ { \\mathrm { C F } } \\}$ for the validation set, according to the covariate Euclidean nearest-neighbour (NN) from the training set. We select the $\\alpha$ that minimizes the validation loss $\\mathcal { L } = \\mathcal { L } _ { \\mathrm { F } } + \\mathcal { L } _ { \\mathrm { C F } }$ from the set $( 0 , 0 . 1 , 1 , 1 0 , 1 0 0 )$ . ", + "bbox": [ + 173, + 660, + 825, + 760 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Baselines We consider the following competitive baseline approaches: $( i )$ propensity weighted CoxPH (Schemper et al., 2009; Buchanan et al., 2014; Rosenbaum & Rubin, 1983); $( i i )$ IPM (5) regularized AFT (log-Normal and Weibull) models; $( i i i )$ an IPM (5) regularized deterministic semisupervised regression (SR) model with accuracy objective from (Chapfuwa et al., 2018), as a contrast for the proposed stochastic predictors (CSA and CSA-INFO); and $( i v )$ survival Bayesian additive regression trees (Surv-BART) (Sparapani et al., 2016). For CoxPH, we consider three normalized weigh2009), $\\begin{array} { r } { \\mathrm { I P W } _ { i } = \\frac { a _ { i } } { \\hat { e } _ { i } } + \\frac { 1 - a _ { i } } { 1 - \\hat { e } _ { i } } ; i i ) } \\end{array}$ $( i )$ weilappi ng (IPW) (Horvitz & Thompson, 1952; Cao et al.,weights (OW) (Crump et al., 2006; Li et al., 2018), $\\mathrm { O W } _ { i } = a _ { i } \\cdot \\left( 1 - \\hat { e } _ { i } \\right) + \\left( 1 - a _ { i } \\right) \\cdot \\hat { e } _ { i }$ $i i i$ \nlinear logistic model $\\hat { e } _ { i } = \\sigma ( x _ { i } ; w )$ , is used as an approximation, $\\boldsymbol { \\hat { e } } _ { i }$ , to the unknown propensity score $P ( A = { \\bar { 1 } } | X = x )$ . See the SM for a details of the baselines. ", + "bbox": [ + 173, + 763, + 826, + 921 + ], + "page_idx": 6 + }, + { + "type": "table", + "img_path": "images/02f7c367f232f20c3ad47f3b482bcff2a58309728cb6b371a98f07c24827d815.jpg", + "table_caption": [ + "Table 3: Performance comparisons on FRAMINGHAM data, with $9 5 \\%$ $\\mathrm { H R } ( t )$ confidence interval. Test set NN assignment of $y _ { \\mathrm { C F } }$ and $\\delta _ { \\mathrm { C F } }$ yields biased $\\mathrm { H R ( t ) } = 1 . 2 3 _ { ( 1 . 1 7 , 1 . 2 5 ) }$ , while previous large scale longitudinal RCT studies estimated $\\mathrm { H R } ( \\mathrm { t } ) = 0 . 7 5$ (0.64,0.88) (Yusuf et al., 2016). " + ], + "table_footnote": [], + "table_body": "
MethodCausal HR(t)C-Index (A=0, A=1)Factual Mean COVC-Slope (A=0,A=1)
CoxPH-Uniform1.69(1.38,2.07)NANANA
CoxPH-IPW1.09(0.76,1.57)NANANA
CoxPH-OW0.88(0.73,1.08)NANANA
Surv-BART14.99(14.9,14.9e8)(0.629,0.630)0.003(0.232,0.084)
AFT-Weibull1.09(1.09,1.09)(0.734,0.395)8.609(0.857,0.89)
AFT-log-Normal1.55(1.46,1.55)(0.68,0.56)10.415(0.979,0.732)
SR0.58(0.53,0.71)(0.601, 0.57)0(0.491,0.63)
CSA(proposed)1.04(1.00,1.09)(0.763,0.728)0.161(0.891,0.81)
CSA-INFO (proposed)0.81(0.77,0.83)(0.752,0.651)0.156(0.907, 0.881)
", + "bbox": [ + 222, + 148, + 772, + 276 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Datasets We consider the following datasets: (i) FRAMINGHAM, is an EHR-based longitudinal cardiovascular cohort study that we use to evaluate the effect of statins on future coronary heart disease outcomes (Benjamin et al., 1994); $( i i )$ ACTG, is a longitudinal RCT study comparing monotherapy with Zidovudine or Didanosine with combination therapy in HIV patients (Hammer et al., 1996); and $( i i i )$ ACTG", + "bbox": [ + 174, + 289, + 517, + 400 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/662567b84ec6b8412fd1a2eb1f6e306f029c8bde26af3d28de506fa5a1d065e3.jpg", + "table_caption": [ + "Table 2: Summary statistics of the datasets. " + ], + "table_footnote": [], + "table_body": "
FRAMINGHAMACTGACTG-SYNTHETIC
Events (%)26.026.948.9
Treatment (%)10.449.555.9
N3,4351,0542,139
p322323
Missing (%)0.231.411.38
tmax (days)7,2791,2311,313
", + "bbox": [ + 532, + 303, + 820, + 375 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "SYNTHETIC, is a semi-synthetic dataset based on ACTG covariates. We simulate potential outcomes according to a Gompertz-Cox distribution (Bender et al., 2005) with selection bias from a simple logistic model for $\\bar { P ( A = 1 | X = x ) }$ and AFT-based censoring mechanism. The generative process is detailed in the SM. Table 2 summarizes the datasets according to $( i )$ covariates of size $p$ ; $( i i )$ proportion of non-censored events, treated units, and missing entries in the $N \\times p$ covariate matrix; and $( i i i )$ time range $t _ { \\mathrm { m a x } }$ for both censored and non-censored events. Missing entries are imputed with median or mode if continuous or categorical, respectively. ", + "bbox": [ + 174, + 400, + 825, + 497 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Quantitative Results Experimental results for two data-sets in Tables 1 and 3, illustrate that AFTbased methods are high variance, inferior in calibration and C-Index than accuracy-based methods (SR, CSA, CSA-INFO). Surv-BART is the least calibrated but low variance method. CSA-INFO and CSA outperform all methods across all factual metrics, whereas CSA-INFO is better calibrated, low variance but slightly lower C-Index than CSA. Note that we fit CoxPH using the entire dataset; since it does not support counterfactual inference, we do not present factual metrics. By properly adjusting for both informative censoring and selection bias, CSA-INFO significantly outperforms all methods in treatment effect estimation according to $\\mathrm { H R } ( t )$ and $\\epsilon _ { \\mathrm { P E H E } }$ , across non-RCT datasets, while remaining comparable to AFT-Weibull on the RCT dataset (see the SM). Further, RCT-based results on ACTG data in the SM illustrate comparable $\\mathrm { H R } ( t )$ across all models except for AFT-log-Normal and Surv-BART, which overestimate, and SR, which underestimates risk. For non-RCT datasets (ACTG-SYNTHENTIC and FRAMINGHAM), CoxPH-OW has a clear advantage over all CoxPH based methods, mostly credited to the well-behaved bounded propensity weights $\\in [ 0 , 1 ]$ . Interestingly, the FRAMINGHAM observational data exhibits a common paradox, where without proper adjustment of selection and censoring bias, naive approaches would result in a counter-intuitive treatment effect from statins. However, there is severe confounding from covariates such as age, BMI, diabetes, CAD, PAD, MI, stroke, etc., that influence both treatment likelihood and survival time. Table 3, demonstrates that CSA-INFO is clearly the best performing approach. Specifically, its $\\mathrm { H R } ( t )$ , reverses the biased observational treatment effect, to demonstrate positive treatment from statins, which is consistent with prior large RCT longitudinal findings (Yusuf et al., 2016). ", + "bbox": [ + 173, + 503, + 825, + 781 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Qualitative Results Figure 2a demonstrates that CSA-INFO matches the ground truth population hazard, $\\mathrm { H R } ( t )$ , better than alternative methods on ACTG-SYNTHETIC data. See the SM for ACTG and FRAMINGHAM. Figure 2b shows sub-population log hazard ratios for four patient clusters obtained via hierarchical clustering on the individual log hazard ratios, $\\log \\mathrm { H R } ( t | x )$ , of the test set of FRAMINGHAM data. Interestingly, these clusters stratify treatment effects into: positive (2), negative (1 and 3), and neutral (4) sub-populations. Moreover, the estimated density of median $\\log \\mathrm { H } \\bar { \\mathrm { R } } ( t | x )$ values in Figure $_ { 2 \\mathrm { c } }$ illustrates that nearly $7 0 \\%$ of the testing set individuals have $\\log \\mathrm { H R } ( t | x ) < 0$ , thus may benefit from taking statins. Further, we isolated the extreme top and bottom quantiles, $\\mathrm { H R } ( t | x ) < 0 . 0 2 4$ and $\\mathrm { H R } ( \\bar { t } | x ) > 1 . 9 1 6$ , respectively, of the median $\\log \\mathrm { \\bar { H R } } ( t | x )$ values for the test set of FRAMINGHAM, as shown in Figure 2c. After comparing their covariates, we found that individuals with the following characteristics may benefit from taking statins: young, male, diabetic, without prior history (CAD, PAD, stroke or MI), high BMI, cholesterol, triglycerides, fasting glucose, and low high-density lipoprotein. There seem to be consensus that diabetics and high-cholesterol patients benefit from statins (Cheung et al., 2004; Wilt et al., 2004). See SM for additional results. ", + "bbox": [ + 173, + 797, + 825, + 924 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/4c89c16905f48a1bdcc35b1ade4f0f866e4e151f0c05e3f1b2a559591a62f777.jpg", + "image_caption": [ + "Figure 2: (a) Inferred population $\\mathrm { H R } ( t )$ compared against ground truth (EMP) on ACTG-SYNTHETIC data. CSA-INFO-based (b) cluster-specific average l $\\operatorname { \\Pi } ^ { } \\operatorname { \\Pi } ^ { } \\operatorname { \\mathrm { ) g } } \\operatorname { \\mathrm { { H R } } } ( t | x )$ curves and (c) estimated density of median $\\log \\mathrm { H R } ( t | x )$ values on the test set of the FRAMINGHAM dataset. Clusters assignment were obtained via hierarchical clustering of individualized $\\log \\mathrm { H R } ( t | x )$ traces. " + ], + "image_footnote": [], + "bbox": [ + 187, + 94, + 799, + 223 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 299, + 826, + 369 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "6 CONCLUSIONS ", + "text_level": 1, + "bbox": [ + 174, + 380, + 330, + 395 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "We have proposed a unified counterfactual inference framework for survival analysis. Our approach adjusts for bias from two unknown sources, namely, confounding due to covariate dependent selection bias and censoring (informative or non-informative). Relative to competitive alternatives, we demonstrate superior performance for both survival-outcome prediction and treatment-effect estimation, across three diverse datasets, including a semi-synthetic dataset which we introduce. Moreover, we formulate a model-free nonparametric hazard ratio metric for comparing treatment effects or leveraging prior randomized real-world experiments in longitudinal studies. ", + "bbox": [ + 173, + 400, + 826, + 498 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "REFERENCES ", + "text_level": 1, + "bbox": [ + 174, + 102, + 287, + 117 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Peter C Austin. Propensity-score matching in the cardiovascular surgery literature from 2004 to 2006: a systematic review and suggestions for improvement. The Journal of Thoracic and Cardiovascular Surgery, 2007. ", + "bbox": [ + 174, + 126, + 826, + 167 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Peter C Austin. The use of propensity score methods with survival or time-to-event outcomes: reporting measures of effect similar to those used in randomized experiments. Statistics in Medicine, 2014. ", + "bbox": [ + 173, + 179, + 823, + 220 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Elias Bareinboim and Judea Pearl. Controlling selection bias in causal inference. In AISTATS, 2012. ", + "bbox": [ + 173, + 231, + 823, + 247 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Ralf Bender, Thomas Augustin, and Maria Blettner. Generating survival times to simulate cox proportional hazards models. Statistics in medicine, 2005. ", + "bbox": [ + 176, + 256, + 821, + 286 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Emelia J Benjamin, Daniel Levy, Sonya M Vaziri, Ralph B D’Agostino, Albert J Belanger, and Philip A Wolf. Independent risk factors for atrial fibrillation in a population-based cohort: the framingham heart study. Jama, 1994. ", + "bbox": [ + 174, + 295, + 825, + 338 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Ashley L Buchanan, Michael G Hudgens, Stephen R Cole, Bryan Lau, Adaora A Adimora, and Women’s Interagency HIV Study. Worth the weight: using inverse probability weighted cox models in aids research. AIDS research and human retroviruses, 2014. ", + "bbox": [ + 174, + 348, + 826, + 391 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Weihua Cao, Anastasios A Tsiatis, and Marie Davidian. Improving efficiency and robustness of the doubly robust estimator for a population mean with incomplete data. Biometrika, 2009. ", + "bbox": [ + 173, + 400, + 825, + 430 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "P. Chapfuwa, C. Tao, C. Li, I. Khan, K. J. Chandross, M. J. Pencina, L. Carin, and R. Henao. Calibration and uncertainty in neural time-to-event modeling. IEEE Transactions on Neural Networks and Learning Systems, 2020. ", + "bbox": [ + 179, + 439, + 825, + 482 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Paidamoyo Chapfuwa, Chenyang Tao, Chunyuan Li, Courtney Page, Benjamin Goldstein, Lawrence Carin, and Ricardo Henao. Adversarial time-to-event modeling. In ICML, 2018. ", + "bbox": [ + 171, + 492, + 823, + 522 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Bernard MY Cheung, Ian J Lauder, Chu-Pak Lau, and Cyrus R Kumana. Meta-analysis of large randomized controlled trials to evaluate the impact of statins on cardiovascular outcomes. British journal of clinical pharmacology, 2004. ", + "bbox": [ + 174, + 531, + 825, + 574 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Hugh A Chipman, Edward I George, Robert E McCulloch, et al. Bart: Bayesian additive regression trees. The Annals of Applied Statistics, 2010. ", + "bbox": [ + 173, + 583, + 823, + 613 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Stephen R Cole and Miguel A Hernán. Adjusted survival curves with inverse probability weights. Computer methods and programs in biomedicine, 2004. ", + "bbox": [ + 173, + 622, + 825, + 652 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "David R Cox. Regression models and life-tables. Journal of the Royal Statistical Society: Series B (Methodological), 1972. ", + "bbox": [ + 173, + 661, + 825, + 690 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Richard K Crump, V Joseph Hotz, Guido W Imbens, and Oscar A Mitnik. Moving the goalposts: Addressing limited overlap in the estimation of average treatment effects by changing the estimand. Technical report, National Bureau of Economic Research, 2006. ", + "bbox": [ + 173, + 700, + 826, + 743 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Yifan Cui, Michael R Kosorok, Stefan Wager, and Ruoqing Zhu. Estimating heterogeneous treatment effects with right-censored data via causal survival forests. arXiv, 2020. ", + "bbox": [ + 171, + 753, + 823, + 782 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Marco Cuturi. Sinkhorn distances: Lightspeed computation of optimal transport. In NeurIPS, 2013. ", + "bbox": [ + 173, + 792, + 825, + 808 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Iván Díaz. Statistical inference for data-adaptive doubly robust estimators with survival outcomes. Statistics in Medicine, 2019. ", + "bbox": [ + 171, + 818, + 825, + 845 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Jennifer Frankovich, Christopher A Longhurst, and Scott M Sutherland. Evidence-based medicine in the emr era. N Engl J Med, 2011. ", + "bbox": [ + 169, + 856, + 823, + 886 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Saurabh Gombar, Alison Callahan, Robert Califf, Robert Harrington, and Nigam H Shah. It is time to learn from patients like mine. NPJ digital medicine, 2019. ", + "bbox": [ + 174, + 895, + 821, + 924 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Scott M Hammer, David A Katzenstein, Michael D Hughes, Holly Gundacker, Robert T Schooley, Richard H Haubrich, W Keith Henry, Michael M Lederman, John P Phair, Manette Niu, et al. A trial comparing nucleoside monotherapy with combination therapy in hiv-infected adults with cd4 cell counts from 200 to 500 per cubic millimeter. New England Journal of Medicine, 1996. ", + "bbox": [ + 176, + 103, + 825, + 160 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Frank E Harrell Jr, Kerry L Lee, Robert M Califf, David B Pryor, and Robert A Rosati. Regression modelling strategies for improved prognostic prediction. Statistics in medicine, 1984. ", + "bbox": [ + 174, + 170, + 823, + 198 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Kristiina Häyrinen, Kaija Saranto, and Pirkko Nykänen. Definition, structure, content, use and impacts of electronic health records: a review of the research literature. International Journal of Medical Informatics, 2008. ", + "bbox": [ + 174, + 208, + 826, + 251 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Nicholas C Henderson, Thomas A Louis, Gary L Rosner, and Ravi Varadhan. Individualized treatment effects with censored data via fully nonparametric bayesian accelerated failure time models. Biostatistics, 2020. ", + "bbox": [ + 173, + 260, + 826, + 303 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Miguel A Hernán and James M Robins. Causal inference: what if. Boca Raton: Chapman & Hill/CRC, 2020. ", + "bbox": [ + 173, + 313, + 823, + 342 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Miguel A Hernán, Stephen R Cole, Joseph Margolick, Mardge Cohen, and James M Robins. Structural accelerated failure time models for survival analysis in studies with time-varying treatments. Pharmacoepidemiology and Drug Safety, 2005. ", + "bbox": [ + 174, + 352, + 823, + 395 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Miguel Ángel Hernán, Babette Brumback, and James M Robins. Marginal structural models to estimate the causal effect of zidovudine on the survival of hiv-positive men. Epidemiology, 2000. ", + "bbox": [ + 171, + 405, + 825, + 434 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Jennifer L Hill. Bayesian nonparametric modeling for causal inference. Journal of Computational and Graphical Statistics, 2011. ", + "bbox": [ + 171, + 443, + 823, + 472 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Daniel G Horvitz and Donovan J Thompson. A generalization of sampling without replacement from a finite universe. Journal of the American statistical Association, 1952. ", + "bbox": [ + 174, + 482, + 823, + 511 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Liangyuan Hu, Jiayi Ji, and Fan Li. Estimating heterogeneous survival treatment effect in observational data using machine learning. arXiv, 2020. ", + "bbox": [ + 173, + 520, + 823, + 550 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Hemant Ishwaran, Udaya B Kogalur, Eugene H Blackstone, Michael S Lauer, et al. Random survival forests. The annals of applied statistics, 2008. ", + "bbox": [ + 173, + 559, + 823, + 588 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Ashish K Jha, Catherine M DesRoches, Eric G Campbell, Karen Donelan, Sowmya R Rao, Timothy G Ferris, Alexandra Shields, Sara Rosenbaum, and David Blumenthal. Use of electronic health records in us hospitals. New England Journal of Medicine, 2009. ", + "bbox": [ + 174, + 597, + 825, + 640 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Edward L Kaplan and Paul Meier. Nonparametric estimation from incomplete observations. Journal of the American statistical association, 1958. ", + "bbox": [ + 173, + 650, + 825, + 679 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Jared L Katzman, Uri Shaham, Alexander Cloninger, Jonathan Bates, Tingting Jiang, and Yuval Kluger. Deepsurv: personalized treatment recommender system using a cox proportional hazards deep neural network. BMC medical research methodology, 2018. ", + "bbox": [ + 174, + 689, + 825, + 732 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "David G Kleinbaum and Mitchel Klein. Survival analysis. Springer, 2010. ", + "bbox": [ + 174, + 741, + 663, + 757 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Fan Li, Kari Lock Morgan, and Alan M Zaslavsky. Balancing covariates via propensity score weighting. Journal of the American Statistical Association, 2018. ", + "bbox": [ + 173, + 765, + 823, + 795 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Christopher A Longhurst, Robert A Harrington, and Nigam H Shah. A ‘green button’for using aggregate patient data at the point of care. Health affairs, 2014. ", + "bbox": [ + 173, + 804, + 825, + 833 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "B Mihaylova, J Emberson, L Blackwell, A Keech, J Simes, EH Barnes, M Voysey, 3A Gray, R Collins, and C Baigent. The effects of lowering ldl cholesterol with statin therapy in people at low risk of vascular disease: meta-analysis of individual data from 27 randomised trials., 2012. ", + "bbox": [ + 178, + 842, + 823, + 886 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Xenia Miscouridou, Adler Perotte, Noémie Elhadad, and Rajesh Ranganath. Deep survival analysis: Nonparametrics and missingness. In Machine Learning for Healthcare Conference, 2018. ", + "bbox": [ + 176, + 895, + 823, + 924 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Alfred Müller. Integral probability metrics and their generating classes of functions. Advances in Applied Probability, 1997. ", + "bbox": [ + 171, + 103, + 825, + 132 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Judea Pearl and Elias Bareinboim. External validity: From do-calculus to transportability across populations. Statistical Science, 2014. ", + "bbox": [ + 173, + 140, + 823, + 170 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Danilo Jimenez Rezende and Shakir Mohamed. Variational inference with normalizing flows. In ICML, 2015. ", + "bbox": [ + 173, + 178, + 823, + 207 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "James Robins. A new approach to causal inference in mortality studies with a sustained exposure period—application to control of the healthy worker survivor effect. Mathematical modelling, 1986. ", + "bbox": [ + 174, + 215, + 825, + 257 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Paul R Rosenbaum and Donald B Rubin. The central role of the propensity score in observational studies for causal effects. Biometrika, 1983. ", + "bbox": [ + 173, + 266, + 825, + 295 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Donald B Rubin. Causal inference using potential outcomes. Journal of the American Statistical Association, 2005. ", + "bbox": [ + 174, + 304, + 823, + 333 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Michael Schemper, Samo Wakounig, and Georg Heinze. The estimation of average hazard ratios by weighted cox regression. Statistics in medicine, 2009. ", + "bbox": [ + 173, + 340, + 825, + 369 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Uri Shalit, Fredrik D Johansson, and David Sontag. Estimating individual treatment effect: generalization bounds and algorithms. In ICML, 2017. ", + "bbox": [ + 171, + 378, + 825, + 407 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Jincheng Shen, Lu Wang, Stephanie Daignault, Daniel E Spratt, Todd M Morgan, and Jeremy MG Taylor. Estimating the optimal personalized treatment strategy based on selected variables to prolong survival via random survival forest with weighted bootstrap. Journal of biopharmaceutical statistics, 2018. ", + "bbox": [ + 173, + 415, + 826, + 472 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Rodney A Sparapani, Brent R Logan, Robert E McCulloch, and Purushottam W Laud. Nonparametric survival analysis using bayesian additive regression trees (bart). Statistics in medicine, 2016. ", + "bbox": [ + 171, + 479, + 825, + 510 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Bharath K Sriperumbudur, Kenji Fukumizu, Arthur Gretton, Bernhard Schölkopf, Gert RG Lanckriet, et al. On the empirical estimation of integral probability metrics. Electronic Journal of Statistics, 2012. ", + "bbox": [ + 174, + 517, + 825, + 560 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Ludovic Trinquart, Justine Jacot, Sarah C Conner, and Raphaël Porcher. Comparison of treatment effects measured by the hazard ratio and by the ratio of restricted mean survival times in oncology randomized controlled trials. Journal of Clinical Oncology, 2016. ", + "bbox": [ + 173, + 569, + 826, + 612 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Anastasios Tsiatis. Semiparametric theory and missing data. Springer Science & Business Media, 2007. ", + "bbox": [ + 173, + 619, + 823, + 648 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Mark J van der Laan and James M Robins. Unified approach for causal inference and censored data. In Unified Methods for Censored Longitudinal Data and Causality. Springer, 2003. ", + "bbox": [ + 174, + 656, + 823, + 688 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Mark J Van der Laan and Sherri Rose. Targeted learning: causal inference for observational and experimental data. Springer Science & Business Media, 2011. ", + "bbox": [ + 173, + 694, + 825, + 724 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Cédric Villani. Optimal transport: old and new. Springer Science & Business Media, 2008. ", + "bbox": [ + 171, + 731, + 776, + 747 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Stefan Wager and Susan Athey. Estimation and inference of heterogeneous treatment effects using random forests. Journal of the American Statistical Association, 2018. ", + "bbox": [ + 171, + 756, + 823, + 785 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Lee-Jen Wei. The accelerated failure time model: a useful alternative to the cox regression model in survival analysis. Statistics in medicine, 1992. ", + "bbox": [ + 171, + 792, + 823, + 821 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Timothy J Wilt, Hanna E Bloomfield, Roderick MacDonald, David Nelson, Indulis Rutks, Michael Ho, Gregory Larsen, Anthony McCall, Sandra Pineros, and Anne Sales. Effectiveness of statin therapy in adults with coronary heart disease. Archives of internal medicine, 2004. ", + "bbox": [ + 174, + 829, + 825, + 873 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Salim Yusuf, Jackie Bosch, Gilles Dagenais, Jun Zhu, Denis Xavier, Lisheng Liu, Prem Pais, Patricio López-Jaramillo, Lawrence A Leiter, Antonio Dans, et al. Cholesterol lowering in intermediate-risk persons without cardiovascular disease. New England Journal of Medicine, 2016. ", + "bbox": [ + 176, + 881, + 823, + 924 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Yao Zhang, Alexis Bellot, and Mihaela van der Schaar. Learning overlapping representations for the estimation of individualized treatment effects. In AISTATS, 2020. ", + "bbox": [ + 173, + 103, + 823, + 132 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Lihui Zhao, Lu Tian, Hajime Uno, Scott D Solomon, Marc A Pfeffer, Jerald S Schindler, and Lee Jen Wei. Utilizing the integrated difference of two survival functions to quantify the treatment contrast for designing, monitoring, and analyzing a comparative clinical study. Clinical trials, 2012. ", + "bbox": [ + 178, + 141, + 823, + 184 + ], + "page_idx": 12 + } +] \ No newline at end of file diff --git a/parse/train/3ZeGLibhFo0/3ZeGLibhFo0_middle.json b/parse/train/3ZeGLibhFo0/3ZeGLibhFo0_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..a10f3fb4082a2a9d778b2e9b4631a62673fb177f --- /dev/null +++ b/parse/train/3ZeGLibhFo0/3ZeGLibhFo0_middle.json @@ -0,0 +1,42022 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 78, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 78, + 505, + 98 + ], + "spans": [ + { + "bbox": [ + 106, + 78, + 505, + 98 + ], + "score": 1.0, + "content": "ENABLING COUNTERFACTUAL SURVIVAL ANALYSIS", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 99, + 380, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 99, + 380, + 118 + ], + "score": 1.0, + "content": "WITH BALANCED REPRESENTATIONS", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 112, + 135, + 244, + 157 + ], + "lines": [ + { + "bbox": [ + 113, + 135, + 201, + 147 + ], + "spans": [ + { + "bbox": [ + 113, + 135, + 201, + 147 + ], + "score": 1.0, + "content": "Anonymous authors", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 111, + 146, + 245, + 158 + ], + "spans": [ + { + "bbox": [ + 111, + 146, + 245, + 158 + ], + "score": 1.0, + "content": "Paper under double-blind review", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 278, + 186, + 333, + 199 + ], + "lines": [ + { + "bbox": [ + 276, + 185, + 336, + 200 + ], + "spans": [ + { + "bbox": [ + 276, + 185, + 336, + 200 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 143, + 212, + 469, + 365 + ], + "lines": [ + { + "bbox": [ + 141, + 212, + 470, + 224 + ], + "spans": [ + { + "bbox": [ + 141, + 212, + 470, + 224 + ], + "score": 1.0, + "content": "Balanced representation learning methods have been applied successfully to coun-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 141, + 222, + 470, + 235 + ], + "spans": [ + { + "bbox": [ + 141, + 222, + 470, + 235 + ], + "score": 1.0, + "content": "terfactual inference from observational data. However, approaches that account for", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 233, + 470, + 246 + ], + "spans": [ + { + "bbox": [ + 141, + 233, + 470, + 246 + ], + "score": 1.0, + "content": "survival outcomes are relatively limited. Survival data are frequently encountered", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 244, + 470, + 257 + ], + "spans": [ + { + "bbox": [ + 141, + 244, + 470, + 257 + ], + "score": 1.0, + "content": "across diverse medical applications, i.e., drug development, risk profiling, and clini-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 255, + 469, + 268 + ], + "spans": [ + { + "bbox": [ + 141, + 255, + 469, + 268 + ], + "score": 1.0, + "content": "cal trials, and such data are also relevant in fields like manufacturing (for equipment", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 267, + 469, + 279 + ], + "spans": [ + { + "bbox": [ + 141, + 267, + 469, + 279 + ], + "score": 1.0, + "content": "monitoring). When the outcome of interest is time-to-event, special precautions", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 277, + 470, + 290 + ], + "spans": [ + { + "bbox": [ + 141, + 277, + 470, + 290 + ], + "score": 1.0, + "content": "for handling censored events need to be taken, as ignoring censored outcomes may", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 289, + 470, + 301 + ], + "spans": [ + { + "bbox": [ + 141, + 289, + 470, + 301 + ], + "score": 1.0, + "content": "lead to biased estimates. We propose a theoretically grounded unified framework", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 299, + 470, + 312 + ], + "spans": [ + { + "bbox": [ + 141, + 299, + 470, + 312 + ], + "score": 1.0, + "content": "for counterfactual inference applicable to survival outcomes. Further, we formulate", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 311, + 469, + 322 + ], + "spans": [ + { + "bbox": [ + 141, + 311, + 469, + 322 + ], + "score": 1.0, + "content": "a nonparametric hazard ratio metric for evaluating average and individualized", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 321, + 470, + 334 + ], + "spans": [ + { + "bbox": [ + 141, + 321, + 470, + 334 + ], + "score": 1.0, + "content": "treatment effects. Experimental results on real-world and semi-synthetic datasets,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 332, + 469, + 345 + ], + "spans": [ + { + "bbox": [ + 141, + 332, + 469, + 345 + ], + "score": 1.0, + "content": "the latter which we introduce, demonstrate that the proposed approach significantly", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 344, + 469, + 355 + ], + "spans": [ + { + "bbox": [ + 141, + 344, + 469, + 355 + ], + "score": 1.0, + "content": "outperforms competitive alternatives in both survival-outcome predictions and", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 142, + 355, + 254, + 366 + ], + "spans": [ + { + "bbox": [ + 142, + 355, + 254, + 366 + ], + "score": 1.0, + "content": "treatment-effect estimation.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 11.5 + }, + { + "type": "title", + "bbox": [ + 109, + 388, + 206, + 400 + ], + "lines": [ + { + "bbox": [ + 105, + 386, + 208, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 386, + 208, + 402 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 412, + 505, + 512 + ], + "lines": [ + { + "bbox": [ + 106, + 413, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 106, + 413, + 505, + 425 + ], + "score": 1.0, + "content": "Survival analysis or time-to-event studies focus on modeling the time of a future event, such as", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 424, + 506, + 436 + ], + "spans": [ + { + "bbox": [ + 106, + 424, + 506, + 436 + ], + "score": 1.0, + "content": "death or failure, and investigate its relationship with covariates or predictors of interest. Specifically,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 435, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 505, + 447 + ], + "score": 1.0, + "content": "we may be interested in the causal effect of a given intervention or treatment on survival time. A", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 446, + 506, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 506, + 459 + ], + "score": 1.0, + "content": "typical question may be: will a given therapy increase the chances of survival of an individual or", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 456, + 505, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 505, + 470 + ], + "score": 1.0, + "content": "population? Such causal inquiries on survival outcomes are common in the fields of epidemiology", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "and medicine (Robins, 1986; Hammer et al., 1996; Yusuf et al., 2016). As an important current", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 478, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 506, + 492 + ], + "score": 1.0, + "content": "example, the COVID-19 pandemic is creating a demand for methodological development to address", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "score": 1.0, + "content": "such questions, specifically, when evaluating the effectiveness of a potential vaccine or therapeutic", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 500, + 284, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 284, + 514 + ], + "score": 1.0, + "content": "outside randomized controlled trial settings.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 517, + 505, + 649 + ], + "lines": [ + { + "bbox": [ + 105, + 516, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 505, + 530 + ], + "score": 1.0, + "content": "Traditional causal survival analysis is typically carried out in the context of a randomized controlled", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 528, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 106, + 528, + 505, + 541 + ], + "score": 1.0, + "content": "trial (RCT), where the treatment assignment is controlled by researchers. Though they are the gold", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 539, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 505, + 552 + ], + "score": 1.0, + "content": "standard for causal inference, RCTs are usually long-term engagements, expensive and limited in", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 550, + 506, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 506, + 563 + ], + "score": 1.0, + "content": "sample size. Alternatively, the availability of observational data with comprehensive information", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 562, + 505, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 562, + 505, + 573 + ], + "score": 1.0, + "content": "about patients, such as electronic health records (EHRs), constitutes a more accessible but also", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 572, + 506, + 584 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 506, + 584 + ], + "score": 1.0, + "content": "more challenging source for estimating causal effects (Häyrinen et al., 2008; Jha et al., 2009). Such", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 584, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 584, + 505, + 595 + ], + "score": 1.0, + "content": "observational data may be used to augment and verify an RCT, after a particular treatment is approved", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "and in use (Gombar et al., 2019; Frankovich et al., 2011; Longhurst et al., 2014). Moreover, the", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 605, + 506, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 506, + 617 + ], + "score": 1.0, + "content": "wealth of information from observational data also allows for the estimation of the individualized", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "score": 1.0, + "content": "treatment effect (ITE), namely, the causal effect of an intervention at the individual level. In this", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 627, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 505, + 639 + ], + "score": 1.0, + "content": "work, we develop a novel framework for counterfactual time-to-event prediction to estimate the ITE", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 638, + 361, + 649 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 361, + 649 + ], + "score": 1.0, + "content": "for survival or time-to-event outcomes from observational data.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 107, + 655, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "score": 1.0, + "content": "Estimating the causal effect for survival outcomes in observational data manifests two principal", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 664, + 506, + 679 + ], + "spans": [ + { + "bbox": [ + 105, + 664, + 506, + 679 + ], + "score": 1.0, + "content": "challenges. First, the treatment assignment mechanism is not known a priori. Therefore, there may", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "score": 1.0, + "content": "be variables, known as confounders, affecting both the treatment and survival time, which lead to", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "selection bias (Bareinboim & Pearl, 2012), i.e., that the distributions across treatment groups are not", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "the same. In this work, we focus on selection biases due to confounding, but other sources may also", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "be considered. For instance, patients who are severely ill are likely to receive more aggressive therapy,", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 720, + 506, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 506, + 732 + ], + "score": 1.0, + "content": "however, their health status may also inevitably influence survival. Traditional survival analysis", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 44 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 308, + 761 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 308, + 761 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 307, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 78, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 78, + 505, + 98 + ], + "spans": [ + { + "bbox": [ + 106, + 78, + 505, + 98 + ], + "score": 1.0, + "content": "ENABLING COUNTERFACTUAL SURVIVAL ANALYSIS", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 99, + 380, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 99, + 380, + 118 + ], + "score": 1.0, + "content": "WITH BALANCED REPRESENTATIONS", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 112, + 135, + 244, + 157 + ], + "lines": [ + { + "bbox": [ + 113, + 135, + 201, + 147 + ], + "spans": [ + { + "bbox": [ + 113, + 135, + 201, + 147 + ], + "score": 1.0, + "content": "Anonymous authors", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 111, + 146, + 245, + 158 + ], + "spans": [ + { + "bbox": [ + 111, + 146, + 245, + 158 + ], + "score": 1.0, + "content": "Paper under double-blind review", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5, + "bbox_fs": [ + 111, + 135, + 245, + 158 + ] + }, + { + "type": "title", + "bbox": [ + 278, + 186, + 333, + 199 + ], + "lines": [ + { + "bbox": [ + 276, + 185, + 336, + 200 + ], + "spans": [ + { + "bbox": [ + 276, + 185, + 336, + 200 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 143, + 212, + 469, + 365 + ], + "lines": [ + { + "bbox": [ + 141, + 212, + 470, + 224 + ], + "spans": [ + { + "bbox": [ + 141, + 212, + 470, + 224 + ], + "score": 1.0, + "content": "Balanced representation learning methods have been applied successfully to coun-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 141, + 222, + 470, + 235 + ], + "spans": [ + { + "bbox": [ + 141, + 222, + 470, + 235 + ], + "score": 1.0, + "content": "terfactual inference from observational data. However, approaches that account for", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 233, + 470, + 246 + ], + "spans": [ + { + "bbox": [ + 141, + 233, + 470, + 246 + ], + "score": 1.0, + "content": "survival outcomes are relatively limited. Survival data are frequently encountered", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 244, + 470, + 257 + ], + "spans": [ + { + "bbox": [ + 141, + 244, + 470, + 257 + ], + "score": 1.0, + "content": "across diverse medical applications, i.e., drug development, risk profiling, and clini-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 255, + 469, + 268 + ], + "spans": [ + { + "bbox": [ + 141, + 255, + 469, + 268 + ], + "score": 1.0, + "content": "cal trials, and such data are also relevant in fields like manufacturing (for equipment", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 267, + 469, + 279 + ], + "spans": [ + { + "bbox": [ + 141, + 267, + 469, + 279 + ], + "score": 1.0, + "content": "monitoring). When the outcome of interest is time-to-event, special precautions", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 277, + 470, + 290 + ], + "spans": [ + { + "bbox": [ + 141, + 277, + 470, + 290 + ], + "score": 1.0, + "content": "for handling censored events need to be taken, as ignoring censored outcomes may", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 289, + 470, + 301 + ], + "spans": [ + { + "bbox": [ + 141, + 289, + 470, + 301 + ], + "score": 1.0, + "content": "lead to biased estimates. We propose a theoretically grounded unified framework", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 299, + 470, + 312 + ], + "spans": [ + { + "bbox": [ + 141, + 299, + 470, + 312 + ], + "score": 1.0, + "content": "for counterfactual inference applicable to survival outcomes. Further, we formulate", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 311, + 469, + 322 + ], + "spans": [ + { + "bbox": [ + 141, + 311, + 469, + 322 + ], + "score": 1.0, + "content": "a nonparametric hazard ratio metric for evaluating average and individualized", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 321, + 470, + 334 + ], + "spans": [ + { + "bbox": [ + 141, + 321, + 470, + 334 + ], + "score": 1.0, + "content": "treatment effects. Experimental results on real-world and semi-synthetic datasets,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 332, + 469, + 345 + ], + "spans": [ + { + "bbox": [ + 141, + 332, + 469, + 345 + ], + "score": 1.0, + "content": "the latter which we introduce, demonstrate that the proposed approach significantly", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 344, + 469, + 355 + ], + "spans": [ + { + "bbox": [ + 141, + 344, + 469, + 355 + ], + "score": 1.0, + "content": "outperforms competitive alternatives in both survival-outcome predictions and", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 142, + 355, + 254, + 366 + ], + "spans": [ + { + "bbox": [ + 142, + 355, + 254, + 366 + ], + "score": 1.0, + "content": "treatment-effect estimation.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 11.5, + "bbox_fs": [ + 141, + 212, + 470, + 366 + ] + }, + { + "type": "title", + "bbox": [ + 109, + 388, + 206, + 400 + ], + "lines": [ + { + "bbox": [ + 105, + 386, + 208, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 386, + 208, + 402 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 412, + 505, + 512 + ], + "lines": [ + { + "bbox": [ + 106, + 413, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 106, + 413, + 505, + 425 + ], + "score": 1.0, + "content": "Survival analysis or time-to-event studies focus on modeling the time of a future event, such as", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 424, + 506, + 436 + ], + "spans": [ + { + "bbox": [ + 106, + 424, + 506, + 436 + ], + "score": 1.0, + "content": "death or failure, and investigate its relationship with covariates or predictors of interest. Specifically,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 435, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 505, + 447 + ], + "score": 1.0, + "content": "we may be interested in the causal effect of a given intervention or treatment on survival time. A", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 446, + 506, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 506, + 459 + ], + "score": 1.0, + "content": "typical question may be: will a given therapy increase the chances of survival of an individual or", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 456, + 505, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 505, + 470 + ], + "score": 1.0, + "content": "population? Such causal inquiries on survival outcomes are common in the fields of epidemiology", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "and medicine (Robins, 1986; Hammer et al., 1996; Yusuf et al., 2016). As an important current", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 478, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 506, + 492 + ], + "score": 1.0, + "content": "example, the COVID-19 pandemic is creating a demand for methodological development to address", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 505, + 502 + ], + "score": 1.0, + "content": "such questions, specifically, when evaluating the effectiveness of a potential vaccine or therapeutic", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 500, + 284, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 284, + 514 + ], + "score": 1.0, + "content": "outside randomized controlled trial settings.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 413, + 506, + 514 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 517, + 505, + 649 + ], + "lines": [ + { + "bbox": [ + 105, + 516, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 505, + 530 + ], + "score": 1.0, + "content": "Traditional causal survival analysis is typically carried out in the context of a randomized controlled", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 528, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 106, + 528, + 505, + 541 + ], + "score": 1.0, + "content": "trial (RCT), where the treatment assignment is controlled by researchers. Though they are the gold", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 539, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 505, + 552 + ], + "score": 1.0, + "content": "standard for causal inference, RCTs are usually long-term engagements, expensive and limited in", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 550, + 506, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 506, + 563 + ], + "score": 1.0, + "content": "sample size. Alternatively, the availability of observational data with comprehensive information", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 562, + 505, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 562, + 505, + 573 + ], + "score": 1.0, + "content": "about patients, such as electronic health records (EHRs), constitutes a more accessible but also", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 572, + 506, + 584 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 506, + 584 + ], + "score": 1.0, + "content": "more challenging source for estimating causal effects (Häyrinen et al., 2008; Jha et al., 2009). Such", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 584, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 584, + 505, + 595 + ], + "score": 1.0, + "content": "observational data may be used to augment and verify an RCT, after a particular treatment is approved", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "and in use (Gombar et al., 2019; Frankovich et al., 2011; Longhurst et al., 2014). Moreover, the", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 605, + 506, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 506, + 617 + ], + "score": 1.0, + "content": "wealth of information from observational data also allows for the estimation of the individualized", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 505, + 628 + ], + "score": 1.0, + "content": "treatment effect (ITE), namely, the causal effect of an intervention at the individual level. In this", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 627, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 505, + 639 + ], + "score": 1.0, + "content": "work, we develop a novel framework for counterfactual time-to-event prediction to estimate the ITE", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 638, + 361, + 649 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 361, + 649 + ], + "score": 1.0, + "content": "for survival or time-to-event outcomes from observational data.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 34.5, + "bbox_fs": [ + 105, + 516, + 506, + 649 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 655, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "score": 1.0, + "content": "Estimating the causal effect for survival outcomes in observational data manifests two principal", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 664, + 506, + 679 + ], + "spans": [ + { + "bbox": [ + 105, + 664, + 506, + 679 + ], + "score": 1.0, + "content": "challenges. First, the treatment assignment mechanism is not known a priori. Therefore, there may", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "score": 1.0, + "content": "be variables, known as confounders, affecting both the treatment and survival time, which lead to", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "selection bias (Bareinboim & Pearl, 2012), i.e., that the distributions across treatment groups are not", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "the same. In this work, we focus on selection biases due to confounding, but other sources may also", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "be considered. For instance, patients who are severely ill are likely to receive more aggressive therapy,", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 720, + 506, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 506, + 732 + ], + "score": 1.0, + "content": "however, their health status may also inevitably influence survival. Traditional survival analysis", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 83, + 505, + 93 + ], + "spans": [ + { + "bbox": [ + 106, + 83, + 505, + 93 + ], + "score": 1.0, + "content": "neglects such bias, leading to incorrect causal estimation. Second, the exact time-to-event is not", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "always observed, i.e., sometimes we only know that an event has not occurred up to a certain point in", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 102, + 506, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 102, + 506, + 119 + ], + "score": 1.0, + "content": "time. This is known as the censoring problem. Moreover, censoring might be informative depending", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "on the characteristics of the individuals and their treatment assignments, thus proper adjustment is", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 410, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 410, + 139 + ], + "score": 1.0, + "content": "required for accurate causal estimation (Cole & Hernán, 2004; Díaz, 2019).", + "type": "text", + "cross_page": true + } + ], + "index": 4 + } + ], + "index": 44, + "bbox_fs": [ + 105, + 654, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 137 + ], + "lines": [ + { + "bbox": [ + 106, + 83, + 505, + 93 + ], + "spans": [ + { + "bbox": [ + 106, + 83, + 505, + 93 + ], + "score": 1.0, + "content": "neglects such bias, leading to incorrect causal estimation. Second, the exact time-to-event is not", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "always observed, i.e., sometimes we only know that an event has not occurred up to a certain point in", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 102, + 506, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 102, + 506, + 119 + ], + "score": 1.0, + "content": "time. This is known as the censoring problem. Moreover, censoring might be informative depending", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "on the characteristics of the individuals and their treatment assignments, thus proper adjustment is", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 410, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 410, + 139 + ], + "score": 1.0, + "content": "required for accurate causal estimation (Cole & Hernán, 2004; Díaz, 2019).", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 107, + 144, + 505, + 297 + ], + "lines": [ + { + "bbox": [ + 106, + 142, + 507, + 156 + ], + "spans": [ + { + "bbox": [ + 106, + 142, + 507, + 156 + ], + "score": 1.0, + "content": "Traditional causal survival-analysis approaches typically model the effect of the treatment or co-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 155, + 505, + 166 + ], + "spans": [ + { + "bbox": [ + 106, + 155, + 505, + 166 + ], + "score": 1.0, + "content": "variates (not time or survival) in a parametric manner. Two commonly used models are the Cox", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 164, + 506, + 178 + ], + "spans": [ + { + "bbox": [ + 105, + 164, + 506, + 178 + ], + "score": 1.0, + "content": "proportional hazards (CoxPH) model (Cox, 1972) and the accelerated failure time (AFT) model (Wei,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 176, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 506, + 189 + ], + "score": 1.0, + "content": "1992), which presume a linear relationship between the covariates and survival probability. Further,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 104, + 187, + 506, + 200 + ], + "spans": [ + { + "bbox": [ + 104, + 187, + 506, + 200 + ], + "score": 1.0, + "content": "proper weighting for each individual has been employed to account for confounding bias from these", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 198, + 506, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 506, + 211 + ], + "score": 1.0, + "content": "models (Austin, 2007; 2014; Hernán et al., 2005). For instance, probability weighting schemes", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 208, + 506, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 208, + 506, + 222 + ], + "score": 1.0, + "content": "that account for both selection bias and covariate dependent censoring have been considered for", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 218, + 506, + 234 + ], + "spans": [ + { + "bbox": [ + 105, + 218, + 506, + 234 + ], + "score": 1.0, + "content": "adjusted survival curves (Cole & Hernán, 2004; Díaz, 2019). Moreover, such probability weighting", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 229, + 506, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 229, + 506, + 245 + ], + "score": 1.0, + "content": "schemes have been applied to causal survival-analysis under time-varying treatment and confounding", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 241, + 506, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 506, + 255 + ], + "score": 1.0, + "content": "(Robins, 1986; Hernán et al., 2000). See van der Laan & Robins (2003); Tsiatis (2007); Van der", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 253, + 505, + 265 + ], + "spans": [ + { + "bbox": [ + 105, + 253, + 505, + 265 + ], + "score": 1.0, + "content": "Laan & Rose (2011); Hernán & Robins (2020) for an overview. Such linear specification makes", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 263, + 505, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 505, + 277 + ], + "score": 1.0, + "content": "these models interpretable but compromises their flexibility, and makes it difficult to adapt them", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 274, + 505, + 287 + ], + "spans": [ + { + "bbox": [ + 105, + 274, + 505, + 287 + ], + "score": 1.0, + "content": "for high-dimensional data or to capture complex interactions among covariates. Importantly, these", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 286, + 506, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 286, + 506, + 298 + ], + "score": 1.0, + "content": "methods lack a counterfactual prediction mechanism, which is key for ITE estimation (see Section 2).", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 107, + 303, + 505, + 489 + ], + "lines": [ + { + "bbox": [ + 106, + 303, + 505, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 505, + 315 + ], + "score": 1.0, + "content": "Fortunately, recent advances in machine learning, such as representation learning or generative", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 313, + 505, + 326 + ], + "spans": [ + { + "bbox": [ + 106, + 313, + 505, + 326 + ], + "score": 1.0, + "content": "modeling, have enabled causal inference methods to handle high-dimensional data and to characterize", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 325, + 505, + 337 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 505, + 337 + ], + "score": 1.0, + "content": "complex interactions effectively. For instance, there has been recent interest in tree-based (Chipman", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 336, + 506, + 348 + ], + "spans": [ + { + "bbox": [ + 106, + 336, + 506, + 348 + ], + "score": 1.0, + "content": "et al., 2010; Wager & Athey, 2018) and neural-network-based (Shalit et al., 2017; Zhang et al.,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 345, + 506, + 359 + ], + "spans": [ + { + "bbox": [ + 105, + 345, + 506, + 359 + ], + "score": 1.0, + "content": "2020) approaches. For pre-specified time-horizons, the nonparametric Random Survival Forest (RSF)", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 357, + 506, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 357, + 506, + 370 + ], + "score": 1.0, + "content": "(Ishwaran et al., 2008) and Bayesian Additive regression trees (BART) (Chipman et al., 2010) have", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 369, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 369, + 505, + 380 + ], + "score": 1.0, + "content": "been extended to causal survival analysis. RSF has been applied to causal survival forests with", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 379, + 506, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 379, + 506, + 392 + ], + "score": 1.0, + "content": "weighted bootstrap inference (Shen et al., 2018; Cui et al., 2020) while a BART is extended to account", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "score": 1.0, + "content": "for survival outcomes in Surv-BART (Sparapani et al., 2016), and AFT-BART (Henderson et al.,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 401, + 506, + 414 + ], + "spans": [ + { + "bbox": [ + 106, + 401, + 506, + 414 + ], + "score": 1.0, + "content": "2020). See Hu et al. (2020) for an extensive investigation of the causal survival tree-based methods.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 413, + 505, + 424 + ], + "spans": [ + { + "bbox": [ + 106, + 413, + 505, + 424 + ], + "score": 1.0, + "content": "Alternatively, when estimating the ITE, neural-network-based methods propose to regularize the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 423, + 506, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 506, + 436 + ], + "score": 1.0, + "content": "transformed covariates or representations for an individual to have balanced distributions across", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 434, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 506, + 447 + ], + "score": 1.0, + "content": "treatment groups, thus accounting for the confounding bias and improving ITE prediction. However,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 445, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 505, + 457 + ], + "score": 1.0, + "content": "most approaches employing representation learning techniques for counterfactual inference deal", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 457, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 106, + 457, + 505, + 468 + ], + "score": 1.0, + "content": "with continuous or binary outcomes, instead of time-to-event outcomes with censoring (informative", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 106, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "or non-informative). Hence, a principled generalization to the context of counterfactual survival", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 479, + 183, + 490 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 183, + 490 + ], + "score": 1.0, + "content": "analysis is needed.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 107, + 495, + 505, + 539 + ], + "lines": [ + { + "bbox": [ + 105, + 495, + 506, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 506, + 508 + ], + "score": 1.0, + "content": "In this work we leverage balanced (latent) representation learning to estimate ITE via counterfactual", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 506, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 505, + 518 + ], + "score": 1.0, + "content": "prediction of survival outcomes in observational studies. We develop a framework to predict event", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 518, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 505, + 529 + ], + "score": 1.0, + "content": "times from a low-dimensional transformation of the original covariate space. To address the specific", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 528, + 497, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 497, + 541 + ], + "score": 1.0, + "content": "challenges associated with counterfactual survival analysis, we make the following contributions:", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 108, + 545, + 506, + 677 + ], + "lines": [ + { + "bbox": [ + 106, + 543, + 506, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 506, + 558 + ], + "score": 1.0, + "content": "• We develop an optimization objective incorporating adjustments for informative censoring, as well", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 556, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 115, + 556, + 505, + 569 + ], + "score": 1.0, + "content": "as a balanced regularization term bounding the generalization error for ITE prediction. For the", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 114, + 567, + 507, + 580 + ], + "spans": [ + { + "bbox": [ + 114, + 567, + 507, + 580 + ], + "score": 1.0, + "content": "latter, we repurpose a recently proposed bound (Shalit et al., 2017) for our time-to-event scenario.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 107, + 577, + 507, + 591 + ], + "spans": [ + { + "bbox": [ + 107, + 577, + 507, + 591 + ], + "score": 1.0, + "content": "• We propose a generative model for event times to relax restrictive survival linear and parametric as-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 589, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 115, + 589, + 505, + 601 + ], + "score": 1.0, + "content": "sumptions, thus allowing for more flexible modeling. Our approach can also provide nonparametric", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 601, + 303, + 612 + ], + "spans": [ + { + "bbox": [ + 115, + 601, + 303, + 612 + ], + "score": 1.0, + "content": "uncertainty quantification for ITE predictions.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 107, + 610, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 107, + 610, + 505, + 623 + ], + "score": 1.0, + "content": "• We provide survival-specific evaluation metrics, including a new nonparametric hazard ratio", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 115, + 622, + 506, + 634 + ], + "spans": [ + { + "bbox": [ + 115, + 622, + 506, + 634 + ], + "score": 1.0, + "content": "estimator, and discuss how to perform model selection for survival outcomes. The proposed model", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 115, + 632, + 506, + 645 + ], + "spans": [ + { + "bbox": [ + 115, + 632, + 506, + 645 + ], + "score": 1.0, + "content": "demonstrates superior performance relative to the commonly used baselines in real-world and", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 114, + 644, + 212, + 655 + ], + "spans": [ + { + "bbox": [ + 114, + 644, + 212, + 655 + ], + "score": 1.0, + "content": "semi-synthetic datasets.", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 108, + 653, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 108, + 653, + 506, + 668 + ], + "score": 1.0, + "content": "• We introduce a survival-specific semi-synthetic dataset and demonstrate an approach for leveraging", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 115, + 666, + 416, + 678 + ], + "spans": [ + { + "bbox": [ + 115, + 666, + 416, + 678 + ], + "score": 1.0, + "content": "prior randomized experiments in longitudinal studies for model validation.", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 45.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 306, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 137 + ], + "lines": [], + "index": 2, + "bbox_fs": [ + 105, + 83, + 506, + 139 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 144, + 505, + 297 + ], + "lines": [ + { + "bbox": [ + 106, + 142, + 507, + 156 + ], + "spans": [ + { + "bbox": [ + 106, + 142, + 507, + 156 + ], + "score": 1.0, + "content": "Traditional causal survival-analysis approaches typically model the effect of the treatment or co-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 155, + 505, + 166 + ], + "spans": [ + { + "bbox": [ + 106, + 155, + 505, + 166 + ], + "score": 1.0, + "content": "variates (not time or survival) in a parametric manner. Two commonly used models are the Cox", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 164, + 506, + 178 + ], + "spans": [ + { + "bbox": [ + 105, + 164, + 506, + 178 + ], + "score": 1.0, + "content": "proportional hazards (CoxPH) model (Cox, 1972) and the accelerated failure time (AFT) model (Wei,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 176, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 506, + 189 + ], + "score": 1.0, + "content": "1992), which presume a linear relationship between the covariates and survival probability. Further,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 104, + 187, + 506, + 200 + ], + "spans": [ + { + "bbox": [ + 104, + 187, + 506, + 200 + ], + "score": 1.0, + "content": "proper weighting for each individual has been employed to account for confounding bias from these", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 198, + 506, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 506, + 211 + ], + "score": 1.0, + "content": "models (Austin, 2007; 2014; Hernán et al., 2005). For instance, probability weighting schemes", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 208, + 506, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 208, + 506, + 222 + ], + "score": 1.0, + "content": "that account for both selection bias and covariate dependent censoring have been considered for", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 218, + 506, + 234 + ], + "spans": [ + { + "bbox": [ + 105, + 218, + 506, + 234 + ], + "score": 1.0, + "content": "adjusted survival curves (Cole & Hernán, 2004; Díaz, 2019). Moreover, such probability weighting", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 229, + 506, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 229, + 506, + 245 + ], + "score": 1.0, + "content": "schemes have been applied to causal survival-analysis under time-varying treatment and confounding", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 241, + 506, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 506, + 255 + ], + "score": 1.0, + "content": "(Robins, 1986; Hernán et al., 2000). See van der Laan & Robins (2003); Tsiatis (2007); Van der", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 253, + 505, + 265 + ], + "spans": [ + { + "bbox": [ + 105, + 253, + 505, + 265 + ], + "score": 1.0, + "content": "Laan & Rose (2011); Hernán & Robins (2020) for an overview. Such linear specification makes", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 263, + 505, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 505, + 277 + ], + "score": 1.0, + "content": "these models interpretable but compromises their flexibility, and makes it difficult to adapt them", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 274, + 505, + 287 + ], + "spans": [ + { + "bbox": [ + 105, + 274, + 505, + 287 + ], + "score": 1.0, + "content": "for high-dimensional data or to capture complex interactions among covariates. Importantly, these", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 286, + 506, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 286, + 506, + 298 + ], + "score": 1.0, + "content": "methods lack a counterfactual prediction mechanism, which is key for ITE estimation (see Section 2).", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 11.5, + "bbox_fs": [ + 104, + 142, + 507, + 298 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 303, + 505, + 489 + ], + "lines": [ + { + "bbox": [ + 106, + 303, + 505, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 505, + 315 + ], + "score": 1.0, + "content": "Fortunately, recent advances in machine learning, such as representation learning or generative", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 313, + 505, + 326 + ], + "spans": [ + { + "bbox": [ + 106, + 313, + 505, + 326 + ], + "score": 1.0, + "content": "modeling, have enabled causal inference methods to handle high-dimensional data and to characterize", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 325, + 505, + 337 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 505, + 337 + ], + "score": 1.0, + "content": "complex interactions effectively. For instance, there has been recent interest in tree-based (Chipman", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 336, + 506, + 348 + ], + "spans": [ + { + "bbox": [ + 106, + 336, + 506, + 348 + ], + "score": 1.0, + "content": "et al., 2010; Wager & Athey, 2018) and neural-network-based (Shalit et al., 2017; Zhang et al.,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 345, + 506, + 359 + ], + "spans": [ + { + "bbox": [ + 105, + 345, + 506, + 359 + ], + "score": 1.0, + "content": "2020) approaches. For pre-specified time-horizons, the nonparametric Random Survival Forest (RSF)", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 357, + 506, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 357, + 506, + 370 + ], + "score": 1.0, + "content": "(Ishwaran et al., 2008) and Bayesian Additive regression trees (BART) (Chipman et al., 2010) have", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 369, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 369, + 505, + 380 + ], + "score": 1.0, + "content": "been extended to causal survival analysis. RSF has been applied to causal survival forests with", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 379, + 506, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 379, + 506, + 392 + ], + "score": 1.0, + "content": "weighted bootstrap inference (Shen et al., 2018; Cui et al., 2020) while a BART is extended to account", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "score": 1.0, + "content": "for survival outcomes in Surv-BART (Sparapani et al., 2016), and AFT-BART (Henderson et al.,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 401, + 506, + 414 + ], + "spans": [ + { + "bbox": [ + 106, + 401, + 506, + 414 + ], + "score": 1.0, + "content": "2020). See Hu et al. (2020) for an extensive investigation of the causal survival tree-based methods.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 413, + 505, + 424 + ], + "spans": [ + { + "bbox": [ + 106, + 413, + 505, + 424 + ], + "score": 1.0, + "content": "Alternatively, when estimating the ITE, neural-network-based methods propose to regularize the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 423, + 506, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 506, + 436 + ], + "score": 1.0, + "content": "transformed covariates or representations for an individual to have balanced distributions across", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 434, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 506, + 447 + ], + "score": 1.0, + "content": "treatment groups, thus accounting for the confounding bias and improving ITE prediction. However,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 445, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 505, + 457 + ], + "score": 1.0, + "content": "most approaches employing representation learning techniques for counterfactual inference deal", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 457, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 106, + 457, + 505, + 468 + ], + "score": 1.0, + "content": "with continuous or binary outcomes, instead of time-to-event outcomes with censoring (informative", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 106, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "or non-informative). Hence, a principled generalization to the context of counterfactual survival", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 479, + 183, + 490 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 183, + 490 + ], + "score": 1.0, + "content": "analysis is needed.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 27, + "bbox_fs": [ + 105, + 303, + 506, + 490 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 495, + 505, + 539 + ], + "lines": [ + { + "bbox": [ + 105, + 495, + 506, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 506, + 508 + ], + "score": 1.0, + "content": "In this work we leverage balanced (latent) representation learning to estimate ITE via counterfactual", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 506, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 505, + 518 + ], + "score": 1.0, + "content": "prediction of survival outcomes in observational studies. We develop a framework to predict event", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 518, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 505, + 529 + ], + "score": 1.0, + "content": "times from a low-dimensional transformation of the original covariate space. To address the specific", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 528, + 497, + 541 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 497, + 541 + ], + "score": 1.0, + "content": "challenges associated with counterfactual survival analysis, we make the following contributions:", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 495, + 506, + 541 + ] + }, + { + "type": "list", + "bbox": [ + 108, + 545, + 506, + 677 + ], + "lines": [ + { + "bbox": [ + 106, + 543, + 506, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 506, + 558 + ], + "score": 1.0, + "content": "• We develop an optimization objective incorporating adjustments for informative censoring, as well", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 556, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 115, + 556, + 505, + 569 + ], + "score": 1.0, + "content": "as a balanced regularization term bounding the generalization error for ITE prediction. For the", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 114, + 567, + 507, + 580 + ], + "spans": [ + { + "bbox": [ + 114, + 567, + 507, + 580 + ], + "score": 1.0, + "content": "latter, we repurpose a recently proposed bound (Shalit et al., 2017) for our time-to-event scenario.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 107, + 577, + 507, + 591 + ], + "spans": [ + { + "bbox": [ + 107, + 577, + 507, + 591 + ], + "score": 1.0, + "content": "• We propose a generative model for event times to relax restrictive survival linear and parametric as-", + "type": "text" + } + ], + "index": 43, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 589, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 115, + 589, + 505, + 601 + ], + "score": 1.0, + "content": "sumptions, thus allowing for more flexible modeling. Our approach can also provide nonparametric", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 601, + 303, + 612 + ], + "spans": [ + { + "bbox": [ + 115, + 601, + 303, + 612 + ], + "score": 1.0, + "content": "uncertainty quantification for ITE predictions.", + "type": "text" + } + ], + "index": 45, + "is_list_end_line": true + }, + { + "bbox": [ + 107, + 610, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 107, + 610, + 505, + 623 + ], + "score": 1.0, + "content": "• We provide survival-specific evaluation metrics, including a new nonparametric hazard ratio", + "type": "text" + } + ], + "index": 46, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 622, + 506, + 634 + ], + "spans": [ + { + "bbox": [ + 115, + 622, + 506, + 634 + ], + "score": 1.0, + "content": "estimator, and discuss how to perform model selection for survival outcomes. The proposed model", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 115, + 632, + 506, + 645 + ], + "spans": [ + { + "bbox": [ + 115, + 632, + 506, + 645 + ], + "score": 1.0, + "content": "demonstrates superior performance relative to the commonly used baselines in real-world and", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 114, + 644, + 212, + 655 + ], + "spans": [ + { + "bbox": [ + 114, + 644, + 212, + 655 + ], + "score": 1.0, + "content": "semi-synthetic datasets.", + "type": "text" + } + ], + "index": 49, + "is_list_end_line": true + }, + { + "bbox": [ + 108, + 653, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 108, + 653, + 506, + 668 + ], + "score": 1.0, + "content": "• We introduce a survival-specific semi-synthetic dataset and demonstrate an approach for leveraging", + "type": "text" + } + ], + "index": 50, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 666, + 416, + 678 + ], + "spans": [ + { + "bbox": [ + 115, + 666, + 416, + 678 + ], + "score": 1.0, + "content": "prior randomized experiments in longitudinal studies for model validation.", + "type": "text" + } + ], + "index": 51, + "is_list_end_line": true + } + ], + "index": 45.5, + "bbox_fs": [ + 106, + 543, + 507, + 678 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 123, + 73, + 483, + 164 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 123, + 73, + 483, + 164 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 123, + 73, + 483, + 164 + ], + "spans": [ + { + "bbox": [ + 123, + 73, + 483, + 164 + ], + "score": 0.967, + "type": "image", + "image_path": "482e9df6a7b5d031a444403a4be41d381d10c980808568d8cbf42f012f3882e6.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 123, + 73, + 483, + 103.33333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 123, + 103.33333333333333, + 483, + 133.66666666666666 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 123, + 133.66666666666666, + 483, + 164.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 168, + 506, + 234 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 168, + 505, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 168, + 482, + 180 + ], + "score": 1.0, + "content": "Figure 1: (a) Illustration of the proposed counterfactual survival analysis (CSA). Covariates", + "type": "text" + }, + { + "bbox": [ + 483, + 168, + 505, + 178 + ], + "score": 0.84, + "content": "X =", + "type": "inline_equation" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 178, + 505, + 191 + ], + "spans": [ + { + "bbox": [ + 106, + 181, + 114, + 189 + ], + "score": 0.69, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 114, + 178, + 275, + 191 + ], + "score": 1.0, + "content": "are mapped into latent representation", + "type": "text" + }, + { + "bbox": [ + 275, + 181, + 281, + 189 + ], + "score": 0.74, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 178, + 395, + 191 + ], + "score": 1.0, + "content": "via deterministic mapping", + "type": "text" + }, + { + "bbox": [ + 395, + 179, + 439, + 190 + ], + "score": 0.92, + "content": "r \\ = \\ \\Phi ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 178, + 505, + 191 + ], + "score": 1.0, + "content": ". The potential", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 189, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 106, + 190, + 221, + 201 + ], + "score": 1.0, + "content": "outcomes are sampled from", + "type": "text" + }, + { + "bbox": [ + 222, + 189, + 300, + 201 + ], + "score": 0.92, + "content": "\\bar { t } _ { a } \\sim p ( T _ { A } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 190, + 316, + 201 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 316, + 190, + 343, + 200 + ], + "score": 0.91, + "content": "A = a", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 190, + 440, + 201 + ], + "score": 1.0, + "content": "via stochastic mapping", + "type": "text" + }, + { + "bbox": [ + 440, + 190, + 473, + 201 + ], + "score": 0.92, + "content": "h _ { A } ( r , \\tilde { \\epsilon } )", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 190, + 505, + 201 + ], + "score": 1.0, + "content": ", where", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 201, + 505, + 212 + ], + "spans": [ + { + "bbox": [ + 106, + 201, + 331, + 212 + ], + "score": 1.0, + "content": "randomness is induced with a flow-based transformation,", + "type": "text" + }, + { + "bbox": [ + 331, + 201, + 337, + 211 + ], + "score": 0.45, + "content": "\\tilde { \\epsilon }", + "type": "inline_equation" + }, + { + "bbox": [ + 337, + 201, + 433, + 212 + ], + "score": 1.0, + "content": ", of a simple distribution", + "type": "text" + }, + { + "bbox": [ + 433, + 201, + 451, + 212 + ], + "score": 0.9, + "content": "p ( \\epsilon )", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 201, + 505, + 212 + ], + "score": 1.0, + "content": ", i.e., uniform", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 212, + 505, + 223 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 505, + 223 + ], + "score": 1.0, + "content": "or Gaussian. (b) and (c) show the proposed causal graphs for non-informative and informative", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 223, + 202, + 236 + ], + "spans": [ + { + "bbox": [ + 105, + 223, + 202, + 236 + ], + "score": 1.0, + "content": "censoring, respectively.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 5.5 + } + ], + "index": 3.25 + }, + { + "type": "title", + "bbox": [ + 108, + 244, + 256, + 257 + ], + "lines": [ + { + "bbox": [ + 105, + 242, + 258, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 258, + 258 + ], + "score": 1.0, + "content": "2 PROBLEM FORMULATION", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 106, + 268, + 506, + 369 + ], + "lines": [ + { + "bbox": [ + 106, + 269, + 507, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 269, + 507, + 282 + ], + "score": 1.0, + "content": "We first introduce the basic setup for performing causal survival analysis in observational studies.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 280, + 507, + 293 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 177, + 293 + ], + "score": 1.0, + "content": "Suppose we have", + "type": "text" + }, + { + "bbox": [ + 177, + 281, + 187, + 290 + ], + "score": 0.78, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 280, + 232, + 293 + ], + "score": 1.0, + "content": "units, with", + "type": "text" + }, + { + "bbox": [ + 233, + 281, + 246, + 291 + ], + "score": 0.89, + "content": "N _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 280, + 340, + 293 + ], + "score": 1.0, + "content": "units being treated and", + "type": "text" + }, + { + "bbox": [ + 340, + 281, + 354, + 291 + ], + "score": 0.89, + "content": "N _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 280, + 437, + 293 + ], + "score": 1.0, + "content": "in the control group", + "type": "text" + }, + { + "bbox": [ + 438, + 280, + 502, + 291 + ], + "score": 0.88, + "content": "( N = N _ { 1 } + N _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 280, + 507, + 293 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 291, + 506, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 296, + 304 + ], + "score": 1.0, + "content": "For each unit (individual), we have covariates", + "type": "text" + }, + { + "bbox": [ + 297, + 292, + 307, + 301 + ], + "score": 0.82, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 291, + 506, + 304 + ], + "score": 1.0, + "content": ", which can be heterogeneous, e.g., a mixture of", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 302, + 506, + 315 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 506, + 315 + ], + "score": 1.0, + "content": "categorical and continuous covariates which, in the context of medicine, may include labs, vitals,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 312, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 312, + 347, + 326 + ], + "score": 1.0, + "content": "procedure codes, etc. We also have a treatment indicator", + "type": "text" + }, + { + "bbox": [ + 348, + 313, + 356, + 323 + ], + "score": 0.77, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 312, + 389, + 326 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 389, + 313, + 419, + 323 + ], + "score": 0.9, + "content": "A = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 312, + 506, + 326 + ], + "score": 1.0, + "content": "for the controls and", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 107, + 323, + 506, + 338 + ], + "spans": [ + { + "bbox": [ + 107, + 324, + 134, + 334 + ], + "score": 0.9, + "content": "A = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 323, + 368, + 338 + ], + "score": 1.0, + "content": "for the treated, as well as the outcome (event) of interest", + "type": "text" + }, + { + "bbox": [ + 368, + 325, + 376, + 334 + ], + "score": 0.76, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 323, + 506, + 338 + ], + "score": 1.0, + "content": ". Under the potential-outcomes", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 334, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 334, + 223, + 348 + ], + "score": 1.0, + "content": "framework (Rubin, 2005), let", + "type": "text" + }, + { + "bbox": [ + 223, + 335, + 235, + 346 + ], + "score": 0.88, + "content": "T _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 334, + 253, + 348 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 253, + 335, + 264, + 346 + ], + "score": 0.88, + "content": "T _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 265, + 334, + 505, + 348 + ], + "score": 1.0, + "content": "be the potential event times for a given subject under control", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 344, + 507, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 507, + 360 + ], + "score": 1.0, + "content": "and treatment, respectively. In practice we only observe one realization of the potential outcomes,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 357, + 464, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 357, + 205, + 370 + ], + "score": 1.0, + "content": "i.e., the factual outcome", + "type": "text" + }, + { + "bbox": [ + 205, + 357, + 239, + 368 + ], + "score": 0.91, + "content": "T = T _ { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 357, + 379, + 370 + ], + "score": 1.0, + "content": ", while the counterfactual outcome", + "type": "text" + }, + { + "bbox": [ + 380, + 357, + 403, + 368 + ], + "score": 0.92, + "content": "T _ { 1 - A }", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 357, + 464, + 370 + ], + "score": 1.0, + "content": "is unobserved.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 106, + 373, + 505, + 429 + ], + "lines": [ + { + "bbox": [ + 106, + 373, + 506, + 387 + ], + "spans": [ + { + "bbox": [ + 106, + 373, + 506, + 387 + ], + "score": 1.0, + "content": "In survival analysis, the problem becomes more difficult because we do not always observe the exact", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 384, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 505, + 397 + ], + "score": 1.0, + "content": "event time for each individual, but rather the time up to which we are certain that the event has not", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 396, + 506, + 409 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 506, + 409 + ], + "score": 1.0, + "content": "occurred; specifically, we have a (right) censoring problem, most likely due to the loss of follow-up.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 406, + 505, + 420 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 238, + 420 + ], + "score": 1.0, + "content": "We denote the censoring time as", + "type": "text" + }, + { + "bbox": [ + 238, + 407, + 247, + 417 + ], + "score": 0.82, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 406, + 355, + 420 + ], + "score": 1.0, + "content": "and censoring indicator as", + "type": "text" + }, + { + "bbox": [ + 356, + 407, + 398, + 419 + ], + "score": 0.92, + "content": "\\delta \\in \\{ 0 , 1 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 406, + 505, + 420 + ], + "score": 1.0, + "content": ". The actual observed time", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 417, + 460, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 116, + 430 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 116, + 418, + 187, + 430 + ], + "score": 0.93, + "content": "Y = \\operatorname* { m i n } ( T _ { A } , C )", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 417, + 378, + 430 + ], + "score": 1.0, + "content": ", i.e., the outcome is observed (non-censored) if", + "type": "text" + }, + { + "bbox": [ + 378, + 418, + 413, + 429 + ], + "score": 0.91, + "content": "T _ { A } < C", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 417, + 431, + 430 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 431, + 418, + 455, + 428 + ], + "score": 0.9, + "content": "\\delta = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 417, + 460, + 430 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 434, + 505, + 512 + ], + "lines": [ + { + "bbox": [ + 104, + 433, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 104, + 433, + 396, + 447 + ], + "score": 1.0, + "content": "In this work, we are interested in the expected difference between the", + "type": "text" + }, + { + "bbox": [ + 397, + 435, + 408, + 446 + ], + "score": 0.88, + "content": "T _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 433, + 428, + 447 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 428, + 435, + 439, + 446 + ], + "score": 0.87, + "content": "T _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 433, + 506, + 447 + ], + "score": 1.0, + "content": "conditioned on", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 107, + 446, + 506, + 458 + ], + "spans": [ + { + "bbox": [ + 107, + 446, + 117, + 456 + ], + "score": 0.82, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 446, + 506, + 458 + ], + "score": 1.0, + "content": "for a given unit (individual), which is commonly known as the individualized treatment effect", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 456, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 444, + 469 + ], + "score": 1.0, + "content": "(ITE). Specifically, we wish to perform inference on the conditional distributions of", + "type": "text" + }, + { + "bbox": [ + 444, + 457, + 456, + 468 + ], + "score": 0.89, + "content": "T _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 456, + 474, + 469 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 474, + 457, + 485, + 468 + ], + "score": 0.87, + "content": "T _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 456, + 506, + 469 + ], + "score": 1.0, + "content": ", i.e.,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 107, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 107, + 468, + 142, + 479 + ], + "score": 0.9, + "content": "p ( T _ { 1 } | X )", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 467, + 160, + 480 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 160, + 468, + 196, + 479 + ], + "score": 0.92, + "content": "p ( T _ { 0 } | X )", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 467, + 435, + 480 + ], + "score": 1.0, + "content": ", respectively, as shown in Figure 1a. In practice, we observe", + "type": "text" + }, + { + "bbox": [ + 435, + 468, + 446, + 477 + ], + "score": 0.83, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "realizations of", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 477, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 106, + 479, + 154, + 490 + ], + "score": 0.91, + "content": "( Y , \\delta , X , A )", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 477, + 506, + 492 + ], + "score": 1.0, + "content": "for observed time, censoring indicator, covariates and treatment indicator, respectively;", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 103, + 486, + 508, + 505 + ], + "spans": [ + { + "bbox": [ + 103, + 486, + 355, + 505 + ], + "score": 1.0, + "content": "hence, from an observational study the dataset takes the form", + "type": "text" + }, + { + "bbox": [ + 356, + 489, + 456, + 502 + ], + "score": 0.92, + "content": "\\mathcal { D } = \\{ ( y _ { i } , \\delta _ { i } , x _ { i } , a _ { i } ) \\} _ { i = 1 } ^ { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 486, + 508, + 505 + ], + "score": 1.0, + "content": ". Below, we", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 500, + 372, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 372, + 513 + ], + "score": 1.0, + "content": "discuss several common choices of estimands in survival analysis.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 106, + 518, + 506, + 621 + ], + "lines": [ + { + "bbox": [ + 105, + 518, + 506, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 506, + 532 + ], + "score": 1.0, + "content": "Estimands of Interest We begin by considering survival analysis in the absence of an intervening", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 531, + 506, + 544 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 178, + 544 + ], + "score": 1.0, + "content": "treatment choice,", + "type": "text" + }, + { + "bbox": [ + 178, + 532, + 187, + 541 + ], + "score": 0.68, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 531, + 207, + 544 + ], + "score": 1.0, + "content": ". Let", + "type": "text" + }, + { + "bbox": [ + 207, + 531, + 320, + 543 + ], + "score": 0.9, + "content": "F ( t | x ) \\triangleq P ( T \\leq t | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 531, + 506, + 544 + ], + "score": 1.0, + "content": "be the cumulative distribution function of the", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 543, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 185, + 555 + ], + "score": 1.0, + "content": "event (failure) time,", + "type": "text" + }, + { + "bbox": [ + 186, + 544, + 190, + 553 + ], + "score": 0.7, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 543, + 335, + 555 + ], + "score": 1.0, + "content": ", given a realization of the covariates,", + "type": "text" + }, + { + "bbox": [ + 336, + 545, + 342, + 552 + ], + "score": 0.71, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 543, + 505, + 555 + ], + "score": 1.0, + "content": ". Survival analysis is primarily concerned", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 554, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 386, + 567 + ], + "score": 1.0, + "content": "with characterization of the survival function conditioned on covariates", + "type": "text" + }, + { + "bbox": [ + 386, + 555, + 470, + 567 + ], + "score": 0.91, + "content": "S ( t | x ) \\triangleq 1 - F ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 554, + 506, + 567 + ], + "score": 1.0, + "content": ", and the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 565, + 505, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 226, + 578 + ], + "score": 1.0, + "content": "hazard function or risk score,", + "type": "text" + }, + { + "bbox": [ + 227, + 566, + 253, + 578 + ], + "score": 0.92, + "content": "\\lambda ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 565, + 318, + 578 + ], + "score": 1.0, + "content": ", defined below.", + "type": "text" + }, + { + "bbox": [ + 319, + 567, + 346, + 578 + ], + "score": 0.92, + "content": "S ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 565, + 505, + 578 + ], + "score": 1.0, + "content": "is a monotonically decreasing function", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 577, + 505, + 589 + ], + "spans": [ + { + "bbox": [ + 106, + 577, + 300, + 589 + ], + "score": 1.0, + "content": "indicating the probability of survival up to time", + "type": "text" + }, + { + "bbox": [ + 300, + 579, + 304, + 586 + ], + "score": 0.73, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 577, + 505, + 589 + ], + "score": 1.0, + "content": ". The hazard function measures the instantaneous", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 587, + 506, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 284, + 600 + ], + "score": 1.0, + "content": "probability of the event occurring between", + "type": "text" + }, + { + "bbox": [ + 284, + 589, + 330, + 600 + ], + "score": 0.91, + "content": "\\{ t , t + \\Delta t \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 587, + 357, + 600 + ], + "score": 1.0, + "content": "given", + "type": "text" + }, + { + "bbox": [ + 358, + 588, + 384, + 598 + ], + "score": 0.87, + "content": "T > t", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 587, + 403, + 600 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 403, + 588, + 438, + 597 + ], + "score": 0.86, + "content": "\\Delta t 0", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 587, + 506, + 600 + ], + "score": 1.0, + "content": ". From standard", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 598, + 506, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 506, + 611 + ], + "score": 1.0, + "content": "definitions (Kleinbaum & Klein, 2010), the relationship between cumulative and hazard function is", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 609, + 164, + 621 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 164, + 621 + ], + "score": 1.0, + "content": "formulated as", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 35 + }, + { + "type": "interline_equation", + "bbox": [ + 161, + 624, + 451, + 652 + ], + "lines": [ + { + "bbox": [ + 161, + 624, + 451, + 652 + ], + "spans": [ + { + "bbox": [ + 161, + 624, + 451, + 652 + ], + "score": 0.95, + "content": "\\lambda ( t | x ) = \\operatorname* { l i m } _ { d t 0 } { \\frac { P ( t < T < t + d t | X = x ) } { P ( T > t | X = x ) d t } } = - { \\frac { d \\log S ( t | x ) } { d t } } = { \\frac { f ( t | x ) } { S ( t | x ) } } .", + "type": "interline_equation", + "image_path": "89816b6b4ab5614eea0ade34840cd19c6b9c6faa995393a64ef37e851560bce4.jpg" + } + ] + } + ], + "index": 41, + "virtual_lines": [ + { + "bbox": [ + 161, + 624, + 451, + 633.3333333333334 + ], + "spans": [], + "index": 40 + }, + { + "bbox": [ + 161, + 633.3333333333334, + 451, + 642.6666666666667 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 161, + 642.6666666666667, + 451, + 652.0000000000001 + ], + "spans": [], + "index": 42 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 659, + 504, + 682 + ], + "lines": [ + { + "bbox": [ + 105, + 658, + 506, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 197, + 673 + ], + "score": 1.0, + "content": "From (1) we see that", + "type": "text" + }, + { + "bbox": [ + 197, + 660, + 382, + 672 + ], + "score": 0.92, + "content": "f ( t | x ) \\triangleq P ( T = t | X = x ) = \\lambda ( t | x ) S ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 382, + 658, + 506, + 673 + ], + "score": 1.0, + "content": ", is the conditional event time", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 670, + 289, + 683 + ], + "spans": [ + { + "bbox": [ + 106, + 670, + 223, + 683 + ], + "score": 1.0, + "content": "density function (Kleinbaum", + "type": "text" + }, + { + "bbox": [ + 223, + 673, + 231, + 681 + ], + "score": 0.69, + "content": "\\&", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 670, + 289, + 683 + ], + "score": 1.0, + "content": "Klein, 2010).", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5 + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 688, + 506, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 213, + 699 + ], + "score": 1.0, + "content": "Given the binary treatment", + "type": "text" + }, + { + "bbox": [ + 214, + 690, + 221, + 698 + ], + "score": 0.87, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 688, + 506, + 699 + ], + "score": 1.0, + "content": ", we are interested in its impact on the survival time. For ITE estimation,", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 699, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 407, + 711 + ], + "score": 1.0, + "content": "we are also interested in the difference between the two potential outcomes", + "type": "text" + }, + { + "bbox": [ + 407, + 701, + 433, + 710 + ], + "score": 0.88, + "content": "T _ { 1 } , T _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 699, + 453, + 711 + ], + "score": 1.0, + "content": ". Let", + "type": "text" + }, + { + "bbox": [ + 453, + 700, + 487, + 711 + ], + "score": 0.91, + "content": "S _ { A } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 699, + 506, + 711 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 107, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 107, + 711, + 140, + 722 + ], + "score": 0.92, + "content": "\\lambda _ { A } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 709, + 424, + 722 + ], + "score": 1.0, + "content": "denote the survival and hazard functions for the potential outcomes", + "type": "text" + }, + { + "bbox": [ + 424, + 711, + 437, + 721 + ], + "score": 0.88, + "content": "T _ { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 709, + 460, + 722 + ], + "score": 1.0, + "content": ", i.e.,", + "type": "text" + }, + { + "bbox": [ + 460, + 711, + 471, + 721 + ], + "score": 0.87, + "content": "T _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 709, + 491, + 722 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 492, + 711, + 502, + 721 + ], + "score": 0.87, + "content": "T _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 709, + 506, + 722 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 719, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 719, + 505, + 734 + ], + "score": 1.0, + "content": "Several common estimands of interest include (Zhao et al., 2012; Trinquart et al., 2016): difference", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 46.5 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 307, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 123, + 73, + 483, + 164 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 123, + 73, + 483, + 164 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 123, + 73, + 483, + 164 + ], + "spans": [ + { + "bbox": [ + 123, + 73, + 483, + 164 + ], + "score": 0.967, + "type": "image", + "image_path": "482e9df6a7b5d031a444403a4be41d381d10c980808568d8cbf42f012f3882e6.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 123, + 73, + 483, + 103.33333333333333 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 123, + 103.33333333333333, + 483, + 133.66666666666666 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 123, + 133.66666666666666, + 483, + 164.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 168, + 506, + 234 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 168, + 505, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 168, + 482, + 180 + ], + "score": 1.0, + "content": "Figure 1: (a) Illustration of the proposed counterfactual survival analysis (CSA). Covariates", + "type": "text" + }, + { + "bbox": [ + 483, + 168, + 505, + 178 + ], + "score": 0.84, + "content": "X =", + "type": "inline_equation" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 178, + 505, + 191 + ], + "spans": [ + { + "bbox": [ + 106, + 181, + 114, + 189 + ], + "score": 0.69, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 114, + 178, + 275, + 191 + ], + "score": 1.0, + "content": "are mapped into latent representation", + "type": "text" + }, + { + "bbox": [ + 275, + 181, + 281, + 189 + ], + "score": 0.74, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 178, + 395, + 191 + ], + "score": 1.0, + "content": "via deterministic mapping", + "type": "text" + }, + { + "bbox": [ + 395, + 179, + 439, + 190 + ], + "score": 0.92, + "content": "r \\ = \\ \\Phi ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 178, + 505, + 191 + ], + "score": 1.0, + "content": ". The potential", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 189, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 106, + 190, + 221, + 201 + ], + "score": 1.0, + "content": "outcomes are sampled from", + "type": "text" + }, + { + "bbox": [ + 222, + 189, + 300, + 201 + ], + "score": 0.92, + "content": "\\bar { t } _ { a } \\sim p ( T _ { A } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 190, + 316, + 201 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 316, + 190, + 343, + 200 + ], + "score": 0.91, + "content": "A = a", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 190, + 440, + 201 + ], + "score": 1.0, + "content": "via stochastic mapping", + "type": "text" + }, + { + "bbox": [ + 440, + 190, + 473, + 201 + ], + "score": 0.92, + "content": "h _ { A } ( r , \\tilde { \\epsilon } )", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 190, + 505, + 201 + ], + "score": 1.0, + "content": ", where", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 201, + 505, + 212 + ], + "spans": [ + { + "bbox": [ + 106, + 201, + 331, + 212 + ], + "score": 1.0, + "content": "randomness is induced with a flow-based transformation,", + "type": "text" + }, + { + "bbox": [ + 331, + 201, + 337, + 211 + ], + "score": 0.45, + "content": "\\tilde { \\epsilon }", + "type": "inline_equation" + }, + { + "bbox": [ + 337, + 201, + 433, + 212 + ], + "score": 1.0, + "content": ", of a simple distribution", + "type": "text" + }, + { + "bbox": [ + 433, + 201, + 451, + 212 + ], + "score": 0.9, + "content": "p ( \\epsilon )", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 201, + 505, + 212 + ], + "score": 1.0, + "content": ", i.e., uniform", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 212, + 505, + 223 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 505, + 223 + ], + "score": 1.0, + "content": "or Gaussian. (b) and (c) show the proposed causal graphs for non-informative and informative", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 223, + 202, + 236 + ], + "spans": [ + { + "bbox": [ + 105, + 223, + 202, + 236 + ], + "score": 1.0, + "content": "censoring, respectively.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 5.5 + } + ], + "index": 3.25 + }, + { + "type": "title", + "bbox": [ + 108, + 244, + 256, + 257 + ], + "lines": [ + { + "bbox": [ + 105, + 242, + 258, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 258, + 258 + ], + "score": 1.0, + "content": "2 PROBLEM FORMULATION", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 106, + 268, + 506, + 369 + ], + "lines": [ + { + "bbox": [ + 106, + 269, + 507, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 269, + 507, + 282 + ], + "score": 1.0, + "content": "We first introduce the basic setup for performing causal survival analysis in observational studies.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 280, + 507, + 293 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 177, + 293 + ], + "score": 1.0, + "content": "Suppose we have", + "type": "text" + }, + { + "bbox": [ + 177, + 281, + 187, + 290 + ], + "score": 0.78, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 280, + 232, + 293 + ], + "score": 1.0, + "content": "units, with", + "type": "text" + }, + { + "bbox": [ + 233, + 281, + 246, + 291 + ], + "score": 0.89, + "content": "N _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 280, + 340, + 293 + ], + "score": 1.0, + "content": "units being treated and", + "type": "text" + }, + { + "bbox": [ + 340, + 281, + 354, + 291 + ], + "score": 0.89, + "content": "N _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 280, + 437, + 293 + ], + "score": 1.0, + "content": "in the control group", + "type": "text" + }, + { + "bbox": [ + 438, + 280, + 502, + 291 + ], + "score": 0.88, + "content": "( N = N _ { 1 } + N _ { 0 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 280, + 507, + 293 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 291, + 506, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 296, + 304 + ], + "score": 1.0, + "content": "For each unit (individual), we have covariates", + "type": "text" + }, + { + "bbox": [ + 297, + 292, + 307, + 301 + ], + "score": 0.82, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 291, + 506, + 304 + ], + "score": 1.0, + "content": ", which can be heterogeneous, e.g., a mixture of", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 302, + 506, + 315 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 506, + 315 + ], + "score": 1.0, + "content": "categorical and continuous covariates which, in the context of medicine, may include labs, vitals,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 312, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 312, + 347, + 326 + ], + "score": 1.0, + "content": "procedure codes, etc. We also have a treatment indicator", + "type": "text" + }, + { + "bbox": [ + 348, + 313, + 356, + 323 + ], + "score": 0.77, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 312, + 389, + 326 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 389, + 313, + 419, + 323 + ], + "score": 0.9, + "content": "A = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 312, + 506, + 326 + ], + "score": 1.0, + "content": "for the controls and", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 107, + 323, + 506, + 338 + ], + "spans": [ + { + "bbox": [ + 107, + 324, + 134, + 334 + ], + "score": 0.9, + "content": "A = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 323, + 368, + 338 + ], + "score": 1.0, + "content": "for the treated, as well as the outcome (event) of interest", + "type": "text" + }, + { + "bbox": [ + 368, + 325, + 376, + 334 + ], + "score": 0.76, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 323, + 506, + 338 + ], + "score": 1.0, + "content": ". Under the potential-outcomes", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 334, + 505, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 334, + 223, + 348 + ], + "score": 1.0, + "content": "framework (Rubin, 2005), let", + "type": "text" + }, + { + "bbox": [ + 223, + 335, + 235, + 346 + ], + "score": 0.88, + "content": "T _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 334, + 253, + 348 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 253, + 335, + 264, + 346 + ], + "score": 0.88, + "content": "T _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 265, + 334, + 505, + 348 + ], + "score": 1.0, + "content": "be the potential event times for a given subject under control", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 344, + 507, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 507, + 360 + ], + "score": 1.0, + "content": "and treatment, respectively. In practice we only observe one realization of the potential outcomes,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 357, + 464, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 357, + 205, + 370 + ], + "score": 1.0, + "content": "i.e., the factual outcome", + "type": "text" + }, + { + "bbox": [ + 205, + 357, + 239, + 368 + ], + "score": 0.91, + "content": "T = T _ { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 357, + 379, + 370 + ], + "score": 1.0, + "content": ", while the counterfactual outcome", + "type": "text" + }, + { + "bbox": [ + 380, + 357, + 403, + 368 + ], + "score": 0.92, + "content": "T _ { 1 - A }", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 357, + 464, + 370 + ], + "score": 1.0, + "content": "is unobserved.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 269, + 507, + 370 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 373, + 505, + 429 + ], + "lines": [ + { + "bbox": [ + 106, + 373, + 506, + 387 + ], + "spans": [ + { + "bbox": [ + 106, + 373, + 506, + 387 + ], + "score": 1.0, + "content": "In survival analysis, the problem becomes more difficult because we do not always observe the exact", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 384, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 505, + 397 + ], + "score": 1.0, + "content": "event time for each individual, but rather the time up to which we are certain that the event has not", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 396, + 506, + 409 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 506, + 409 + ], + "score": 1.0, + "content": "occurred; specifically, we have a (right) censoring problem, most likely due to the loss of follow-up.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 406, + 505, + 420 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 238, + 420 + ], + "score": 1.0, + "content": "We denote the censoring time as", + "type": "text" + }, + { + "bbox": [ + 238, + 407, + 247, + 417 + ], + "score": 0.82, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 406, + 355, + 420 + ], + "score": 1.0, + "content": "and censoring indicator as", + "type": "text" + }, + { + "bbox": [ + 356, + 407, + 398, + 419 + ], + "score": 0.92, + "content": "\\delta \\in \\{ 0 , 1 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 406, + 505, + 420 + ], + "score": 1.0, + "content": ". The actual observed time", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 417, + 460, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 116, + 430 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 116, + 418, + 187, + 430 + ], + "score": 0.93, + "content": "Y = \\operatorname* { m i n } ( T _ { A } , C )", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 417, + 378, + 430 + ], + "score": 1.0, + "content": ", i.e., the outcome is observed (non-censored) if", + "type": "text" + }, + { + "bbox": [ + 378, + 418, + 413, + 429 + ], + "score": 0.91, + "content": "T _ { A } < C", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 417, + 431, + 430 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 431, + 418, + 455, + 428 + ], + "score": 0.9, + "content": "\\delta = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 417, + 460, + 430 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21, + "bbox_fs": [ + 105, + 373, + 506, + 430 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 434, + 505, + 512 + ], + "lines": [ + { + "bbox": [ + 104, + 433, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 104, + 433, + 396, + 447 + ], + "score": 1.0, + "content": "In this work, we are interested in the expected difference between the", + "type": "text" + }, + { + "bbox": [ + 397, + 435, + 408, + 446 + ], + "score": 0.88, + "content": "T _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 433, + 428, + 447 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 428, + 435, + 439, + 446 + ], + "score": 0.87, + "content": "T _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 433, + 506, + 447 + ], + "score": 1.0, + "content": "conditioned on", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 107, + 446, + 506, + 458 + ], + "spans": [ + { + "bbox": [ + 107, + 446, + 117, + 456 + ], + "score": 0.82, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 446, + 506, + 458 + ], + "score": 1.0, + "content": "for a given unit (individual), which is commonly known as the individualized treatment effect", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 456, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 444, + 469 + ], + "score": 1.0, + "content": "(ITE). Specifically, we wish to perform inference on the conditional distributions of", + "type": "text" + }, + { + "bbox": [ + 444, + 457, + 456, + 468 + ], + "score": 0.89, + "content": "T _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 456, + 474, + 469 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 474, + 457, + 485, + 468 + ], + "score": 0.87, + "content": "T _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 456, + 506, + 469 + ], + "score": 1.0, + "content": ", i.e.,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 107, + 467, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 107, + 468, + 142, + 479 + ], + "score": 0.9, + "content": "p ( T _ { 1 } | X )", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 467, + 160, + 480 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 160, + 468, + 196, + 479 + ], + "score": 0.92, + "content": "p ( T _ { 0 } | X )", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 467, + 435, + 480 + ], + "score": 1.0, + "content": ", respectively, as shown in Figure 1a. In practice, we observe", + "type": "text" + }, + { + "bbox": [ + 435, + 468, + 446, + 477 + ], + "score": 0.83, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 467, + 506, + 480 + ], + "score": 1.0, + "content": "realizations of", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 477, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 106, + 479, + 154, + 490 + ], + "score": 0.91, + "content": "( Y , \\delta , X , A )", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 477, + 506, + 492 + ], + "score": 1.0, + "content": "for observed time, censoring indicator, covariates and treatment indicator, respectively;", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 103, + 486, + 508, + 505 + ], + "spans": [ + { + "bbox": [ + 103, + 486, + 355, + 505 + ], + "score": 1.0, + "content": "hence, from an observational study the dataset takes the form", + "type": "text" + }, + { + "bbox": [ + 356, + 489, + 456, + 502 + ], + "score": 0.92, + "content": "\\mathcal { D } = \\{ ( y _ { i } , \\delta _ { i } , x _ { i } , a _ { i } ) \\} _ { i = 1 } ^ { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 486, + 508, + 505 + ], + "score": 1.0, + "content": ". Below, we", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 500, + 372, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 372, + 513 + ], + "score": 1.0, + "content": "discuss several common choices of estimands in survival analysis.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27, + "bbox_fs": [ + 103, + 433, + 508, + 513 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 518, + 506, + 621 + ], + "lines": [ + { + "bbox": [ + 105, + 518, + 506, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 506, + 532 + ], + "score": 1.0, + "content": "Estimands of Interest We begin by considering survival analysis in the absence of an intervening", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 531, + 506, + 544 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 178, + 544 + ], + "score": 1.0, + "content": "treatment choice,", + "type": "text" + }, + { + "bbox": [ + 178, + 532, + 187, + 541 + ], + "score": 0.68, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 531, + 207, + 544 + ], + "score": 1.0, + "content": ". Let", + "type": "text" + }, + { + "bbox": [ + 207, + 531, + 320, + 543 + ], + "score": 0.9, + "content": "F ( t | x ) \\triangleq P ( T \\leq t | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 531, + 506, + 544 + ], + "score": 1.0, + "content": "be the cumulative distribution function of the", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 543, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 185, + 555 + ], + "score": 1.0, + "content": "event (failure) time,", + "type": "text" + }, + { + "bbox": [ + 186, + 544, + 190, + 553 + ], + "score": 0.7, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 543, + 335, + 555 + ], + "score": 1.0, + "content": ", given a realization of the covariates,", + "type": "text" + }, + { + "bbox": [ + 336, + 545, + 342, + 552 + ], + "score": 0.71, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 543, + 505, + 555 + ], + "score": 1.0, + "content": ". Survival analysis is primarily concerned", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 554, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 386, + 567 + ], + "score": 1.0, + "content": "with characterization of the survival function conditioned on covariates", + "type": "text" + }, + { + "bbox": [ + 386, + 555, + 470, + 567 + ], + "score": 0.91, + "content": "S ( t | x ) \\triangleq 1 - F ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 554, + 506, + 567 + ], + "score": 1.0, + "content": ", and the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 565, + 505, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 226, + 578 + ], + "score": 1.0, + "content": "hazard function or risk score,", + "type": "text" + }, + { + "bbox": [ + 227, + 566, + 253, + 578 + ], + "score": 0.92, + "content": "\\lambda ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 565, + 318, + 578 + ], + "score": 1.0, + "content": ", defined below.", + "type": "text" + }, + { + "bbox": [ + 319, + 567, + 346, + 578 + ], + "score": 0.92, + "content": "S ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 565, + 505, + 578 + ], + "score": 1.0, + "content": "is a monotonically decreasing function", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 577, + 505, + 589 + ], + "spans": [ + { + "bbox": [ + 106, + 577, + 300, + 589 + ], + "score": 1.0, + "content": "indicating the probability of survival up to time", + "type": "text" + }, + { + "bbox": [ + 300, + 579, + 304, + 586 + ], + "score": 0.73, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 577, + 505, + 589 + ], + "score": 1.0, + "content": ". The hazard function measures the instantaneous", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 587, + 506, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 284, + 600 + ], + "score": 1.0, + "content": "probability of the event occurring between", + "type": "text" + }, + { + "bbox": [ + 284, + 589, + 330, + 600 + ], + "score": 0.91, + "content": "\\{ t , t + \\Delta t \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 587, + 357, + 600 + ], + "score": 1.0, + "content": "given", + "type": "text" + }, + { + "bbox": [ + 358, + 588, + 384, + 598 + ], + "score": 0.87, + "content": "T > t", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 587, + 403, + 600 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 403, + 588, + 438, + 597 + ], + "score": 0.86, + "content": "\\Delta t 0", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 587, + 506, + 600 + ], + "score": 1.0, + "content": ". From standard", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 598, + 506, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 506, + 611 + ], + "score": 1.0, + "content": "definitions (Kleinbaum & Klein, 2010), the relationship between cumulative and hazard function is", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 609, + 164, + 621 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 164, + 621 + ], + "score": 1.0, + "content": "formulated as", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 35, + "bbox_fs": [ + 105, + 518, + 506, + 621 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 161, + 624, + 451, + 652 + ], + "lines": [ + { + "bbox": [ + 161, + 624, + 451, + 652 + ], + "spans": [ + { + "bbox": [ + 161, + 624, + 451, + 652 + ], + "score": 0.95, + "content": "\\lambda ( t | x ) = \\operatorname* { l i m } _ { d t 0 } { \\frac { P ( t < T < t + d t | X = x ) } { P ( T > t | X = x ) d t } } = - { \\frac { d \\log S ( t | x ) } { d t } } = { \\frac { f ( t | x ) } { S ( t | x ) } } .", + "type": "interline_equation", + "image_path": "89816b6b4ab5614eea0ade34840cd19c6b9c6faa995393a64ef37e851560bce4.jpg" + } + ] + } + ], + "index": 41, + "virtual_lines": [ + { + "bbox": [ + 161, + 624, + 451, + 633.3333333333334 + ], + "spans": [], + "index": 40 + }, + { + "bbox": [ + 161, + 633.3333333333334, + 451, + 642.6666666666667 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 161, + 642.6666666666667, + 451, + 652.0000000000001 + ], + "spans": [], + "index": 42 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 659, + 504, + 682 + ], + "lines": [ + { + "bbox": [ + 105, + 658, + 506, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 197, + 673 + ], + "score": 1.0, + "content": "From (1) we see that", + "type": "text" + }, + { + "bbox": [ + 197, + 660, + 382, + 672 + ], + "score": 0.92, + "content": "f ( t | x ) \\triangleq P ( T = t | X = x ) = \\lambda ( t | x ) S ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 382, + 658, + 506, + 673 + ], + "score": 1.0, + "content": ", is the conditional event time", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 670, + 289, + 683 + ], + "spans": [ + { + "bbox": [ + 106, + 670, + 223, + 683 + ], + "score": 1.0, + "content": "density function (Kleinbaum", + "type": "text" + }, + { + "bbox": [ + 223, + 673, + 231, + 681 + ], + "score": 0.69, + "content": "\\&", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 670, + 289, + 683 + ], + "score": 1.0, + "content": "Klein, 2010).", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5, + "bbox_fs": [ + 105, + 658, + 506, + 683 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 688, + 506, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 213, + 699 + ], + "score": 1.0, + "content": "Given the binary treatment", + "type": "text" + }, + { + "bbox": [ + 214, + 690, + 221, + 698 + ], + "score": 0.87, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 688, + 506, + 699 + ], + "score": 1.0, + "content": ", we are interested in its impact on the survival time. For ITE estimation,", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 699, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 407, + 711 + ], + "score": 1.0, + "content": "we are also interested in the difference between the two potential outcomes", + "type": "text" + }, + { + "bbox": [ + 407, + 701, + 433, + 710 + ], + "score": 0.88, + "content": "T _ { 1 } , T _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 699, + 453, + 711 + ], + "score": 1.0, + "content": ". Let", + "type": "text" + }, + { + "bbox": [ + 453, + 700, + 487, + 711 + ], + "score": 0.91, + "content": "S _ { A } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 699, + 506, + 711 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 107, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 107, + 711, + 140, + 722 + ], + "score": 0.92, + "content": "\\lambda _ { A } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 709, + 424, + 722 + ], + "score": 1.0, + "content": "denote the survival and hazard functions for the potential outcomes", + "type": "text" + }, + { + "bbox": [ + 424, + 711, + 437, + 721 + ], + "score": 0.88, + "content": "T _ { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 709, + 460, + 722 + ], + "score": 1.0, + "content": ", i.e.,", + "type": "text" + }, + { + "bbox": [ + 460, + 711, + 471, + 721 + ], + "score": 0.87, + "content": "T _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 709, + 491, + 722 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 492, + 711, + 502, + 721 + ], + "score": 0.87, + "content": "T _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 709, + 506, + 722 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 719, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 719, + 505, + 734 + ], + "score": 1.0, + "content": "Several common estimands of interest include (Zhao et al., 2012; Trinquart et al., 2016): difference", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 102, + 77, + 509, + 100 + ], + "spans": [ + { + "bbox": [ + 102, + 77, + 188, + 100 + ], + "score": 1.0, + "content": "in expected lifetime:", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 189, + 81, + 450, + 95 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\mathrm { { I T E } } ( t , x ) = \\int _ { 0 } ^ { t _ { \\operatorname* { m a x } } } \\{ S _ { 1 } ( t | x ) - S _ { 0 } ( t | x ) \\} \\mathrm { d } t = \\mathbb { E } \\{ T _ { 1 } - T _ { 0 } | X = x \\} . } \\end{array}", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 450, + 77, + 509, + 100 + ], + "score": 1.0, + "content": ", difference in", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 507, + 108 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 181, + 108 + ], + "score": 1.0, + "content": "survival function:", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 181, + 95, + 307, + 106 + ], + "score": 0.9, + "content": "\\mathrm { I T E } ( t , x ) = S _ { 1 } ( t | x ) - S _ { 0 } ( t | x )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 307, + 93, + 385, + 108 + ], + "score": 1.0, + "content": ", and hazard ratio:", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 385, + 94, + 503, + 106 + ], + "score": 0.89, + "content": "\\mathrm { I T E } ( t , x ) = \\lambda _ { 1 } ( t | x ) / \\lambda _ { 0 } ( t | x )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 503, + 93, + 507, + 108 + ], + "score": 1.0, + "content": ".", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 106, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 506, + 117 + ], + "score": 1.0, + "content": "The inference difficulties associated with the above estimands from observational data are two-fold.", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 117, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 117, + 505, + 128 + ], + "score": 1.0, + "content": "First, there are confounders affecting both the treatment assignment and outcomes, which stem from", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 127, + 507, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 507, + 141 + ], + "score": 1.0, + "content": "selection bias, i.e., the treatment and control covariate distributions are not necessarily the same.", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 507, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 507, + 151 + ], + "score": 1.0, + "content": "Also, we do not have direct knowledge of the conditional treatment assignment mechanism, i.e.,", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 107, + 148, + 506, + 162 + ], + "spans": [ + { + "bbox": [ + 107, + 149, + 179, + 162 + ], + "score": 0.9, + "content": "P ( A = a | X = x )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 180, + 148, + 506, + 162 + ], + "score": 1.0, + "content": ", also known as the propensity score. Let ⊥⊥ denote statistical independence. For", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 160, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 160, + 431, + 173 + ], + "score": 1.0, + "content": "estimands to be identifiable from observational data, we make two assumptions: (i)", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 431, + 160, + 502, + 172 + ], + "score": 0.85, + "content": "\\{ T _ { 1 } , T _ { 0 } \\} \\perp \\perp A | X", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 503, + 160, + 506, + 173 + ], + "score": 1.0, + "content": ",", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 171, + 506, + 183 + ], + "spans": [ + { + "bbox": [ + 104, + 171, + 312, + 183 + ], + "score": 1.0, + "content": "i.e., no unobserved confounders or ignorability, and", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 312, + 172, + 327, + 182 + ], + "score": 0.44, + "content": "( i i )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 327, + 171, + 455, + 183 + ], + "score": 1.0, + "content": "overlap in the covariate support", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 455, + 172, + 506, + 183 + ], + "score": 0.88, + "content": "0 < P ( A =", + "type": "inline_equation", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 182, + 506, + 195 + ], + "spans": [ + { + "bbox": [ + 106, + 182, + 165, + 194 + ], + "score": 0.9, + "content": "1 | X = x ) < 1", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 165, + 182, + 232, + 195 + ], + "score": 1.0, + "content": "almost surely if", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 232, + 182, + 293, + 194 + ], + "score": 0.92, + "content": "p ( X = x ) > 0", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 293, + 182, + 506, + 195 + ], + "score": 1.0, + "content": ". Second, the censoring mechanism is also unknown", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 193, + 506, + 206 + ], + "spans": [ + { + "bbox": [ + 105, + 193, + 506, + 206 + ], + "score": 1.0, + "content": "and may lead to bias without proper adjustment. We consider two censoring mechanisms in our", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 104, + 203, + 506, + 217 + ], + "spans": [ + { + "bbox": [ + 104, + 203, + 132, + 217 + ], + "score": 1.0, + "content": "work,", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 133, + 205, + 144, + 216 + ], + "score": 0.7, + "content": "( i )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 145, + 203, + 363, + 217 + ], + "score": 1.0, + "content": "conditionally independent or informative censoring:", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 363, + 204, + 421, + 216 + ], + "score": 0.91, + "content": "T \\perp \\perp C | X , A", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 422, + 203, + 443, + 217 + ], + "score": 1.0, + "content": ", and", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 444, + 205, + 458, + 215 + ], + "score": 0.8, + "content": "( i i )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 459, + 203, + 506, + 217 + ], + "score": 1.0, + "content": "random or", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 215, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 219, + 228 + ], + "score": 1.0, + "content": "non-informative censoring:", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 220, + 215, + 253, + 226 + ], + "score": 0.32, + "content": "T \\perp \\perp C", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 253, + 215, + 506, + 228 + ], + "score": 1.0, + "content": ". Note that for informative censoring, we also have to consider", + "type": "text", + "cross_page": true + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 226, + 506, + 239 + ], + "spans": [ + { + "bbox": [ + 106, + 226, + 213, + 239 + ], + "score": 1.0, + "content": "potential censoring times", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 213, + 227, + 226, + 237 + ], + "score": 0.88, + "content": "C _ { 1 }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 227, + 226, + 246, + 239 + ], + "score": 1.0, + "content": "and", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 246, + 226, + 259, + 237 + ], + "score": 0.88, + "content": "C _ { 0 }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 259, + 226, + 353, + 239 + ], + "score": 1.0, + "content": "and their conditionals", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 354, + 226, + 391, + 238 + ], + "score": 0.93, + "content": "p ( C _ { 1 } | X )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 392, + 226, + 411, + 239 + ], + "score": 1.0, + "content": "and", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 411, + 226, + 448, + 238 + ], + "score": 0.93, + "content": "p ( C _ { 0 } | X )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 449, + 226, + 506, + 239 + ], + "score": 1.0, + "content": ", respectively.", + "type": "text", + "cross_page": true + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 237, + 389, + 250 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 389, + 250 + ], + "score": 1.0, + "content": "Figure 1 shows causal graphs illustrating these modeling assumptions.", + "type": "text", + "cross_page": true + } + ], + "index": 14 + } + ], + "index": 46.5, + "bbox_fs": [ + 105, + 688, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 80, + 506, + 249 + ], + "lines": [ + { + "bbox": [ + 102, + 77, + 509, + 100 + ], + "spans": [ + { + "bbox": [ + 102, + 77, + 188, + 100 + ], + "score": 1.0, + "content": "in expected lifetime:", + "type": "text" + }, + { + "bbox": [ + 189, + 81, + 450, + 95 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\mathrm { { I T E } } ( t , x ) = \\int _ { 0 } ^ { t _ { \\operatorname* { m a x } } } \\{ S _ { 1 } ( t | x ) - S _ { 0 } ( t | x ) \\} \\mathrm { d } t = \\mathbb { E } \\{ T _ { 1 } - T _ { 0 } | X = x \\} . } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 77, + 509, + 100 + ], + "score": 1.0, + "content": ", difference in", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 507, + 108 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 181, + 108 + ], + "score": 1.0, + "content": "survival function:", + "type": "text" + }, + { + "bbox": [ + 181, + 95, + 307, + 106 + ], + "score": 0.9, + "content": "\\mathrm { I T E } ( t , x ) = S _ { 1 } ( t | x ) - S _ { 0 } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 93, + 385, + 108 + ], + "score": 1.0, + "content": ", and hazard ratio:", + "type": "text" + }, + { + "bbox": [ + 385, + 94, + 503, + 106 + ], + "score": 0.89, + "content": "\\mathrm { I T E } ( t , x ) = \\lambda _ { 1 } ( t | x ) / \\lambda _ { 0 } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 93, + 507, + 108 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 106, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 506, + 117 + ], + "score": 1.0, + "content": "The inference difficulties associated with the above estimands from observational data are two-fold.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 117, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 117, + 505, + 128 + ], + "score": 1.0, + "content": "First, there are confounders affecting both the treatment assignment and outcomes, which stem from", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 127, + 507, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 507, + 141 + ], + "score": 1.0, + "content": "selection bias, i.e., the treatment and control covariate distributions are not necessarily the same.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 507, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 507, + 151 + ], + "score": 1.0, + "content": "Also, we do not have direct knowledge of the conditional treatment assignment mechanism, i.e.,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 107, + 148, + 506, + 162 + ], + "spans": [ + { + "bbox": [ + 107, + 149, + 179, + 162 + ], + "score": 0.9, + "content": "P ( A = a | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 148, + 506, + 162 + ], + "score": 1.0, + "content": ", also known as the propensity score. Let ⊥⊥ denote statistical independence. For", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 160, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 160, + 431, + 173 + ], + "score": 1.0, + "content": "estimands to be identifiable from observational data, we make two assumptions: (i)", + "type": "text" + }, + { + "bbox": [ + 431, + 160, + 502, + 172 + ], + "score": 0.85, + "content": "\\{ T _ { 1 } , T _ { 0 } \\} \\perp \\perp A | X", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 160, + 506, + 173 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 171, + 506, + 183 + ], + "spans": [ + { + "bbox": [ + 104, + 171, + 312, + 183 + ], + "score": 1.0, + "content": "i.e., no unobserved confounders or ignorability, and", + "type": "text" + }, + { + "bbox": [ + 312, + 172, + 327, + 182 + ], + "score": 0.44, + "content": "( i i )", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 171, + 455, + 183 + ], + "score": 1.0, + "content": "overlap in the covariate support", + "type": "text" + }, + { + "bbox": [ + 455, + 172, + 506, + 183 + ], + "score": 0.88, + "content": "0 < P ( A =", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 182, + 506, + 195 + ], + "spans": [ + { + "bbox": [ + 106, + 182, + 165, + 194 + ], + "score": 0.9, + "content": "1 | X = x ) < 1", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 182, + 232, + 195 + ], + "score": 1.0, + "content": "almost surely if", + "type": "text" + }, + { + "bbox": [ + 232, + 182, + 293, + 194 + ], + "score": 0.92, + "content": "p ( X = x ) > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 182, + 506, + 195 + ], + "score": 1.0, + "content": ". Second, the censoring mechanism is also unknown", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 193, + 506, + 206 + ], + "spans": [ + { + "bbox": [ + 105, + 193, + 506, + 206 + ], + "score": 1.0, + "content": "and may lead to bias without proper adjustment. We consider two censoring mechanisms in our", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 104, + 203, + 506, + 217 + ], + "spans": [ + { + "bbox": [ + 104, + 203, + 132, + 217 + ], + "score": 1.0, + "content": "work,", + "type": "text" + }, + { + "bbox": [ + 133, + 205, + 144, + 216 + ], + "score": 0.7, + "content": "( i )", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 203, + 363, + 217 + ], + "score": 1.0, + "content": "conditionally independent or informative censoring:", + "type": "text" + }, + { + "bbox": [ + 363, + 204, + 421, + 216 + ], + "score": 0.91, + "content": "T \\perp \\perp C | X , A", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 203, + 443, + 217 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 444, + 205, + 458, + 215 + ], + "score": 0.8, + "content": "( i i )", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 203, + 506, + 217 + ], + "score": 1.0, + "content": "random or", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 215, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 219, + 228 + ], + "score": 1.0, + "content": "non-informative censoring:", + "type": "text" + }, + { + "bbox": [ + 220, + 215, + 253, + 226 + ], + "score": 0.32, + "content": "T \\perp \\perp C", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 215, + 506, + 228 + ], + "score": 1.0, + "content": ". Note that for informative censoring, we also have to consider", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 226, + 506, + 239 + ], + "spans": [ + { + "bbox": [ + 106, + 226, + 213, + 239 + ], + "score": 1.0, + "content": "potential censoring times", + "type": "text" + }, + { + "bbox": [ + 213, + 227, + 226, + 237 + ], + "score": 0.88, + "content": "C _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 226, + 246, + 239 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 246, + 226, + 259, + 237 + ], + "score": 0.88, + "content": "C _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 226, + 353, + 239 + ], + "score": 1.0, + "content": "and their conditionals", + "type": "text" + }, + { + "bbox": [ + 354, + 226, + 391, + 238 + ], + "score": 0.93, + "content": "p ( C _ { 1 } | X )", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 226, + 411, + 239 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 411, + 226, + 448, + 238 + ], + "score": 0.93, + "content": "p ( C _ { 0 } | X )", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 226, + 506, + 239 + ], + "score": 1.0, + "content": ", respectively.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 237, + 389, + 250 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 389, + 250 + ], + "score": 1.0, + "content": "Figure 1 shows causal graphs illustrating these modeling assumptions.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 7 + }, + { + "type": "title", + "bbox": [ + 107, + 262, + 184, + 275 + ], + "lines": [ + { + "bbox": [ + 104, + 260, + 186, + 278 + ], + "spans": [ + { + "bbox": [ + 104, + 260, + 186, + 278 + ], + "score": 1.0, + "content": "3 MODELING", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 107, + 284, + 505, + 428 + ], + "lines": [ + { + "bbox": [ + 105, + 284, + 506, + 297 + ], + "spans": [ + { + "bbox": [ + 105, + 284, + 506, + 297 + ], + "score": 1.0, + "content": "To overcome the above challenges and adjust for observational biases, we propose a unified frame-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 295, + 506, + 308 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 506, + 308 + ], + "score": 1.0, + "content": "work for counterfactual survival analysis (CSA). Specifically, we repurpose the counterfactual bound", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 306, + 506, + 319 + ], + "spans": [ + { + "bbox": [ + 106, + 306, + 506, + 319 + ], + "score": 1.0, + "content": "in Shalit et al. (2017) for our time-to-event scenario and introduce a nonparametric approach for", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 317, + 505, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 505, + 330 + ], + "score": 1.0, + "content": "stochastic survival outcome predictions. Below we formulate a theoretically grounded and unified", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 328, + 507, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 207, + 342 + ], + "score": 1.0, + "content": "approach for estimating", + "type": "text" + }, + { + "bbox": [ + 207, + 329, + 219, + 340 + ], + "score": 0.77, + "content": "( i )", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 328, + 307, + 342 + ], + "score": 1.0, + "content": "the encoder function", + "type": "text" + }, + { + "bbox": [ + 307, + 329, + 348, + 339 + ], + "score": 0.9, + "content": "r = \\Phi ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 328, + 507, + 342 + ], + "score": 1.0, + "content": ", which deterministically maps covari-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 338, + 506, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 125, + 353 + ], + "score": 1.0, + "content": "ates", + "type": "text" + }, + { + "bbox": [ + 125, + 341, + 132, + 349 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 338, + 314, + 353 + ], + "score": 1.0, + "content": "to their corresponding latent representation", + "type": "text" + }, + { + "bbox": [ + 315, + 339, + 347, + 349 + ], + "score": 0.86, + "content": "r \\in { \\mathbb { R } } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 338, + 369, + 353 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 370, + 339, + 385, + 351 + ], + "score": 0.79, + "content": "( i i )", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 338, + 506, + 353 + ], + "score": 1.0, + "content": "two stochastic time-to-event", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 350, + 506, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 194, + 363 + ], + "score": 1.0, + "content": "generative functions,", + "type": "text" + }, + { + "bbox": [ + 194, + 350, + 218, + 362 + ], + "score": 0.91, + "content": "h _ { A } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 350, + 506, + 363 + ], + "score": 1.0, + "content": ", to implicitly draw samples from both potential outcome conditionals", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 360, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 106, + 361, + 200, + 373 + ], + "score": 0.9, + "content": "\\bar { t } _ { a } \\sim p _ { h , \\Phi } ( T _ { A } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 360, + 218, + 375 + ], + "score": 1.0, + "content": ", for", + "type": "text" + }, + { + "bbox": [ + 219, + 361, + 266, + 373 + ], + "score": 0.93, + "content": "A = \\{ 1 , 0 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 360, + 314, + 375 + ], + "score": 1.0, + "content": ", and where", + "type": "text" + }, + { + "bbox": [ + 315, + 362, + 324, + 372 + ], + "score": 0.87, + "content": "t _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 360, + 434, + 375 + ], + "score": 1.0, + "content": "indicates the sample from", + "type": "text" + }, + { + "bbox": [ + 434, + 361, + 505, + 374 + ], + "score": 0.91, + "content": "p _ { h , \\Phi } ( T _ { A } | X = x )", + "type": "inline_equation" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 371, + 506, + 386 + ], + "spans": [ + { + "bbox": [ + 104, + 371, + 131, + 386 + ], + "score": 1.0, + "content": "is for", + "type": "text" + }, + { + "bbox": [ + 131, + 374, + 160, + 383 + ], + "score": 0.88, + "content": "A = a", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 371, + 506, + 386 + ], + "score": 1.0, + "content": ". Further, we formulate a general extension that accounts for informative censoring", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 383, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 383, + 358, + 396 + ], + "score": 1.0, + "content": "by introducing two stochastic censoring generative functions,", + "type": "text" + }, + { + "bbox": [ + 358, + 383, + 380, + 395 + ], + "score": 0.9, + "content": "\\nu _ { A } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 383, + 505, + 396 + ], + "score": 1.0, + "content": ", to draw samples for potential", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 394, + 506, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 394, + 172, + 408 + ], + "score": 1.0, + "content": "censoring times", + "type": "text" + }, + { + "bbox": [ + 173, + 394, + 266, + 407 + ], + "score": 0.91, + "content": "c _ { a } \\sim p _ { \\nu , \\Phi } ( C _ { A } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 394, + 404, + 408 + ], + "score": 1.0, + "content": ". The model-specifying functions,", + "type": "text" + }, + { + "bbox": [ + 405, + 394, + 487, + 406 + ], + "score": 0.92, + "content": "\\{ h _ { A } ( \\cdot ) , \\nu _ { A } ( \\cdot ) , \\Phi ( \\cdot ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 394, + 506, + 408 + ], + "score": 1.0, + "content": ", are", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 405, + 506, + 419 + ], + "spans": [ + { + "bbox": [ + 104, + 405, + 506, + 419 + ], + "score": 1.0, + "content": "parameterized via neural networks. See the Supplementary Material (SM) for details. Figure 1a", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 417, + 254, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 254, + 428 + ], + "score": 1.0, + "content": "summarizes our modeling approach.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 434, + 503, + 457 + ], + "lines": [ + { + "bbox": [ + 106, + 434, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 106, + 434, + 232, + 446 + ], + "score": 1.0, + "content": "Accounting for selection bias", + "type": "text" + }, + { + "bbox": [ + 237, + 434, + 505, + 447 + ], + "score": 1.0, + "content": "We wish to estimate the potential outcomes, i.e., event times, which", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 444, + 410, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 444, + 333, + 459 + ], + "score": 1.0, + "content": "are sampled by distributions parameterized by functions", + "type": "text" + }, + { + "bbox": [ + 333, + 446, + 388, + 458 + ], + "score": 0.93, + "content": "\\{ h _ { A } ( \\cdot ) , \\Phi ( \\cdot ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 444, + 410, + 459 + ], + "score": 1.0, + "content": ", i.e.,", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5 + }, + { + "type": "interline_equation", + "bbox": [ + 246, + 462, + 365, + 493 + ], + "lines": [ + { + "bbox": [ + 246, + 462, + 365, + 493 + ], + "spans": [ + { + "bbox": [ + 246, + 462, + 365, + 493 + ], + "score": 0.92, + "content": "\\begin{array} { c } { { t \\sim p _ { h , \\Phi } ( T | X = x , A = a ) } } \\\\ { { { \\it t } _ { a } \\sim p _ { h , \\Phi } ( T _ { a } | X = x ) } } \\end{array}", + "type": "interline_equation", + "image_path": "9e497767f6b296dac4c8cf83d111947941f15982093bbf8fbe2814a7e006faf0.jpg" + } + ] + } + ], + "index": 31.5, + "virtual_lines": [ + { + "bbox": [ + 246, + 462, + 365, + 477.5 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 246, + 477.5, + 365, + 493.0 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 495, + 505, + 608 + ], + "lines": [ + { + "bbox": [ + 106, + 495, + 506, + 508 + ], + "spans": [ + { + "bbox": [ + 106, + 495, + 393, + 508 + ], + "score": 1.0, + "content": "We obtain (3) from (2) via the strong ignorability assumption, i.e.,", + "type": "text" + }, + { + "bbox": [ + 393, + 496, + 429, + 508 + ], + "score": 0.84, + "content": "\\{ T _ { 0 } , T _ { 1 } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 495, + 448, + 508 + ], + "score": 1.0, + "content": "⊥⊥", + "type": "text" + }, + { + "bbox": [ + 449, + 496, + 470, + 508 + ], + "score": 0.84, + "content": "A | X", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 495, + 506, + 508 + ], + "score": 1.0, + "content": "(consis-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 506, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 335, + 519 + ], + "score": 1.0, + "content": "tent with the causal graphs in Figure 1b and 1c) and", + "type": "text" + }, + { + "bbox": [ + 335, + 507, + 466, + 519 + ], + "score": 0.9, + "content": "0 < P ( A = \\mathsf { \\bar { a } } | X \\doteq x ) < 1 .", + "type": "inline_equation" + }, + { + "bbox": [ + 467, + 506, + 505, + 519 + ], + "score": 1.0, + "content": ", and the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 517, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 106, + 517, + 227, + 530 + ], + "score": 1.0, + "content": "consistency assumption, i.e.,", + "type": "text" + }, + { + "bbox": [ + 227, + 518, + 295, + 530 + ], + "score": 0.91, + "content": "T = T _ { A } | A = a", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 517, + 505, + 530 + ], + "score": 1.0, + "content": ". A similar argument can be made for informative", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 528, + 505, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 324, + 542 + ], + "score": 1.0, + "content": "censoring based on Figure 1c, so we can also write", + "type": "text" + }, + { + "bbox": [ + 325, + 529, + 424, + 541 + ], + "score": 0.91, + "content": "c _ { a } \\sim p _ { \\nu , \\Phi } ( C _ { A } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 528, + 505, + 542 + ], + "score": 1.0, + "content": ". Given (3), model", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 538, + 506, + 553 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 148, + 553 + ], + "score": 1.0, + "content": "functions", + "type": "text" + }, + { + "bbox": [ + 149, + 540, + 204, + 552 + ], + "score": 0.93, + "content": "\\{ h _ { A } ( \\cdot ) , \\Phi ( \\cdot ) \\bar \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 538, + 225, + 553 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 225, + 540, + 248, + 552 + ], + "score": 0.9, + "content": "\\nu _ { A } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 538, + 506, + 553 + ], + "score": 1.0, + "content": "for informative censoring can be learned by leveraging stan-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 550, + 505, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 411, + 564 + ], + "score": 1.0, + "content": "dard statistical optimization approaches, that minimize a loss hypothesis", + "type": "text" + }, + { + "bbox": [ + 411, + 551, + 419, + 561 + ], + "score": 0.81, + "content": "\\mathcal { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 550, + 505, + 564 + ], + "score": 1.0, + "content": "given samples from", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 561, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 213, + 574 + ], + "score": 1.0, + "content": "the empirical distribution", + "type": "text" + }, + { + "bbox": [ + 213, + 562, + 322, + 573 + ], + "score": 0.9, + "content": "( y , \\delta , x , a ) \\sim p ( Y , \\delta , X , A )", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 561, + 399, + 574 + ], + "score": 1.0, + "content": ", i.e., from dataset", + "type": "text" + }, + { + "bbox": [ + 399, + 562, + 408, + 572 + ], + "score": 0.8, + "content": "\\mathcal { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 561, + 505, + 574 + ], + "score": 1.0, + "content": ". Specifically, we write", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 571, + 506, + 588 + ], + "spans": [ + { + "bbox": [ + 106, + 573, + 271, + 586 + ], + "score": 0.86, + "content": "\\mathcal { L } = \\mathbb { E } _ { ( y , \\delta , x , a ) \\sim p ( Y , \\delta , X , A ) } \\left[ \\ell _ { h , \\Phi } ( t _ { a } , y , \\delta ) \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 571, + 303, + 588 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 303, + 573, + 356, + 585 + ], + "score": 0.93, + "content": "\\ell _ { h , \\Phi } ( t _ { a } , y , \\delta )", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 571, + 506, + 588 + ], + "score": 1.0, + "content": "is a loss function that measures the", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 584, + 505, + 597 + ], + "spans": [ + { + "bbox": [ + 105, + 584, + 163, + 597 + ], + "score": 1.0, + "content": "agreement of", + "type": "text" + }, + { + "bbox": [ + 163, + 585, + 256, + 597 + ], + "score": 0.89, + "content": "t _ { a } \\sim p _ { h , \\Phi } ( T _ { A } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 584, + 278, + 597 + ], + "score": 1.0, + "content": "(and", + "type": "text" + }, + { + "bbox": [ + 279, + 585, + 373, + 597 + ], + "score": 0.9, + "content": "c _ { a } \\sim p _ { \\nu , \\Phi } ( C _ { A } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 584, + 505, + 597 + ], + "score": 1.0, + "content": "for informative censoring) with", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 595, + 411, + 609 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 159, + 609 + ], + "score": 1.0, + "content": "ground truth", + "type": "text" + }, + { + "bbox": [ + 159, + 596, + 184, + 608 + ], + "score": 0.92, + "content": "\\{ y , \\delta \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 595, + 411, + 609 + ], + "score": 1.0, + "content": ", the observed time and censoring indicator, respectively.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 106, + 612, + 505, + 681 + ], + "lines": [ + { + "bbox": [ + 106, + 612, + 506, + 625 + ], + "spans": [ + { + "bbox": [ + 106, + 612, + 357, + 625 + ], + "score": 1.0, + "content": "For some parametric formulations of event time distribution", + "type": "text" + }, + { + "bbox": [ + 357, + 613, + 429, + 625 + ], + "score": 0.9, + "content": "p _ { h , \\Phi } ( T _ { A } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 612, + 506, + 625 + ], + "score": 1.0, + "content": ", e.g., exponential,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 623, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 623, + 443, + 637 + ], + "score": 1.0, + "content": "Weibull, log-Normal, etc., and provided the censoring mechanism is non-informative,", + "type": "text" + }, + { + "bbox": [ + 443, + 624, + 505, + 636 + ], + "score": 0.92, + "content": "- \\ell _ { h , \\Phi } ( t _ { a } , y , \\delta )", + "type": "inline_equation" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 636, + 506, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 308, + 649 + ], + "score": 1.0, + "content": "is the closed form log likelihood. Specifically,", + "type": "text" + }, + { + "bbox": [ + 308, + 636, + 501, + 649 + ], + "score": 0.9, + "content": "- \\ell _ { h , \\Phi } ( t _ { a } , y , \\delta ) \\triangleq \\log p _ { h , \\Phi } ( T _ { a } | X = x ) = \\delta", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 636, + 506, + 649 + ], + "score": 1.0, + "content": "·", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 646, + 506, + 661 + ], + "spans": [ + { + "bbox": [ + 106, + 647, + 272, + 660 + ], + "score": 0.91, + "content": "\\log f _ { h , \\Phi } ( t _ { a } | x ) + ( 1 - \\bar { \\delta } ) \\cdot \\log S _ { h , \\Phi } ( t _ { a } | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 646, + 506, + 661 + ], + "score": 1.0, + "content": ", which implies that the conditional event time density and", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 658, + 505, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 395, + 671 + ], + "score": 1.0, + "content": "survival functions can be calculated in closed form from transformations", + "type": "text" + }, + { + "bbox": [ + 395, + 658, + 451, + 671 + ], + "score": 0.93, + "content": "\\{ h _ { A } ( \\cdot ) , \\Phi ( \\cdot ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 658, + 462, + 671 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 463, + 660, + 469, + 668 + ], + "score": 0.76, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 658, + 505, + 671 + ], + "score": 1.0, + "content": ". See the", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 668, + 395, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 234, + 683 + ], + "score": 1.0, + "content": "SM for parametric examples of", + "type": "text" + }, + { + "bbox": [ + 234, + 670, + 241, + 679 + ], + "score": 0.82, + "content": "\\mathcal { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 668, + 395, + 683 + ], + "score": 1.0, + "content": "accounting for informative censoring.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 45.5 + }, + { + "type": "text", + "bbox": [ + 107, + 686, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 685, + 504, + 699 + ], + "spans": [ + { + "bbox": [ + 105, + 685, + 386, + 699 + ], + "score": 1.0, + "content": "We further define the expected loss for a given realization of covariates", + "type": "text" + }, + { + "bbox": [ + 387, + 689, + 393, + 696 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 685, + 497, + 699 + ], + "score": 1.0, + "content": "and treatment assignment", + "type": "text" + }, + { + "bbox": [ + 498, + 689, + 504, + 696 + ], + "score": 0.72, + "content": "a", + "type": "inline_equation" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 190, + 711 + ], + "score": 1.0, + "content": "over observed times", + "type": "text" + }, + { + "bbox": [ + 190, + 700, + 197, + 710 + ], + "score": 0.76, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 698, + 433, + 711 + ], + "score": 1.0, + "content": "(censored and non-censored), and the censoring indicator", + "type": "text" + }, + { + "bbox": [ + 433, + 699, + 439, + 709 + ], + "score": 0.77, + "content": "\\delta", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 698, + 451, + 711 + ], + "score": 1.0, + "content": "as", + "type": "text" + }, + { + "bbox": [ + 452, + 698, + 505, + 711 + ], + "score": 0.9, + "content": "\\zeta _ { h , \\Phi } ( x , a ) \\triangleq", + "type": "inline_equation" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 708, + 507, + 724 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 227, + 723 + ], + "score": 0.9, + "content": "\\mathbb { E } _ { ( y , \\delta , x ) \\sim p ( Y , \\delta | X ) } \\ell _ { h , \\Phi } ( t _ { a } , y , \\delta )", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 708, + 479, + 724 + ], + "score": 1.0, + "content": "as in Shalit et al. (2017). For a given subject with covariates", + "type": "text" + }, + { + "bbox": [ + 479, + 712, + 487, + 720 + ], + "score": 0.8, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 708, + 507, + 724 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 720, + 502, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 192, + 734 + ], + "score": 1.0, + "content": "treatment assignment", + "type": "text" + }, + { + "bbox": [ + 192, + 724, + 199, + 730 + ], + "score": 0.77, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 720, + 451, + 734 + ], + "score": 1.0, + "content": ", we wish to minimize both the factual and counterfactual losses,", + "type": "text" + }, + { + "bbox": [ + 452, + 721, + 465, + 732 + ], + "score": 0.89, + "content": "\\mathcal { L } _ { \\mathrm { F } }", + "type": "inline_equation" + }, + { + "bbox": [ + 466, + 720, + 483, + 734 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 483, + 721, + 502, + 732 + ], + "score": 0.9, + "content": "\\mathcal { L } _ { \\mathrm { C F } }", + "type": "inline_equation" + } + ], + "index": 52 + } + ], + "index": 50.5 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 759 + ], + "lines": [] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 307, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 80, + 506, + 249 + ], + "lines": [], + "index": 7, + "bbox_fs": [ + 102, + 77, + 509, + 250 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 107, + 262, + 184, + 275 + ], + "lines": [ + { + "bbox": [ + 104, + 260, + 186, + 278 + ], + "spans": [ + { + "bbox": [ + 104, + 260, + 186, + 278 + ], + "score": 1.0, + "content": "3 MODELING", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 107, + 284, + 505, + 428 + ], + "lines": [ + { + "bbox": [ + 105, + 284, + 506, + 297 + ], + "spans": [ + { + "bbox": [ + 105, + 284, + 506, + 297 + ], + "score": 1.0, + "content": "To overcome the above challenges and adjust for observational biases, we propose a unified frame-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 295, + 506, + 308 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 506, + 308 + ], + "score": 1.0, + "content": "work for counterfactual survival analysis (CSA). Specifically, we repurpose the counterfactual bound", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 306, + 506, + 319 + ], + "spans": [ + { + "bbox": [ + 106, + 306, + 506, + 319 + ], + "score": 1.0, + "content": "in Shalit et al. (2017) for our time-to-event scenario and introduce a nonparametric approach for", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 317, + 505, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 505, + 330 + ], + "score": 1.0, + "content": "stochastic survival outcome predictions. Below we formulate a theoretically grounded and unified", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 328, + 507, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 207, + 342 + ], + "score": 1.0, + "content": "approach for estimating", + "type": "text" + }, + { + "bbox": [ + 207, + 329, + 219, + 340 + ], + "score": 0.77, + "content": "( i )", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 328, + 307, + 342 + ], + "score": 1.0, + "content": "the encoder function", + "type": "text" + }, + { + "bbox": [ + 307, + 329, + 348, + 339 + ], + "score": 0.9, + "content": "r = \\Phi ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 328, + 507, + 342 + ], + "score": 1.0, + "content": ", which deterministically maps covari-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 338, + 506, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 125, + 353 + ], + "score": 1.0, + "content": "ates", + "type": "text" + }, + { + "bbox": [ + 125, + 341, + 132, + 349 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 338, + 314, + 353 + ], + "score": 1.0, + "content": "to their corresponding latent representation", + "type": "text" + }, + { + "bbox": [ + 315, + 339, + 347, + 349 + ], + "score": 0.86, + "content": "r \\in { \\mathbb { R } } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 338, + 369, + 353 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 370, + 339, + 385, + 351 + ], + "score": 0.79, + "content": "( i i )", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 338, + 506, + 353 + ], + "score": 1.0, + "content": "two stochastic time-to-event", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 350, + 506, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 194, + 363 + ], + "score": 1.0, + "content": "generative functions,", + "type": "text" + }, + { + "bbox": [ + 194, + 350, + 218, + 362 + ], + "score": 0.91, + "content": "h _ { A } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 350, + 506, + 363 + ], + "score": 1.0, + "content": ", to implicitly draw samples from both potential outcome conditionals", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 360, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 106, + 361, + 200, + 373 + ], + "score": 0.9, + "content": "\\bar { t } _ { a } \\sim p _ { h , \\Phi } ( T _ { A } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 360, + 218, + 375 + ], + "score": 1.0, + "content": ", for", + "type": "text" + }, + { + "bbox": [ + 219, + 361, + 266, + 373 + ], + "score": 0.93, + "content": "A = \\{ 1 , 0 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 360, + 314, + 375 + ], + "score": 1.0, + "content": ", and where", + "type": "text" + }, + { + "bbox": [ + 315, + 362, + 324, + 372 + ], + "score": 0.87, + "content": "t _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 360, + 434, + 375 + ], + "score": 1.0, + "content": "indicates the sample from", + "type": "text" + }, + { + "bbox": [ + 434, + 361, + 505, + 374 + ], + "score": 0.91, + "content": "p _ { h , \\Phi } ( T _ { A } | X = x )", + "type": "inline_equation" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 371, + 506, + 386 + ], + "spans": [ + { + "bbox": [ + 104, + 371, + 131, + 386 + ], + "score": 1.0, + "content": "is for", + "type": "text" + }, + { + "bbox": [ + 131, + 374, + 160, + 383 + ], + "score": 0.88, + "content": "A = a", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 371, + 506, + 386 + ], + "score": 1.0, + "content": ". Further, we formulate a general extension that accounts for informative censoring", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 383, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 383, + 358, + 396 + ], + "score": 1.0, + "content": "by introducing two stochastic censoring generative functions,", + "type": "text" + }, + { + "bbox": [ + 358, + 383, + 380, + 395 + ], + "score": 0.9, + "content": "\\nu _ { A } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 383, + 505, + 396 + ], + "score": 1.0, + "content": ", to draw samples for potential", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 394, + 506, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 394, + 172, + 408 + ], + "score": 1.0, + "content": "censoring times", + "type": "text" + }, + { + "bbox": [ + 173, + 394, + 266, + 407 + ], + "score": 0.91, + "content": "c _ { a } \\sim p _ { \\nu , \\Phi } ( C _ { A } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 394, + 404, + 408 + ], + "score": 1.0, + "content": ". The model-specifying functions,", + "type": "text" + }, + { + "bbox": [ + 405, + 394, + 487, + 406 + ], + "score": 0.92, + "content": "\\{ h _ { A } ( \\cdot ) , \\nu _ { A } ( \\cdot ) , \\Phi ( \\cdot ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 394, + 506, + 408 + ], + "score": 1.0, + "content": ", are", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 104, + 405, + 506, + 419 + ], + "spans": [ + { + "bbox": [ + 104, + 405, + 506, + 419 + ], + "score": 1.0, + "content": "parameterized via neural networks. See the Supplementary Material (SM) for details. Figure 1a", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 417, + 254, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 254, + 428 + ], + "score": 1.0, + "content": "summarizes our modeling approach.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 22, + "bbox_fs": [ + 104, + 284, + 507, + 428 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 434, + 503, + 457 + ], + "lines": [ + { + "bbox": [ + 106, + 434, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 106, + 434, + 232, + 446 + ], + "score": 1.0, + "content": "Accounting for selection bias", + "type": "text" + }, + { + "bbox": [ + 237, + 434, + 505, + 447 + ], + "score": 1.0, + "content": "We wish to estimate the potential outcomes, i.e., event times, which", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 444, + 410, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 444, + 333, + 459 + ], + "score": 1.0, + "content": "are sampled by distributions parameterized by functions", + "type": "text" + }, + { + "bbox": [ + 333, + 446, + 388, + 458 + ], + "score": 0.93, + "content": "\\{ h _ { A } ( \\cdot ) , \\Phi ( \\cdot ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 444, + 410, + 459 + ], + "score": 1.0, + "content": ", i.e.,", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 434, + 505, + 459 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 246, + 462, + 365, + 493 + ], + "lines": [ + { + "bbox": [ + 246, + 462, + 365, + 493 + ], + "spans": [ + { + "bbox": [ + 246, + 462, + 365, + 493 + ], + "score": 0.92, + "content": "\\begin{array} { c } { { t \\sim p _ { h , \\Phi } ( T | X = x , A = a ) } } \\\\ { { { \\it t } _ { a } \\sim p _ { h , \\Phi } ( T _ { a } | X = x ) } } \\end{array}", + "type": "interline_equation", + "image_path": "9e497767f6b296dac4c8cf83d111947941f15982093bbf8fbe2814a7e006faf0.jpg" + } + ] + } + ], + "index": 31.5, + "virtual_lines": [ + { + "bbox": [ + 246, + 462, + 365, + 477.5 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 246, + 477.5, + 365, + 493.0 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 495, + 505, + 608 + ], + "lines": [ + { + "bbox": [ + 106, + 495, + 506, + 508 + ], + "spans": [ + { + "bbox": [ + 106, + 495, + 393, + 508 + ], + "score": 1.0, + "content": "We obtain (3) from (2) via the strong ignorability assumption, i.e.,", + "type": "text" + }, + { + "bbox": [ + 393, + 496, + 429, + 508 + ], + "score": 0.84, + "content": "\\{ T _ { 0 } , T _ { 1 } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 495, + 448, + 508 + ], + "score": 1.0, + "content": "⊥⊥", + "type": "text" + }, + { + "bbox": [ + 449, + 496, + 470, + 508 + ], + "score": 0.84, + "content": "A | X", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 495, + 506, + 508 + ], + "score": 1.0, + "content": "(consis-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 506, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 335, + 519 + ], + "score": 1.0, + "content": "tent with the causal graphs in Figure 1b and 1c) and", + "type": "text" + }, + { + "bbox": [ + 335, + 507, + 466, + 519 + ], + "score": 0.9, + "content": "0 < P ( A = \\mathsf { \\bar { a } } | X \\doteq x ) < 1 .", + "type": "inline_equation" + }, + { + "bbox": [ + 467, + 506, + 505, + 519 + ], + "score": 1.0, + "content": ", and the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 517, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 106, + 517, + 227, + 530 + ], + "score": 1.0, + "content": "consistency assumption, i.e.,", + "type": "text" + }, + { + "bbox": [ + 227, + 518, + 295, + 530 + ], + "score": 0.91, + "content": "T = T _ { A } | A = a", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 517, + 505, + 530 + ], + "score": 1.0, + "content": ". A similar argument can be made for informative", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 528, + 505, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 324, + 542 + ], + "score": 1.0, + "content": "censoring based on Figure 1c, so we can also write", + "type": "text" + }, + { + "bbox": [ + 325, + 529, + 424, + 541 + ], + "score": 0.91, + "content": "c _ { a } \\sim p _ { \\nu , \\Phi } ( C _ { A } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 528, + 505, + 542 + ], + "score": 1.0, + "content": ". Given (3), model", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 538, + 506, + 553 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 148, + 553 + ], + "score": 1.0, + "content": "functions", + "type": "text" + }, + { + "bbox": [ + 149, + 540, + 204, + 552 + ], + "score": 0.93, + "content": "\\{ h _ { A } ( \\cdot ) , \\Phi ( \\cdot ) \\bar \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 538, + 225, + 553 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 225, + 540, + 248, + 552 + ], + "score": 0.9, + "content": "\\nu _ { A } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 538, + 506, + 553 + ], + "score": 1.0, + "content": "for informative censoring can be learned by leveraging stan-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 550, + 505, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 411, + 564 + ], + "score": 1.0, + "content": "dard statistical optimization approaches, that minimize a loss hypothesis", + "type": "text" + }, + { + "bbox": [ + 411, + 551, + 419, + 561 + ], + "score": 0.81, + "content": "\\mathcal { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 550, + 505, + 564 + ], + "score": 1.0, + "content": "given samples from", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 561, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 213, + 574 + ], + "score": 1.0, + "content": "the empirical distribution", + "type": "text" + }, + { + "bbox": [ + 213, + 562, + 322, + 573 + ], + "score": 0.9, + "content": "( y , \\delta , x , a ) \\sim p ( Y , \\delta , X , A )", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 561, + 399, + 574 + ], + "score": 1.0, + "content": ", i.e., from dataset", + "type": "text" + }, + { + "bbox": [ + 399, + 562, + 408, + 572 + ], + "score": 0.8, + "content": "\\mathcal { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 561, + 505, + 574 + ], + "score": 1.0, + "content": ". Specifically, we write", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 571, + 506, + 588 + ], + "spans": [ + { + "bbox": [ + 106, + 573, + 271, + 586 + ], + "score": 0.86, + "content": "\\mathcal { L } = \\mathbb { E } _ { ( y , \\delta , x , a ) \\sim p ( Y , \\delta , X , A ) } \\left[ \\ell _ { h , \\Phi } ( t _ { a } , y , \\delta ) \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 571, + 303, + 588 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 303, + 573, + 356, + 585 + ], + "score": 0.93, + "content": "\\ell _ { h , \\Phi } ( t _ { a } , y , \\delta )", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 571, + 506, + 588 + ], + "score": 1.0, + "content": "is a loss function that measures the", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 584, + 505, + 597 + ], + "spans": [ + { + "bbox": [ + 105, + 584, + 163, + 597 + ], + "score": 1.0, + "content": "agreement of", + "type": "text" + }, + { + "bbox": [ + 163, + 585, + 256, + 597 + ], + "score": 0.89, + "content": "t _ { a } \\sim p _ { h , \\Phi } ( T _ { A } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 584, + 278, + 597 + ], + "score": 1.0, + "content": "(and", + "type": "text" + }, + { + "bbox": [ + 279, + 585, + 373, + 597 + ], + "score": 0.9, + "content": "c _ { a } \\sim p _ { \\nu , \\Phi } ( C _ { A } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 584, + 505, + 597 + ], + "score": 1.0, + "content": "for informative censoring) with", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 595, + 411, + 609 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 159, + 609 + ], + "score": 1.0, + "content": "ground truth", + "type": "text" + }, + { + "bbox": [ + 159, + 596, + 184, + 608 + ], + "score": 0.92, + "content": "\\{ y , \\delta \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 595, + 411, + 609 + ], + "score": 1.0, + "content": ", the observed time and censoring indicator, respectively.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 495, + 506, + 609 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 612, + 505, + 681 + ], + "lines": [ + { + "bbox": [ + 106, + 612, + 506, + 625 + ], + "spans": [ + { + "bbox": [ + 106, + 612, + 357, + 625 + ], + "score": 1.0, + "content": "For some parametric formulations of event time distribution", + "type": "text" + }, + { + "bbox": [ + 357, + 613, + 429, + 625 + ], + "score": 0.9, + "content": "p _ { h , \\Phi } ( T _ { A } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 612, + 506, + 625 + ], + "score": 1.0, + "content": ", e.g., exponential,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 623, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 623, + 443, + 637 + ], + "score": 1.0, + "content": "Weibull, log-Normal, etc., and provided the censoring mechanism is non-informative,", + "type": "text" + }, + { + "bbox": [ + 443, + 624, + 505, + 636 + ], + "score": 0.92, + "content": "- \\ell _ { h , \\Phi } ( t _ { a } , y , \\delta )", + "type": "inline_equation" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 636, + 506, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 308, + 649 + ], + "score": 1.0, + "content": "is the closed form log likelihood. Specifically,", + "type": "text" + }, + { + "bbox": [ + 308, + 636, + 501, + 649 + ], + "score": 0.9, + "content": "- \\ell _ { h , \\Phi } ( t _ { a } , y , \\delta ) \\triangleq \\log p _ { h , \\Phi } ( T _ { a } | X = x ) = \\delta", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 636, + 506, + 649 + ], + "score": 1.0, + "content": "·", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 646, + 506, + 661 + ], + "spans": [ + { + "bbox": [ + 106, + 647, + 272, + 660 + ], + "score": 0.91, + "content": "\\log f _ { h , \\Phi } ( t _ { a } | x ) + ( 1 - \\bar { \\delta } ) \\cdot \\log S _ { h , \\Phi } ( t _ { a } | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 646, + 506, + 661 + ], + "score": 1.0, + "content": ", which implies that the conditional event time density and", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 658, + 505, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 395, + 671 + ], + "score": 1.0, + "content": "survival functions can be calculated in closed form from transformations", + "type": "text" + }, + { + "bbox": [ + 395, + 658, + 451, + 671 + ], + "score": 0.93, + "content": "\\{ h _ { A } ( \\cdot ) , \\Phi ( \\cdot ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 658, + 462, + 671 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 463, + 660, + 469, + 668 + ], + "score": 0.76, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 658, + 505, + 671 + ], + "score": 1.0, + "content": ". See the", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 668, + 395, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 234, + 683 + ], + "score": 1.0, + "content": "SM for parametric examples of", + "type": "text" + }, + { + "bbox": [ + 234, + 670, + 241, + 679 + ], + "score": 0.82, + "content": "\\mathcal { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 668, + 395, + 683 + ], + "score": 1.0, + "content": "accounting for informative censoring.", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 45.5, + "bbox_fs": [ + 105, + 612, + 506, + 683 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 686, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 685, + 504, + 699 + ], + "spans": [ + { + "bbox": [ + 105, + 685, + 386, + 699 + ], + "score": 1.0, + "content": "We further define the expected loss for a given realization of covariates", + "type": "text" + }, + { + "bbox": [ + 387, + 689, + 393, + 696 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 685, + 497, + 699 + ], + "score": 1.0, + "content": "and treatment assignment", + "type": "text" + }, + { + "bbox": [ + 498, + 689, + 504, + 696 + ], + "score": 0.72, + "content": "a", + "type": "inline_equation" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 190, + 711 + ], + "score": 1.0, + "content": "over observed times", + "type": "text" + }, + { + "bbox": [ + 190, + 700, + 197, + 710 + ], + "score": 0.76, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 698, + 433, + 711 + ], + "score": 1.0, + "content": "(censored and non-censored), and the censoring indicator", + "type": "text" + }, + { + "bbox": [ + 433, + 699, + 439, + 709 + ], + "score": 0.77, + "content": "\\delta", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 698, + 451, + 711 + ], + "score": 1.0, + "content": "as", + "type": "text" + }, + { + "bbox": [ + 452, + 698, + 505, + 711 + ], + "score": 0.9, + "content": "\\zeta _ { h , \\Phi } ( x , a ) \\triangleq", + "type": "inline_equation" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 708, + 507, + 724 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 227, + 723 + ], + "score": 0.9, + "content": "\\mathbb { E } _ { ( y , \\delta , x ) \\sim p ( Y , \\delta | X ) } \\ell _ { h , \\Phi } ( t _ { a } , y , \\delta )", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 708, + 479, + 724 + ], + "score": 1.0, + "content": "as in Shalit et al. (2017). For a given subject with covariates", + "type": "text" + }, + { + "bbox": [ + 479, + 712, + 487, + 720 + ], + "score": 0.8, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 708, + 507, + 724 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 720, + 502, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 192, + 734 + ], + "score": 1.0, + "content": "treatment assignment", + "type": "text" + }, + { + "bbox": [ + 192, + 724, + 199, + 730 + ], + "score": 0.77, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 720, + 451, + 734 + ], + "score": 1.0, + "content": ", we wish to minimize both the factual and counterfactual losses,", + "type": "text" + }, + { + "bbox": [ + 452, + 721, + 465, + 732 + ], + "score": 0.89, + "content": "\\mathcal { L } _ { \\mathrm { F } }", + "type": "inline_equation" + }, + { + "bbox": [ + 466, + 720, + 483, + 734 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 483, + 721, + 502, + 732 + ], + "score": 0.9, + "content": "\\mathcal { L } _ { \\mathrm { C F } }", + "type": "inline_equation" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 81, + 336, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 227, + 96 + ], + "score": 1.0, + "content": "respectively, by decomposing", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 227, + 83, + 291, + 94 + ], + "score": 0.91, + "content": "\\mathcal { L } = \\mathcal { L } _ { \\mathrm { F } } + \\mathcal { L } _ { \\mathrm { C F } }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 291, + 81, + 336, + 96 + ], + "score": 1.0, + "content": "as follows", + "type": "text", + "cross_page": true + } + ], + "index": 0 + } + ], + "index": 50.5, + "bbox_fs": [ + 105, + 685, + 507, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 335, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 336, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 227, + 96 + ], + "score": 1.0, + "content": "respectively, by decomposing", + "type": "text" + }, + { + "bbox": [ + 227, + 83, + 291, + 94 + ], + "score": 0.91, + "content": "\\mathcal { L } = \\mathcal { L } _ { \\mathrm { F } } + \\mathcal { L } _ { \\mathrm { C F } }", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 81, + 336, + 96 + ], + "score": 1.0, + "content": "as follows", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "interline_equation", + "bbox": [ + 162, + 102, + 448, + 117 + ], + "lines": [ + { + "bbox": [ + 162, + 102, + 448, + 117 + ], + "spans": [ + { + "bbox": [ + 162, + 102, + 448, + 117 + ], + "score": 0.89, + "content": "\\begin{array} { r } { \\mathcal { L } _ { \\mathrm { F } } = \\mathbb { E } _ { ( x , a ) \\sim p ( A , X ) } \\zeta _ { h , \\Phi } ( x , a ) , \\quad \\mathcal { L } _ { \\mathrm { C F } } = \\mathbb { E } _ { ( x , a ) \\sim p ( 1 - A , X ) } \\zeta _ { h , \\Phi } ( x , a ) . } \\end{array}", + "type": "interline_equation", + "image_path": "141da49929ef287744dea443052c2291ced1fb2cbc977e821917e72659db41f1.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 162, + 102, + 448, + 117 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 126, + 505, + 195 + ], + "lines": [ + { + "bbox": [ + 104, + 124, + 506, + 140 + ], + "spans": [ + { + "bbox": [ + 104, + 124, + 124, + 140 + ], + "score": 1.0, + "content": "Let", + "type": "text" + }, + { + "bbox": [ + 124, + 126, + 194, + 138 + ], + "score": 0.91, + "content": "u \\triangleq P ( A = 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 124, + 506, + 140 + ], + "score": 1.0, + "content": "denote the marginal probability of treatment assignment. We can readily", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 104, + 136, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 104, + 136, + 479, + 151 + ], + "score": 1.0, + "content": "decompose the losses in (4) according to treatment assignments. The decomposed factual", + "type": "text" + }, + { + "bbox": [ + 479, + 138, + 505, + 148 + ], + "score": 0.86, + "content": "\\mathcal { L } _ { \\mathrm { F } } =", + "type": "inline_equation" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 142, + 501, + 167 + ], + "spans": [ + { + "bbox": [ + 106, + 147, + 210, + 161 + ], + "score": 0.91, + "content": "u \\cdot \\mathcal { L } _ { \\mathrm { F } } ^ { A = 1 } + \\left( 1 - u \\right) \\cdot \\mathcal { L } _ { \\mathrm { F } } ^ { A = 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 142, + 393, + 167 + ], + "score": 1.0, + "content": ", and similarly, the decomposed counterfactual", + "type": "text" + }, + { + "bbox": [ + 393, + 148, + 501, + 161 + ], + "score": 0.9, + "content": "\\mathcal { L } _ { \\mathrm { C F } } = \\dot { ( } 1 - u ) \\cdot \\mathcal { L } _ { \\mathrm { C F } } ^ { A = 1 } + u", + "type": "inline_equation" + } + ], + "index": 4 + }, + { + "bbox": [ + 107, + 154, + 509, + 179 + ], + "spans": [ + { + "bbox": [ + 107, + 160, + 131, + 173 + ], + "score": 0.9, + "content": "\\mathcal { L } _ { \\mathrm { C F } } ^ { A = 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 131, + 154, + 509, + 179 + ], + "score": 1.0, + "content": "F CF . In practice, only factual outcomes are observed, hence, for a non-randomized non-controlled", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 104, + 172, + 506, + 184 + ], + "spans": [ + { + "bbox": [ + 104, + 172, + 335, + 184 + ], + "score": 1.0, + "content": "experiment, we cannot obtain an unbiased estimate of", + "type": "text" + }, + { + "bbox": [ + 336, + 172, + 355, + 183 + ], + "score": 0.91, + "content": "\\mathcal { L } _ { \\mathrm { C F } }", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 172, + 506, + 184 + ], + "score": 1.0, + "content": "from data due to selection bias (or", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 182, + 445, + 195 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 252, + 195 + ], + "score": 1.0, + "content": "confounding). Therefore, we bound", + "type": "text" + }, + { + "bbox": [ + 252, + 183, + 271, + 194 + ], + "score": 0.9, + "content": "\\mathcal { L } _ { \\mathrm { C F } }", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 182, + 289, + 195 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 289, + 183, + 297, + 192 + ], + "score": 0.82, + "content": "\\mathcal { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 182, + 445, + 195 + ], + "score": 1.0, + "content": "below following Shalit et al. (2017).", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 106, + 204, + 506, + 240 + ], + "lines": [ + { + "bbox": [ + 106, + 203, + 507, + 218 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 194, + 218 + ], + "score": 1.0, + "content": "Corollary 1 Assume", + "type": "text" + }, + { + "bbox": [ + 195, + 205, + 213, + 217 + ], + "score": 0.9, + "content": "\\Phi ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 203, + 322, + 218 + ], + "score": 1.0, + "content": "is an invertible map, and", + "type": "text" + }, + { + "bbox": [ + 323, + 204, + 404, + 217 + ], + "score": 0.93, + "content": "\\alpha ^ { - 1 } \\zeta _ { h , \\Phi } ( x , a ) \\in G", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 203, + 437, + 218 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 437, + 205, + 446, + 215 + ], + "score": 0.78, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 203, + 507, + 218 + ], + "score": 1.0, + "content": "is a family of", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 102, + 212, + 510, + 235 + ], + "spans": [ + { + "bbox": [ + 102, + 212, + 147, + 235 + ], + "score": 1.0, + "content": "functions,", + "type": "text" + }, + { + "bbox": [ + 147, + 217, + 239, + 230 + ], + "score": 0.93, + "content": "p _ { \\Phi } ^ { A = a } \\triangleq p _ { \\Phi } ( R | A = a )", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 212, + 376, + 235 + ], + "score": 1.0, + "content": "is the latent distribution for group", + "type": "text" + }, + { + "bbox": [ + 377, + 218, + 404, + 228 + ], + "score": 0.89, + "content": "A = a", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 212, + 425, + 235 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 425, + 218, + 451, + 228 + ], + "score": 0.89, + "content": "\\alpha > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 212, + 510, + 235 + ], + "score": 1.0, + "content": "is a constant.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 227, + 169, + 241 + ], + "spans": [ + { + "bbox": [ + 105, + 227, + 169, + 241 + ], + "score": 1.0, + "content": "Then, we have:", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + }, + { + "type": "interline_equation", + "bbox": [ + 180, + 240, + 430, + 275 + ], + "lines": [ + { + "bbox": [ + 180, + 240, + 430, + 275 + ], + "spans": [ + { + "bbox": [ + 180, + 240, + 430, + 275 + ], + "score": 0.91, + "content": "\\begin{array} { r l } & { \\mathcal { L } _ { \\mathrm { C F } } \\leq ( 1 - u ) \\cdot \\mathcal { L } _ { \\mathrm { F } } ^ { A = 1 } + u \\cdot \\mathcal { L } _ { \\mathrm { F } } ^ { A = 0 } + \\alpha \\cdot \\mathrm { I P M } _ { G } ( p _ { \\Phi } ^ { A = 1 } , p _ { \\Phi } ^ { A = 0 } ) } \\\\ & { \\quad \\mathcal { L } \\leq \\mathcal { L } _ { \\mathrm { F } } ^ { A = 1 } + \\mathcal { L } _ { \\mathrm { F } } ^ { A = 0 } + \\alpha \\cdot \\mathrm { I P M } _ { G } ( p _ { \\Phi } ^ { A = 1 } , p _ { \\Phi } ^ { A = 0 } ) . } \\end{array}", + "type": "interline_equation", + "image_path": "370f28aa2b443822fd375b3828179499d1816978110177146821204aef6fa36a.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 180, + 240, + 430, + 251.66666666666666 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 180, + 251.66666666666666, + 430, + 263.3333333333333 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 180, + 263.3333333333333, + 430, + 275.0 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 285, + 506, + 378 + ], + "lines": [ + { + "bbox": [ + 105, + 286, + 506, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 286, + 506, + 299 + ], + "score": 1.0, + "content": "The integral probability metric (IPM) (Müller, 1997; Sriperumbudur et al., 2012) measures the", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 297, + 507, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 297, + 297, + 311 + ], + "score": 1.0, + "content": "distance between two probability distributions", + "type": "text" + }, + { + "bbox": [ + 298, + 300, + 304, + 309 + ], + "score": 0.81, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 297, + 322, + 311 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 323, + 299, + 329, + 309 + ], + "score": 0.79, + "content": "q", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 297, + 383, + 311 + ], + "score": 1.0, + "content": "defined over", + "type": "text" + }, + { + "bbox": [ + 384, + 298, + 395, + 308 + ], + "score": 0.74, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 297, + 494, + 311 + ], + "score": 1.0, + "content": ", i.e., the latent space of", + "type": "text" + }, + { + "bbox": [ + 495, + 298, + 503, + 308 + ], + "score": 0.82, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 297, + 507, + 311 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 104, + 308, + 508, + 326 + ], + "spans": [ + { + "bbox": [ + 104, + 308, + 148, + 326 + ], + "score": 1.0, + "content": "Formally,", + "type": "text" + }, + { + "bbox": [ + 149, + 309, + 370, + 323 + ], + "score": 0.86, + "content": "\\begin{array} { r } { \\mathrm { I P M } _ { G } ( p , q ) \\triangleq \\operatorname* { s u p } _ { g \\in G } \\vert \\int _ { M } g ( m ) \\left( p ( m ) - q ( m ) \\right) d m \\vert } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 308, + 401, + 326 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 401, + 310, + 450, + 321 + ], + "score": 0.9, + "content": "g : m \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 308, + 508, + 326 + ], + "score": 1.0, + "content": ", represents a", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 320, + 506, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 330, + 334 + ], + "score": 1.0, + "content": "class of real-valued bounded measurable functions on", + "type": "text" + }, + { + "bbox": [ + 331, + 322, + 343, + 331 + ], + "score": 0.79, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 320, + 506, + 334 + ], + "score": 1.0, + "content": "(Shalit et al., 2017). Therefore, model", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 331, + 505, + 346 + ], + "spans": [ + { + "bbox": [ + 105, + 331, + 148, + 346 + ], + "score": 1.0, + "content": "functions", + "type": "text" + }, + { + "bbox": [ + 149, + 331, + 203, + 344 + ], + "score": 0.93, + "content": "\\{ h _ { a } ( \\cdot ) , \\Phi ( \\cdot ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 331, + 493, + 346 + ], + "score": 1.0, + "content": "can be learned by minimizing the upper bound in (5) consisting of", + "type": "text" + }, + { + "bbox": [ + 493, + 332, + 505, + 344 + ], + "score": 0.65, + "content": "( i )", + "type": "inline_equation" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 343, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 105, + 343, + 342, + 356 + ], + "score": 1.0, + "content": "only factual losses under both treatment assignments and", + "type": "text" + }, + { + "bbox": [ + 343, + 343, + 357, + 354 + ], + "score": 0.81, + "content": "( i i )", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 343, + 506, + 356 + ], + "score": 1.0, + "content": "an IPM regularizer enforcing latent", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 353, + 506, + 367 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 506, + 367 + ], + "score": 1.0, + "content": "distributional equivalence between the treatment groups. Note that if the data originates from a RCT", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 102, + 362, + 341, + 381 + ], + "spans": [ + { + "bbox": [ + 102, + 362, + 235, + 381 + ], + "score": 1.0, + "content": "it follows (by construction) that", + "type": "text" + }, + { + "bbox": [ + 235, + 364, + 336, + 377 + ], + "score": 0.91, + "content": "\\mathrm { I P M } _ { G } ( p _ { \\Phi } ^ { A = 1 } , p _ { \\Phi } ^ { \\tilde { A = 0 } } ) \\stackrel { . } { = } 0", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 362, + 341, + 381 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 106, + 385, + 506, + 562 + ], + "lines": [ + { + "bbox": [ + 106, + 385, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 106, + 385, + 481, + 398 + ], + "score": 1.0, + "content": "Accounting for censoring bias Below we formulate an approach for estimating functions", + "type": "text" + }, + { + "bbox": [ + 481, + 386, + 505, + 398 + ], + "score": 0.89, + "content": "h _ { A } ( \\cdot )", + "type": "inline_equation" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 396, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 124, + 410 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 397, + 147, + 408 + ], + "score": 0.89, + "content": "\\nu _ { A } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 396, + 324, + 410 + ], + "score": 1.0, + "content": "for synthesizing (sampling) non-censored", + "type": "text" + }, + { + "bbox": [ + 325, + 397, + 422, + 409 + ], + "score": 0.91, + "content": "t _ { a } \\sim \\stackrel { \\sim } { p } _ { h , \\Phi } ( T _ { A } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 396, + 481, + 410 + ], + "score": 1.0, + "content": "and censored", + "type": "text" + }, + { + "bbox": [ + 482, + 399, + 505, + 408 + ], + "score": 0.84, + "content": "c _ { a } \\sim", + "type": "inline_equation" + } + ], + "index": 23 + }, + { + "bbox": [ + 107, + 407, + 504, + 421 + ], + "spans": [ + { + "bbox": [ + 107, + 408, + 180, + 420 + ], + "score": 0.9, + "content": "p _ { \\nu , \\Phi } ( C _ { A } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 407, + 432, + 421 + ], + "score": 1.0, + "content": "times, respectively. While some parametric assumptions for", + "type": "text" + }, + { + "bbox": [ + 433, + 408, + 504, + 420 + ], + "score": 0.9, + "content": "p _ { h , \\Phi } ( T _ { A } | X = x )", + "type": "inline_equation" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 419, + 505, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 419, + 273, + 431 + ], + "score": 1.0, + "content": "yield easy-to-evaluate closed forms for", + "type": "text" + }, + { + "bbox": [ + 273, + 419, + 318, + 431 + ], + "score": 0.94, + "content": "S _ { h , \\Phi } ( t _ { a } | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 319, + 419, + 505, + 431 + ], + "score": 1.0, + "content": "that can be used as likelihood for censored", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 430, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 430, + 505, + 441 + ], + "score": 1.0, + "content": "observations, they are restrictive, and have been shown to generate unrealistic high variance samples", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 441, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 106, + 441, + 505, + 452 + ], + "score": 1.0, + "content": "(Chapfuwa et al., 2018). So motivated, we seek a nonparametric likelihood-based approach that", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 451, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 484, + 464 + ], + "score": 1.0, + "content": "can model a flexible family of distributions, with an easy-to-sample approach for event times", + "type": "text" + }, + { + "bbox": [ + 484, + 453, + 505, + 463 + ], + "score": 0.83, + "content": "t _ { a } \\sim", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 461, + 506, + 476 + ], + "spans": [ + { + "bbox": [ + 106, + 462, + 175, + 475 + ], + "score": 0.9, + "content": "p _ { h , \\Phi } ( T _ { a } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 461, + 485, + 476 + ], + "score": 1.0, + "content": ". We model the event time generation process with a source of randomness,", + "type": "text" + }, + { + "bbox": [ + 485, + 464, + 502, + 475 + ], + "score": 0.91, + "content": "p ( \\epsilon )", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 461, + 506, + 476 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 474, + 507, + 485 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 507, + 485 + ], + "score": 1.0, + "content": "e.g. Gaussian or uniform, which is obtained from a neural-network-based nonlinear transformation.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 483, + 506, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 397, + 497 + ], + "score": 1.0, + "content": "In the experiments we use a planar flow formulation parameterized by", + "type": "text" + }, + { + "bbox": [ + 397, + 486, + 452, + 497 + ], + "score": 0.93, + "content": "\\{ U _ { h } , W _ { h } , b _ { h } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 483, + 506, + 497 + ], + "score": 1.0, + "content": "(Rezende &", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 494, + 506, + 508 + ], + "spans": [ + { + "bbox": [ + 104, + 494, + 506, + 508 + ], + "score": 1.0, + "content": "Mohamed, 2015), however, other specifications can also be used. Note that Miscouridou et al. (2018)", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 505, + 505, + 520 + ], + "spans": [ + { + "bbox": [ + 104, + 505, + 505, + 520 + ], + "score": 1.0, + "content": "has previously leveraged normalizing flows for survival analysis, however, our approach is very", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 516, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 267, + 530 + ], + "score": 1.0, + "content": "different in that it focuses on formulating", + "type": "text" + }, + { + "bbox": [ + 268, + 519, + 272, + 528 + ], + "score": 0.63, + "content": "\\romannumeral 1", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 516, + 505, + 530 + ], + "score": 1.0, + "content": ") a counterfactual survival analysis framework that accounts", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 528, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 440, + 540 + ], + "score": 1.0, + "content": "for informative or non-informative censoring mechanisms and confounding, and", + "type": "text" + }, + { + "bbox": [ + 440, + 531, + 447, + 538 + ], + "score": 0.83, + "content": "_ { i i }", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 528, + 505, + 540 + ], + "score": 1.0, + "content": ") model event", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 538, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 505, + 551 + ], + "score": 1.0, + "content": "times as a continuous variable instead of discretizing them. Specifically, we transform the source of", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 551, + 343, + 562 + ], + "spans": [ + { + "bbox": [ + 106, + 551, + 159, + 562 + ], + "score": 1.0, + "content": "randomness,", + "type": "text" + }, + { + "bbox": [ + 159, + 554, + 164, + 560 + ], + "score": 0.62, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 551, + 343, + 562 + ], + "score": 1.0, + "content": ", using a single layer specification as follows", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 29.5 + }, + { + "type": "interline_equation", + "bbox": [ + 138, + 570, + 473, + 584 + ], + "lines": [ + { + "bbox": [ + 138, + 570, + 473, + 584 + ], + "spans": [ + { + "bbox": [ + 138, + 570, + 473, + 584 + ], + "score": 0.82, + "content": "{ \\widetilde { \\epsilon } } _ { h } = \\epsilon + U _ { h } \\operatorname { t a n h } ( W _ { h } \\epsilon + b _ { h } ) , \\quad \\epsilon \\sim \\operatorname { U n i f o r m } ( 0 , 1 ) , t _ { a } = h _ { A } ( r , { \\widetilde { \\epsilon } } _ { h } ) , \\quad r = \\Phi ( x )", + "type": "interline_equation", + "image_path": "5d0fe636d034b8d80f53f9ef96d49f1ce924a440335a5cca19c9ccfdb29ad614.jpg" + } + ] + } + ], + "index": 38, + "virtual_lines": [ + { + "bbox": [ + 138, + 570, + 473, + 584 + ], + "spans": [], + "index": 38 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 592, + 505, + 671 + ], + "lines": [ + { + "bbox": [ + 105, + 590, + 506, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 135, + 607 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 592, + 214, + 605 + ], + "score": 0.88, + "content": "\\{ U _ { h } , W _ { h } \\} \\in \\mathbb { R } ^ { d \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 590, + 219, + 607 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 219, + 592, + 276, + 605 + ], + "score": 0.81, + "content": "\\{ b _ { h } , \\epsilon \\} \\in \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 590, + 280, + 607 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 281, + 594, + 288, + 603 + ], + "score": 0.65, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 590, + 506, + 607 + ], + "score": 1.0, + "content": "is the dimensionality of the normalizing flow; each", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 604, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 168, + 617 + ], + "score": 1.0, + "content": "component of", + "type": "text" + }, + { + "bbox": [ + 168, + 607, + 173, + 614 + ], + "score": 0.74, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 604, + 337, + 617 + ], + "score": 1.0, + "content": "is drawn independently from Uniform", + "type": "text" + }, + { + "bbox": [ + 338, + 604, + 360, + 616 + ], + "score": 0.8, + "content": "( 0 , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 604, + 383, + 617 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 384, + 605, + 394, + 615 + ], + "score": 0.88, + "content": "\\tilde { \\epsilon } _ { h }", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 604, + 505, + 617 + ], + "score": 1.0, + "content": "may be viewed as a skip", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 615, + 506, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 240, + 629 + ], + "score": 1.0, + "content": "connection with stochasticity in", + "type": "text" + }, + { + "bbox": [ + 240, + 618, + 246, + 625 + ], + "score": 0.64, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 615, + 287, + 629 + ], + "score": 1.0, + "content": ". Further,", + "type": "text" + }, + { + "bbox": [ + 287, + 615, + 326, + 627 + ], + "score": 0.93, + "content": "h _ { A } ( r , \\tilde { \\epsilon } _ { h } )", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 615, + 346, + 629 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 346, + 616, + 368, + 627 + ], + "score": 0.92, + "content": "\\Phi ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 615, + 506, + 629 + ], + "score": 1.0, + "content": "are time-to-event generative and", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 626, + 506, + 638 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 506, + 638 + ], + "score": 1.0, + "content": "encoding functions, respectively, parameterized as neural networks. For simplicity, the dimensions of", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 107, + 638, + 505, + 649 + ], + "spans": [ + { + "bbox": [ + 107, + 640, + 113, + 647 + ], + "score": 0.73, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 113, + 638, + 130, + 649 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 131, + 639, + 136, + 647 + ], + "score": 0.74, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 137, + 638, + 176, + 649 + ], + "score": 1.0, + "content": "are set to", + "type": "text" + }, + { + "bbox": [ + 176, + 638, + 182, + 647 + ], + "score": 0.79, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 638, + 505, + 649 + ], + "score": 1.0, + "content": ", however, they can be set independently if desired. In practice, we are interested", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 649, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 505, + 660 + ], + "score": 1.0, + "content": "in generating realistic event-time samples; therefore, we account for both censored and non-censored", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 659, + 438, + 672 + ], + "spans": [ + { + "bbox": [ + 106, + 659, + 438, + 672 + ], + "score": 1.0, + "content": "observations by adopting the objective from Chapfuwa et al. (2018), formulated as", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 42 + }, + { + "type": "interline_equation", + "bbox": [ + 141, + 678, + 469, + 695 + ], + "lines": [ + { + "bbox": [ + 141, + 678, + 469, + 695 + ], + "spans": [ + { + "bbox": [ + 141, + 678, + 469, + 695 + ], + "score": 0.88, + "content": "\\mathcal { L } _ { \\mathrm { F } } ^ { \\mathrm { C S A } } \\triangleq \\mathbb { E } _ { ( y , \\delta , x , a ) \\sim p ( Y , \\delta , X , A ) , \\epsilon \\sim p ( \\epsilon ) } \\left[ \\delta \\cdot \\left( \\left| y - t _ { a } \\right| \\right) + \\left( 1 - \\delta \\right) \\cdot \\left( \\operatorname* { m a x } ( 0 , y - t _ { a } ) \\right) \\right] ,", + "type": "interline_equation", + "image_path": "7fde54abcbf1fd3fd611de6768c944eaf23863c85ea57f28571b691d6e808338.jpg" + } + ] + } + ], + "index": 46, + "virtual_lines": [ + { + "bbox": [ + 141, + 678, + 469, + 695 + ], + "spans": [], + "index": 46 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 702, + 506, + 735 + ], + "lines": [ + { + "bbox": [ + 105, + 702, + 506, + 715 + ], + "spans": [ + { + "bbox": [ + 105, + 702, + 313, + 715 + ], + "score": 1.0, + "content": "where the first term encourages sampled event times", + "type": "text" + }, + { + "bbox": [ + 313, + 703, + 322, + 713 + ], + "score": 0.87, + "content": "t _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 702, + 378, + 715 + ], + "score": 1.0, + "content": "to be close to", + "type": "text" + }, + { + "bbox": [ + 378, + 704, + 384, + 713 + ], + "score": 0.8, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 702, + 506, + 715 + ], + "score": 1.0, + "content": ", the ground truth for observed", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 713, + 506, + 725 + ], + "spans": [ + { + "bbox": [ + 106, + 713, + 152, + 725 + ], + "score": 1.0, + "content": "events, i.e.,", + "type": "text" + }, + { + "bbox": [ + 153, + 713, + 177, + 723 + ], + "score": 0.9, + "content": "\\delta = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 713, + 248, + 725 + ], + "score": 1.0, + "content": ", while penalizing", + "type": "text" + }, + { + "bbox": [ + 248, + 714, + 257, + 724 + ], + "score": 0.86, + "content": "t _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 713, + 444, + 725 + ], + "score": 1.0, + "content": "for being smaller than the censoring time when", + "type": "text" + }, + { + "bbox": [ + 444, + 713, + 469, + 723 + ], + "score": 0.89, + "content": "\\delta = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 713, + 506, + 725 + ], + "score": 1.0, + "content": ". Further,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 724, + 480, + 736 + ], + "spans": [ + { + "bbox": [ + 106, + 724, + 423, + 736 + ], + "score": 1.0, + "content": "the expectation is taken over samples (a minibatch) from empirical distribution", + "type": "text" + }, + { + "bbox": [ + 424, + 724, + 477, + 736 + ], + "score": 0.92, + "content": "p ( Y , \\delta , X , A )", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 724, + 480, + 736 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 48 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 106, + 26, + 307, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 308, + 39 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 335, + 94 + ], + "lines": [], + "index": 0, + "bbox_fs": [ + 105, + 81, + 336, + 96 + ], + "lines_deleted": true + }, + { + "type": "interline_equation", + "bbox": [ + 162, + 102, + 448, + 117 + ], + "lines": [ + { + "bbox": [ + 162, + 102, + 448, + 117 + ], + "spans": [ + { + "bbox": [ + 162, + 102, + 448, + 117 + ], + "score": 0.89, + "content": "\\begin{array} { r } { \\mathcal { L } _ { \\mathrm { F } } = \\mathbb { E } _ { ( x , a ) \\sim p ( A , X ) } \\zeta _ { h , \\Phi } ( x , a ) , \\quad \\mathcal { L } _ { \\mathrm { C F } } = \\mathbb { E } _ { ( x , a ) \\sim p ( 1 - A , X ) } \\zeta _ { h , \\Phi } ( x , a ) . } \\end{array}", + "type": "interline_equation", + "image_path": "141da49929ef287744dea443052c2291ced1fb2cbc977e821917e72659db41f1.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 162, + 102, + 448, + 117 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 126, + 505, + 195 + ], + "lines": [ + { + "bbox": [ + 104, + 124, + 506, + 140 + ], + "spans": [ + { + "bbox": [ + 104, + 124, + 124, + 140 + ], + "score": 1.0, + "content": "Let", + "type": "text" + }, + { + "bbox": [ + 124, + 126, + 194, + 138 + ], + "score": 0.91, + "content": "u \\triangleq P ( A = 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 124, + 506, + 140 + ], + "score": 1.0, + "content": "denote the marginal probability of treatment assignment. We can readily", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 104, + 136, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 104, + 136, + 479, + 151 + ], + "score": 1.0, + "content": "decompose the losses in (4) according to treatment assignments. The decomposed factual", + "type": "text" + }, + { + "bbox": [ + 479, + 138, + 505, + 148 + ], + "score": 0.86, + "content": "\\mathcal { L } _ { \\mathrm { F } } =", + "type": "inline_equation" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 142, + 501, + 167 + ], + "spans": [ + { + "bbox": [ + 106, + 147, + 210, + 161 + ], + "score": 0.91, + "content": "u \\cdot \\mathcal { L } _ { \\mathrm { F } } ^ { A = 1 } + \\left( 1 - u \\right) \\cdot \\mathcal { L } _ { \\mathrm { F } } ^ { A = 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 142, + 393, + 167 + ], + "score": 1.0, + "content": ", and similarly, the decomposed counterfactual", + "type": "text" + }, + { + "bbox": [ + 393, + 148, + 501, + 161 + ], + "score": 0.9, + "content": "\\mathcal { L } _ { \\mathrm { C F } } = \\dot { ( } 1 - u ) \\cdot \\mathcal { L } _ { \\mathrm { C F } } ^ { A = 1 } + u", + "type": "inline_equation" + } + ], + "index": 4 + }, + { + "bbox": [ + 107, + 154, + 509, + 179 + ], + "spans": [ + { + "bbox": [ + 107, + 160, + 131, + 173 + ], + "score": 0.9, + "content": "\\mathcal { L } _ { \\mathrm { C F } } ^ { A = 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 131, + 154, + 509, + 179 + ], + "score": 1.0, + "content": "F CF . In practice, only factual outcomes are observed, hence, for a non-randomized non-controlled", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 104, + 172, + 506, + 184 + ], + "spans": [ + { + "bbox": [ + 104, + 172, + 335, + 184 + ], + "score": 1.0, + "content": "experiment, we cannot obtain an unbiased estimate of", + "type": "text" + }, + { + "bbox": [ + 336, + 172, + 355, + 183 + ], + "score": 0.91, + "content": "\\mathcal { L } _ { \\mathrm { C F } }", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 172, + 506, + 184 + ], + "score": 1.0, + "content": "from data due to selection bias (or", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 182, + 445, + 195 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 252, + 195 + ], + "score": 1.0, + "content": "confounding). Therefore, we bound", + "type": "text" + }, + { + "bbox": [ + 252, + 183, + 271, + 194 + ], + "score": 0.9, + "content": "\\mathcal { L } _ { \\mathrm { C F } }", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 182, + 289, + 195 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 289, + 183, + 297, + 192 + ], + "score": 0.82, + "content": "\\mathcal { L }", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 182, + 445, + 195 + ], + "score": 1.0, + "content": "below following Shalit et al. (2017).", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 4.5, + "bbox_fs": [ + 104, + 124, + 509, + 195 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 204, + 506, + 240 + ], + "lines": [ + { + "bbox": [ + 106, + 203, + 507, + 218 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 194, + 218 + ], + "score": 1.0, + "content": "Corollary 1 Assume", + "type": "text" + }, + { + "bbox": [ + 195, + 205, + 213, + 217 + ], + "score": 0.9, + "content": "\\Phi ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 203, + 322, + 218 + ], + "score": 1.0, + "content": "is an invertible map, and", + "type": "text" + }, + { + "bbox": [ + 323, + 204, + 404, + 217 + ], + "score": 0.93, + "content": "\\alpha ^ { - 1 } \\zeta _ { h , \\Phi } ( x , a ) \\in G", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 203, + 437, + 218 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 437, + 205, + 446, + 215 + ], + "score": 0.78, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 203, + 507, + 218 + ], + "score": 1.0, + "content": "is a family of", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 102, + 212, + 510, + 235 + ], + "spans": [ + { + "bbox": [ + 102, + 212, + 147, + 235 + ], + "score": 1.0, + "content": "functions,", + "type": "text" + }, + { + "bbox": [ + 147, + 217, + 239, + 230 + ], + "score": 0.93, + "content": "p _ { \\Phi } ^ { A = a } \\triangleq p _ { \\Phi } ( R | A = a )", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 212, + 376, + 235 + ], + "score": 1.0, + "content": "is the latent distribution for group", + "type": "text" + }, + { + "bbox": [ + 377, + 218, + 404, + 228 + ], + "score": 0.89, + "content": "A = a", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 212, + 425, + 235 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 425, + 218, + 451, + 228 + ], + "score": 0.89, + "content": "\\alpha > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 212, + 510, + 235 + ], + "score": 1.0, + "content": "is a constant.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 227, + 169, + 241 + ], + "spans": [ + { + "bbox": [ + 105, + 227, + 169, + 241 + ], + "score": 1.0, + "content": "Then, we have:", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9, + "bbox_fs": [ + 102, + 203, + 510, + 241 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 180, + 240, + 430, + 275 + ], + "lines": [ + { + "bbox": [ + 180, + 240, + 430, + 275 + ], + "spans": [ + { + "bbox": [ + 180, + 240, + 430, + 275 + ], + "score": 0.91, + "content": "\\begin{array} { r l } & { \\mathcal { L } _ { \\mathrm { C F } } \\leq ( 1 - u ) \\cdot \\mathcal { L } _ { \\mathrm { F } } ^ { A = 1 } + u \\cdot \\mathcal { L } _ { \\mathrm { F } } ^ { A = 0 } + \\alpha \\cdot \\mathrm { I P M } _ { G } ( p _ { \\Phi } ^ { A = 1 } , p _ { \\Phi } ^ { A = 0 } ) } \\\\ & { \\quad \\mathcal { L } \\leq \\mathcal { L } _ { \\mathrm { F } } ^ { A = 1 } + \\mathcal { L } _ { \\mathrm { F } } ^ { A = 0 } + \\alpha \\cdot \\mathrm { I P M } _ { G } ( p _ { \\Phi } ^ { A = 1 } , p _ { \\Phi } ^ { A = 0 } ) . } \\end{array}", + "type": "interline_equation", + "image_path": "370f28aa2b443822fd375b3828179499d1816978110177146821204aef6fa36a.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 180, + 240, + 430, + 251.66666666666666 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 180, + 251.66666666666666, + 430, + 263.3333333333333 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 180, + 263.3333333333333, + 430, + 275.0 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 285, + 506, + 378 + ], + "lines": [ + { + "bbox": [ + 105, + 286, + 506, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 286, + 506, + 299 + ], + "score": 1.0, + "content": "The integral probability metric (IPM) (Müller, 1997; Sriperumbudur et al., 2012) measures the", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 297, + 507, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 297, + 297, + 311 + ], + "score": 1.0, + "content": "distance between two probability distributions", + "type": "text" + }, + { + "bbox": [ + 298, + 300, + 304, + 309 + ], + "score": 0.81, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 297, + 322, + 311 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 323, + 299, + 329, + 309 + ], + "score": 0.79, + "content": "q", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 297, + 383, + 311 + ], + "score": 1.0, + "content": "defined over", + "type": "text" + }, + { + "bbox": [ + 384, + 298, + 395, + 308 + ], + "score": 0.74, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 297, + 494, + 311 + ], + "score": 1.0, + "content": ", i.e., the latent space of", + "type": "text" + }, + { + "bbox": [ + 495, + 298, + 503, + 308 + ], + "score": 0.82, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 297, + 507, + 311 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 104, + 308, + 508, + 326 + ], + "spans": [ + { + "bbox": [ + 104, + 308, + 148, + 326 + ], + "score": 1.0, + "content": "Formally,", + "type": "text" + }, + { + "bbox": [ + 149, + 309, + 370, + 323 + ], + "score": 0.86, + "content": "\\begin{array} { r } { \\mathrm { I P M } _ { G } ( p , q ) \\triangleq \\operatorname* { s u p } _ { g \\in G } \\vert \\int _ { M } g ( m ) \\left( p ( m ) - q ( m ) \\right) d m \\vert } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 308, + 401, + 326 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 401, + 310, + 450, + 321 + ], + "score": 0.9, + "content": "g : m \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 308, + 508, + 326 + ], + "score": 1.0, + "content": ", represents a", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 320, + 506, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 330, + 334 + ], + "score": 1.0, + "content": "class of real-valued bounded measurable functions on", + "type": "text" + }, + { + "bbox": [ + 331, + 322, + 343, + 331 + ], + "score": 0.79, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 320, + 506, + 334 + ], + "score": 1.0, + "content": "(Shalit et al., 2017). Therefore, model", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 331, + 505, + 346 + ], + "spans": [ + { + "bbox": [ + 105, + 331, + 148, + 346 + ], + "score": 1.0, + "content": "functions", + "type": "text" + }, + { + "bbox": [ + 149, + 331, + 203, + 344 + ], + "score": 0.93, + "content": "\\{ h _ { a } ( \\cdot ) , \\Phi ( \\cdot ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 331, + 493, + 346 + ], + "score": 1.0, + "content": "can be learned by minimizing the upper bound in (5) consisting of", + "type": "text" + }, + { + "bbox": [ + 493, + 332, + 505, + 344 + ], + "score": 0.65, + "content": "( i )", + "type": "inline_equation" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 343, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 105, + 343, + 342, + 356 + ], + "score": 1.0, + "content": "only factual losses under both treatment assignments and", + "type": "text" + }, + { + "bbox": [ + 343, + 343, + 357, + 354 + ], + "score": 0.81, + "content": "( i i )", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 343, + 506, + 356 + ], + "score": 1.0, + "content": "an IPM regularizer enforcing latent", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 353, + 506, + 367 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 506, + 367 + ], + "score": 1.0, + "content": "distributional equivalence between the treatment groups. Note that if the data originates from a RCT", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 102, + 362, + 341, + 381 + ], + "spans": [ + { + "bbox": [ + 102, + 362, + 235, + 381 + ], + "score": 1.0, + "content": "it follows (by construction) that", + "type": "text" + }, + { + "bbox": [ + 235, + 364, + 336, + 377 + ], + "score": 0.91, + "content": "\\mathrm { I P M } _ { G } ( p _ { \\Phi } ^ { A = 1 } , p _ { \\Phi } ^ { \\tilde { A = 0 } } ) \\stackrel { . } { = } 0", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 362, + 341, + 381 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 17.5, + "bbox_fs": [ + 102, + 286, + 508, + 381 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 385, + 506, + 562 + ], + "lines": [ + { + "bbox": [ + 106, + 385, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 106, + 385, + 481, + 398 + ], + "score": 1.0, + "content": "Accounting for censoring bias Below we formulate an approach for estimating functions", + "type": "text" + }, + { + "bbox": [ + 481, + 386, + 505, + 398 + ], + "score": 0.89, + "content": "h _ { A } ( \\cdot )", + "type": "inline_equation" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 396, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 124, + 410 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 397, + 147, + 408 + ], + "score": 0.89, + "content": "\\nu _ { A } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 396, + 324, + 410 + ], + "score": 1.0, + "content": "for synthesizing (sampling) non-censored", + "type": "text" + }, + { + "bbox": [ + 325, + 397, + 422, + 409 + ], + "score": 0.91, + "content": "t _ { a } \\sim \\stackrel { \\sim } { p } _ { h , \\Phi } ( T _ { A } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 396, + 481, + 410 + ], + "score": 1.0, + "content": "and censored", + "type": "text" + }, + { + "bbox": [ + 482, + 399, + 505, + 408 + ], + "score": 0.84, + "content": "c _ { a } \\sim", + "type": "inline_equation" + } + ], + "index": 23 + }, + { + "bbox": [ + 107, + 407, + 504, + 421 + ], + "spans": [ + { + "bbox": [ + 107, + 408, + 180, + 420 + ], + "score": 0.9, + "content": "p _ { \\nu , \\Phi } ( C _ { A } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 407, + 432, + 421 + ], + "score": 1.0, + "content": "times, respectively. While some parametric assumptions for", + "type": "text" + }, + { + "bbox": [ + 433, + 408, + 504, + 420 + ], + "score": 0.9, + "content": "p _ { h , \\Phi } ( T _ { A } | X = x )", + "type": "inline_equation" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 419, + 505, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 419, + 273, + 431 + ], + "score": 1.0, + "content": "yield easy-to-evaluate closed forms for", + "type": "text" + }, + { + "bbox": [ + 273, + 419, + 318, + 431 + ], + "score": 0.94, + "content": "S _ { h , \\Phi } ( t _ { a } | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 319, + 419, + 505, + 431 + ], + "score": 1.0, + "content": "that can be used as likelihood for censored", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 430, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 430, + 505, + 441 + ], + "score": 1.0, + "content": "observations, they are restrictive, and have been shown to generate unrealistic high variance samples", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 441, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 106, + 441, + 505, + 452 + ], + "score": 1.0, + "content": "(Chapfuwa et al., 2018). So motivated, we seek a nonparametric likelihood-based approach that", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 451, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 484, + 464 + ], + "score": 1.0, + "content": "can model a flexible family of distributions, with an easy-to-sample approach for event times", + "type": "text" + }, + { + "bbox": [ + 484, + 453, + 505, + 463 + ], + "score": 0.83, + "content": "t _ { a } \\sim", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 461, + 506, + 476 + ], + "spans": [ + { + "bbox": [ + 106, + 462, + 175, + 475 + ], + "score": 0.9, + "content": "p _ { h , \\Phi } ( T _ { a } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 461, + 485, + 476 + ], + "score": 1.0, + "content": ". We model the event time generation process with a source of randomness,", + "type": "text" + }, + { + "bbox": [ + 485, + 464, + 502, + 475 + ], + "score": 0.91, + "content": "p ( \\epsilon )", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 461, + 506, + 476 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 474, + 507, + 485 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 507, + 485 + ], + "score": 1.0, + "content": "e.g. Gaussian or uniform, which is obtained from a neural-network-based nonlinear transformation.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 483, + 506, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 397, + 497 + ], + "score": 1.0, + "content": "In the experiments we use a planar flow formulation parameterized by", + "type": "text" + }, + { + "bbox": [ + 397, + 486, + 452, + 497 + ], + "score": 0.93, + "content": "\\{ U _ { h } , W _ { h } , b _ { h } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 483, + 506, + 497 + ], + "score": 1.0, + "content": "(Rezende &", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 494, + 506, + 508 + ], + "spans": [ + { + "bbox": [ + 104, + 494, + 506, + 508 + ], + "score": 1.0, + "content": "Mohamed, 2015), however, other specifications can also be used. Note that Miscouridou et al. (2018)", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 505, + 505, + 520 + ], + "spans": [ + { + "bbox": [ + 104, + 505, + 505, + 520 + ], + "score": 1.0, + "content": "has previously leveraged normalizing flows for survival analysis, however, our approach is very", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 516, + 505, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 267, + 530 + ], + "score": 1.0, + "content": "different in that it focuses on formulating", + "type": "text" + }, + { + "bbox": [ + 268, + 519, + 272, + 528 + ], + "score": 0.63, + "content": "\\romannumeral 1", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 516, + 505, + 530 + ], + "score": 1.0, + "content": ") a counterfactual survival analysis framework that accounts", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 528, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 440, + 540 + ], + "score": 1.0, + "content": "for informative or non-informative censoring mechanisms and confounding, and", + "type": "text" + }, + { + "bbox": [ + 440, + 531, + 447, + 538 + ], + "score": 0.83, + "content": "_ { i i }", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 528, + 505, + 540 + ], + "score": 1.0, + "content": ") model event", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 538, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 505, + 551 + ], + "score": 1.0, + "content": "times as a continuous variable instead of discretizing them. Specifically, we transform the source of", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 551, + 343, + 562 + ], + "spans": [ + { + "bbox": [ + 106, + 551, + 159, + 562 + ], + "score": 1.0, + "content": "randomness,", + "type": "text" + }, + { + "bbox": [ + 159, + 554, + 164, + 560 + ], + "score": 0.62, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 551, + 343, + 562 + ], + "score": 1.0, + "content": ", using a single layer specification as follows", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 29.5, + "bbox_fs": [ + 104, + 385, + 507, + 562 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 138, + 570, + 473, + 584 + ], + "lines": [ + { + "bbox": [ + 138, + 570, + 473, + 584 + ], + "spans": [ + { + "bbox": [ + 138, + 570, + 473, + 584 + ], + "score": 0.82, + "content": "{ \\widetilde { \\epsilon } } _ { h } = \\epsilon + U _ { h } \\operatorname { t a n h } ( W _ { h } \\epsilon + b _ { h } ) , \\quad \\epsilon \\sim \\operatorname { U n i f o r m } ( 0 , 1 ) , t _ { a } = h _ { A } ( r , { \\widetilde { \\epsilon } } _ { h } ) , \\quad r = \\Phi ( x )", + "type": "interline_equation", + "image_path": "5d0fe636d034b8d80f53f9ef96d49f1ce924a440335a5cca19c9ccfdb29ad614.jpg" + } + ] + } + ], + "index": 38, + "virtual_lines": [ + { + "bbox": [ + 138, + 570, + 473, + 584 + ], + "spans": [], + "index": 38 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 592, + 505, + 671 + ], + "lines": [ + { + "bbox": [ + 105, + 590, + 506, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 135, + 607 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 592, + 214, + 605 + ], + "score": 0.88, + "content": "\\{ U _ { h } , W _ { h } \\} \\in \\mathbb { R } ^ { d \\times d }", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 590, + 219, + 607 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 219, + 592, + 276, + 605 + ], + "score": 0.81, + "content": "\\{ b _ { h } , \\epsilon \\} \\in \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 590, + 280, + 607 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 281, + 594, + 288, + 603 + ], + "score": 0.65, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 590, + 506, + 607 + ], + "score": 1.0, + "content": "is the dimensionality of the normalizing flow; each", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 604, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 168, + 617 + ], + "score": 1.0, + "content": "component of", + "type": "text" + }, + { + "bbox": [ + 168, + 607, + 173, + 614 + ], + "score": 0.74, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 604, + 337, + 617 + ], + "score": 1.0, + "content": "is drawn independently from Uniform", + "type": "text" + }, + { + "bbox": [ + 338, + 604, + 360, + 616 + ], + "score": 0.8, + "content": "( 0 , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 604, + 383, + 617 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 384, + 605, + 394, + 615 + ], + "score": 0.88, + "content": "\\tilde { \\epsilon } _ { h }", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 604, + 505, + 617 + ], + "score": 1.0, + "content": "may be viewed as a skip", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 615, + 506, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 240, + 629 + ], + "score": 1.0, + "content": "connection with stochasticity in", + "type": "text" + }, + { + "bbox": [ + 240, + 618, + 246, + 625 + ], + "score": 0.64, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 615, + 287, + 629 + ], + "score": 1.0, + "content": ". Further,", + "type": "text" + }, + { + "bbox": [ + 287, + 615, + 326, + 627 + ], + "score": 0.93, + "content": "h _ { A } ( r , \\tilde { \\epsilon } _ { h } )", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 615, + 346, + 629 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 346, + 616, + 368, + 627 + ], + "score": 0.92, + "content": "\\Phi ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 615, + 506, + 629 + ], + "score": 1.0, + "content": "are time-to-event generative and", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 626, + 506, + 638 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 506, + 638 + ], + "score": 1.0, + "content": "encoding functions, respectively, parameterized as neural networks. For simplicity, the dimensions of", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 107, + 638, + 505, + 649 + ], + "spans": [ + { + "bbox": [ + 107, + 640, + 113, + 647 + ], + "score": 0.73, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 113, + 638, + 130, + 649 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 131, + 639, + 136, + 647 + ], + "score": 0.74, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 137, + 638, + 176, + 649 + ], + "score": 1.0, + "content": "are set to", + "type": "text" + }, + { + "bbox": [ + 176, + 638, + 182, + 647 + ], + "score": 0.79, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 638, + 505, + 649 + ], + "score": 1.0, + "content": ", however, they can be set independently if desired. In practice, we are interested", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 649, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 505, + 660 + ], + "score": 1.0, + "content": "in generating realistic event-time samples; therefore, we account for both censored and non-censored", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 659, + 438, + 672 + ], + "spans": [ + { + "bbox": [ + 106, + 659, + 438, + 672 + ], + "score": 1.0, + "content": "observations by adopting the objective from Chapfuwa et al. (2018), formulated as", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 42, + "bbox_fs": [ + 105, + 590, + 506, + 672 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 141, + 678, + 469, + 695 + ], + "lines": [ + { + "bbox": [ + 141, + 678, + 469, + 695 + ], + "spans": [ + { + "bbox": [ + 141, + 678, + 469, + 695 + ], + "score": 0.88, + "content": "\\mathcal { L } _ { \\mathrm { F } } ^ { \\mathrm { C S A } } \\triangleq \\mathbb { E } _ { ( y , \\delta , x , a ) \\sim p ( Y , \\delta , X , A ) , \\epsilon \\sim p ( \\epsilon ) } \\left[ \\delta \\cdot \\left( \\left| y - t _ { a } \\right| \\right) + \\left( 1 - \\delta \\right) \\cdot \\left( \\operatorname* { m a x } ( 0 , y - t _ { a } ) \\right) \\right] ,", + "type": "interline_equation", + "image_path": "7fde54abcbf1fd3fd611de6768c944eaf23863c85ea57f28571b691d6e808338.jpg" + } + ] + } + ], + "index": 46, + "virtual_lines": [ + { + "bbox": [ + 141, + 678, + 469, + 695 + ], + "spans": [], + "index": 46 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 702, + 506, + 735 + ], + "lines": [ + { + "bbox": [ + 105, + 702, + 506, + 715 + ], + "spans": [ + { + "bbox": [ + 105, + 702, + 313, + 715 + ], + "score": 1.0, + "content": "where the first term encourages sampled event times", + "type": "text" + }, + { + "bbox": [ + 313, + 703, + 322, + 713 + ], + "score": 0.87, + "content": "t _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 702, + 378, + 715 + ], + "score": 1.0, + "content": "to be close to", + "type": "text" + }, + { + "bbox": [ + 378, + 704, + 384, + 713 + ], + "score": 0.8, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 702, + 506, + 715 + ], + "score": 1.0, + "content": ", the ground truth for observed", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 713, + 506, + 725 + ], + "spans": [ + { + "bbox": [ + 106, + 713, + 152, + 725 + ], + "score": 1.0, + "content": "events, i.e.,", + "type": "text" + }, + { + "bbox": [ + 153, + 713, + 177, + 723 + ], + "score": 0.9, + "content": "\\delta = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 713, + 248, + 725 + ], + "score": 1.0, + "content": ", while penalizing", + "type": "text" + }, + { + "bbox": [ + 248, + 714, + 257, + 724 + ], + "score": 0.86, + "content": "t _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 713, + 444, + 725 + ], + "score": 1.0, + "content": "for being smaller than the censoring time when", + "type": "text" + }, + { + "bbox": [ + 444, + 713, + 469, + 723 + ], + "score": 0.89, + "content": "\\delta = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 713, + 506, + 725 + ], + "score": 1.0, + "content": ". Further,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 724, + 480, + 736 + ], + "spans": [ + { + "bbox": [ + 106, + 724, + 423, + 736 + ], + "score": 1.0, + "content": "the expectation is taken over samples (a minibatch) from empirical distribution", + "type": "text" + }, + { + "bbox": [ + 424, + 724, + 477, + 736 + ], + "score": 0.92, + "content": "p ( Y , \\delta , X , A )", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 724, + 480, + 736 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 48, + "bbox_fs": [ + 105, + 702, + 506, + 736 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 505, + 138 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "Informative censoring We model informative censoring similar to (7) but mirroring the censoring", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 92, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 92, + 344, + 107 + ], + "score": 1.0, + "content": "indicators to encourage accurate censoring time samples", + "type": "text" + }, + { + "bbox": [ + 345, + 95, + 355, + 105 + ], + "score": 0.85, + "content": "c _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 92, + 373, + 107 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 373, + 94, + 401, + 104 + ], + "score": 0.9, + "content": "\\delta = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 92, + 478, + 107 + ], + "score": 1.0, + "content": ", while penalizing", + "type": "text" + }, + { + "bbox": [ + 478, + 95, + 488, + 105 + ], + "score": 0.83, + "content": "c _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 92, + 506, + 107 + ], + "score": 1.0, + "content": "for", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 104, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 104, + 187, + 118 + ], + "score": 1.0, + "content": "being smaller than", + "type": "text" + }, + { + "bbox": [ + 187, + 106, + 195, + 116 + ], + "score": 0.79, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 104, + 212, + 118 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 212, + 105, + 241, + 115 + ], + "score": 0.9, + "content": "\\delta \\ : = \\ : 1", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 104, + 505, + 118 + ], + "score": 1.0, + "content": "(observed events). Specifically, we set an independent source", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 303, + 128 + ], + "score": 1.0, + "content": "of randomness like in (6) but parameterized by", + "type": "text" + }, + { + "bbox": [ + 303, + 115, + 358, + 127 + ], + "score": 0.91, + "content": "\\{ U _ { \\nu } , W _ { \\nu } , b _ { \\nu } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 115, + 506, + 128 + ], + "score": 1.0, + "content": "and censoring generative functions", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 107, + 126, + 466, + 140 + ], + "spans": [ + { + "bbox": [ + 107, + 126, + 144, + 138 + ], + "score": 0.93, + "content": "\\nu _ { A } ( r , \\tilde { \\epsilon } _ { \\nu } )", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 126, + 313, + 140 + ], + "score": 1.0, + "content": ", parameterized as neural networks, where", + "type": "text" + }, + { + "bbox": [ + 314, + 127, + 406, + 139 + ], + "score": 0.91, + "content": "c _ { a } \\sim p _ { \\nu , \\Phi } ( C _ { A } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 126, + 466, + 140 + ], + "score": 1.0, + "content": "formulated as", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "interline_equation", + "bbox": [ + 135, + 143, + 476, + 158 + ], + "lines": [ + { + "bbox": [ + 135, + 143, + 476, + 158 + ], + "spans": [ + { + "bbox": [ + 135, + 143, + 476, + 158 + ], + "score": 0.88, + "content": "\\ell _ { c } ( \\nu , \\Phi ) = \\mathbb { E } _ { ( y , \\delta , x , a ) \\sim p ( y , \\delta , X , A ) , \\epsilon \\sim p ( \\epsilon ) } \\left[ ( 1 - \\delta ) \\cdot ( | y - c _ { a } | ) + \\delta \\cdot ( \\operatorname* { m a x } ( 0 , y - c _ { a } ) ) \\right] .", + "type": "interline_equation", + "image_path": "30f94ae655fc0f98b5a994dfdb2c983552b9b03f08d4eae0320432c33861540e.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 135, + 143, + 476, + 158 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 162, + 504, + 185 + ], + "lines": [ + { + "bbox": [ + 105, + 161, + 505, + 174 + ], + "spans": [ + { + "bbox": [ + 105, + 161, + 505, + 174 + ], + "score": 1.0, + "content": "Further, we introduce an additional time-order-consistency loss that enforces the correct order of the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 172, + 494, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 172, + 320, + 185 + ], + "score": 1.0, + "content": "observed time relative to the censoring indicator, i.e.,", + "type": "text" + }, + { + "bbox": [ + 320, + 174, + 352, + 184 + ], + "score": 0.92, + "content": "c _ { a } < t _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 172, + 362, + 185 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 362, + 173, + 387, + 183 + ], + "score": 0.9, + "content": "\\delta = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 172, + 404, + 185 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 404, + 174, + 436, + 184 + ], + "score": 0.92, + "content": "t \\sb a < c \\sb a", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 172, + 446, + 185 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 447, + 173, + 470, + 183 + ], + "score": 0.9, + "content": "\\delta = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 172, + 494, + 185 + ], + "score": 1.0, + "content": ", thus", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "interline_equation", + "bbox": [ + 109, + 189, + 483, + 204 + ], + "lines": [ + { + "bbox": [ + 109, + 189, + 483, + 204 + ], + "spans": [ + { + "bbox": [ + 109, + 189, + 483, + 204 + ], + "score": 0.88, + "content": "\\ell _ { \\mathrm { T C } } ( h , \\nu , \\Phi ) = \\mathbb { E } _ { ( \\delta , x , a ) \\sim p ( \\delta , X , A ) , \\epsilon \\sim p ( \\epsilon ) } \\left[ \\delta \\cdot ( \\operatorname* { m a x } ( 0 , t _ { a } - c _ { a } ) ) + ( 1 - \\delta ) \\cdot ( \\operatorname* { m a x } ( 0 , c _ { a } - t _ { a } ) ) \\right] .", + "type": "interline_equation", + "image_path": "5a8c95490dd905c65beac4f71a501b50581c5ec62609efa122fed7c912a3ba02.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 109, + 189, + 483, + 204 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 209, + 505, + 245 + ], + "lines": [ + { + "bbox": [ + 104, + 207, + 506, + 224 + ], + "spans": [ + { + "bbox": [ + 104, + 207, + 148, + 224 + ], + "score": 1.0, + "content": "Note that", + "type": "text" + }, + { + "bbox": [ + 149, + 210, + 200, + 222 + ], + "score": 0.93, + "content": "\\ell _ { \\mathrm { T C } } ( h , \\nu , \\Phi )", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 207, + 506, + 224 + ], + "score": 1.0, + "content": "does not depend on the observed event times but only on the censoring", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 219, + 505, + 234 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 505, + 234 + ], + "score": 1.0, + "content": "indicators. Finally, we write the consolidated CSA loss for informative censoring (CSA-INFO) by", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 102, + 226, + 372, + 251 + ], + "spans": [ + { + "bbox": [ + 102, + 226, + 228, + 251 + ], + "score": 1.0, + "content": "aggregating (7), (8) and (9) as", + "type": "text" + }, + { + "bbox": [ + 228, + 231, + 365, + 245 + ], + "score": 0.82, + "content": "\\mathcal { L } _ { \\mathrm { F } } ^ { \\mathrm { C S A - I N F O } } \\triangleq \\mathcal { L } _ { \\mathrm { F } } ^ { \\mathrm { C S A } } + \\ell _ { \\mathrm { c } } + \\ell _ { \\mathrm { T C } }", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 226, + 372, + 251 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 250, + 506, + 319 + ], + "lines": [ + { + "bbox": [ + 105, + 250, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 250, + 227, + 264 + ], + "score": 1.0, + "content": "Learning Model functions", + "type": "text" + }, + { + "bbox": [ + 228, + 250, + 310, + 263 + ], + "score": 0.92, + "content": "\\{ h _ { A } ( \\cdot ) , \\Phi ( \\cdot ) , \\nu _ { A } ( \\cdot ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 250, + 506, + 264 + ], + "score": 1.0, + "content": "are learned by minimizing the bound (5), via", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 104, + 259, + 507, + 276 + ], + "spans": [ + { + "bbox": [ + 104, + 259, + 304, + 276 + ], + "score": 1.0, + "content": "stochastic gradient descent on minibatches from", + "type": "text" + }, + { + "bbox": [ + 304, + 263, + 313, + 272 + ], + "score": 0.82, + "content": "\\mathcal { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 259, + 338, + 276 + ], + "score": 1.0, + "content": ", with", + "type": "text" + }, + { + "bbox": [ + 338, + 261, + 362, + 274 + ], + "score": 0.92, + "content": "\\mathcal { L } _ { \\mathrm { F } } ^ { \\mathrm { C S A } }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 259, + 507, + 276 + ], + "score": 1.0, + "content": "for non-informative censoring and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 107, + 268, + 509, + 293 + ], + "spans": [ + { + "bbox": [ + 107, + 273, + 157, + 287 + ], + "score": 0.88, + "content": "\\mathcal { L } _ { \\mathrm { F } } ^ { \\mathrm { C S A - I N F O } }", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 268, + 509, + 293 + ], + "score": 1.0, + "content": "for informative censoring. Further, for the IPM regularization loss in (5), we optimize", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 285, + 506, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 506, + 299 + ], + "score": 1.0, + "content": "the dual formulation of the Wasserstein distance, via the regularized optimal transport (Villani, 2008;", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 296, + 506, + 309 + ], + "spans": [ + { + "bbox": [ + 106, + 296, + 293, + 309 + ], + "score": 1.0, + "content": "Cuturi, 2013). Consequently, we only require", + "type": "text" + }, + { + "bbox": [ + 294, + 296, + 352, + 309 + ], + "score": 0.93, + "content": "\\alpha ^ { - 1 } \\zeta _ { h , \\Phi } ( \\bar { x , } a )", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 296, + 506, + 309 + ], + "score": 1.0, + "content": "to be 1-Lipschitz (Shalit et al., 2017)", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 307, + 474, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 123, + 320 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 123, + 309, + 131, + 317 + ], + "score": 0.78, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 307, + 474, + 320 + ], + "score": 1.0, + "content": "is selected by grid search on the validation set using only factual data (details below).", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14.5 + }, + { + "type": "title", + "bbox": [ + 107, + 335, + 174, + 347 + ], + "lines": [ + { + "bbox": [ + 104, + 333, + 176, + 350 + ], + "spans": [ + { + "bbox": [ + 104, + 333, + 176, + 350 + ], + "score": 1.0, + "content": "4 METRICS", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 351, + 506, + 461 + ], + "lines": [ + { + "bbox": [ + 106, + 352, + 507, + 364 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 507, + 364 + ], + "score": 1.0, + "content": "We propose a comprehensive evaluation approach that accounts for both factual and causal metrics.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 362, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 506, + 375 + ], + "score": 1.0, + "content": "Factual survival outcome predictions are evaluated according to standard survival metrics that measure", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 373, + 506, + 386 + ], + "spans": [ + { + "bbox": [ + 106, + 373, + 506, + 386 + ], + "score": 1.0, + "content": "diverse performance characteristics, such as concordance index (C-Index) (Harrell Jr et al., 1984),", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 384, + 506, + 397 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 506, + 397 + ], + "score": 1.0, + "content": "mean coefficient of variation (COV) and calibration slope (C-slope) (Chapfuwa et al., 2020). See the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 394, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 394, + 505, + 408 + ], + "score": 1.0, + "content": "SM for more details on these metrics. For causal metrics, defined below, we introduce a nonparametric", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 406, + 506, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 506, + 419 + ], + "score": 1.0, + "content": "hazard ratio (HR) between treatment outcomes, and adopt the conventional precision in estimation of", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 417, + 507, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 507, + 430 + ], + "score": 1.0, + "content": "heterogeneous effect (PEHE) and average treatment effect (ATE) performance metrics (Hill, 2011).", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 428, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 505, + 441 + ], + "score": 1.0, + "content": "Note that PEHE and ATE require ground truth counterfactual event times, which is only possible in", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 439, + 506, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 506, + 452 + ], + "score": 1.0, + "content": "(semi-)synthetic data. For HR, we compare our findings with those independently reported in the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 449, + 268, + 463 + ], + "spans": [ + { + "bbox": [ + 105, + 449, + 268, + 463 + ], + "score": 1.0, + "content": "literature from gold-standard RCT data.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 106, + 465, + 506, + 625 + ], + "lines": [ + { + "bbox": [ + 105, + 465, + 506, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 439, + 479 + ], + "score": 1.0, + "content": "Nonparametric Hazard Ratio In a medical setting, the population hazard ratio", + "type": "text" + }, + { + "bbox": [ + 440, + 465, + 467, + 477 + ], + "score": 0.88, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 467, + 465, + 506, + 479 + ], + "score": 1.0, + "content": "between", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 476, + 506, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 476, + 506, + 489 + ], + "score": 1.0, + "content": "treatment groups is considered informative thus has been widely used in drug development and RCT", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 487, + 506, + 500 + ], + "spans": [ + { + "bbox": [ + 106, + 487, + 330, + 500 + ], + "score": 1.0, + "content": "(Yusuf et al., 2016; Mihaylova et al., 2012). For example,", + "type": "text" + }, + { + "bbox": [ + 331, + 487, + 396, + 499 + ], + "score": 0.9, + "content": "\\mathrm { H R } ( t ) < 1 , > 1", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 487, + 409, + 500 + ], + "score": 1.0, + "content": ", or", + "type": "text" + }, + { + "bbox": [ + 410, + 488, + 426, + 497 + ], + "score": 0.81, + "content": "\\approx 1", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 487, + 506, + 500 + ], + "score": 1.0, + "content": "indicate population", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 497, + 506, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 329, + 512 + ], + "score": 1.0, + "content": "positive, negative and neutral treatment effects at time", + "type": "text" + }, + { + "bbox": [ + 330, + 500, + 335, + 508 + ], + "score": 0.62, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 497, + 438, + 512 + ], + "score": 1.0, + "content": ", respectively. Moreover,", + "type": "text" + }, + { + "bbox": [ + 438, + 498, + 465, + 510 + ], + "score": 0.76, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 497, + 506, + 512 + ], + "score": 1.0, + "content": "naturally", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 509, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 477, + 522 + ], + "score": 1.0, + "content": "accounts for both censored and non-censored outcomes. Standard approaches for computing", + "type": "text" + }, + { + "bbox": [ + 477, + 509, + 505, + 521 + ], + "score": 0.8, + "content": "\\mathrm { H R } ( { \\dot { t } } )", + "type": "inline_equation" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 520, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 505, + 533 + ], + "score": 1.0, + "content": "rely on the restrictive proportional hazard assumption from CoxPH (Cox, 1972), which is constituted", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 531, + 505, + 544 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 246, + 544 + ], + "score": 1.0, + "content": "as a semi-parametric linear model", + "type": "text" + }, + { + "bbox": [ + 247, + 531, + 345, + 543 + ], + "score": 0.92, + "content": "\\lambda ( t | a ) = \\lambda _ { \\mathrm { b } } \\bar { ( } t ) \\exp ( a \\beta )", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 531, + 505, + 544 + ], + "score": 1.0, + "content": ". However, the constant covariate (time", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 542, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 542, + 505, + 554 + ], + "score": 1.0, + "content": "independent) effect is often violated in practice (see Figure 2b). For CoxPH, the marginal HR between", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 552, + 506, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 365, + 566 + ], + "score": 1.0, + "content": "treatment and control can be obtained from regression coefficient", + "type": "text" + }, + { + "bbox": [ + 366, + 554, + 373, + 564 + ], + "score": 0.84, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 552, + 506, + 566 + ], + "score": 1.0, + "content": "learned via maximum likelihood", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 563, + 506, + 581 + ], + "spans": [ + { + "bbox": [ + 104, + 563, + 316, + 581 + ], + "score": 1.0, + "content": "without the need for specifying the baseline hazard", + "type": "text" + }, + { + "bbox": [ + 322, + 564, + 487, + 581 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\mathrm { ~ \\cdot ~ } _ { \\mathrm { b } } ( t ) \\colon \\mathrm { H R } _ { \\mathrm { C o x P H } } ( t ) = \\frac { \\lambda ( t | a = 1 ) } { \\lambda ( t | a = 0 ) } = \\exp ( \\beta ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 488, + 563, + 506, + 581 + ], + "score": 1.0, + "content": ". So", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 580, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 580, + 411, + 592 + ], + "score": 1.0, + "content": "motivated, we propose a nonparametric, model-free approach for computing", + "type": "text" + }, + { + "bbox": [ + 411, + 581, + 438, + 592 + ], + "score": 0.85, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 580, + 506, + 592 + ], + "score": 1.0, + "content": ", in which we do", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 590, + 506, + 605 + ], + "spans": [ + { + "bbox": [ + 104, + 590, + 506, + 605 + ], + "score": 1.0, + "content": "not assume a parametric form for the event time distribution or the proportional hazard assumption", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 601, + 506, + 615 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 506, + 615 + ], + "score": 1.0, + "content": "from CoxPH. This approach only relies on samples from the conditional event time density functions,", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 107, + 613, + 291, + 626 + ], + "spans": [ + { + "bbox": [ + 107, + 613, + 138, + 625 + ], + "score": 0.92, + "content": "f ( t _ { 1 } | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 613, + 156, + 626 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 156, + 613, + 187, + 625 + ], + "score": 0.92, + "content": "f ( t _ { 0 } | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 613, + 205, + 626 + ], + "score": 1.0, + "content": ", via", + "type": "text" + }, + { + "bbox": [ + 206, + 613, + 251, + 625 + ], + "score": 0.92, + "content": "t _ { a } = h _ { A } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 613, + 291, + 626 + ], + "score": 1.0, + "content": "from (6).", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 105, + 629, + 504, + 642 + ], + "lines": [ + { + "bbox": [ + 105, + 627, + 505, + 644 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 462, + 644 + ], + "score": 1.0, + "content": "Definition 1 We define the nonparametric marginal Hazard Ratio and its approximation,", + "type": "text" + }, + { + "bbox": [ + 463, + 628, + 489, + 642 + ], + "score": 0.9, + "content": "\\mathrm { \\hat { H R } } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 490, + 627, + 505, + 644 + ], + "score": 1.0, + "content": ", as", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 43 + }, + { + "type": "interline_equation", + "bbox": [ + 171, + 647, + 439, + 677 + ], + "lines": [ + { + "bbox": [ + 171, + 647, + 439, + 677 + ], + "spans": [ + { + "bbox": [ + 171, + 647, + 439, + 677 + ], + "score": 0.95, + "content": "\\mathrm { H R } ( t ) = \\frac { \\lambda _ { 1 } ( t ) } { \\lambda _ { 0 } ( t ) } = \\frac { S _ { 0 } ( t ) } { S _ { 1 } ( t ) } \\cdot \\frac { S _ { 1 } ^ { \\prime } ( t ) } { S _ { 0 } ^ { \\prime } ( t ) } , \\quad \\mathrm { H R } ( t ) = \\frac { \\hat { S } _ { 0 } ^ { \\mathrm { P K M } } ( t ) } { \\hat { S } _ { 1 } ^ { \\mathrm { P K M } } ( t ) } \\cdot \\frac { m _ { 1 } ( t ) } { m _ { 0 } ( t ) } ,", + "type": "interline_equation", + "image_path": "d9ca3513b198afb6432afe6930b65739952ee5cac99ad41dbc5cf652e4ac4216.jpg" + } + ] + } + ], + "index": 45, + "virtual_lines": [ + { + "bbox": [ + 171, + 647, + 439, + 657.0 + ], + "spans": [], + "index": 44 + }, + { + "bbox": [ + 171, + 657.0, + 439, + 667.0 + ], + "spans": [], + "index": 45 + }, + { + "bbox": [ + 171, + 667.0, + 439, + 677.0 + ], + "spans": [], + "index": 46 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 686, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 685, + 507, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 685, + 146, + 702 + ], + "score": 1.0, + "content": "where for", + "type": "text" + }, + { + "bbox": [ + 147, + 687, + 174, + 699 + ], + "score": 0.88, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 685, + 317, + 702 + ], + "score": 1.0, + "content": "we leveraged (1) to obtain (10) and", + "type": "text" + }, + { + "bbox": [ + 318, + 687, + 389, + 700 + ], + "score": 0.91, + "content": "S ^ { \\prime } ( t ) \\triangleq d S ( t ) / d t", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 685, + 507, + 702 + ], + "score": 1.0, + "content": ". The nonparametric assump-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 140, + 712 + ], + "score": 1.0, + "content": "tion for", + "type": "text" + }, + { + "bbox": [ + 140, + 699, + 159, + 711 + ], + "score": 0.91, + "content": "S ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 698, + 270, + 712 + ], + "score": 1.0, + "content": "makes the computation of", + "type": "text" + }, + { + "bbox": [ + 271, + 700, + 292, + 711 + ], + "score": 0.92, + "content": "S ^ { \\prime } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 698, + 406, + 712 + ], + "score": 1.0, + "content": "challenging. Provided that", + "type": "text" + }, + { + "bbox": [ + 406, + 699, + 425, + 711 + ], + "score": 0.92, + "content": "S ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "is a monotonically", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 335, + 722 + ], + "score": 1.0, + "content": "decreasing function, for simplicity, we fit a linear function", + "type": "text" + }, + { + "bbox": [ + 336, + 710, + 403, + 722 + ], + "score": 0.91, + "content": "S ( t ) = m \\cdot t + c .", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 710, + 435, + 722 + ], + "score": 1.0, + "content": ", and set", + "type": "text" + }, + { + "bbox": [ + 435, + 709, + 479, + 722 + ], + "score": 0.92, + "content": "S ^ { \\prime } ( t ) \\approx m", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 710, + 505, + 722 + ], + "score": 1.0, + "content": ". Note", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 720, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 305, + 734 + ], + "score": 1.0, + "content": "that the linear model is only used for estimating", + "type": "text" + }, + { + "bbox": [ + 305, + 721, + 326, + 732 + ], + "score": 0.91, + "content": "S ^ { \\prime } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 720, + 484, + 734 + ], + "score": 1.0, + "content": "from the nonparametric estimation of", + "type": "text" + }, + { + "bbox": [ + 484, + 721, + 502, + 732 + ], + "score": 0.92, + "content": "S ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 720, + 506, + 734 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 48.5 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 307, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 310, + 761 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 310, + 761 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 505, + 138 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "Informative censoring We model informative censoring similar to (7) but mirroring the censoring", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 92, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 92, + 344, + 107 + ], + "score": 1.0, + "content": "indicators to encourage accurate censoring time samples", + "type": "text" + }, + { + "bbox": [ + 345, + 95, + 355, + 105 + ], + "score": 0.85, + "content": "c _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 92, + 373, + 107 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 373, + 94, + 401, + 104 + ], + "score": 0.9, + "content": "\\delta = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 92, + 478, + 107 + ], + "score": 1.0, + "content": ", while penalizing", + "type": "text" + }, + { + "bbox": [ + 478, + 95, + 488, + 105 + ], + "score": 0.83, + "content": "c _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 92, + 506, + 107 + ], + "score": 1.0, + "content": "for", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 104, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 104, + 187, + 118 + ], + "score": 1.0, + "content": "being smaller than", + "type": "text" + }, + { + "bbox": [ + 187, + 106, + 195, + 116 + ], + "score": 0.79, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 104, + 212, + 118 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 212, + 105, + 241, + 115 + ], + "score": 0.9, + "content": "\\delta \\ : = \\ : 1", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 104, + 505, + 118 + ], + "score": 1.0, + "content": "(observed events). Specifically, we set an independent source", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 303, + 128 + ], + "score": 1.0, + "content": "of randomness like in (6) but parameterized by", + "type": "text" + }, + { + "bbox": [ + 303, + 115, + 358, + 127 + ], + "score": 0.91, + "content": "\\{ U _ { \\nu } , W _ { \\nu } , b _ { \\nu } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 115, + 506, + 128 + ], + "score": 1.0, + "content": "and censoring generative functions", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 107, + 126, + 466, + 140 + ], + "spans": [ + { + "bbox": [ + 107, + 126, + 144, + 138 + ], + "score": 0.93, + "content": "\\nu _ { A } ( r , \\tilde { \\epsilon } _ { \\nu } )", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 126, + 313, + 140 + ], + "score": 1.0, + "content": ", parameterized as neural networks, where", + "type": "text" + }, + { + "bbox": [ + 314, + 127, + 406, + 139 + ], + "score": 0.91, + "content": "c _ { a } \\sim p _ { \\nu , \\Phi } ( C _ { A } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 126, + 466, + 140 + ], + "score": 1.0, + "content": "formulated as", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 81, + 506, + 140 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 135, + 143, + 476, + 158 + ], + "lines": [ + { + "bbox": [ + 135, + 143, + 476, + 158 + ], + "spans": [ + { + "bbox": [ + 135, + 143, + 476, + 158 + ], + "score": 0.88, + "content": "\\ell _ { c } ( \\nu , \\Phi ) = \\mathbb { E } _ { ( y , \\delta , x , a ) \\sim p ( y , \\delta , X , A ) , \\epsilon \\sim p ( \\epsilon ) } \\left[ ( 1 - \\delta ) \\cdot ( | y - c _ { a } | ) + \\delta \\cdot ( \\operatorname* { m a x } ( 0 , y - c _ { a } ) ) \\right] .", + "type": "interline_equation", + "image_path": "30f94ae655fc0f98b5a994dfdb2c983552b9b03f08d4eae0320432c33861540e.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 135, + 143, + 476, + 158 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 162, + 504, + 185 + ], + "lines": [ + { + "bbox": [ + 105, + 161, + 505, + 174 + ], + "spans": [ + { + "bbox": [ + 105, + 161, + 505, + 174 + ], + "score": 1.0, + "content": "Further, we introduce an additional time-order-consistency loss that enforces the correct order of the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 172, + 494, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 172, + 320, + 185 + ], + "score": 1.0, + "content": "observed time relative to the censoring indicator, i.e.,", + "type": "text" + }, + { + "bbox": [ + 320, + 174, + 352, + 184 + ], + "score": 0.92, + "content": "c _ { a } < t _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 172, + 362, + 185 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 362, + 173, + 387, + 183 + ], + "score": 0.9, + "content": "\\delta = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 172, + 404, + 185 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 404, + 174, + 436, + 184 + ], + "score": 0.92, + "content": "t \\sb a < c \\sb a", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 172, + 446, + 185 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 447, + 173, + 470, + 183 + ], + "score": 0.9, + "content": "\\delta = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 172, + 494, + 185 + ], + "score": 1.0, + "content": ", thus", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5, + "bbox_fs": [ + 105, + 161, + 505, + 185 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 109, + 189, + 483, + 204 + ], + "lines": [ + { + "bbox": [ + 109, + 189, + 483, + 204 + ], + "spans": [ + { + "bbox": [ + 109, + 189, + 483, + 204 + ], + "score": 0.88, + "content": "\\ell _ { \\mathrm { T C } } ( h , \\nu , \\Phi ) = \\mathbb { E } _ { ( \\delta , x , a ) \\sim p ( \\delta , X , A ) , \\epsilon \\sim p ( \\epsilon ) } \\left[ \\delta \\cdot ( \\operatorname* { m a x } ( 0 , t _ { a } - c _ { a } ) ) + ( 1 - \\delta ) \\cdot ( \\operatorname* { m a x } ( 0 , c _ { a } - t _ { a } ) ) \\right] .", + "type": "interline_equation", + "image_path": "5a8c95490dd905c65beac4f71a501b50581c5ec62609efa122fed7c912a3ba02.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 109, + 189, + 483, + 204 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 209, + 505, + 245 + ], + "lines": [ + { + "bbox": [ + 104, + 207, + 506, + 224 + ], + "spans": [ + { + "bbox": [ + 104, + 207, + 148, + 224 + ], + "score": 1.0, + "content": "Note that", + "type": "text" + }, + { + "bbox": [ + 149, + 210, + 200, + 222 + ], + "score": 0.93, + "content": "\\ell _ { \\mathrm { T C } } ( h , \\nu , \\Phi )", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 207, + 506, + 224 + ], + "score": 1.0, + "content": "does not depend on the observed event times but only on the censoring", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 219, + 505, + 234 + ], + "spans": [ + { + "bbox": [ + 105, + 219, + 505, + 234 + ], + "score": 1.0, + "content": "indicators. Finally, we write the consolidated CSA loss for informative censoring (CSA-INFO) by", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 102, + 226, + 372, + 251 + ], + "spans": [ + { + "bbox": [ + 102, + 226, + 228, + 251 + ], + "score": 1.0, + "content": "aggregating (7), (8) and (9) as", + "type": "text" + }, + { + "bbox": [ + 228, + 231, + 365, + 245 + ], + "score": 0.82, + "content": "\\mathcal { L } _ { \\mathrm { F } } ^ { \\mathrm { C S A - I N F O } } \\triangleq \\mathcal { L } _ { \\mathrm { F } } ^ { \\mathrm { C S A } } + \\ell _ { \\mathrm { c } } + \\ell _ { \\mathrm { T C } }", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 226, + 372, + 251 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 102, + 207, + 506, + 251 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 250, + 506, + 319 + ], + "lines": [ + { + "bbox": [ + 105, + 250, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 250, + 227, + 264 + ], + "score": 1.0, + "content": "Learning Model functions", + "type": "text" + }, + { + "bbox": [ + 228, + 250, + 310, + 263 + ], + "score": 0.92, + "content": "\\{ h _ { A } ( \\cdot ) , \\Phi ( \\cdot ) , \\nu _ { A } ( \\cdot ) \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 250, + 506, + 264 + ], + "score": 1.0, + "content": "are learned by minimizing the bound (5), via", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 104, + 259, + 507, + 276 + ], + "spans": [ + { + "bbox": [ + 104, + 259, + 304, + 276 + ], + "score": 1.0, + "content": "stochastic gradient descent on minibatches from", + "type": "text" + }, + { + "bbox": [ + 304, + 263, + 313, + 272 + ], + "score": 0.82, + "content": "\\mathcal { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 259, + 338, + 276 + ], + "score": 1.0, + "content": ", with", + "type": "text" + }, + { + "bbox": [ + 338, + 261, + 362, + 274 + ], + "score": 0.92, + "content": "\\mathcal { L } _ { \\mathrm { F } } ^ { \\mathrm { C S A } }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 259, + 507, + 276 + ], + "score": 1.0, + "content": "for non-informative censoring and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 107, + 268, + 509, + 293 + ], + "spans": [ + { + "bbox": [ + 107, + 273, + 157, + 287 + ], + "score": 0.88, + "content": "\\mathcal { L } _ { \\mathrm { F } } ^ { \\mathrm { C S A - I N F O } }", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 268, + 509, + 293 + ], + "score": 1.0, + "content": "for informative censoring. Further, for the IPM regularization loss in (5), we optimize", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 285, + 506, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 506, + 299 + ], + "score": 1.0, + "content": "the dual formulation of the Wasserstein distance, via the regularized optimal transport (Villani, 2008;", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 296, + 506, + 309 + ], + "spans": [ + { + "bbox": [ + 106, + 296, + 293, + 309 + ], + "score": 1.0, + "content": "Cuturi, 2013). Consequently, we only require", + "type": "text" + }, + { + "bbox": [ + 294, + 296, + 352, + 309 + ], + "score": 0.93, + "content": "\\alpha ^ { - 1 } \\zeta _ { h , \\Phi } ( \\bar { x , } a )", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 296, + 506, + 309 + ], + "score": 1.0, + "content": "to be 1-Lipschitz (Shalit et al., 2017)", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 307, + 474, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 123, + 320 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 123, + 309, + 131, + 317 + ], + "score": 0.78, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 307, + 474, + 320 + ], + "score": 1.0, + "content": "is selected by grid search on the validation set using only factual data (details below).", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14.5, + "bbox_fs": [ + 104, + 250, + 509, + 320 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 335, + 174, + 347 + ], + "lines": [ + { + "bbox": [ + 104, + 333, + 176, + 350 + ], + "spans": [ + { + "bbox": [ + 104, + 333, + 176, + 350 + ], + "score": 1.0, + "content": "4 METRICS", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 351, + 506, + 461 + ], + "lines": [ + { + "bbox": [ + 106, + 352, + 507, + 364 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 507, + 364 + ], + "score": 1.0, + "content": "We propose a comprehensive evaluation approach that accounts for both factual and causal metrics.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 362, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 506, + 375 + ], + "score": 1.0, + "content": "Factual survival outcome predictions are evaluated according to standard survival metrics that measure", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 373, + 506, + 386 + ], + "spans": [ + { + "bbox": [ + 106, + 373, + 506, + 386 + ], + "score": 1.0, + "content": "diverse performance characteristics, such as concordance index (C-Index) (Harrell Jr et al., 1984),", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 384, + 506, + 397 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 506, + 397 + ], + "score": 1.0, + "content": "mean coefficient of variation (COV) and calibration slope (C-slope) (Chapfuwa et al., 2020). See the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 394, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 394, + 505, + 408 + ], + "score": 1.0, + "content": "SM for more details on these metrics. For causal metrics, defined below, we introduce a nonparametric", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 406, + 506, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 506, + 419 + ], + "score": 1.0, + "content": "hazard ratio (HR) between treatment outcomes, and adopt the conventional precision in estimation of", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 417, + 507, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 507, + 430 + ], + "score": 1.0, + "content": "heterogeneous effect (PEHE) and average treatment effect (ATE) performance metrics (Hill, 2011).", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 428, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 505, + 441 + ], + "score": 1.0, + "content": "Note that PEHE and ATE require ground truth counterfactual event times, which is only possible in", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 439, + 506, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 506, + 452 + ], + "score": 1.0, + "content": "(semi-)synthetic data. For HR, we compare our findings with those independently reported in the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 449, + 268, + 463 + ], + "spans": [ + { + "bbox": [ + 105, + 449, + 268, + 463 + ], + "score": 1.0, + "content": "literature from gold-standard RCT data.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 23.5, + "bbox_fs": [ + 105, + 352, + 507, + 463 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 465, + 506, + 625 + ], + "lines": [ + { + "bbox": [ + 105, + 465, + 506, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 439, + 479 + ], + "score": 1.0, + "content": "Nonparametric Hazard Ratio In a medical setting, the population hazard ratio", + "type": "text" + }, + { + "bbox": [ + 440, + 465, + 467, + 477 + ], + "score": 0.88, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 467, + 465, + 506, + 479 + ], + "score": 1.0, + "content": "between", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 476, + 506, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 476, + 506, + 489 + ], + "score": 1.0, + "content": "treatment groups is considered informative thus has been widely used in drug development and RCT", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 487, + 506, + 500 + ], + "spans": [ + { + "bbox": [ + 106, + 487, + 330, + 500 + ], + "score": 1.0, + "content": "(Yusuf et al., 2016; Mihaylova et al., 2012). For example,", + "type": "text" + }, + { + "bbox": [ + 331, + 487, + 396, + 499 + ], + "score": 0.9, + "content": "\\mathrm { H R } ( t ) < 1 , > 1", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 487, + 409, + 500 + ], + "score": 1.0, + "content": ", or", + "type": "text" + }, + { + "bbox": [ + 410, + 488, + 426, + 497 + ], + "score": 0.81, + "content": "\\approx 1", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 487, + 506, + 500 + ], + "score": 1.0, + "content": "indicate population", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 497, + 506, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 329, + 512 + ], + "score": 1.0, + "content": "positive, negative and neutral treatment effects at time", + "type": "text" + }, + { + "bbox": [ + 330, + 500, + 335, + 508 + ], + "score": 0.62, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 497, + 438, + 512 + ], + "score": 1.0, + "content": ", respectively. Moreover,", + "type": "text" + }, + { + "bbox": [ + 438, + 498, + 465, + 510 + ], + "score": 0.76, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 497, + 506, + 512 + ], + "score": 1.0, + "content": "naturally", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 509, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 477, + 522 + ], + "score": 1.0, + "content": "accounts for both censored and non-censored outcomes. Standard approaches for computing", + "type": "text" + }, + { + "bbox": [ + 477, + 509, + 505, + 521 + ], + "score": 0.8, + "content": "\\mathrm { H R } ( { \\dot { t } } )", + "type": "inline_equation" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 520, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 505, + 533 + ], + "score": 1.0, + "content": "rely on the restrictive proportional hazard assumption from CoxPH (Cox, 1972), which is constituted", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 531, + 505, + 544 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 246, + 544 + ], + "score": 1.0, + "content": "as a semi-parametric linear model", + "type": "text" + }, + { + "bbox": [ + 247, + 531, + 345, + 543 + ], + "score": 0.92, + "content": "\\lambda ( t | a ) = \\lambda _ { \\mathrm { b } } \\bar { ( } t ) \\exp ( a \\beta )", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 531, + 505, + 544 + ], + "score": 1.0, + "content": ". However, the constant covariate (time", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 542, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 542, + 505, + 554 + ], + "score": 1.0, + "content": "independent) effect is often violated in practice (see Figure 2b). For CoxPH, the marginal HR between", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 552, + 506, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 365, + 566 + ], + "score": 1.0, + "content": "treatment and control can be obtained from regression coefficient", + "type": "text" + }, + { + "bbox": [ + 366, + 554, + 373, + 564 + ], + "score": 0.84, + "content": "\\beta", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 552, + 506, + 566 + ], + "score": 1.0, + "content": "learned via maximum likelihood", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 563, + 506, + 581 + ], + "spans": [ + { + "bbox": [ + 104, + 563, + 316, + 581 + ], + "score": 1.0, + "content": "without the need for specifying the baseline hazard", + "type": "text" + }, + { + "bbox": [ + 322, + 564, + 487, + 581 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\mathrm { ~ \\cdot ~ } _ { \\mathrm { b } } ( t ) \\colon \\mathrm { H R } _ { \\mathrm { C o x P H } } ( t ) = \\frac { \\lambda ( t | a = 1 ) } { \\lambda ( t | a = 0 ) } = \\exp ( \\beta ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 488, + 563, + 506, + 581 + ], + "score": 1.0, + "content": ". So", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 580, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 580, + 411, + 592 + ], + "score": 1.0, + "content": "motivated, we propose a nonparametric, model-free approach for computing", + "type": "text" + }, + { + "bbox": [ + 411, + 581, + 438, + 592 + ], + "score": 0.85, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 580, + 506, + 592 + ], + "score": 1.0, + "content": ", in which we do", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 590, + 506, + 605 + ], + "spans": [ + { + "bbox": [ + 104, + 590, + 506, + 605 + ], + "score": 1.0, + "content": "not assume a parametric form for the event time distribution or the proportional hazard assumption", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 601, + 506, + 615 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 506, + 615 + ], + "score": 1.0, + "content": "from CoxPH. This approach only relies on samples from the conditional event time density functions,", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 107, + 613, + 291, + 626 + ], + "spans": [ + { + "bbox": [ + 107, + 613, + 138, + 625 + ], + "score": 0.92, + "content": "f ( t _ { 1 } | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 613, + 156, + 626 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 156, + 613, + 187, + 625 + ], + "score": 0.92, + "content": "f ( t _ { 0 } | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 613, + 205, + 626 + ], + "score": 1.0, + "content": ", via", + "type": "text" + }, + { + "bbox": [ + 206, + 613, + 251, + 625 + ], + "score": 0.92, + "content": "t _ { a } = h _ { A } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 613, + 291, + 626 + ], + "score": 1.0, + "content": "from (6).", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 35.5, + "bbox_fs": [ + 104, + 465, + 506, + 626 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 629, + 504, + 642 + ], + "lines": [ + { + "bbox": [ + 105, + 627, + 505, + 644 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 462, + 644 + ], + "score": 1.0, + "content": "Definition 1 We define the nonparametric marginal Hazard Ratio and its approximation,", + "type": "text" + }, + { + "bbox": [ + 463, + 628, + 489, + 642 + ], + "score": 0.9, + "content": "\\mathrm { \\hat { H R } } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 490, + 627, + 505, + 644 + ], + "score": 1.0, + "content": ", as", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 43, + "bbox_fs": [ + 105, + 627, + 505, + 644 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 171, + 647, + 439, + 677 + ], + "lines": [ + { + "bbox": [ + 171, + 647, + 439, + 677 + ], + "spans": [ + { + "bbox": [ + 171, + 647, + 439, + 677 + ], + "score": 0.95, + "content": "\\mathrm { H R } ( t ) = \\frac { \\lambda _ { 1 } ( t ) } { \\lambda _ { 0 } ( t ) } = \\frac { S _ { 0 } ( t ) } { S _ { 1 } ( t ) } \\cdot \\frac { S _ { 1 } ^ { \\prime } ( t ) } { S _ { 0 } ^ { \\prime } ( t ) } , \\quad \\mathrm { H R } ( t ) = \\frac { \\hat { S } _ { 0 } ^ { \\mathrm { P K M } } ( t ) } { \\hat { S } _ { 1 } ^ { \\mathrm { P K M } } ( t ) } \\cdot \\frac { m _ { 1 } ( t ) } { m _ { 0 } ( t ) } ,", + "type": "interline_equation", + "image_path": "d9ca3513b198afb6432afe6930b65739952ee5cac99ad41dbc5cf652e4ac4216.jpg" + } + ] + } + ], + "index": 45, + "virtual_lines": [ + { + "bbox": [ + 171, + 647, + 439, + 657.0 + ], + "spans": [], + "index": 44 + }, + { + "bbox": [ + 171, + 657.0, + 439, + 667.0 + ], + "spans": [], + "index": 45 + }, + { + "bbox": [ + 171, + 667.0, + 439, + 677.0 + ], + "spans": [], + "index": 46 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 686, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 685, + 507, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 685, + 146, + 702 + ], + "score": 1.0, + "content": "where for", + "type": "text" + }, + { + "bbox": [ + 147, + 687, + 174, + 699 + ], + "score": 0.88, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 685, + 317, + 702 + ], + "score": 1.0, + "content": "we leveraged (1) to obtain (10) and", + "type": "text" + }, + { + "bbox": [ + 318, + 687, + 389, + 700 + ], + "score": 0.91, + "content": "S ^ { \\prime } ( t ) \\triangleq d S ( t ) / d t", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 685, + 507, + 702 + ], + "score": 1.0, + "content": ". The nonparametric assump-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 140, + 712 + ], + "score": 1.0, + "content": "tion for", + "type": "text" + }, + { + "bbox": [ + 140, + 699, + 159, + 711 + ], + "score": 0.91, + "content": "S ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 698, + 270, + 712 + ], + "score": 1.0, + "content": "makes the computation of", + "type": "text" + }, + { + "bbox": [ + 271, + 700, + 292, + 711 + ], + "score": 0.92, + "content": "S ^ { \\prime } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 698, + 406, + 712 + ], + "score": 1.0, + "content": "challenging. Provided that", + "type": "text" + }, + { + "bbox": [ + 406, + 699, + 425, + 711 + ], + "score": 0.92, + "content": "S ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "is a monotonically", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 335, + 722 + ], + "score": 1.0, + "content": "decreasing function, for simplicity, we fit a linear function", + "type": "text" + }, + { + "bbox": [ + 336, + 710, + 403, + 722 + ], + "score": 0.91, + "content": "S ( t ) = m \\cdot t + c .", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 710, + 435, + 722 + ], + "score": 1.0, + "content": ", and set", + "type": "text" + }, + { + "bbox": [ + 435, + 709, + 479, + 722 + ], + "score": 0.92, + "content": "S ^ { \\prime } ( t ) \\approx m", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 710, + 505, + 722 + ], + "score": 1.0, + "content": ". Note", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 720, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 305, + 734 + ], + "score": 1.0, + "content": "that the linear model is only used for estimating", + "type": "text" + }, + { + "bbox": [ + 305, + 721, + 326, + 732 + ], + "score": 0.91, + "content": "S ^ { \\prime } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 720, + 484, + 734 + ], + "score": 1.0, + "content": "from the nonparametric estimation of", + "type": "text" + }, + { + "bbox": [ + 484, + 721, + 502, + 732 + ], + "score": 0.92, + "content": "S ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 720, + 506, + 734 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 48.5, + "bbox_fs": [ + 105, + 685, + 507, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 108, + 106, + 503, + 210 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 80, + 504, + 104 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 80, + 505, + 93 + ], + "spans": [ + { + "bbox": [ + 106, + 80, + 378, + 93 + ], + "score": 1.0, + "content": "Table 1: Performance comparisons on ACTG-SYNTHETIC data, with", + "type": "text" + }, + { + "bbox": [ + 378, + 80, + 397, + 91 + ], + "score": 0.75, + "content": "9 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 80, + 425, + 92 + ], + "score": 0.45, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 80, + 505, + 93 + ], + "score": 1.0, + "content": "confidence interval.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 104, + 87, + 371, + 108 + ], + "spans": [ + { + "bbox": [ + 104, + 87, + 270, + 108 + ], + "score": 1.0, + "content": "The ground truth, test set, hazard ratio is", + "type": "text" + }, + { + "bbox": [ + 270, + 91, + 366, + 104 + ], + "score": 0.76, + "content": "\\mathrm { H R ( t ) } = 0 . 5 2 _ { ( 0 . 3 9 , 0 . 7 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 87, + 371, + 108 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table_body", + "bbox": [ + 108, + 106, + 503, + 210 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 106, + 503, + 210 + ], + "spans": [ + { + "bbox": [ + 108, + 106, + 503, + 210 + ], + "score": 0.982, + "html": "
MethodCausalFactualC-Slope (A=0,A=1)
EPEHEEATEHR(t)C-Index (A=0,A=1)Mean COV
CoxPH-UniformNANA0.97(0.86,1.09)NANANA
CoxPH-IPWNANA0.48(0.03,7.21)NANANA
CoxPH-OWNANA0.60(0.53,0.68)NANANA
Surv-BART352.0777.890.0(0.0,0.0)(0.706,0.686)0.001(0.398,0)
AFT-Weibull367.92133.930.47(0.47,0.47)(0.21,0.267)6.209(0.707,0.729)
AFT-log-Normal377.76157.640.47(0.47,0.47)(0.675, 0.556)6.971(0.707,0.729)
SR369.4788.550.38(0.330.65)(0.791,0.744)0(0.985,1.027)
CSA (proposed)358.720.80.45(0.39,0.65)(0.787,0.767)0.131(0.985,1.026)
CSA-INFO (proposed)344.331.190.53(0.41,0.67)(0.78,0.764)0.13(0.999,1.029)
", + "type": "table", + "image_path": "5d3d824a74b40f44971099aa53bc084369e33021afa3cae740ca983e4448a9de.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 108, + 106, + 503, + 140.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 108, + 140.66666666666666, + 503, + 175.33333333333331 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 108, + 175.33333333333331, + 503, + 209.99999999999997 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 1.75 + }, + { + "type": "text", + "bbox": [ + 106, + 219, + 506, + 327 + ], + "lines": [ + { + "bbox": [ + 105, + 218, + 507, + 234 + ], + "spans": [ + { + "bbox": [ + 105, + 218, + 148, + 234 + ], + "score": 1.0, + "content": "Bias from", + "type": "text" + }, + { + "bbox": [ + 148, + 220, + 170, + 232 + ], + "score": 0.92, + "content": "S ^ { \\prime } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 218, + 464, + 234 + ], + "score": 1.0, + "content": "can be reduced by considering more complex function approximations for", + "type": "text" + }, + { + "bbox": [ + 465, + 220, + 483, + 232 + ], + "score": 0.9, + "content": "S ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 218, + 507, + 234 + ], + "score": 1.0, + "content": ", e.g.,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 230, + 506, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 347, + 244 + ], + "score": 1.0, + "content": "polynomial or spline. For the nonparametric estimation of", + "type": "text" + }, + { + "bbox": [ + 347, + 231, + 366, + 243 + ], + "score": 0.91, + "content": "S ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 230, + 506, + 244 + ], + "score": 1.0, + "content": "we leverage the model-free popu-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 243, + 505, + 254 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 505, + 254 + ], + "score": 1.0, + "content": "lation point-estimate-based nonparametric Kaplan-Meier (Kaplan & Meier, 1958) estimator of the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 251, + 507, + 268 + ], + "spans": [ + { + "bbox": [ + 104, + 251, + 176, + 268 + ], + "score": 1.0, + "content": "survival function", + "type": "text" + }, + { + "bbox": [ + 176, + 253, + 214, + 266 + ], + "score": 0.93, + "content": "\\hat { S } ^ { \\mathrm { P K M } } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 251, + 507, + 268 + ], + "score": 1.0, + "content": "in Chapfuwa et al. (2020) to marginalize both factual and counterfactual", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 104, + 266, + 506, + 282 + ], + "spans": [ + { + "bbox": [ + 104, + 266, + 217, + 282 + ], + "score": 1.0, + "content": "predictions given covariates", + "type": "text" + }, + { + "bbox": [ + 218, + 269, + 225, + 277 + ], + "score": 0.73, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 266, + 353, + 282 + ], + "score": 1.0, + "content": ". The approximated hazard ratio,", + "type": "text" + }, + { + "bbox": [ + 354, + 266, + 381, + 279 + ], + "score": 0.9, + "content": "\\mathrm { \\hat { H R } } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 266, + 506, + 282 + ], + "score": 1.0, + "content": ", is thus obtained by combining", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 101, + 275, + 509, + 296 + ], + "spans": [ + { + "bbox": [ + 101, + 275, + 186, + 296 + ], + "score": 1.0, + "content": "the approximations", + "type": "text" + }, + { + "bbox": [ + 186, + 278, + 224, + 292 + ], + "score": 0.93, + "content": "\\hat { S } _ { a } ^ { \\mathrm { P K M } } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 275, + 242, + 296 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 243, + 281, + 257, + 291 + ], + "score": 0.87, + "content": "m _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 275, + 430, + 296 + ], + "score": 1.0, + "content": ". A similar formulation for the conditional,", + "type": "text" + }, + { + "bbox": [ + 430, + 279, + 466, + 292 + ], + "score": 0.92, + "content": "\\mathrm { H } \\mathrm { \\hat { R } } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 466, + 275, + 509, + 296 + ], + "score": 1.0, + "content": ", can also", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 291, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 342, + 306 + ], + "score": 1.0, + "content": "be derived. See the SM for full details on the evaluation or", + "type": "text" + }, + { + "bbox": [ + 342, + 291, + 370, + 305 + ], + "score": 0.9, + "content": "\\mathrm { \\hat { H R } } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 292, + 388, + 306 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 388, + 291, + 423, + 305 + ], + "score": 0.91, + "content": "\\mathrm { H } \\mathrm { \\hat { R } } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 292, + 505, + 306 + ], + "score": 1.0, + "content": ". Note that for some", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 303, + 505, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 300, + 318 + ], + "score": 1.0, + "content": "AFT- or CoxPH-based parametric formulations,", + "type": "text" + }, + { + "bbox": [ + 300, + 304, + 336, + 316 + ], + "score": 0.9, + "content": "\\mathrm { H R } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 303, + 473, + 318 + ], + "score": 1.0, + "content": ", can be readily evaluated because", + "type": "text" + }, + { + "bbox": [ + 473, + 304, + 505, + 316 + ], + "score": 0.93, + "content": "f ( t _ { a } | x )", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 315, + 272, + 327 + ], + "spans": [ + { + "bbox": [ + 106, + 316, + 123, + 327 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 315, + 156, + 327 + ], + "score": 0.95, + "content": "S ( t _ { a } | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 316, + 272, + 327 + ], + "score": 1.0, + "content": "are available in closed form.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 107, + 331, + 505, + 388 + ], + "lines": [ + { + "bbox": [ + 106, + 332, + 504, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 332, + 235, + 344 + ], + "score": 1.0, + "content": "In the experiments, we will use", + "type": "text" + }, + { + "bbox": [ + 236, + 332, + 263, + 344 + ], + "score": 0.87, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 332, + 504, + 344 + ], + "score": 1.0, + "content": "to compare different approaches against results reported in", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 343, + 505, + 356 + ], + "spans": [ + { + "bbox": [ + 105, + 343, + 297, + 356 + ], + "score": 1.0, + "content": "RCTs (see Tables 1 and 3). Further, we will use", + "type": "text" + }, + { + "bbox": [ + 297, + 343, + 333, + 355 + ], + "score": 0.92, + "content": "\\mathrm { H R } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 343, + 505, + 356 + ], + "score": 1.0, + "content": "to illustrate stratified treatment effects (see", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 353, + 506, + 367 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 506, + 367 + ], + "score": 1.0, + "content": "Figure 2). Note that though a neural-based survival recommender system (Katzman et al., 2018)", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 364, + 506, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 364, + 256, + 378 + ], + "score": 1.0, + "content": "has been previously used to estimate", + "type": "text" + }, + { + "bbox": [ + 256, + 365, + 292, + 377 + ], + "score": 0.92, + "content": "\\operatorname { H R } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 364, + 506, + 378 + ], + "score": 1.0, + "content": ", their approach does not account for confounding or", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 376, + 311, + 388 + ], + "spans": [ + { + "bbox": [ + 106, + 376, + 311, + 388 + ], + "score": 1.0, + "content": "informative censoring thus it is susceptible to bias.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 394, + 506, + 461 + ], + "lines": [ + { + "bbox": [ + 106, + 394, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 394, + 505, + 408 + ], + "score": 1.0, + "content": "Precision in Estimation of Heterogeneous Effect (PEHE) A general individualized estimation", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 406, + 506, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 198, + 425 + ], + "score": 1.0, + "content": "error is formulated as", + "type": "text" + }, + { + "bbox": [ + 199, + 406, + 358, + 426 + ], + "score": 0.92, + "content": "\\epsilon _ { \\mathrm { P E H E } } = \\sqrt { \\mathbb { E } _ { X } [ ( \\mathrm { I T E } ( x ) - \\mathrm { I } \\hat { \\mathrm { T E } } ( x ) ) ^ { 2 } ] }", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 410, + 390, + 425 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 390, + 411, + 423, + 424 + ], + "score": 0.87, + "content": "\\operatorname { I T E } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 410, + 506, + 425 + ], + "score": 1.0, + "content": "is the ground truth,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 425, + 507, + 442 + ], + "spans": [ + { + "bbox": [ + 106, + 425, + 270, + 439 + ], + "score": 0.91, + "content": "\\mathrm { I } \\hat { \\mathrm { T E } } ( x ) = \\mathbb { E } _ { T } \\left[ \\gamma \\left( T _ { 1 } \\right) - \\gamma \\left( T _ { 0 } \\right) \\vert X = x \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 425, + 288, + 442 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 288, + 427, + 306, + 439 + ], + "score": 0.9, + "content": "\\gamma ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 425, + 507, + 442 + ], + "score": 1.0, + "content": "is a deterministic transformation. In our experi-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 436, + 507, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 436, + 136, + 452 + ], + "score": 1.0, + "content": "ments,", + "type": "text" + }, + { + "bbox": [ + 136, + 439, + 154, + 450 + ], + "score": 0.89, + "content": "\\gamma ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 436, + 293, + 452 + ], + "score": 1.0, + "content": "is the average over samples from", + "type": "text" + }, + { + "bbox": [ + 293, + 438, + 387, + 450 + ], + "score": 0.92, + "content": "\\dot { t _ { a } } \\sim p _ { h , \\Phi } ( T _ { A } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 436, + 507, + 452 + ], + "score": 1.0, + "content": ". Alternative estimands, e.g.,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 448, + 480, + 462 + ], + "spans": [ + { + "bbox": [ + 106, + 448, + 217, + 462 + ], + "score": 1.0, + "content": "thresholding survival times", + "type": "text" + }, + { + "bbox": [ + 217, + 449, + 303, + 461 + ], + "score": 0.94, + "content": "\\gamma ( T _ { A } ) = I \\{ T _ { A } > \\tau \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 448, + 480, + 462 + ], + "score": 1.0, + "content": ", can also be considered as described above.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 468, + 505, + 493 + ], + "lines": [ + { + "bbox": [ + 107, + 468, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 107, + 468, + 505, + 480 + ], + "score": 1.0, + "content": "Average Treatment Effect (ATE) The population treatment effect estimation error is defined as", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 479, + 484, + 495 + ], + "spans": [ + { + "bbox": [ + 106, + 479, + 200, + 493 + ], + "score": 0.92, + "content": "\\epsilon _ { \\mathrm { A T E } } = | \\mathrm { A T E } - \\mathrm { A \\hat { T } E } |", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 479, + 231, + 495 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 232, + 480, + 317, + 493 + ], + "score": 0.92, + "content": "\\mathrm { A T E } = \\mathbb { E } _ { X } [ \\mathrm { I T E } ( x ) ]", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 479, + 394, + 495 + ], + "score": 1.0, + "content": "(ground truth) and", + "type": "text" + }, + { + "bbox": [ + 394, + 479, + 480, + 493 + ], + "score": 0.92, + "content": "\\mathrm { A } \\hat { \\mathrm { T E } } = \\mathbb { E } _ { X } [ \\hat { \\mathrm { I T E } } ( x ) ]", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 479, + 484, + 495 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5 + }, + { + "type": "title", + "bbox": [ + 108, + 504, + 201, + 516 + ], + "lines": [ + { + "bbox": [ + 104, + 502, + 202, + 518 + ], + "spans": [ + { + "bbox": [ + 104, + 502, + 202, + 518 + ], + "score": 1.0, + "content": "5 EXPERIMENTS", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 106, + 523, + 505, + 602 + ], + "lines": [ + { + "bbox": [ + 106, + 524, + 505, + 536 + ], + "spans": [ + { + "bbox": [ + 106, + 524, + 505, + 536 + ], + "score": 1.0, + "content": "We describe the baselines and datasets that will be used to evaluate the proposed counterfactual", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 534, + 504, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 504, + 547 + ], + "score": 1.0, + "content": "survival analysis methods (CSA and CSA-INFO). Pytorch code including the new semi-synthetic", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 546, + 506, + 558 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 506, + 558 + ], + "score": 1.0, + "content": "dataset (see below) will be made publicly available. Throughout the experiments, we use the standard", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 556, + 506, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 556, + 134, + 568 + ], + "score": 0.7, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 556, + 460, + 569 + ], + "score": 1.0, + "content": "for CoxPH based methods and (10) for all others. The bound in (5) is sensitive to", + "type": "text" + }, + { + "bbox": [ + 461, + 559, + 468, + 567 + ], + "score": 0.74, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 556, + 506, + 569 + ], + "score": 1.0, + "content": ", thus we", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 568, + 505, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 321, + 581 + ], + "score": 1.0, + "content": "propose approximating proxy counterfactual outcomes", + "type": "text" + }, + { + "bbox": [ + 321, + 568, + 369, + 580 + ], + "score": 0.93, + "content": "\\{ Y _ { \\mathrm { C F } } , \\delta _ { \\mathrm { C F } } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 568, + 505, + 581 + ], + "score": 1.0, + "content": "for the validation set, according to", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 579, + 506, + 591 + ], + "spans": [ + { + "bbox": [ + 106, + 579, + 435, + 591 + ], + "score": 1.0, + "content": "the covariate Euclidean nearest-neighbour (NN) from the training set. We select the", + "type": "text" + }, + { + "bbox": [ + 436, + 580, + 444, + 588 + ], + "score": 0.79, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 579, + 506, + 591 + ], + "score": 1.0, + "content": "that minimizes", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 590, + 375, + 603 + ], + "spans": [ + { + "bbox": [ + 106, + 590, + 181, + 603 + ], + "score": 1.0, + "content": "the validation loss", + "type": "text" + }, + { + "bbox": [ + 181, + 590, + 245, + 601 + ], + "score": 0.92, + "content": "\\mathcal { L } = \\mathcal { L } _ { \\mathrm { F } } + \\mathcal { L } _ { \\mathrm { C F } }", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 590, + 298, + 603 + ], + "score": 1.0, + "content": "from the set", + "type": "text" + }, + { + "bbox": [ + 298, + 590, + 371, + 602 + ], + "score": 0.81, + "content": "( 0 , 0 . 1 , 1 , 1 0 , 1 0 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 590, + 375, + 603 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 106, + 605, + 506, + 730 + ], + "lines": [ + { + "bbox": [ + 106, + 606, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 106, + 606, + 406, + 618 + ], + "score": 1.0, + "content": "Baselines We consider the following competitive baseline approaches:", + "type": "text" + }, + { + "bbox": [ + 406, + 606, + 418, + 618 + ], + "score": 0.32, + "content": "( i )", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 606, + 505, + 618 + ], + "score": 1.0, + "content": "propensity weighted", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 616, + 506, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 452, + 630 + ], + "score": 1.0, + "content": "CoxPH (Schemper et al., 2009; Buchanan et al., 2014; Rosenbaum & Rubin, 1983);", + "type": "text" + }, + { + "bbox": [ + 452, + 617, + 468, + 629 + ], + "score": 0.4, + "content": "( i i )", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 616, + 506, + 630 + ], + "score": 1.0, + "content": "IPM (5)", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 627, + 507, + 640 + ], + "spans": [ + { + "bbox": [ + 104, + 627, + 313, + 640 + ], + "score": 1.0, + "content": "regularized AFT (log-Normal and Weibull) models;", + "type": "text" + }, + { + "bbox": [ + 314, + 628, + 332, + 640 + ], + "score": 0.71, + "content": "( i i i )", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 627, + 507, + 640 + ], + "score": 1.0, + "content": "an IPM (5) regularized deterministic semi-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 639, + 506, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 506, + 651 + ], + "score": 1.0, + "content": "supervised regression (SR) model with accuracy objective from (Chapfuwa et al., 2018), as a contrast", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 649, + 506, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 377, + 663 + ], + "score": 1.0, + "content": "for the proposed stochastic predictors (CSA and CSA-INFO); and", + "type": "text" + }, + { + "bbox": [ + 377, + 650, + 394, + 662 + ], + "score": 0.83, + "content": "( i v )", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 649, + 506, + 663 + ], + "score": 1.0, + "content": "survival Bayesian additive", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 661, + 506, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 506, + 672 + ], + "score": 1.0, + "content": "regression trees (Surv-BART) (Sparapani et al., 2016). For CoxPH, we consider three normalized", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 672, + 511, + 707 + ], + "spans": [ + { + "bbox": [ + 104, + 672, + 132, + 707 + ], + "score": 1.0, + "content": "weigh2009),", + "type": "text" + }, + { + "bbox": [ + 160, + 683, + 255, + 697 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\mathrm { I P W } _ { i } = \\frac { a _ { i } } { \\hat { e } _ { i } } + \\frac { 1 - a _ { i } } { 1 - \\hat { e } _ { i } } ; i i ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 672, + 199, + 683 + ], + "score": 0.76, + "content": "( i )", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 672, + 295, + 707 + ], + "score": 1.0, + "content": "weilappi", + "type": "text" + }, + { + "bbox": [ + 307, + 672, + 511, + 707 + ], + "score": 1.0, + "content": "ng (IPW) (Horvitz & Thompson, 1952; Cao et al.,weights (OW) (Crump et al., 2006; Li et al., 2018),", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 132, + 697, + 307, + 708 + ], + "spans": [ + { + "bbox": [ + 132, + 697, + 275, + 708 + ], + "score": 0.86, + "content": "\\mathrm { O W } _ { i } = a _ { i } \\cdot \\left( 1 - \\hat { e } _ { i } \\right) + \\left( 1 - a _ { i } \\right) \\cdot \\hat { e } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 697, + 307, + 707 + ], + "score": 0.52, + "content": "i i i", + "type": "inline_equation" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 707, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 707, + 189, + 722 + ], + "score": 1.0, + "content": "linear logistic model", + "type": "text" + }, + { + "bbox": [ + 189, + 708, + 246, + 720 + ], + "score": 0.91, + "content": "\\hat { e } _ { i } = \\sigma ( x _ { i } ; w )", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 707, + 362, + 722 + ], + "score": 1.0, + "content": ", is used as an approximation,", + "type": "text" + }, + { + "bbox": [ + 363, + 708, + 372, + 719 + ], + "score": 0.87, + "content": "\\boldsymbol { \\hat { e } } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 707, + 506, + 722 + ], + "score": 1.0, + "content": ", to the unknown propensity score", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 107, + 718, + 349, + 731 + ], + "spans": [ + { + "bbox": [ + 107, + 718, + 179, + 731 + ], + "score": 0.91, + "content": "P ( A = { \\bar { 1 } } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 719, + 349, + 730 + ], + "score": 1.0, + "content": ". See the SM for a details of the baselines.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 38.5 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 307, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 108, + 106, + 503, + 210 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 80, + 504, + 104 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 80, + 505, + 93 + ], + "spans": [ + { + "bbox": [ + 106, + 80, + 378, + 93 + ], + "score": 1.0, + "content": "Table 1: Performance comparisons on ACTG-SYNTHETIC data, with", + "type": "text" + }, + { + "bbox": [ + 378, + 80, + 397, + 91 + ], + "score": 0.75, + "content": "9 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 80, + 425, + 92 + ], + "score": 0.45, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 80, + 505, + 93 + ], + "score": 1.0, + "content": "confidence interval.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 104, + 87, + 371, + 108 + ], + "spans": [ + { + "bbox": [ + 104, + 87, + 270, + 108 + ], + "score": 1.0, + "content": "The ground truth, test set, hazard ratio is", + "type": "text" + }, + { + "bbox": [ + 270, + 91, + 366, + 104 + ], + "score": 0.76, + "content": "\\mathrm { H R ( t ) } = 0 . 5 2 _ { ( 0 . 3 9 , 0 . 7 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 87, + 371, + 108 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table_body", + "bbox": [ + 108, + 106, + 503, + 210 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 106, + 503, + 210 + ], + "spans": [ + { + "bbox": [ + 108, + 106, + 503, + 210 + ], + "score": 0.982, + "html": "
MethodCausalFactualC-Slope (A=0,A=1)
EPEHEEATEHR(t)C-Index (A=0,A=1)Mean COV
CoxPH-UniformNANA0.97(0.86,1.09)NANANA
CoxPH-IPWNANA0.48(0.03,7.21)NANANA
CoxPH-OWNANA0.60(0.53,0.68)NANANA
Surv-BART352.0777.890.0(0.0,0.0)(0.706,0.686)0.001(0.398,0)
AFT-Weibull367.92133.930.47(0.47,0.47)(0.21,0.267)6.209(0.707,0.729)
AFT-log-Normal377.76157.640.47(0.47,0.47)(0.675, 0.556)6.971(0.707,0.729)
SR369.4788.550.38(0.330.65)(0.791,0.744)0(0.985,1.027)
CSA (proposed)358.720.80.45(0.39,0.65)(0.787,0.767)0.131(0.985,1.026)
CSA-INFO (proposed)344.331.190.53(0.41,0.67)(0.78,0.764)0.13(0.999,1.029)
", + "type": "table", + "image_path": "5d3d824a74b40f44971099aa53bc084369e33021afa3cae740ca983e4448a9de.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 108, + 106, + 503, + 140.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 108, + 140.66666666666666, + 503, + 175.33333333333331 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 108, + 175.33333333333331, + 503, + 209.99999999999997 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 1.75 + }, + { + "type": "text", + "bbox": [ + 106, + 219, + 506, + 327 + ], + "lines": [ + { + "bbox": [ + 105, + 218, + 507, + 234 + ], + "spans": [ + { + "bbox": [ + 105, + 218, + 148, + 234 + ], + "score": 1.0, + "content": "Bias from", + "type": "text" + }, + { + "bbox": [ + 148, + 220, + 170, + 232 + ], + "score": 0.92, + "content": "S ^ { \\prime } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 218, + 464, + 234 + ], + "score": 1.0, + "content": "can be reduced by considering more complex function approximations for", + "type": "text" + }, + { + "bbox": [ + 465, + 220, + 483, + 232 + ], + "score": 0.9, + "content": "S ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 218, + 507, + 234 + ], + "score": 1.0, + "content": ", e.g.,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 230, + 506, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 347, + 244 + ], + "score": 1.0, + "content": "polynomial or spline. For the nonparametric estimation of", + "type": "text" + }, + { + "bbox": [ + 347, + 231, + 366, + 243 + ], + "score": 0.91, + "content": "S ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 230, + 506, + 244 + ], + "score": 1.0, + "content": "we leverage the model-free popu-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 243, + 505, + 254 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 505, + 254 + ], + "score": 1.0, + "content": "lation point-estimate-based nonparametric Kaplan-Meier (Kaplan & Meier, 1958) estimator of the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 251, + 507, + 268 + ], + "spans": [ + { + "bbox": [ + 104, + 251, + 176, + 268 + ], + "score": 1.0, + "content": "survival function", + "type": "text" + }, + { + "bbox": [ + 176, + 253, + 214, + 266 + ], + "score": 0.93, + "content": "\\hat { S } ^ { \\mathrm { P K M } } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 251, + 507, + 268 + ], + "score": 1.0, + "content": "in Chapfuwa et al. (2020) to marginalize both factual and counterfactual", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 104, + 266, + 506, + 282 + ], + "spans": [ + { + "bbox": [ + 104, + 266, + 217, + 282 + ], + "score": 1.0, + "content": "predictions given covariates", + "type": "text" + }, + { + "bbox": [ + 218, + 269, + 225, + 277 + ], + "score": 0.73, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 266, + 353, + 282 + ], + "score": 1.0, + "content": ". The approximated hazard ratio,", + "type": "text" + }, + { + "bbox": [ + 354, + 266, + 381, + 279 + ], + "score": 0.9, + "content": "\\mathrm { \\hat { H R } } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 266, + 506, + 282 + ], + "score": 1.0, + "content": ", is thus obtained by combining", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 101, + 275, + 509, + 296 + ], + "spans": [ + { + "bbox": [ + 101, + 275, + 186, + 296 + ], + "score": 1.0, + "content": "the approximations", + "type": "text" + }, + { + "bbox": [ + 186, + 278, + 224, + 292 + ], + "score": 0.93, + "content": "\\hat { S } _ { a } ^ { \\mathrm { P K M } } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 275, + 242, + 296 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 243, + 281, + 257, + 291 + ], + "score": 0.87, + "content": "m _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 275, + 430, + 296 + ], + "score": 1.0, + "content": ". A similar formulation for the conditional,", + "type": "text" + }, + { + "bbox": [ + 430, + 279, + 466, + 292 + ], + "score": 0.92, + "content": "\\mathrm { H } \\mathrm { \\hat { R } } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 466, + 275, + 509, + 296 + ], + "score": 1.0, + "content": ", can also", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 291, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 342, + 306 + ], + "score": 1.0, + "content": "be derived. See the SM for full details on the evaluation or", + "type": "text" + }, + { + "bbox": [ + 342, + 291, + 370, + 305 + ], + "score": 0.9, + "content": "\\mathrm { \\hat { H R } } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 292, + 388, + 306 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 388, + 291, + 423, + 305 + ], + "score": 0.91, + "content": "\\mathrm { H } \\mathrm { \\hat { R } } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 292, + 505, + 306 + ], + "score": 1.0, + "content": ". Note that for some", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 303, + 505, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 300, + 318 + ], + "score": 1.0, + "content": "AFT- or CoxPH-based parametric formulations,", + "type": "text" + }, + { + "bbox": [ + 300, + 304, + 336, + 316 + ], + "score": 0.9, + "content": "\\mathrm { H R } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 303, + 473, + 318 + ], + "score": 1.0, + "content": ", can be readily evaluated because", + "type": "text" + }, + { + "bbox": [ + 473, + 304, + 505, + 316 + ], + "score": 0.93, + "content": "f ( t _ { a } | x )", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 315, + 272, + 327 + ], + "spans": [ + { + "bbox": [ + 106, + 316, + 123, + 327 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 315, + 156, + 327 + ], + "score": 0.95, + "content": "S ( t _ { a } | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 316, + 272, + 327 + ], + "score": 1.0, + "content": "are available in closed form.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 9, + "bbox_fs": [ + 101, + 218, + 509, + 327 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 331, + 505, + 388 + ], + "lines": [ + { + "bbox": [ + 106, + 332, + 504, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 332, + 235, + 344 + ], + "score": 1.0, + "content": "In the experiments, we will use", + "type": "text" + }, + { + "bbox": [ + 236, + 332, + 263, + 344 + ], + "score": 0.87, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 332, + 504, + 344 + ], + "score": 1.0, + "content": "to compare different approaches against results reported in", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 343, + 505, + 356 + ], + "spans": [ + { + "bbox": [ + 105, + 343, + 297, + 356 + ], + "score": 1.0, + "content": "RCTs (see Tables 1 and 3). Further, we will use", + "type": "text" + }, + { + "bbox": [ + 297, + 343, + 333, + 355 + ], + "score": 0.92, + "content": "\\mathrm { H R } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 343, + 505, + 356 + ], + "score": 1.0, + "content": "to illustrate stratified treatment effects (see", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 353, + 506, + 367 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 506, + 367 + ], + "score": 1.0, + "content": "Figure 2). Note that though a neural-based survival recommender system (Katzman et al., 2018)", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 364, + 506, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 364, + 256, + 378 + ], + "score": 1.0, + "content": "has been previously used to estimate", + "type": "text" + }, + { + "bbox": [ + 256, + 365, + 292, + 377 + ], + "score": 0.92, + "content": "\\operatorname { H R } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 364, + 506, + 378 + ], + "score": 1.0, + "content": ", their approach does not account for confounding or", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 376, + 311, + 388 + ], + "spans": [ + { + "bbox": [ + 106, + 376, + 311, + 388 + ], + "score": 1.0, + "content": "informative censoring thus it is susceptible to bias.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 332, + 506, + 388 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 394, + 506, + 461 + ], + "lines": [ + { + "bbox": [ + 106, + 394, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 394, + 505, + 408 + ], + "score": 1.0, + "content": "Precision in Estimation of Heterogeneous Effect (PEHE) A general individualized estimation", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 406, + 506, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 198, + 425 + ], + "score": 1.0, + "content": "error is formulated as", + "type": "text" + }, + { + "bbox": [ + 199, + 406, + 358, + 426 + ], + "score": 0.92, + "content": "\\epsilon _ { \\mathrm { P E H E } } = \\sqrt { \\mathbb { E } _ { X } [ ( \\mathrm { I T E } ( x ) - \\mathrm { I } \\hat { \\mathrm { T E } } ( x ) ) ^ { 2 } ] }", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 410, + 390, + 425 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 390, + 411, + 423, + 424 + ], + "score": 0.87, + "content": "\\operatorname { I T E } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 410, + 506, + 425 + ], + "score": 1.0, + "content": "is the ground truth,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 425, + 507, + 442 + ], + "spans": [ + { + "bbox": [ + 106, + 425, + 270, + 439 + ], + "score": 0.91, + "content": "\\mathrm { I } \\hat { \\mathrm { T E } } ( x ) = \\mathbb { E } _ { T } \\left[ \\gamma \\left( T _ { 1 } \\right) - \\gamma \\left( T _ { 0 } \\right) \\vert X = x \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 425, + 288, + 442 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 288, + 427, + 306, + 439 + ], + "score": 0.9, + "content": "\\gamma ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 425, + 507, + 442 + ], + "score": 1.0, + "content": "is a deterministic transformation. In our experi-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 436, + 507, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 436, + 136, + 452 + ], + "score": 1.0, + "content": "ments,", + "type": "text" + }, + { + "bbox": [ + 136, + 439, + 154, + 450 + ], + "score": 0.89, + "content": "\\gamma ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 436, + 293, + 452 + ], + "score": 1.0, + "content": "is the average over samples from", + "type": "text" + }, + { + "bbox": [ + 293, + 438, + 387, + 450 + ], + "score": 0.92, + "content": "\\dot { t _ { a } } \\sim p _ { h , \\Phi } ( T _ { A } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 436, + 507, + 452 + ], + "score": 1.0, + "content": ". Alternative estimands, e.g.,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 448, + 480, + 462 + ], + "spans": [ + { + "bbox": [ + 106, + 448, + 217, + 462 + ], + "score": 1.0, + "content": "thresholding survival times", + "type": "text" + }, + { + "bbox": [ + 217, + 449, + 303, + 461 + ], + "score": 0.94, + "content": "\\gamma ( T _ { A } ) = I \\{ T _ { A } > \\tau \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 448, + 480, + 462 + ], + "score": 1.0, + "content": ", can also be considered as described above.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21, + "bbox_fs": [ + 105, + 394, + 507, + 462 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 468, + 505, + 493 + ], + "lines": [ + { + "bbox": [ + 107, + 468, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 107, + 468, + 505, + 480 + ], + "score": 1.0, + "content": "Average Treatment Effect (ATE) The population treatment effect estimation error is defined as", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 479, + 484, + 495 + ], + "spans": [ + { + "bbox": [ + 106, + 479, + 200, + 493 + ], + "score": 0.92, + "content": "\\epsilon _ { \\mathrm { A T E } } = | \\mathrm { A T E } - \\mathrm { A \\hat { T } E } |", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 479, + 231, + 495 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 232, + 480, + 317, + 493 + ], + "score": 0.92, + "content": "\\mathrm { A T E } = \\mathbb { E } _ { X } [ \\mathrm { I T E } ( x ) ]", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 479, + 394, + 495 + ], + "score": 1.0, + "content": "(ground truth) and", + "type": "text" + }, + { + "bbox": [ + 394, + 479, + 480, + 493 + ], + "score": 0.92, + "content": "\\mathrm { A } \\hat { \\mathrm { T E } } = \\mathbb { E } _ { X } [ \\hat { \\mathrm { I T E } } ( x ) ]", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 479, + 484, + 495 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5, + "bbox_fs": [ + 106, + 468, + 505, + 495 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 504, + 201, + 516 + ], + "lines": [ + { + "bbox": [ + 104, + 502, + 202, + 518 + ], + "spans": [ + { + "bbox": [ + 104, + 502, + 202, + 518 + ], + "score": 1.0, + "content": "5 EXPERIMENTS", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 106, + 523, + 505, + 602 + ], + "lines": [ + { + "bbox": [ + 106, + 524, + 505, + 536 + ], + "spans": [ + { + "bbox": [ + 106, + 524, + 505, + 536 + ], + "score": 1.0, + "content": "We describe the baselines and datasets that will be used to evaluate the proposed counterfactual", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 534, + 504, + 547 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 504, + 547 + ], + "score": 1.0, + "content": "survival analysis methods (CSA and CSA-INFO). Pytorch code including the new semi-synthetic", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 546, + 506, + 558 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 506, + 558 + ], + "score": 1.0, + "content": "dataset (see below) will be made publicly available. Throughout the experiments, we use the standard", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 556, + 506, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 556, + 134, + 568 + ], + "score": 0.7, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 556, + 460, + 569 + ], + "score": 1.0, + "content": "for CoxPH based methods and (10) for all others. The bound in (5) is sensitive to", + "type": "text" + }, + { + "bbox": [ + 461, + 559, + 468, + 567 + ], + "score": 0.74, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 469, + 556, + 506, + 569 + ], + "score": 1.0, + "content": ", thus we", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 568, + 505, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 321, + 581 + ], + "score": 1.0, + "content": "propose approximating proxy counterfactual outcomes", + "type": "text" + }, + { + "bbox": [ + 321, + 568, + 369, + 580 + ], + "score": 0.93, + "content": "\\{ Y _ { \\mathrm { C F } } , \\delta _ { \\mathrm { C F } } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 568, + 505, + 581 + ], + "score": 1.0, + "content": "for the validation set, according to", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 579, + 506, + 591 + ], + "spans": [ + { + "bbox": [ + 106, + 579, + 435, + 591 + ], + "score": 1.0, + "content": "the covariate Euclidean nearest-neighbour (NN) from the training set. We select the", + "type": "text" + }, + { + "bbox": [ + 436, + 580, + 444, + 588 + ], + "score": 0.79, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 579, + 506, + 591 + ], + "score": 1.0, + "content": "that minimizes", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 590, + 375, + 603 + ], + "spans": [ + { + "bbox": [ + 106, + 590, + 181, + 603 + ], + "score": 1.0, + "content": "the validation loss", + "type": "text" + }, + { + "bbox": [ + 181, + 590, + 245, + 601 + ], + "score": 0.92, + "content": "\\mathcal { L } = \\mathcal { L } _ { \\mathrm { F } } + \\mathcal { L } _ { \\mathrm { C F } }", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 590, + 298, + 603 + ], + "score": 1.0, + "content": "from the set", + "type": "text" + }, + { + "bbox": [ + 298, + 590, + 371, + 602 + ], + "score": 0.81, + "content": "( 0 , 0 . 1 , 1 , 1 0 , 1 0 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 590, + 375, + 603 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 524, + 506, + 603 + ] + }, + { + "type": "list", + "bbox": [ + 106, + 605, + 506, + 730 + ], + "lines": [ + { + "bbox": [ + 106, + 606, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 106, + 606, + 406, + 618 + ], + "score": 1.0, + "content": "Baselines We consider the following competitive baseline approaches:", + "type": "text" + }, + { + "bbox": [ + 406, + 606, + 418, + 618 + ], + "score": 0.32, + "content": "( i )", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 606, + 505, + 618 + ], + "score": 1.0, + "content": "propensity weighted", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 616, + 506, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 452, + 630 + ], + "score": 1.0, + "content": "CoxPH (Schemper et al., 2009; Buchanan et al., 2014; Rosenbaum & Rubin, 1983);", + "type": "text" + }, + { + "bbox": [ + 452, + 617, + 468, + 629 + ], + "score": 0.4, + "content": "( i i )", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 616, + 506, + 630 + ], + "score": 1.0, + "content": "IPM (5)", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 627, + 507, + 640 + ], + "spans": [ + { + "bbox": [ + 104, + 627, + 313, + 640 + ], + "score": 1.0, + "content": "regularized AFT (log-Normal and Weibull) models;", + "type": "text" + }, + { + "bbox": [ + 314, + 628, + 332, + 640 + ], + "score": 0.71, + "content": "( i i i )", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 627, + 507, + 640 + ], + "score": 1.0, + "content": "an IPM (5) regularized deterministic semi-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 639, + 506, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 506, + 651 + ], + "score": 1.0, + "content": "supervised regression (SR) model with accuracy objective from (Chapfuwa et al., 2018), as a contrast", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 649, + 506, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 377, + 663 + ], + "score": 1.0, + "content": "for the proposed stochastic predictors (CSA and CSA-INFO); and", + "type": "text" + }, + { + "bbox": [ + 377, + 650, + 394, + 662 + ], + "score": 0.83, + "content": "( i v )", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 649, + 506, + 663 + ], + "score": 1.0, + "content": "survival Bayesian additive", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 661, + 506, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 506, + 672 + ], + "score": 1.0, + "content": "regression trees (Surv-BART) (Sparapani et al., 2016). For CoxPH, we consider three normalized", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 672, + 511, + 707 + ], + "spans": [ + { + "bbox": [ + 104, + 672, + 132, + 707 + ], + "score": 1.0, + "content": "weigh2009),", + "type": "text" + }, + { + "bbox": [ + 160, + 683, + 255, + 697 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\mathrm { I P W } _ { i } = \\frac { a _ { i } } { \\hat { e } _ { i } } + \\frac { 1 - a _ { i } } { 1 - \\hat { e } _ { i } } ; i i ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 672, + 199, + 683 + ], + "score": 0.76, + "content": "( i )", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 672, + 295, + 707 + ], + "score": 1.0, + "content": "weilappi", + "type": "text" + }, + { + "bbox": [ + 307, + 672, + 511, + 707 + ], + "score": 1.0, + "content": "ng (IPW) (Horvitz & Thompson, 1952; Cao et al.,weights (OW) (Crump et al., 2006; Li et al., 2018),", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 132, + 697, + 307, + 708 + ], + "spans": [ + { + "bbox": [ + 132, + 697, + 275, + 708 + ], + "score": 0.86, + "content": "\\mathrm { O W } _ { i } = a _ { i } \\cdot \\left( 1 - \\hat { e } _ { i } \\right) + \\left( 1 - a _ { i } \\right) \\cdot \\hat { e } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 697, + 307, + 707 + ], + "score": 0.52, + "content": "i i i", + "type": "inline_equation" + } + ], + "index": 41, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 707, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 707, + 189, + 722 + ], + "score": 1.0, + "content": "linear logistic model", + "type": "text" + }, + { + "bbox": [ + 189, + 708, + 246, + 720 + ], + "score": 0.91, + "content": "\\hat { e } _ { i } = \\sigma ( x _ { i } ; w )", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 707, + 362, + 722 + ], + "score": 1.0, + "content": ", is used as an approximation,", + "type": "text" + }, + { + "bbox": [ + 363, + 708, + 372, + 719 + ], + "score": 0.87, + "content": "\\boldsymbol { \\hat { e } } _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 707, + 506, + 722 + ], + "score": 1.0, + "content": ", to the unknown propensity score", + "type": "text" + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 107, + 718, + 349, + 731 + ], + "spans": [ + { + "bbox": [ + 107, + 718, + 179, + 731 + ], + "score": 0.91, + "content": "P ( A = { \\bar { 1 } } | X = x )", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 719, + 349, + 730 + ], + "score": 1.0, + "content": ". See the SM for a details of the baselines.", + "type": "text" + } + ], + "index": 43, + "is_list_end_line": true + } + ], + "index": 38.5, + "bbox_fs": [ + 104, + 606, + 511, + 731 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 136, + 118, + 473, + 219 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 80, + 505, + 115 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 79, + 506, + 93 + ], + "spans": [ + { + "bbox": [ + 105, + 79, + 373, + 93 + ], + "score": 1.0, + "content": "Table 3: Performance comparisons on FRAMINGHAM data, with", + "type": "text" + }, + { + "bbox": [ + 374, + 80, + 394, + 91 + ], + "score": 0.78, + "content": "9 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 80, + 423, + 92 + ], + "score": 0.56, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 79, + 506, + 93 + ], + "score": 1.0, + "content": "confidence interval.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 104, + 89, + 506, + 105 + ], + "spans": [ + { + "bbox": [ + 104, + 89, + 213, + 105 + ], + "score": 1.0, + "content": "Test set NN assignment of", + "type": "text" + }, + { + "bbox": [ + 214, + 93, + 231, + 103 + ], + "score": 0.86, + "content": "y _ { \\mathrm { C F } }", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 89, + 249, + 105 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 249, + 91, + 266, + 102 + ], + "score": 0.89, + "content": "\\delta _ { \\mathrm { C F } }", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 89, + 321, + 105 + ], + "score": 1.0, + "content": "yields biased", + "type": "text" + }, + { + "bbox": [ + 322, + 91, + 417, + 103 + ], + "score": 0.74, + "content": "\\mathrm { H R ( t ) } = 1 . 2 3 _ { ( 1 . 1 7 , 1 . 2 5 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 89, + 506, + 105 + ], + "score": 1.0, + "content": ", while previous large", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 103, + 452, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 272, + 118 + ], + "score": 1.0, + "content": "scale longitudinal RCT studies estimated", + "type": "text" + }, + { + "bbox": [ + 272, + 103, + 331, + 115 + ], + "score": 0.86, + "content": "\\mathrm { H R } ( \\mathrm { t } ) = 0 . 7 5", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 103, + 452, + 118 + ], + "score": 1.0, + "content": "(0.64,0.88) (Yusuf et al., 2016).", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "table_body", + "bbox": [ + 136, + 118, + 473, + 219 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 136, + 118, + 473, + 219 + ], + "spans": [ + { + "bbox": [ + 136, + 118, + 473, + 219 + ], + "score": 0.978, + "html": "
MethodCausal HR(t)C-Index (A=0, A=1)Factual Mean COVC-Slope (A=0,A=1)
CoxPH-Uniform1.69(1.38,2.07)NANANA
CoxPH-IPW1.09(0.76,1.57)NANANA
CoxPH-OW0.88(0.73,1.08)NANANA
Surv-BART14.99(14.9,14.9e8)(0.629,0.630)0.003(0.232,0.084)
AFT-Weibull1.09(1.09,1.09)(0.734,0.395)8.609(0.857,0.89)
AFT-log-Normal1.55(1.46,1.55)(0.68,0.56)10.415(0.979,0.732)
SR0.58(0.53,0.71)(0.601, 0.57)0(0.491,0.63)
CSA(proposed)1.04(1.00,1.09)(0.763,0.728)0.161(0.891,0.81)
CSA-INFO (proposed)0.81(0.77,0.83)(0.752,0.651)0.156(0.907, 0.881)
", + "type": "table", + "image_path": "02f7c367f232f20c3ad47f3b482bcff2a58309728cb6b371a98f07c24827d815.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 136, + 118, + 473, + 151.66666666666666 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 136, + 151.66666666666666, + 473, + 185.33333333333331 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 136, + 185.33333333333331, + 473, + 218.99999999999997 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 107, + 229, + 317, + 317 + ], + "lines": [ + { + "bbox": [ + 105, + 227, + 317, + 242 + ], + "spans": [ + { + "bbox": [ + 105, + 227, + 317, + 242 + ], + "score": 1.0, + "content": "Datasets We consider the following datasets: (i)", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 240, + 318, + 252 + ], + "spans": [ + { + "bbox": [ + 106, + 240, + 318, + 252 + ], + "score": 1.0, + "content": "FRAMINGHAM, is an EHR-based longitudinal car-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 251, + 317, + 263 + ], + "spans": [ + { + "bbox": [ + 106, + 251, + 317, + 263 + ], + "score": 1.0, + "content": "diovascular cohort study that we use to evaluate the", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 262, + 318, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 262, + 318, + 273 + ], + "score": 1.0, + "content": "effect of statins on future coronary heart disease out-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 273, + 318, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 231, + 285 + ], + "score": 1.0, + "content": "comes (Benjamin et al., 1994);", + "type": "text" + }, + { + "bbox": [ + 232, + 273, + 247, + 285 + ], + "score": 0.4, + "content": "( i i )", + "type": "inline_equation" + }, + { + "bbox": [ + 247, + 273, + 318, + 285 + ], + "score": 1.0, + "content": "ACTG, is a longi-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 284, + 318, + 296 + ], + "spans": [ + { + "bbox": [ + 106, + 284, + 318, + 296 + ], + "score": 1.0, + "content": "tudinal RCT study comparing monotherapy with Zi-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 294, + 317, + 307 + ], + "spans": [ + { + "bbox": [ + 106, + 294, + 317, + 307 + ], + "score": 1.0, + "content": "dovudine or Didanosine with combination therapy in", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 306, + 318, + 318 + ], + "spans": [ + { + "bbox": [ + 106, + 306, + 269, + 318 + ], + "score": 1.0, + "content": "HIV patients (Hammer et al., 1996); and", + "type": "text" + }, + { + "bbox": [ + 270, + 306, + 288, + 318 + ], + "score": 0.65, + "content": "( i i i )", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 306, + 318, + 318 + ], + "score": 1.0, + "content": "ACTG-", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 9.5 + }, + { + "type": "table", + "bbox": [ + 326, + 240, + 502, + 297 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 328, + 225, + 500, + 236 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 326, + 225, + 502, + 237 + ], + "spans": [ + { + "bbox": [ + 326, + 225, + 502, + 237 + ], + "score": 1.0, + "content": "Table 2: Summary statistics of the datasets.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "table_body", + "bbox": [ + 326, + 240, + 502, + 297 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 326, + 240, + 502, + 297 + ], + "spans": [ + { + "bbox": [ + 326, + 240, + 502, + 297 + ], + "score": 0.968, + "html": "
FRAMINGHAMACTGACTG-SYNTHETIC
Events (%)26.026.948.9
Treatment (%)10.449.555.9
N3,4351,0542,139
p322323
Missing (%)0.231.411.38
tmax (days)7,2791,2311,313
", + "type": "table", + "image_path": "662567b84ec6b8412fd1a2eb1f6e306f029c8bde26af3d28de506fa5a1d065e3.jpg" + } + ] + } + ], + "index": 16.5, + "virtual_lines": [ + { + "bbox": [ + 326, + 240, + 502, + 254.25 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 326, + 254.25, + 502, + 268.5 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 326, + 268.5, + 502, + 282.75 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 326, + 282.75, + 502, + 297.0 + ], + "spans": [], + "index": 18 + } + ] + } + ], + "index": 15.25 + }, + { + "type": "text", + "bbox": [ + 107, + 317, + 505, + 394 + ], + "lines": [ + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "score": 1.0, + "content": "SYNTHETIC, is a semi-synthetic dataset based on ACTG covariates. We simulate potential outcomes", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 327, + 504, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 504, + 340 + ], + "score": 1.0, + "content": "according to a Gompertz-Cox distribution (Bender et al., 2005) with selection bias from a simple", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 338, + 506, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 180, + 352 + ], + "score": 1.0, + "content": "logistic model for", + "type": "text" + }, + { + "bbox": [ + 181, + 339, + 253, + 351 + ], + "score": 0.91, + "content": "\\bar { P ( A = 1 | X = x ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 338, + 506, + 352 + ], + "score": 1.0, + "content": "and AFT-based censoring mechanism. The generative process", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 349, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 349, + 389, + 363 + ], + "score": 1.0, + "content": "is detailed in the SM. Table 2 summarizes the datasets according to", + "type": "text" + }, + { + "bbox": [ + 389, + 350, + 401, + 362 + ], + "score": 0.59, + "content": "( i )", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 349, + 477, + 363 + ], + "score": 1.0, + "content": "covariates of size", + "type": "text" + }, + { + "bbox": [ + 478, + 351, + 484, + 362 + ], + "score": 0.62, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 349, + 489, + 363 + ], + "score": 1.0, + "content": ";", + "type": "text" + }, + { + "bbox": [ + 490, + 350, + 505, + 362 + ], + "score": 0.34, + "content": "( i i )", + "type": "inline_equation" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 361, + 506, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 407, + 374 + ], + "score": 1.0, + "content": "proportion of non-censored events, treated units, and missing entries in the", + "type": "text" + }, + { + "bbox": [ + 407, + 361, + 435, + 372 + ], + "score": 0.91, + "content": "N \\times p", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 361, + 506, + 374 + ], + "score": 1.0, + "content": "covariate matrix;", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 371, + 506, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 371, + 124, + 385 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 372, + 142, + 383 + ], + "score": 0.84, + "content": "( i i i )", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 371, + 189, + 385 + ], + "score": 1.0, + "content": "time range", + "type": "text" + }, + { + "bbox": [ + 190, + 372, + 210, + 383 + ], + "score": 0.9, + "content": "t _ { \\mathrm { m a x } }", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 371, + 506, + 385 + ], + "score": 1.0, + "content": "for both censored and non-censored events. Missing entries are imputed", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 382, + 359, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 382, + 359, + 396 + ], + "score": 1.0, + "content": "with median or mode if continuous or categorical, respectively.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 106, + 399, + 505, + 619 + ], + "lines": [ + { + "bbox": [ + 106, + 399, + 506, + 411 + ], + "spans": [ + { + "bbox": [ + 106, + 399, + 506, + 411 + ], + "score": 1.0, + "content": "Quantitative Results Experimental results for two data-sets in Tables 1 and 3, illustrate that AFT-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 411, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 106, + 411, + 505, + 422 + ], + "score": 1.0, + "content": "based methods are high variance, inferior in calibration and C-Index than accuracy-based methods", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 421, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 506, + 433 + ], + "score": 1.0, + "content": "(SR, CSA, CSA-INFO). Surv-BART is the least calibrated but low variance method. CSA-INFO and", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 432, + 505, + 444 + ], + "spans": [ + { + "bbox": [ + 106, + 432, + 505, + 444 + ], + "score": 1.0, + "content": "CSA outperform all methods across all factual metrics, whereas CSA-INFO is better calibrated, low", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 442, + 506, + 457 + ], + "spans": [ + { + "bbox": [ + 104, + 442, + 506, + 457 + ], + "score": 1.0, + "content": "variance but slightly lower C-Index than CSA. Note that we fit CoxPH using the entire dataset; since", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 453, + 506, + 468 + ], + "spans": [ + { + "bbox": [ + 104, + 453, + 506, + 468 + ], + "score": 1.0, + "content": "it does not support counterfactual inference, we do not present factual metrics. By properly adjusting", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 466, + 505, + 477 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 505, + 477 + ], + "score": 1.0, + "content": "for both informative censoring and selection bias, CSA-INFO significantly outperforms all methods", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 476, + 506, + 489 + ], + "spans": [ + { + "bbox": [ + 104, + 476, + 288, + 489 + ], + "score": 1.0, + "content": "in treatment effect estimation according to", + "type": "text" + }, + { + "bbox": [ + 288, + 476, + 315, + 488 + ], + "score": 0.74, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 476, + 336, + 489 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 336, + 477, + 365, + 487 + ], + "score": 0.81, + "content": "\\epsilon _ { \\mathrm { P E H E } }", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 476, + 506, + 489 + ], + "score": 1.0, + "content": ", across non-RCT datasets, while", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 487, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 505, + 500 + ], + "score": 1.0, + "content": "remaining comparable to AFT-Weibull on the RCT dataset (see the SM). Further, RCT-based results", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 498, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 498, + 292, + 511 + ], + "score": 1.0, + "content": "on ACTG data in the SM illustrate comparable", + "type": "text" + }, + { + "bbox": [ + 292, + 498, + 319, + 510 + ], + "score": 0.82, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 498, + 505, + 511 + ], + "score": 1.0, + "content": "across all models except for AFT-log-Normal", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 510, + 505, + 521 + ], + "spans": [ + { + "bbox": [ + 106, + 510, + 505, + 521 + ], + "score": 1.0, + "content": "and Surv-BART, which overestimate, and SR, which underestimates risk. For non-RCT datasets", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 520, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 505, + 532 + ], + "score": 1.0, + "content": "(ACTG-SYNTHENTIC and FRAMINGHAM), CoxPH-OW has a clear advantage over all CoxPH based", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 531, + 506, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 401, + 545 + ], + "score": 1.0, + "content": "methods, mostly credited to the well-behaved bounded propensity weights", + "type": "text" + }, + { + "bbox": [ + 402, + 531, + 432, + 543 + ], + "score": 0.89, + "content": "\\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 531, + 506, + 545 + ], + "score": 1.0, + "content": ". Interestingly, the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 541, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 506, + 555 + ], + "score": 1.0, + "content": "FRAMINGHAM observational data exhibits a common paradox, where without proper adjustment of", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 553, + 506, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 506, + 566 + ], + "score": 1.0, + "content": "selection and censoring bias, naive approaches would result in a counter-intuitive treatment effect", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 562, + 507, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 562, + 507, + 577 + ], + "score": 1.0, + "content": "from statins. However, there is severe confounding from covariates such as age, BMI, diabetes,", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 575, + 505, + 586 + ], + "spans": [ + { + "bbox": [ + 106, + 575, + 505, + 586 + ], + "score": 1.0, + "content": "CAD, PAD, MI, stroke, etc., that influence both treatment likelihood and survival time. Table", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 585, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 475, + 599 + ], + "score": 1.0, + "content": "3, demonstrates that CSA-INFO is clearly the best performing approach. Specifically, its", + "type": "text" + }, + { + "bbox": [ + 475, + 586, + 502, + 598 + ], + "score": 0.78, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 585, + 506, + 599 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 596, + 507, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 507, + 610 + ], + "score": 1.0, + "content": "reverses the biased observational treatment effect, to demonstrate positive treatment from statins,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 608, + 437, + 620 + ], + "spans": [ + { + "bbox": [ + 106, + 608, + 437, + 620 + ], + "score": 1.0, + "content": "which is consistent with prior large RCT longitudinal findings (Yusuf et al., 2016).", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 106, + 632, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 632, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 106, + 632, + 505, + 645 + ], + "score": 1.0, + "content": "Qualitative Results Figure 2a demonstrates that CSA-INFO matches the ground truth population", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 643, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 138, + 657 + ], + "score": 1.0, + "content": "hazard,", + "type": "text" + }, + { + "bbox": [ + 138, + 644, + 166, + 656 + ], + "score": 0.73, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 643, + 505, + 657 + ], + "score": 1.0, + "content": ", better than alternative methods on ACTG-SYNTHETIC data. See the SM for ACTG", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 654, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 506, + 668 + ], + "score": 1.0, + "content": "and FRAMINGHAM. Figure 2b shows sub-population log hazard ratios for four patient clusters", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 665, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 385, + 678 + ], + "score": 1.0, + "content": "obtained via hierarchical clustering on the individual log hazard ratios,", + "type": "text" + }, + { + "bbox": [ + 386, + 666, + 436, + 678 + ], + "score": 0.89, + "content": "\\log \\mathrm { H R } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 436, + 665, + 506, + 678 + ], + "score": 1.0, + "content": ", of the test set of", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 676, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 690 + ], + "score": 1.0, + "content": "FRAMINGHAM data. Interestingly, these clusters stratify treatment effects into: positive (2), negative", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 687, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 455, + 701 + ], + "score": 1.0, + "content": "(1 and 3), and neutral (4) sub-populations. Moreover, the estimated density of median", + "type": "text" + }, + { + "bbox": [ + 455, + 687, + 505, + 699 + ], + "score": 0.88, + "content": "\\log \\mathrm { H } \\bar { \\mathrm { R } } ( t | x )", + "type": "inline_equation" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 173, + 712 + ], + "score": 1.0, + "content": "values in Figure", + "type": "text" + }, + { + "bbox": [ + 174, + 699, + 185, + 709 + ], + "score": 0.25, + "content": "_ { 2 \\mathrm { c } }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 699, + 274, + 712 + ], + "score": 1.0, + "content": "illustrates that nearly", + "type": "text" + }, + { + "bbox": [ + 274, + 699, + 294, + 709 + ], + "score": 0.88, + "content": "7 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 699, + 433, + 712 + ], + "score": 1.0, + "content": "of the testing set individuals have", + "type": "text" + }, + { + "bbox": [ + 434, + 699, + 502, + 711 + ], + "score": 0.9, + "content": "\\log \\mathrm { H R } ( t | x ) < 0", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 699, + 506, + 712 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "thus may benefit from taking statins. Further, we isolated the extreme top and bottom quantiles,", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 107, + 719, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 107, + 720, + 182, + 732 + ], + "score": 0.91, + "content": "\\mathrm { H R } ( t | x ) < 0 . 0 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 719, + 200, + 734 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 201, + 720, + 275, + 732 + ], + "score": 0.92, + "content": "\\mathrm { H R } ( \\bar { t } | x ) > 1 . 9 1 6", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 719, + 394, + 734 + ], + "score": 1.0, + "content": ", respectively, of the median", + "type": "text" + }, + { + "bbox": [ + 394, + 720, + 444, + 732 + ], + "score": 0.91, + "content": "\\log \\mathrm { \\bar { H R } } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 719, + 506, + 734 + ], + "score": 1.0, + "content": "values for the", + "type": "text" + } + ], + "index": 54 + } + ], + "index": 50 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 307, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "spans": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "score": 1.0, + "content": "8", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 136, + 118, + 473, + 219 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 80, + 505, + 115 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 79, + 506, + 93 + ], + "spans": [ + { + "bbox": [ + 105, + 79, + 373, + 93 + ], + "score": 1.0, + "content": "Table 3: Performance comparisons on FRAMINGHAM data, with", + "type": "text" + }, + { + "bbox": [ + 374, + 80, + 394, + 91 + ], + "score": 0.78, + "content": "9 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 80, + 423, + 92 + ], + "score": 0.56, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 79, + 506, + 93 + ], + "score": 1.0, + "content": "confidence interval.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 104, + 89, + 506, + 105 + ], + "spans": [ + { + "bbox": [ + 104, + 89, + 213, + 105 + ], + "score": 1.0, + "content": "Test set NN assignment of", + "type": "text" + }, + { + "bbox": [ + 214, + 93, + 231, + 103 + ], + "score": 0.86, + "content": "y _ { \\mathrm { C F } }", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 89, + 249, + 105 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 249, + 91, + 266, + 102 + ], + "score": 0.89, + "content": "\\delta _ { \\mathrm { C F } }", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 89, + 321, + 105 + ], + "score": 1.0, + "content": "yields biased", + "type": "text" + }, + { + "bbox": [ + 322, + 91, + 417, + 103 + ], + "score": 0.74, + "content": "\\mathrm { H R ( t ) } = 1 . 2 3 _ { ( 1 . 1 7 , 1 . 2 5 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 89, + 506, + 105 + ], + "score": 1.0, + "content": ", while previous large", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 103, + 452, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 272, + 118 + ], + "score": 1.0, + "content": "scale longitudinal RCT studies estimated", + "type": "text" + }, + { + "bbox": [ + 272, + 103, + 331, + 115 + ], + "score": 0.86, + "content": "\\mathrm { H R } ( \\mathrm { t } ) = 0 . 7 5", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 103, + 452, + 118 + ], + "score": 1.0, + "content": "(0.64,0.88) (Yusuf et al., 2016).", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "table_body", + "bbox": [ + 136, + 118, + 473, + 219 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 136, + 118, + 473, + 219 + ], + "spans": [ + { + "bbox": [ + 136, + 118, + 473, + 219 + ], + "score": 0.978, + "html": "
MethodCausal HR(t)C-Index (A=0, A=1)Factual Mean COVC-Slope (A=0,A=1)
CoxPH-Uniform1.69(1.38,2.07)NANANA
CoxPH-IPW1.09(0.76,1.57)NANANA
CoxPH-OW0.88(0.73,1.08)NANANA
Surv-BART14.99(14.9,14.9e8)(0.629,0.630)0.003(0.232,0.084)
AFT-Weibull1.09(1.09,1.09)(0.734,0.395)8.609(0.857,0.89)
AFT-log-Normal1.55(1.46,1.55)(0.68,0.56)10.415(0.979,0.732)
SR0.58(0.53,0.71)(0.601, 0.57)0(0.491,0.63)
CSA(proposed)1.04(1.00,1.09)(0.763,0.728)0.161(0.891,0.81)
CSA-INFO (proposed)0.81(0.77,0.83)(0.752,0.651)0.156(0.907, 0.881)
", + "type": "table", + "image_path": "02f7c367f232f20c3ad47f3b482bcff2a58309728cb6b371a98f07c24827d815.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 136, + 118, + 473, + 151.66666666666666 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 136, + 151.66666666666666, + 473, + 185.33333333333331 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 136, + 185.33333333333331, + 473, + 218.99999999999997 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 107, + 229, + 317, + 317 + ], + "lines": [ + { + "bbox": [ + 105, + 227, + 317, + 242 + ], + "spans": [ + { + "bbox": [ + 105, + 227, + 317, + 242 + ], + "score": 1.0, + "content": "Datasets We consider the following datasets: (i)", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 240, + 318, + 252 + ], + "spans": [ + { + "bbox": [ + 106, + 240, + 318, + 252 + ], + "score": 1.0, + "content": "FRAMINGHAM, is an EHR-based longitudinal car-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 251, + 317, + 263 + ], + "spans": [ + { + "bbox": [ + 106, + 251, + 317, + 263 + ], + "score": 1.0, + "content": "diovascular cohort study that we use to evaluate the", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 262, + 318, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 262, + 318, + 273 + ], + "score": 1.0, + "content": "effect of statins on future coronary heart disease out-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 273, + 318, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 231, + 285 + ], + "score": 1.0, + "content": "comes (Benjamin et al., 1994);", + "type": "text" + }, + { + "bbox": [ + 232, + 273, + 247, + 285 + ], + "score": 0.4, + "content": "( i i )", + "type": "inline_equation" + }, + { + "bbox": [ + 247, + 273, + 318, + 285 + ], + "score": 1.0, + "content": "ACTG, is a longi-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 284, + 318, + 296 + ], + "spans": [ + { + "bbox": [ + 106, + 284, + 318, + 296 + ], + "score": 1.0, + "content": "tudinal RCT study comparing monotherapy with Zi-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 294, + 317, + 307 + ], + "spans": [ + { + "bbox": [ + 106, + 294, + 317, + 307 + ], + "score": 1.0, + "content": "dovudine or Didanosine with combination therapy in", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 306, + 318, + 318 + ], + "spans": [ + { + "bbox": [ + 106, + 306, + 269, + 318 + ], + "score": 1.0, + "content": "HIV patients (Hammer et al., 1996); and", + "type": "text" + }, + { + "bbox": [ + 270, + 306, + 288, + 318 + ], + "score": 0.65, + "content": "( i i i )", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 306, + 318, + 318 + ], + "score": 1.0, + "content": "ACTG-", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 227, + 318, + 318 + ] + }, + { + "type": "table", + "bbox": [ + 326, + 240, + 502, + 297 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 328, + 225, + 500, + 236 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 326, + 225, + 502, + 237 + ], + "spans": [ + { + "bbox": [ + 326, + 225, + 502, + 237 + ], + "score": 1.0, + "content": "Table 2: Summary statistics of the datasets.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "table_body", + "bbox": [ + 326, + 240, + 502, + 297 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 326, + 240, + 502, + 297 + ], + "spans": [ + { + "bbox": [ + 326, + 240, + 502, + 297 + ], + "score": 0.968, + "html": "
FRAMINGHAMACTGACTG-SYNTHETIC
Events (%)26.026.948.9
Treatment (%)10.449.555.9
N3,4351,0542,139
p322323
Missing (%)0.231.411.38
tmax (days)7,2791,2311,313
", + "type": "table", + "image_path": "662567b84ec6b8412fd1a2eb1f6e306f029c8bde26af3d28de506fa5a1d065e3.jpg" + } + ] + } + ], + "index": 16.5, + "virtual_lines": [ + { + "bbox": [ + 326, + 240, + 502, + 254.25 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 326, + 254.25, + 502, + 268.5 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 326, + 268.5, + 502, + 282.75 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 326, + 282.75, + 502, + 297.0 + ], + "spans": [], + "index": 18 + } + ] + } + ], + "index": 15.25 + }, + { + "type": "text", + "bbox": [ + 107, + 317, + 505, + 394 + ], + "lines": [ + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 505, + 329 + ], + "score": 1.0, + "content": "SYNTHETIC, is a semi-synthetic dataset based on ACTG covariates. We simulate potential outcomes", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 327, + 504, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 504, + 340 + ], + "score": 1.0, + "content": "according to a Gompertz-Cox distribution (Bender et al., 2005) with selection bias from a simple", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 338, + 506, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 180, + 352 + ], + "score": 1.0, + "content": "logistic model for", + "type": "text" + }, + { + "bbox": [ + 181, + 339, + 253, + 351 + ], + "score": 0.91, + "content": "\\bar { P ( A = 1 | X = x ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 338, + 506, + 352 + ], + "score": 1.0, + "content": "and AFT-based censoring mechanism. The generative process", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 349, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 349, + 389, + 363 + ], + "score": 1.0, + "content": "is detailed in the SM. Table 2 summarizes the datasets according to", + "type": "text" + }, + { + "bbox": [ + 389, + 350, + 401, + 362 + ], + "score": 0.59, + "content": "( i )", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 349, + 477, + 363 + ], + "score": 1.0, + "content": "covariates of size", + "type": "text" + }, + { + "bbox": [ + 478, + 351, + 484, + 362 + ], + "score": 0.62, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 349, + 489, + 363 + ], + "score": 1.0, + "content": ";", + "type": "text" + }, + { + "bbox": [ + 490, + 350, + 505, + 362 + ], + "score": 0.34, + "content": "( i i )", + "type": "inline_equation" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 361, + 506, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 407, + 374 + ], + "score": 1.0, + "content": "proportion of non-censored events, treated units, and missing entries in the", + "type": "text" + }, + { + "bbox": [ + 407, + 361, + 435, + 372 + ], + "score": 0.91, + "content": "N \\times p", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 361, + 506, + 374 + ], + "score": 1.0, + "content": "covariate matrix;", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 371, + 506, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 371, + 124, + 385 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 372, + 142, + 383 + ], + "score": 0.84, + "content": "( i i i )", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 371, + 189, + 385 + ], + "score": 1.0, + "content": "time range", + "type": "text" + }, + { + "bbox": [ + 190, + 372, + 210, + 383 + ], + "score": 0.9, + "content": "t _ { \\mathrm { m a x } }", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 371, + 506, + 385 + ], + "score": 1.0, + "content": "for both censored and non-censored events. Missing entries are imputed", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 382, + 359, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 382, + 359, + 396 + ], + "score": 1.0, + "content": "with median or mode if continuous or categorical, respectively.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 317, + 506, + 396 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 399, + 505, + 619 + ], + "lines": [ + { + "bbox": [ + 106, + 399, + 506, + 411 + ], + "spans": [ + { + "bbox": [ + 106, + 399, + 506, + 411 + ], + "score": 1.0, + "content": "Quantitative Results Experimental results for two data-sets in Tables 1 and 3, illustrate that AFT-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 411, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 106, + 411, + 505, + 422 + ], + "score": 1.0, + "content": "based methods are high variance, inferior in calibration and C-Index than accuracy-based methods", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 421, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 506, + 433 + ], + "score": 1.0, + "content": "(SR, CSA, CSA-INFO). Surv-BART is the least calibrated but low variance method. CSA-INFO and", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 432, + 505, + 444 + ], + "spans": [ + { + "bbox": [ + 106, + 432, + 505, + 444 + ], + "score": 1.0, + "content": "CSA outperform all methods across all factual metrics, whereas CSA-INFO is better calibrated, low", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 442, + 506, + 457 + ], + "spans": [ + { + "bbox": [ + 104, + 442, + 506, + 457 + ], + "score": 1.0, + "content": "variance but slightly lower C-Index than CSA. Note that we fit CoxPH using the entire dataset; since", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 453, + 506, + 468 + ], + "spans": [ + { + "bbox": [ + 104, + 453, + 506, + 468 + ], + "score": 1.0, + "content": "it does not support counterfactual inference, we do not present factual metrics. By properly adjusting", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 466, + 505, + 477 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 505, + 477 + ], + "score": 1.0, + "content": "for both informative censoring and selection bias, CSA-INFO significantly outperforms all methods", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 476, + 506, + 489 + ], + "spans": [ + { + "bbox": [ + 104, + 476, + 288, + 489 + ], + "score": 1.0, + "content": "in treatment effect estimation according to", + "type": "text" + }, + { + "bbox": [ + 288, + 476, + 315, + 488 + ], + "score": 0.74, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 476, + 336, + 489 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 336, + 477, + 365, + 487 + ], + "score": 0.81, + "content": "\\epsilon _ { \\mathrm { P E H E } }", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 476, + 506, + 489 + ], + "score": 1.0, + "content": ", across non-RCT datasets, while", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 487, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 505, + 500 + ], + "score": 1.0, + "content": "remaining comparable to AFT-Weibull on the RCT dataset (see the SM). Further, RCT-based results", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 498, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 498, + 292, + 511 + ], + "score": 1.0, + "content": "on ACTG data in the SM illustrate comparable", + "type": "text" + }, + { + "bbox": [ + 292, + 498, + 319, + 510 + ], + "score": 0.82, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 498, + 505, + 511 + ], + "score": 1.0, + "content": "across all models except for AFT-log-Normal", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 510, + 505, + 521 + ], + "spans": [ + { + "bbox": [ + 106, + 510, + 505, + 521 + ], + "score": 1.0, + "content": "and Surv-BART, which overestimate, and SR, which underestimates risk. For non-RCT datasets", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 520, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 505, + 532 + ], + "score": 1.0, + "content": "(ACTG-SYNTHENTIC and FRAMINGHAM), CoxPH-OW has a clear advantage over all CoxPH based", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 531, + 506, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 401, + 545 + ], + "score": 1.0, + "content": "methods, mostly credited to the well-behaved bounded propensity weights", + "type": "text" + }, + { + "bbox": [ + 402, + 531, + 432, + 543 + ], + "score": 0.89, + "content": "\\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 531, + 506, + 545 + ], + "score": 1.0, + "content": ". Interestingly, the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 541, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 506, + 555 + ], + "score": 1.0, + "content": "FRAMINGHAM observational data exhibits a common paradox, where without proper adjustment of", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 553, + 506, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 506, + 566 + ], + "score": 1.0, + "content": "selection and censoring bias, naive approaches would result in a counter-intuitive treatment effect", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 562, + 507, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 562, + 507, + 577 + ], + "score": 1.0, + "content": "from statins. However, there is severe confounding from covariates such as age, BMI, diabetes,", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 575, + 505, + 586 + ], + "spans": [ + { + "bbox": [ + 106, + 575, + 505, + 586 + ], + "score": 1.0, + "content": "CAD, PAD, MI, stroke, etc., that influence both treatment likelihood and survival time. Table", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 585, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 475, + 599 + ], + "score": 1.0, + "content": "3, demonstrates that CSA-INFO is clearly the best performing approach. Specifically, its", + "type": "text" + }, + { + "bbox": [ + 475, + 586, + 502, + 598 + ], + "score": 0.78, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 585, + 506, + 599 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 596, + 507, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 507, + 610 + ], + "score": 1.0, + "content": "reverses the biased observational treatment effect, to demonstrate positive treatment from statins,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 608, + 437, + 620 + ], + "spans": [ + { + "bbox": [ + 106, + 608, + 437, + 620 + ], + "score": 1.0, + "content": "which is consistent with prior large RCT longitudinal findings (Yusuf et al., 2016).", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 35.5, + "bbox_fs": [ + 104, + 399, + 507, + 620 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 632, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 632, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 106, + 632, + 505, + 645 + ], + "score": 1.0, + "content": "Qualitative Results Figure 2a demonstrates that CSA-INFO matches the ground truth population", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 643, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 138, + 657 + ], + "score": 1.0, + "content": "hazard,", + "type": "text" + }, + { + "bbox": [ + 138, + 644, + 166, + 656 + ], + "score": 0.73, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 643, + 505, + 657 + ], + "score": 1.0, + "content": ", better than alternative methods on ACTG-SYNTHETIC data. See the SM for ACTG", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 654, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 506, + 668 + ], + "score": 1.0, + "content": "and FRAMINGHAM. Figure 2b shows sub-population log hazard ratios for four patient clusters", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 665, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 385, + 678 + ], + "score": 1.0, + "content": "obtained via hierarchical clustering on the individual log hazard ratios,", + "type": "text" + }, + { + "bbox": [ + 386, + 666, + 436, + 678 + ], + "score": 0.89, + "content": "\\log \\mathrm { H R } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 436, + 665, + 506, + 678 + ], + "score": 1.0, + "content": ", of the test set of", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 676, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 690 + ], + "score": 1.0, + "content": "FRAMINGHAM data. Interestingly, these clusters stratify treatment effects into: positive (2), negative", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 687, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 455, + 701 + ], + "score": 1.0, + "content": "(1 and 3), and neutral (4) sub-populations. Moreover, the estimated density of median", + "type": "text" + }, + { + "bbox": [ + 455, + 687, + 505, + 699 + ], + "score": 0.88, + "content": "\\log \\mathrm { H } \\bar { \\mathrm { R } } ( t | x )", + "type": "inline_equation" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 173, + 712 + ], + "score": 1.0, + "content": "values in Figure", + "type": "text" + }, + { + "bbox": [ + 174, + 699, + 185, + 709 + ], + "score": 0.25, + "content": "_ { 2 \\mathrm { c } }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 699, + 274, + 712 + ], + "score": 1.0, + "content": "illustrates that nearly", + "type": "text" + }, + { + "bbox": [ + 274, + 699, + 294, + 709 + ], + "score": 0.88, + "content": "7 0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 699, + 433, + 712 + ], + "score": 1.0, + "content": "of the testing set individuals have", + "type": "text" + }, + { + "bbox": [ + 434, + 699, + 502, + 711 + ], + "score": 0.9, + "content": "\\log \\mathrm { H R } ( t | x ) < 0", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 699, + 506, + 712 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "thus may benefit from taking statins. Further, we isolated the extreme top and bottom quantiles,", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 107, + 719, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 107, + 720, + 182, + 732 + ], + "score": 0.91, + "content": "\\mathrm { H R } ( t | x ) < 0 . 0 2 4", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 719, + 200, + 734 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 201, + 720, + 275, + 732 + ], + "score": 0.92, + "content": "\\mathrm { H R } ( \\bar { t } | x ) > 1 . 9 1 6", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 719, + 394, + 734 + ], + "score": 1.0, + "content": ", respectively, of the median", + "type": "text" + }, + { + "bbox": [ + 394, + 720, + 444, + 732 + ], + "score": 0.91, + "content": "\\log \\mathrm { \\bar { H R } } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 444, + 719, + 506, + 734 + ], + "score": 1.0, + "content": "values for the", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 106, + 237, + 505, + 250 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 505, + 250 + ], + "score": 1.0, + "content": "test set of FRAMINGHAM, as shown in Figure 2c. After comparing their covariates, we found that", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 248, + 506, + 262 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 506, + 262 + ], + "score": 1.0, + "content": "individuals with the following characteristics may benefit from taking statins: young, male, diabetic,", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 259, + 507, + 274 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 507, + 274 + ], + "score": 1.0, + "content": "without prior history (CAD, PAD, stroke or MI), high BMI, cholesterol, triglycerides, fasting glucose,", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 271, + 505, + 283 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 505, + 283 + ], + "score": 1.0, + "content": "and low high-density lipoprotein. There seem to be consensus that diabetics and high-cholesterol", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 282, + 500, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 282, + 500, + 295 + ], + "score": 1.0, + "content": "patients benefit from statins (Cheung et al., 2004; Wilt et al., 2004). See SM for additional results.", + "type": "text", + "cross_page": true + } + ], + "index": 11 + } + ], + "index": 50, + "bbox_fs": [ + 105, + 632, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 115, + 75, + 489, + 177 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 115, + 75, + 489, + 177 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 115, + 75, + 489, + 177 + ], + "spans": [ + { + "bbox": [ + 115, + 75, + 489, + 177 + ], + "score": 0.969, + "type": "image", + "image_path": "4c89c16905f48a1bdcc35b1ade4f0f866e4e151f0c05e3f1b2a559591a62f777.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 115, + 75, + 489, + 109.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 115, + 109.0, + 489, + 143.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 115, + 143.0, + 489, + 177.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 181, + 505, + 225 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 180, + 505, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 181, + 236, + 194 + ], + "score": 1.0, + "content": "Figure 2: (a) Inferred population", + "type": "text" + }, + { + "bbox": [ + 236, + 180, + 263, + 192 + ], + "score": 0.66, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 181, + 505, + 194 + ], + "score": 1.0, + "content": "compared against ground truth (EMP) on ACTG-SYNTHETIC", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 191, + 506, + 205 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 318, + 205 + ], + "score": 1.0, + "content": "data. CSA-INFO-based (b) cluster-specific average l", + "type": "text" + }, + { + "bbox": [ + 318, + 192, + 362, + 204 + ], + "score": 0.85, + "content": "\\operatorname { \\Pi } ^ { } \\operatorname { \\Pi } ^ { } \\operatorname { \\mathrm { ) g } } \\operatorname { \\mathrm { { H R } } } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 191, + 506, + 205 + ], + "score": 1.0, + "content": "curves and (c) estimated density of", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 203, + 505, + 215 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 138, + 215 + ], + "score": 1.0, + "content": "median", + "type": "text" + }, + { + "bbox": [ + 139, + 203, + 189, + 214 + ], + "score": 0.9, + "content": "\\log \\mathrm { H R } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 203, + 505, + 215 + ], + "score": 1.0, + "content": "values on the test set of the FRAMINGHAM dataset. Clusters assignment were", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 213, + 399, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 319, + 227 + ], + "score": 1.0, + "content": "obtained via hierarchical clustering of individualized", + "type": "text" + }, + { + "bbox": [ + 319, + 213, + 370, + 225 + ], + "score": 0.89, + "content": "\\log \\mathrm { H R } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 213, + 399, + 227 + ], + "score": 1.0, + "content": "traces.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 107, + 237, + 506, + 293 + ], + "lines": [ + { + "bbox": [ + 106, + 237, + 505, + 250 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 505, + 250 + ], + "score": 1.0, + "content": "test set of FRAMINGHAM, as shown in Figure 2c. After comparing their covariates, we found that", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 248, + 506, + 262 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 506, + 262 + ], + "score": 1.0, + "content": "individuals with the following characteristics may benefit from taking statins: young, male, diabetic,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 259, + 507, + 274 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 507, + 274 + ], + "score": 1.0, + "content": "without prior history (CAD, PAD, stroke or MI), high BMI, cholesterol, triglycerides, fasting glucose,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 271, + 505, + 283 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 505, + 283 + ], + "score": 1.0, + "content": "and low high-density lipoprotein. There seem to be consensus that diabetics and high-cholesterol", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 282, + 500, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 282, + 500, + 295 + ], + "score": 1.0, + "content": "patients benefit from statins (Cheung et al., 2004; Wilt et al., 2004). See SM for additional results.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9 + }, + { + "type": "title", + "bbox": [ + 107, + 301, + 202, + 313 + ], + "lines": [ + { + "bbox": [ + 105, + 299, + 203, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 299, + 203, + 316 + ], + "score": 1.0, + "content": "6 CONCLUSIONS", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 317, + 506, + 395 + ], + "lines": [ + { + "bbox": [ + 106, + 317, + 505, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 505, + 330 + ], + "score": 1.0, + "content": "We have proposed a unified counterfactual inference framework for survival analysis. Our approach", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 328, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 106, + 328, + 505, + 340 + ], + "score": 1.0, + "content": "adjusts for bias from two unknown sources, namely, confounding due to covariate dependent selection", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 340, + 506, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 340, + 506, + 352 + ], + "score": 1.0, + "content": "bias and censoring (informative or non-informative). Relative to competitive alternatives, we demon-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 351, + 506, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 506, + 363 + ], + "score": 1.0, + "content": "strate superior performance for both survival-outcome prediction and treatment-effect estimation,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 361, + 506, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 506, + 374 + ], + "score": 1.0, + "content": "across three diverse datasets, including a semi-synthetic dataset which we introduce. Moreover,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 372, + 506, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 506, + 385 + ], + "score": 1.0, + "content": "we formulate a model-free nonparametric hazard ratio metric for comparing treatment effects or", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 384, + 410, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 410, + 396 + ], + "score": 1.0, + "content": "leveraging prior randomized real-world experiments in longitudinal studies.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 16 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 307, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 115, + 75, + 489, + 177 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 115, + 75, + 489, + 177 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 115, + 75, + 489, + 177 + ], + "spans": [ + { + "bbox": [ + 115, + 75, + 489, + 177 + ], + "score": 0.969, + "type": "image", + "image_path": "4c89c16905f48a1bdcc35b1ade4f0f866e4e151f0c05e3f1b2a559591a62f777.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 115, + 75, + 489, + 109.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 115, + 109.0, + 489, + 143.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 115, + 143.0, + 489, + 177.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 181, + 505, + 225 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 180, + 505, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 181, + 236, + 194 + ], + "score": 1.0, + "content": "Figure 2: (a) Inferred population", + "type": "text" + }, + { + "bbox": [ + 236, + 180, + 263, + 192 + ], + "score": 0.66, + "content": "\\mathrm { H R } ( t )", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 181, + 505, + 194 + ], + "score": 1.0, + "content": "compared against ground truth (EMP) on ACTG-SYNTHETIC", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 191, + 506, + 205 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 318, + 205 + ], + "score": 1.0, + "content": "data. CSA-INFO-based (b) cluster-specific average l", + "type": "text" + }, + { + "bbox": [ + 318, + 192, + 362, + 204 + ], + "score": 0.85, + "content": "\\operatorname { \\Pi } ^ { } \\operatorname { \\Pi } ^ { } \\operatorname { \\mathrm { ) g } } \\operatorname { \\mathrm { { H R } } } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 191, + 506, + 205 + ], + "score": 1.0, + "content": "curves and (c) estimated density of", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 203, + 505, + 215 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 138, + 215 + ], + "score": 1.0, + "content": "median", + "type": "text" + }, + { + "bbox": [ + 139, + 203, + 189, + 214 + ], + "score": 0.9, + "content": "\\log \\mathrm { H R } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 203, + 505, + 215 + ], + "score": 1.0, + "content": "values on the test set of the FRAMINGHAM dataset. Clusters assignment were", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 213, + 399, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 319, + 227 + ], + "score": 1.0, + "content": "obtained via hierarchical clustering of individualized", + "type": "text" + }, + { + "bbox": [ + 319, + 213, + 370, + 225 + ], + "score": 0.89, + "content": "\\log \\mathrm { H R } ( t | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 213, + 399, + 227 + ], + "score": 1.0, + "content": "traces.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 107, + 237, + 506, + 293 + ], + "lines": [], + "index": 9, + "bbox_fs": [ + 105, + 237, + 507, + 295 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 107, + 301, + 202, + 313 + ], + "lines": [ + { + "bbox": [ + 105, + 299, + 203, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 299, + 203, + 316 + ], + "score": 1.0, + "content": "6 CONCLUSIONS", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 317, + 506, + 395 + ], + "lines": [ + { + "bbox": [ + 106, + 317, + 505, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 505, + 330 + ], + "score": 1.0, + "content": "We have proposed a unified counterfactual inference framework for survival analysis. Our approach", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 328, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 106, + 328, + 505, + 340 + ], + "score": 1.0, + "content": "adjusts for bias from two unknown sources, namely, confounding due to covariate dependent selection", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 340, + 506, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 340, + 506, + 352 + ], + "score": 1.0, + "content": "bias and censoring (informative or non-informative). Relative to competitive alternatives, we demon-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 351, + 506, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 506, + 363 + ], + "score": 1.0, + "content": "strate superior performance for both survival-outcome prediction and treatment-effect estimation,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 361, + 506, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 506, + 374 + ], + "score": 1.0, + "content": "across three diverse datasets, including a semi-synthetic dataset which we introduce. Moreover,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 372, + 506, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 506, + 385 + ], + "score": 1.0, + "content": "we formulate a model-free nonparametric hazard ratio metric for comparing treatment effects or", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 384, + 410, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 410, + 396 + ], + "score": 1.0, + "content": "leveraging prior randomized real-world experiments in longitudinal studies.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 317, + 506, + 396 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 176, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 176, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 176, + 95 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 100, + 506, + 133 + ], + "lines": [ + { + "bbox": [ + 106, + 100, + 506, + 113 + ], + "spans": [ + { + "bbox": [ + 106, + 100, + 506, + 113 + ], + "score": 1.0, + "content": "Peter C Austin. Propensity-score matching in the cardiovascular surgery literature from 2004 to 2006:", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 112, + 505, + 124 + ], + "spans": [ + { + "bbox": [ + 115, + 112, + 505, + 124 + ], + "score": 1.0, + "content": "a systematic review and suggestions for improvement. The Journal of Thoracic and Cardiovascular", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 122, + 178, + 136 + ], + "spans": [ + { + "bbox": [ + 115, + 122, + 178, + 136 + ], + "score": 1.0, + "content": "Surgery, 2007.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 106, + 142, + 504, + 175 + ], + "lines": [ + { + "bbox": [ + 105, + 141, + 506, + 155 + ], + "spans": [ + { + "bbox": [ + 105, + 141, + 506, + 155 + ], + "score": 1.0, + "content": "Peter C Austin. The use of propensity score methods with survival or time-to-event outcomes:", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 153, + 505, + 165 + ], + "spans": [ + { + "bbox": [ + 115, + 153, + 505, + 165 + ], + "score": 1.0, + "content": "reporting measures of effect similar to those used in randomized experiments. Statistics in", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 164, + 183, + 176 + ], + "spans": [ + { + "bbox": [ + 115, + 164, + 183, + 176 + ], + "score": 1.0, + "content": "Medicine, 2014.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 183, + 504, + 196 + ], + "lines": [ + { + "bbox": [ + 106, + 183, + 505, + 195 + ], + "spans": [ + { + "bbox": [ + 106, + 183, + 505, + 195 + ], + "score": 1.0, + "content": "Elias Bareinboim and Judea Pearl. Controlling selection bias in causal inference. In AISTATS, 2012.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 108, + 203, + 503, + 227 + ], + "lines": [ + { + "bbox": [ + 106, + 203, + 505, + 216 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 505, + 216 + ], + "score": 1.0, + "content": "Ralf Bender, Thomas Augustin, and Maria Blettner. Generating survival times to simulate cox", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 215, + 351, + 227 + ], + "spans": [ + { + "bbox": [ + 115, + 215, + 351, + 227 + ], + "score": 1.0, + "content": "proportional hazards models. Statistics in medicine, 2005.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 107, + 234, + 505, + 268 + ], + "lines": [ + { + "bbox": [ + 106, + 234, + 505, + 248 + ], + "spans": [ + { + "bbox": [ + 106, + 234, + 505, + 248 + ], + "score": 1.0, + "content": "Emelia J Benjamin, Daniel Levy, Sonya M Vaziri, Ralph B D’Agostino, Albert J Belanger, and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 245, + 504, + 257 + ], + "spans": [ + { + "bbox": [ + 116, + 245, + 504, + 257 + ], + "score": 1.0, + "content": "Philip A Wolf. Independent risk factors for atrial fibrillation in a population-based cohort: the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 116, + 257, + 268, + 269 + ], + "spans": [ + { + "bbox": [ + 116, + 257, + 268, + 269 + ], + "score": 1.0, + "content": "framingham heart study. Jama, 1994.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 107, + 276, + 506, + 310 + ], + "lines": [ + { + "bbox": [ + 106, + 276, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 505, + 289 + ], + "score": 1.0, + "content": "Ashley L Buchanan, Michael G Hudgens, Stephen R Cole, Bryan Lau, Adaora A Adimora, and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 116, + 287, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 116, + 287, + 505, + 300 + ], + "score": 1.0, + "content": "Women’s Interagency HIV Study. Worth the weight: using inverse probability weighted cox", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 116, + 298, + 401, + 311 + ], + "spans": [ + { + "bbox": [ + 116, + 298, + 401, + 311 + ], + "score": 1.0, + "content": "models in aids research. AIDS research and human retroviruses, 2014.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 106, + 317, + 505, + 341 + ], + "lines": [ + { + "bbox": [ + 106, + 317, + 505, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 505, + 330 + ], + "score": 1.0, + "content": "Weihua Cao, Anastasios A Tsiatis, and Marie Davidian. Improving efficiency and robustness of the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 116, + 329, + 468, + 342 + ], + "spans": [ + { + "bbox": [ + 116, + 329, + 468, + 342 + ], + "score": 1.0, + "content": "doubly robust estimator for a population mean with incomplete data. Biometrika, 2009.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "text", + "bbox": [ + 110, + 348, + 505, + 382 + ], + "lines": [ + { + "bbox": [ + 106, + 348, + 506, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 348, + 506, + 362 + ], + "score": 1.0, + "content": "P. Chapfuwa, C. Tao, C. Li, I. Khan, K. J. Chandross, M. J. Pencina, L. Carin, and R. Henao.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 359, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 115, + 359, + 505, + 373 + ], + "score": 1.0, + "content": "Calibration and uncertainty in neural time-to-event modeling. IEEE Transactions on Neural", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 371, + 274, + 383 + ], + "spans": [ + { + "bbox": [ + 115, + 371, + 274, + 383 + ], + "score": 1.0, + "content": "Networks and Learning Systems, 2020.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 105, + 390, + 504, + 414 + ], + "lines": [ + { + "bbox": [ + 105, + 390, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 505, + 403 + ], + "score": 1.0, + "content": "Paidamoyo Chapfuwa, Chenyang Tao, Chunyuan Li, Courtney Page, Benjamin Goldstein, Lawrence", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 401, + 440, + 414 + ], + "spans": [ + { + "bbox": [ + 116, + 401, + 440, + 414 + ], + "score": 1.0, + "content": "Carin, and Ricardo Henao. Adversarial time-to-event modeling. In ICML, 2018.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 107, + 421, + 505, + 455 + ], + "lines": [ + { + "bbox": [ + 105, + 420, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 505, + 435 + ], + "score": 1.0, + "content": "Bernard MY Cheung, Ian J Lauder, Chu-Pak Lau, and Cyrus R Kumana. Meta-analysis of large", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 432, + 505, + 444 + ], + "spans": [ + { + "bbox": [ + 115, + 432, + 505, + 444 + ], + "score": 1.0, + "content": "randomized controlled trials to evaluate the impact of statins on cardiovascular outcomes. British", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 444, + 279, + 456 + ], + "spans": [ + { + "bbox": [ + 115, + 444, + 279, + 456 + ], + "score": 1.0, + "content": "journal of clinical pharmacology, 2004.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 106, + 462, + 504, + 486 + ], + "lines": [ + { + "bbox": [ + 105, + 462, + 505, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 505, + 477 + ], + "score": 1.0, + "content": "Hugh A Chipman, Edward I George, Robert E McCulloch, et al. Bart: Bayesian additive regression", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 474, + 299, + 486 + ], + "spans": [ + { + "bbox": [ + 115, + 474, + 299, + 486 + ], + "score": 1.0, + "content": "trees. The Annals of Applied Statistics, 2010.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 106, + 493, + 505, + 517 + ], + "lines": [ + { + "bbox": [ + 105, + 492, + 506, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 506, + 507 + ], + "score": 1.0, + "content": "Stephen R Cole and Miguel A Hernán. Adjusted survival curves with inverse probability weights.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 505, + 341, + 517 + ], + "spans": [ + { + "bbox": [ + 115, + 505, + 341, + 517 + ], + "score": 1.0, + "content": "Computer methods and programs in biomedicine, 2004.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 106, + 524, + 505, + 547 + ], + "lines": [ + { + "bbox": [ + 106, + 524, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 106, + 524, + 505, + 537 + ], + "score": 1.0, + "content": "David R Cox. Regression models and life-tables. Journal of the Royal Statistical Society: Series B", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 535, + 216, + 548 + ], + "spans": [ + { + "bbox": [ + 115, + 535, + 216, + 548 + ], + "score": 1.0, + "content": "(Methodological), 1972.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5 + }, + { + "type": "text", + "bbox": [ + 106, + 555, + 506, + 589 + ], + "lines": [ + { + "bbox": [ + 105, + 554, + 506, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 506, + 569 + ], + "score": 1.0, + "content": "Richard K Crump, V Joseph Hotz, Guido W Imbens, and Oscar A Mitnik. Moving the goalposts:", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 116, + 567, + 506, + 579 + ], + "spans": [ + { + "bbox": [ + 116, + 567, + 506, + 579 + ], + "score": 1.0, + "content": "Addressing limited overlap in the estimation of average treatment effects by changing the estimand.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 577, + 374, + 590 + ], + "spans": [ + { + "bbox": [ + 116, + 577, + 374, + 590 + ], + "score": 1.0, + "content": "Technical report, National Bureau of Economic Research, 2006.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 105, + 597, + 504, + 620 + ], + "lines": [ + { + "bbox": [ + 106, + 595, + 505, + 611 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 505, + 611 + ], + "score": 1.0, + "content": "Yifan Cui, Michael R Kosorok, Stefan Wager, and Ruoqing Zhu. Estimating heterogeneous treatment", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 116, + 608, + 406, + 621 + ], + "spans": [ + { + "bbox": [ + 116, + 608, + 406, + 621 + ], + "score": 1.0, + "content": "effects with right-censored data via causal survival forests. arXiv, 2020.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 106, + 628, + 505, + 640 + ], + "lines": [ + { + "bbox": [ + 106, + 627, + 506, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 506, + 641 + ], + "score": 1.0, + "content": "Marco Cuturi. Sinkhorn distances: Lightspeed computation of optimal transport. In NeurIPS, 2013.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 105, + 648, + 505, + 670 + ], + "lines": [ + { + "bbox": [ + 105, + 647, + 507, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 647, + 507, + 661 + ], + "score": 1.0, + "content": "Iván Díaz. Statistical inference for data-adaptive doubly robust estimators with survival outcomes.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 659, + 232, + 670 + ], + "spans": [ + { + "bbox": [ + 116, + 659, + 232, + 670 + ], + "score": 1.0, + "content": "Statistics in Medicine, 2019.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 104, + 678, + 504, + 702 + ], + "lines": [ + { + "bbox": [ + 105, + 677, + 506, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 692 + ], + "score": 1.0, + "content": "Jennifer Frankovich, Christopher A Longhurst, and Scott M Sutherland. Evidence-based medicine in", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 689, + 253, + 702 + ], + "spans": [ + { + "bbox": [ + 116, + 689, + 253, + 702 + ], + "score": 1.0, + "content": "the emr era. N Engl J Med, 2011.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 503, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 708, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 708, + 505, + 722 + ], + "score": 1.0, + "content": "Saurabh Gombar, Alison Callahan, Robert Califf, Robert Harrington, and Nigam H Shah. It is time", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 721, + 362, + 732 + ], + "spans": [ + { + "bbox": [ + 116, + 721, + 362, + 732 + ], + "score": 1.0, + "content": "to learn from patients like mine. NPJ digital medicine, 2019.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42.5 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "10", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 307, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 176, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 176, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 176, + 95 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 100, + 506, + 133 + ], + "lines": [ + { + "bbox": [ + 106, + 100, + 506, + 113 + ], + "spans": [ + { + "bbox": [ + 106, + 100, + 506, + 113 + ], + "score": 1.0, + "content": "Peter C Austin. Propensity-score matching in the cardiovascular surgery literature from 2004 to 2006:", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 112, + 505, + 124 + ], + "spans": [ + { + "bbox": [ + 115, + 112, + 505, + 124 + ], + "score": 1.0, + "content": "a systematic review and suggestions for improvement. The Journal of Thoracic and Cardiovascular", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 122, + 178, + 136 + ], + "spans": [ + { + "bbox": [ + 115, + 122, + 178, + 136 + ], + "score": 1.0, + "content": "Surgery, 2007.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2, + "bbox_fs": [ + 106, + 100, + 506, + 136 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 142, + 504, + 175 + ], + "lines": [ + { + "bbox": [ + 105, + 141, + 506, + 155 + ], + "spans": [ + { + "bbox": [ + 105, + 141, + 506, + 155 + ], + "score": 1.0, + "content": "Peter C Austin. The use of propensity score methods with survival or time-to-event outcomes:", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 153, + 505, + 165 + ], + "spans": [ + { + "bbox": [ + 115, + 153, + 505, + 165 + ], + "score": 1.0, + "content": "reporting measures of effect similar to those used in randomized experiments. Statistics in", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 164, + 183, + 176 + ], + "spans": [ + { + "bbox": [ + 115, + 164, + 183, + 176 + ], + "score": 1.0, + "content": "Medicine, 2014.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 141, + 506, + 176 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 183, + 504, + 196 + ], + "lines": [ + { + "bbox": [ + 106, + 183, + 505, + 195 + ], + "spans": [ + { + "bbox": [ + 106, + 183, + 505, + 195 + ], + "score": 1.0, + "content": "Elias Bareinboim and Judea Pearl. Controlling selection bias in causal inference. In AISTATS, 2012.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7, + "bbox_fs": [ + 106, + 183, + 505, + 195 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 203, + 503, + 227 + ], + "lines": [ + { + "bbox": [ + 106, + 203, + 505, + 216 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 505, + 216 + ], + "score": 1.0, + "content": "Ralf Bender, Thomas Augustin, and Maria Blettner. Generating survival times to simulate cox", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 215, + 351, + 227 + ], + "spans": [ + { + "bbox": [ + 115, + 215, + 351, + 227 + ], + "score": 1.0, + "content": "proportional hazards models. Statistics in medicine, 2005.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 106, + 203, + 505, + 227 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 234, + 505, + 268 + ], + "lines": [ + { + "bbox": [ + 106, + 234, + 505, + 248 + ], + "spans": [ + { + "bbox": [ + 106, + 234, + 505, + 248 + ], + "score": 1.0, + "content": "Emelia J Benjamin, Daniel Levy, Sonya M Vaziri, Ralph B D’Agostino, Albert J Belanger, and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 245, + 504, + 257 + ], + "spans": [ + { + "bbox": [ + 116, + 245, + 504, + 257 + ], + "score": 1.0, + "content": "Philip A Wolf. Independent risk factors for atrial fibrillation in a population-based cohort: the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 116, + 257, + 268, + 269 + ], + "spans": [ + { + "bbox": [ + 116, + 257, + 268, + 269 + ], + "score": 1.0, + "content": "framingham heart study. Jama, 1994.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11, + "bbox_fs": [ + 106, + 234, + 505, + 269 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 276, + 506, + 310 + ], + "lines": [ + { + "bbox": [ + 106, + 276, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 505, + 289 + ], + "score": 1.0, + "content": "Ashley L Buchanan, Michael G Hudgens, Stephen R Cole, Bryan Lau, Adaora A Adimora, and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 116, + 287, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 116, + 287, + 505, + 300 + ], + "score": 1.0, + "content": "Women’s Interagency HIV Study. Worth the weight: using inverse probability weighted cox", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 116, + 298, + 401, + 311 + ], + "spans": [ + { + "bbox": [ + 116, + 298, + 401, + 311 + ], + "score": 1.0, + "content": "models in aids research. AIDS research and human retroviruses, 2014.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14, + "bbox_fs": [ + 106, + 276, + 505, + 311 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 317, + 505, + 341 + ], + "lines": [ + { + "bbox": [ + 106, + 317, + 505, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 505, + 330 + ], + "score": 1.0, + "content": "Weihua Cao, Anastasios A Tsiatis, and Marie Davidian. Improving efficiency and robustness of the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 116, + 329, + 468, + 342 + ], + "spans": [ + { + "bbox": [ + 116, + 329, + 468, + 342 + ], + "score": 1.0, + "content": "doubly robust estimator for a population mean with incomplete data. Biometrika, 2009.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5, + "bbox_fs": [ + 106, + 317, + 505, + 342 + ] + }, + { + "type": "text", + "bbox": [ + 110, + 348, + 505, + 382 + ], + "lines": [ + { + "bbox": [ + 106, + 348, + 506, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 348, + 506, + 362 + ], + "score": 1.0, + "content": "P. Chapfuwa, C. Tao, C. Li, I. Khan, K. J. Chandross, M. J. Pencina, L. Carin, and R. Henao.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 359, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 115, + 359, + 505, + 373 + ], + "score": 1.0, + "content": "Calibration and uncertainty in neural time-to-event modeling. IEEE Transactions on Neural", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 371, + 274, + 383 + ], + "spans": [ + { + "bbox": [ + 115, + 371, + 274, + 383 + ], + "score": 1.0, + "content": "Networks and Learning Systems, 2020.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19, + "bbox_fs": [ + 106, + 348, + 506, + 383 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 390, + 504, + 414 + ], + "lines": [ + { + "bbox": [ + 105, + 390, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 505, + 403 + ], + "score": 1.0, + "content": "Paidamoyo Chapfuwa, Chenyang Tao, Chunyuan Li, Courtney Page, Benjamin Goldstein, Lawrence", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 401, + 440, + 414 + ], + "spans": [ + { + "bbox": [ + 116, + 401, + 440, + 414 + ], + "score": 1.0, + "content": "Carin, and Ricardo Henao. Adversarial time-to-event modeling. In ICML, 2018.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 390, + 505, + 414 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 421, + 505, + 455 + ], + "lines": [ + { + "bbox": [ + 105, + 420, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 505, + 435 + ], + "score": 1.0, + "content": "Bernard MY Cheung, Ian J Lauder, Chu-Pak Lau, and Cyrus R Kumana. Meta-analysis of large", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 432, + 505, + 444 + ], + "spans": [ + { + "bbox": [ + 115, + 432, + 505, + 444 + ], + "score": 1.0, + "content": "randomized controlled trials to evaluate the impact of statins on cardiovascular outcomes. British", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 444, + 279, + 456 + ], + "spans": [ + { + "bbox": [ + 115, + 444, + 279, + 456 + ], + "score": 1.0, + "content": "journal of clinical pharmacology, 2004.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 420, + 505, + 456 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 462, + 504, + 486 + ], + "lines": [ + { + "bbox": [ + 105, + 462, + 505, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 462, + 505, + 477 + ], + "score": 1.0, + "content": "Hugh A Chipman, Edward I George, Robert E McCulloch, et al. Bart: Bayesian additive regression", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 474, + 299, + 486 + ], + "spans": [ + { + "bbox": [ + 115, + 474, + 299, + 486 + ], + "score": 1.0, + "content": "trees. The Annals of Applied Statistics, 2010.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 462, + 505, + 486 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 493, + 505, + 517 + ], + "lines": [ + { + "bbox": [ + 105, + 492, + 506, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 506, + 507 + ], + "score": 1.0, + "content": "Stephen R Cole and Miguel A Hernán. Adjusted survival curves with inverse probability weights.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 505, + 341, + 517 + ], + "spans": [ + { + "bbox": [ + 115, + 505, + 341, + 517 + ], + "score": 1.0, + "content": "Computer methods and programs in biomedicine, 2004.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5, + "bbox_fs": [ + 105, + 492, + 506, + 517 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 524, + 505, + 547 + ], + "lines": [ + { + "bbox": [ + 106, + 524, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 106, + 524, + 505, + 537 + ], + "score": 1.0, + "content": "David R Cox. Regression models and life-tables. Journal of the Royal Statistical Society: Series B", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 535, + 216, + 548 + ], + "spans": [ + { + "bbox": [ + 115, + 535, + 216, + 548 + ], + "score": 1.0, + "content": "(Methodological), 1972.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5, + "bbox_fs": [ + 106, + 524, + 505, + 548 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 555, + 506, + 589 + ], + "lines": [ + { + "bbox": [ + 105, + 554, + 506, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 506, + 569 + ], + "score": 1.0, + "content": "Richard K Crump, V Joseph Hotz, Guido W Imbens, and Oscar A Mitnik. Moving the goalposts:", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 116, + 567, + 506, + 579 + ], + "spans": [ + { + "bbox": [ + 116, + 567, + 506, + 579 + ], + "score": 1.0, + "content": "Addressing limited overlap in the estimation of average treatment effects by changing the estimand.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 577, + 374, + 590 + ], + "spans": [ + { + "bbox": [ + 116, + 577, + 374, + 590 + ], + "score": 1.0, + "content": "Technical report, National Bureau of Economic Research, 2006.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33, + "bbox_fs": [ + 105, + 554, + 506, + 590 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 597, + 504, + 620 + ], + "lines": [ + { + "bbox": [ + 106, + 595, + 505, + 611 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 505, + 611 + ], + "score": 1.0, + "content": "Yifan Cui, Michael R Kosorok, Stefan Wager, and Ruoqing Zhu. Estimating heterogeneous treatment", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 116, + 608, + 406, + 621 + ], + "spans": [ + { + "bbox": [ + 116, + 608, + 406, + 621 + ], + "score": 1.0, + "content": "effects with right-censored data via causal survival forests. arXiv, 2020.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5, + "bbox_fs": [ + 106, + 595, + 505, + 621 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 628, + 505, + 640 + ], + "lines": [ + { + "bbox": [ + 106, + 627, + 506, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 506, + 641 + ], + "score": 1.0, + "content": "Marco Cuturi. Sinkhorn distances: Lightspeed computation of optimal transport. In NeurIPS, 2013.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37, + "bbox_fs": [ + 106, + 627, + 506, + 641 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 648, + 505, + 670 + ], + "lines": [ + { + "bbox": [ + 105, + 647, + 507, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 647, + 507, + 661 + ], + "score": 1.0, + "content": "Iván Díaz. Statistical inference for data-adaptive doubly robust estimators with survival outcomes.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 659, + 232, + 670 + ], + "spans": [ + { + "bbox": [ + 116, + 659, + 232, + 670 + ], + "score": 1.0, + "content": "Statistics in Medicine, 2019.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 105, + 647, + 507, + 670 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 678, + 504, + 702 + ], + "lines": [ + { + "bbox": [ + 105, + 677, + 506, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 692 + ], + "score": 1.0, + "content": "Jennifer Frankovich, Christopher A Longhurst, and Scott M Sutherland. Evidence-based medicine in", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 689, + 253, + 702 + ], + "spans": [ + { + "bbox": [ + 116, + 689, + 253, + 702 + ], + "score": 1.0, + "content": "the emr era. N Engl J Med, 2011.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5, + "bbox_fs": [ + 105, + 677, + 506, + 702 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 503, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 708, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 708, + 505, + 722 + ], + "score": 1.0, + "content": "Saurabh Gombar, Alison Callahan, Robert Califf, Robert Harrington, and Nigam H Shah. It is time", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 721, + 362, + 732 + ], + "spans": [ + { + "bbox": [ + 116, + 721, + 362, + 732 + ], + "score": 1.0, + "content": "to learn from patients like mine. NPJ digital medicine, 2019.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42.5, + "bbox_fs": [ + 106, + 708, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 505, + 127 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 507, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 507, + 96 + ], + "score": 1.0, + "content": "Scott M Hammer, David A Katzenstein, Michael D Hughes, Holly Gundacker, Robert T Schooley,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 94, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 116, + 94, + 506, + 106 + ], + "score": 1.0, + "content": "Richard H Haubrich, W Keith Henry, Michael M Lederman, John P Phair, Manette Niu, et al. A", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 105, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 115, + 105, + 505, + 117 + ], + "score": 1.0, + "content": "trial comparing nucleoside monotherapy with combination therapy in hiv-infected adults with cd4", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 116, + 116, + 481, + 127 + ], + "spans": [ + { + "bbox": [ + 116, + 116, + 481, + 127 + ], + "score": 1.0, + "content": "cell counts from 200 to 500 per cubic millimeter. New England Journal of Medicine, 1996.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 107, + 135, + 504, + 157 + ], + "lines": [ + { + "bbox": [ + 105, + 135, + 505, + 148 + ], + "spans": [ + { + "bbox": [ + 105, + 135, + 505, + 148 + ], + "score": 1.0, + "content": "Frank E Harrell Jr, Kerry L Lee, Robert M Califf, David B Pryor, and Robert A Rosati. Regression", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 145, + 459, + 158 + ], + "spans": [ + { + "bbox": [ + 116, + 145, + 459, + 158 + ], + "score": 1.0, + "content": "modelling strategies for improved prognostic prediction. Statistics in medicine, 1984.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 107, + 165, + 506, + 199 + ], + "lines": [ + { + "bbox": [ + 106, + 165, + 505, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 165, + 505, + 177 + ], + "score": 1.0, + "content": "Kristiina Häyrinen, Kaija Saranto, and Pirkko Nykänen. Definition, structure, content, use and", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 176, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 115, + 176, + 506, + 189 + ], + "score": 1.0, + "content": "impacts of electronic health records: a review of the research literature. International Journal of", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 187, + 227, + 199 + ], + "spans": [ + { + "bbox": [ + 115, + 187, + 227, + 199 + ], + "score": 1.0, + "content": "Medical Informatics, 2008.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 106, + 206, + 506, + 240 + ], + "lines": [ + { + "bbox": [ + 106, + 207, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 106, + 207, + 505, + 219 + ], + "score": 1.0, + "content": "Nicholas C Henderson, Thomas A Louis, Gary L Rosner, and Ravi Varadhan. Individualized", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 116, + 217, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 116, + 217, + 505, + 230 + ], + "score": 1.0, + "content": "treatment effects with censored data via fully nonparametric bayesian accelerated failure time", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 229, + 230, + 240 + ], + "spans": [ + { + "bbox": [ + 116, + 229, + 230, + 240 + ], + "score": 1.0, + "content": "models. Biostatistics, 2020.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 248, + 504, + 271 + ], + "lines": [ + { + "bbox": [ + 105, + 246, + 506, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 506, + 261 + ], + "score": 1.0, + "content": "Miguel A Hernán and James M Robins. Causal inference: what if. Boca Raton: Chapman &", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 116, + 259, + 183, + 271 + ], + "spans": [ + { + "bbox": [ + 116, + 259, + 183, + 271 + ], + "score": 1.0, + "content": "Hill/CRC, 2020.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 107, + 279, + 504, + 313 + ], + "lines": [ + { + "bbox": [ + 106, + 279, + 505, + 291 + ], + "spans": [ + { + "bbox": [ + 106, + 279, + 505, + 291 + ], + "score": 1.0, + "content": "Miguel A Hernán, Stephen R Cole, Joseph Margolick, Mardge Cohen, and James M Robins. Structural", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 290, + 506, + 302 + ], + "spans": [ + { + "bbox": [ + 115, + 290, + 506, + 302 + ], + "score": 1.0, + "content": "accelerated failure time models for survival analysis in studies with time-varying treatments.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 300, + 309, + 314 + ], + "spans": [ + { + "bbox": [ + 115, + 300, + 309, + 314 + ], + "score": 1.0, + "content": "Pharmacoepidemiology and Drug Safety, 2005.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 105, + 321, + 505, + 344 + ], + "lines": [ + { + "bbox": [ + 105, + 320, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 505, + 334 + ], + "score": 1.0, + "content": "Miguel Ángel Hernán, Babette Brumback, and James M Robins. Marginal structural models to", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 332, + 506, + 345 + ], + "spans": [ + { + "bbox": [ + 116, + 332, + 506, + 345 + ], + "score": 1.0, + "content": "estimate the causal effect of zidovudine on the survival of hiv-positive men. Epidemiology, 2000.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 105, + 351, + 504, + 374 + ], + "lines": [ + { + "bbox": [ + 106, + 352, + 505, + 365 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 505, + 365 + ], + "score": 1.0, + "content": "Jennifer L Hill. Bayesian nonparametric modeling for causal inference. Journal of Computational", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 363, + 244, + 375 + ], + "spans": [ + { + "bbox": [ + 116, + 363, + 244, + 375 + ], + "score": 1.0, + "content": "and Graphical Statistics, 2011.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 107, + 382, + 504, + 405 + ], + "lines": [ + { + "bbox": [ + 105, + 381, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 505, + 396 + ], + "score": 1.0, + "content": "Daniel G Horvitz and Donovan J Thompson. A generalization of sampling without replacement from", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 393, + 402, + 405 + ], + "spans": [ + { + "bbox": [ + 115, + 393, + 402, + 405 + ], + "score": 1.0, + "content": "a finite universe. Journal of the American statistical Association, 1952.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 106, + 412, + 504, + 436 + ], + "lines": [ + { + "bbox": [ + 106, + 413, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 106, + 413, + 505, + 425 + ], + "score": 1.0, + "content": "Liangyuan Hu, Jiayi Ji, and Fan Li. Estimating heterogeneous survival treatment effect in observa-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 424, + 311, + 436 + ], + "spans": [ + { + "bbox": [ + 115, + 424, + 311, + 436 + ], + "score": 1.0, + "content": "tional data using machine learning. arXiv, 2020.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 106, + 443, + 504, + 466 + ], + "lines": [ + { + "bbox": [ + 106, + 443, + 505, + 456 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 505, + 456 + ], + "score": 1.0, + "content": "Hemant Ishwaran, Udaya B Kogalur, Eugene H Blackstone, Michael S Lauer, et al. Random survival", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 116, + 455, + 302, + 466 + ], + "spans": [ + { + "bbox": [ + 116, + 455, + 302, + 466 + ], + "score": 1.0, + "content": "forests. The annals of applied statistics, 2008.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 107, + 473, + 505, + 507 + ], + "lines": [ + { + "bbox": [ + 105, + 473, + 506, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 506, + 487 + ], + "score": 1.0, + "content": "Ashish K Jha, Catherine M DesRoches, Eric G Campbell, Karen Donelan, Sowmya R Rao, Timothy G", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 486, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 116, + 486, + 505, + 496 + ], + "score": 1.0, + "content": "Ferris, Alexandra Shields, Sara Rosenbaum, and David Blumenthal. Use of electronic health", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 496, + 378, + 509 + ], + "spans": [ + { + "bbox": [ + 115, + 496, + 378, + 509 + ], + "score": 1.0, + "content": "records in us hospitals. New England Journal of Medicine, 2009.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 515, + 505, + 538 + ], + "lines": [ + { + "bbox": [ + 105, + 514, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 505, + 528 + ], + "score": 1.0, + "content": "Edward L Kaplan and Paul Meier. Nonparametric estimation from incomplete observations. Journal", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 527, + 299, + 538 + ], + "spans": [ + { + "bbox": [ + 116, + 527, + 299, + 538 + ], + "score": 1.0, + "content": "of the American statistical association, 1958.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5 + }, + { + "type": "text", + "bbox": [ + 107, + 546, + 505, + 580 + ], + "lines": [ + { + "bbox": [ + 106, + 546, + 505, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 505, + 558 + ], + "score": 1.0, + "content": "Jared L Katzman, Uri Shaham, Alexander Cloninger, Jonathan Bates, Tingting Jiang, and Yuval", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 556, + 505, + 570 + ], + "spans": [ + { + "bbox": [ + 115, + 556, + 505, + 570 + ], + "score": 1.0, + "content": "Kluger. Deepsurv: personalized treatment recommender system using a cox proportional hazards", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 568, + 379, + 581 + ], + "spans": [ + { + "bbox": [ + 116, + 568, + 379, + 581 + ], + "score": 1.0, + "content": "deep neural network. BMC medical research methodology, 2018.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 587, + 406, + 600 + ], + "lines": [ + { + "bbox": [ + 105, + 586, + 406, + 601 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 406, + 601 + ], + "score": 1.0, + "content": "David G Kleinbaum and Mitchel Klein. Survival analysis. Springer, 2010.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 106, + 606, + 504, + 630 + ], + "lines": [ + { + "bbox": [ + 105, + 605, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 505, + 621 + ], + "score": 1.0, + "content": "Fan Li, Kari Lock Morgan, and Alan M Zaslavsky. Balancing covariates via propensity score", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 618, + 380, + 630 + ], + "spans": [ + { + "bbox": [ + 115, + 618, + 380, + 630 + ], + "score": 1.0, + "content": "weighting. Journal of the American Statistical Association, 2018.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 106, + 637, + 505, + 660 + ], + "lines": [ + { + "bbox": [ + 105, + 635, + 506, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 635, + 506, + 652 + ], + "score": 1.0, + "content": "Christopher A Longhurst, Robert A Harrington, and Nigam H Shah. A ‘green button’for using", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 648, + 373, + 661 + ], + "spans": [ + { + "bbox": [ + 115, + 648, + 373, + 661 + ], + "score": 1.0, + "content": "aggregate patient data at the point of care. Health affairs, 2014.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 109, + 667, + 504, + 702 + ], + "lines": [ + { + "bbox": [ + 106, + 667, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 506, + 681 + ], + "score": 1.0, + "content": "B Mihaylova, J Emberson, L Blackwell, A Keech, J Simes, EH Barnes, M Voysey, 3A Gray, R Collins,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 679, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 116, + 679, + 506, + 691 + ], + "score": 1.0, + "content": "and C Baigent. The effects of lowering ldl cholesterol with statin therapy in people at low risk of", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 690, + 452, + 702 + ], + "spans": [ + { + "bbox": [ + 116, + 690, + 452, + 702 + ], + "score": 1.0, + "content": "vascular disease: meta-analysis of individual data from 27 randomised trials., 2012.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 108, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 708, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 708, + 506, + 722 + ], + "score": 1.0, + "content": "Xenia Miscouridou, Adler Perotte, Noémie Elhadad, and Rajesh Ranganath. Deep survival analysis:", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 720, + 477, + 734 + ], + "spans": [ + { + "bbox": [ + 115, + 720, + 477, + 734 + ], + "score": 1.0, + "content": "Nonparametrics and missingness. In Machine Learning for Healthcare Conference, 2018.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 307, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 505, + 127 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 507, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 507, + 96 + ], + "score": 1.0, + "content": "Scott M Hammer, David A Katzenstein, Michael D Hughes, Holly Gundacker, Robert T Schooley,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 94, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 116, + 94, + 506, + 106 + ], + "score": 1.0, + "content": "Richard H Haubrich, W Keith Henry, Michael M Lederman, John P Phair, Manette Niu, et al. A", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 105, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 115, + 105, + 505, + 117 + ], + "score": 1.0, + "content": "trial comparing nucleoside monotherapy with combination therapy in hiv-infected adults with cd4", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 116, + 116, + 481, + 127 + ], + "spans": [ + { + "bbox": [ + 116, + 116, + 481, + 127 + ], + "score": 1.0, + "content": "cell counts from 200 to 500 per cubic millimeter. New England Journal of Medicine, 1996.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 81, + 507, + 127 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 135, + 504, + 157 + ], + "lines": [ + { + "bbox": [ + 105, + 135, + 505, + 148 + ], + "spans": [ + { + "bbox": [ + 105, + 135, + 505, + 148 + ], + "score": 1.0, + "content": "Frank E Harrell Jr, Kerry L Lee, Robert M Califf, David B Pryor, and Robert A Rosati. Regression", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 145, + 459, + 158 + ], + "spans": [ + { + "bbox": [ + 116, + 145, + 459, + 158 + ], + "score": 1.0, + "content": "modelling strategies for improved prognostic prediction. Statistics in medicine, 1984.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 105, + 135, + 505, + 158 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 165, + 506, + 199 + ], + "lines": [ + { + "bbox": [ + 106, + 165, + 505, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 165, + 505, + 177 + ], + "score": 1.0, + "content": "Kristiina Häyrinen, Kaija Saranto, and Pirkko Nykänen. Definition, structure, content, use and", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 176, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 115, + 176, + 506, + 189 + ], + "score": 1.0, + "content": "impacts of electronic health records: a review of the research literature. International Journal of", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 187, + 227, + 199 + ], + "spans": [ + { + "bbox": [ + 115, + 187, + 227, + 199 + ], + "score": 1.0, + "content": "Medical Informatics, 2008.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7, + "bbox_fs": [ + 106, + 165, + 506, + 199 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 206, + 506, + 240 + ], + "lines": [ + { + "bbox": [ + 106, + 207, + 505, + 219 + ], + "spans": [ + { + "bbox": [ + 106, + 207, + 505, + 219 + ], + "score": 1.0, + "content": "Nicholas C Henderson, Thomas A Louis, Gary L Rosner, and Ravi Varadhan. Individualized", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 116, + 217, + 505, + 230 + ], + "spans": [ + { + "bbox": [ + 116, + 217, + 505, + 230 + ], + "score": 1.0, + "content": "treatment effects with censored data via fully nonparametric bayesian accelerated failure time", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 229, + 230, + 240 + ], + "spans": [ + { + "bbox": [ + 116, + 229, + 230, + 240 + ], + "score": 1.0, + "content": "models. Biostatistics, 2020.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 106, + 207, + 505, + 240 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 248, + 504, + 271 + ], + "lines": [ + { + "bbox": [ + 105, + 246, + 506, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 506, + 261 + ], + "score": 1.0, + "content": "Miguel A Hernán and James M Robins. Causal inference: what if. Boca Raton: Chapman &", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 116, + 259, + 183, + 271 + ], + "spans": [ + { + "bbox": [ + 116, + 259, + 183, + 271 + ], + "score": 1.0, + "content": "Hill/CRC, 2020.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 246, + 506, + 271 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 279, + 504, + 313 + ], + "lines": [ + { + "bbox": [ + 106, + 279, + 505, + 291 + ], + "spans": [ + { + "bbox": [ + 106, + 279, + 505, + 291 + ], + "score": 1.0, + "content": "Miguel A Hernán, Stephen R Cole, Joseph Margolick, Mardge Cohen, and James M Robins. Structural", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 290, + 506, + 302 + ], + "spans": [ + { + "bbox": [ + 115, + 290, + 506, + 302 + ], + "score": 1.0, + "content": "accelerated failure time models for survival analysis in studies with time-varying treatments.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 300, + 309, + 314 + ], + "spans": [ + { + "bbox": [ + 115, + 300, + 309, + 314 + ], + "score": 1.0, + "content": "Pharmacoepidemiology and Drug Safety, 2005.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15, + "bbox_fs": [ + 106, + 279, + 506, + 314 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 321, + 505, + 344 + ], + "lines": [ + { + "bbox": [ + 105, + 320, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 505, + 334 + ], + "score": 1.0, + "content": "Miguel Ángel Hernán, Babette Brumback, and James M Robins. Marginal structural models to", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 332, + 506, + 345 + ], + "spans": [ + { + "bbox": [ + 116, + 332, + 506, + 345 + ], + "score": 1.0, + "content": "estimate the causal effect of zidovudine on the survival of hiv-positive men. Epidemiology, 2000.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 105, + 320, + 506, + 345 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 351, + 504, + 374 + ], + "lines": [ + { + "bbox": [ + 106, + 352, + 505, + 365 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 505, + 365 + ], + "score": 1.0, + "content": "Jennifer L Hill. Bayesian nonparametric modeling for causal inference. Journal of Computational", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 363, + 244, + 375 + ], + "spans": [ + { + "bbox": [ + 116, + 363, + 244, + 375 + ], + "score": 1.0, + "content": "and Graphical Statistics, 2011.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 106, + 352, + 505, + 375 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 382, + 504, + 405 + ], + "lines": [ + { + "bbox": [ + 105, + 381, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 505, + 396 + ], + "score": 1.0, + "content": "Daniel G Horvitz and Donovan J Thompson. A generalization of sampling without replacement from", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 393, + 402, + 405 + ], + "spans": [ + { + "bbox": [ + 115, + 393, + 402, + 405 + ], + "score": 1.0, + "content": "a finite universe. Journal of the American statistical Association, 1952.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 381, + 505, + 405 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 412, + 504, + 436 + ], + "lines": [ + { + "bbox": [ + 106, + 413, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 106, + 413, + 505, + 425 + ], + "score": 1.0, + "content": "Liangyuan Hu, Jiayi Ji, and Fan Li. Estimating heterogeneous survival treatment effect in observa-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 424, + 311, + 436 + ], + "spans": [ + { + "bbox": [ + 115, + 424, + 311, + 436 + ], + "score": 1.0, + "content": "tional data using machine learning. arXiv, 2020.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 106, + 413, + 505, + 436 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 443, + 504, + 466 + ], + "lines": [ + { + "bbox": [ + 106, + 443, + 505, + 456 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 505, + 456 + ], + "score": 1.0, + "content": "Hemant Ishwaran, Udaya B Kogalur, Eugene H Blackstone, Michael S Lauer, et al. Random survival", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 116, + 455, + 302, + 466 + ], + "spans": [ + { + "bbox": [ + 116, + 455, + 302, + 466 + ], + "score": 1.0, + "content": "forests. The annals of applied statistics, 2008.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 106, + 443, + 505, + 466 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 473, + 505, + 507 + ], + "lines": [ + { + "bbox": [ + 105, + 473, + 506, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 506, + 487 + ], + "score": 1.0, + "content": "Ashish K Jha, Catherine M DesRoches, Eric G Campbell, Karen Donelan, Sowmya R Rao, Timothy G", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 486, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 116, + 486, + 505, + 496 + ], + "score": 1.0, + "content": "Ferris, Alexandra Shields, Sara Rosenbaum, and David Blumenthal. Use of electronic health", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 496, + 378, + 509 + ], + "spans": [ + { + "bbox": [ + 115, + 496, + 378, + 509 + ], + "score": 1.0, + "content": "records in us hospitals. New England Journal of Medicine, 2009.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 473, + 506, + 509 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 515, + 505, + 538 + ], + "lines": [ + { + "bbox": [ + 105, + 514, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 505, + 528 + ], + "score": 1.0, + "content": "Edward L Kaplan and Paul Meier. Nonparametric estimation from incomplete observations. Journal", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 527, + 299, + 538 + ], + "spans": [ + { + "bbox": [ + 116, + 527, + 299, + 538 + ], + "score": 1.0, + "content": "of the American statistical association, 1958.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 514, + 505, + 538 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 546, + 505, + 580 + ], + "lines": [ + { + "bbox": [ + 106, + 546, + 505, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 505, + 558 + ], + "score": 1.0, + "content": "Jared L Katzman, Uri Shaham, Alexander Cloninger, Jonathan Bates, Tingting Jiang, and Yuval", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 556, + 505, + 570 + ], + "spans": [ + { + "bbox": [ + 115, + 556, + 505, + 570 + ], + "score": 1.0, + "content": "Kluger. Deepsurv: personalized treatment recommender system using a cox proportional hazards", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 568, + 379, + 581 + ], + "spans": [ + { + "bbox": [ + 116, + 568, + 379, + 581 + ], + "score": 1.0, + "content": "deep neural network. BMC medical research methodology, 2018.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33, + "bbox_fs": [ + 106, + 546, + 505, + 581 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 587, + 406, + 600 + ], + "lines": [ + { + "bbox": [ + 105, + 586, + 406, + 601 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 406, + 601 + ], + "score": 1.0, + "content": "David G Kleinbaum and Mitchel Klein. Survival analysis. Springer, 2010.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35, + "bbox_fs": [ + 105, + 586, + 406, + 601 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 606, + 504, + 630 + ], + "lines": [ + { + "bbox": [ + 105, + 605, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 505, + 621 + ], + "score": 1.0, + "content": "Fan Li, Kari Lock Morgan, and Alan M Zaslavsky. Balancing covariates via propensity score", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 618, + 380, + 630 + ], + "spans": [ + { + "bbox": [ + 115, + 618, + 380, + 630 + ], + "score": 1.0, + "content": "weighting. Journal of the American Statistical Association, 2018.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5, + "bbox_fs": [ + 105, + 605, + 505, + 630 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 637, + 505, + 660 + ], + "lines": [ + { + "bbox": [ + 105, + 635, + 506, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 635, + 506, + 652 + ], + "score": 1.0, + "content": "Christopher A Longhurst, Robert A Harrington, and Nigam H Shah. A ‘green button’for using", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 648, + 373, + 661 + ], + "spans": [ + { + "bbox": [ + 115, + 648, + 373, + 661 + ], + "score": 1.0, + "content": "aggregate patient data at the point of care. Health affairs, 2014.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 105, + 635, + 506, + 661 + ] + }, + { + "type": "text", + "bbox": [ + 109, + 667, + 504, + 702 + ], + "lines": [ + { + "bbox": [ + 106, + 667, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 506, + 681 + ], + "score": 1.0, + "content": "B Mihaylova, J Emberson, L Blackwell, A Keech, J Simes, EH Barnes, M Voysey, 3A Gray, R Collins,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 116, + 679, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 116, + 679, + 506, + 691 + ], + "score": 1.0, + "content": "and C Baigent. The effects of lowering ldl cholesterol with statin therapy in people at low risk of", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 690, + 452, + 702 + ], + "spans": [ + { + "bbox": [ + 116, + 690, + 452, + 702 + ], + "score": 1.0, + "content": "vascular disease: meta-analysis of individual data from 27 randomised trials., 2012.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41, + "bbox_fs": [ + 106, + 667, + 506, + 702 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 708, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 708, + 506, + 722 + ], + "score": 1.0, + "content": "Xenia Miscouridou, Adler Perotte, Noémie Elhadad, and Rajesh Ranganath. Deep survival analysis:", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 720, + 477, + 734 + ], + "spans": [ + { + "bbox": [ + 115, + 720, + 477, + 734 + ], + "score": 1.0, + "content": "Nonparametrics and missingness. In Machine Learning for Healthcare Conference, 2018.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5, + "bbox_fs": [ + 106, + 708, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 505, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Alfred Müller. Integral probability metrics and their generating classes of functions. Advances in", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 114, + 93, + 225, + 105 + ], + "spans": [ + { + "bbox": [ + 114, + 93, + 225, + 105 + ], + "score": 1.0, + "content": "Applied Probability, 1997.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 106, + 111, + 504, + 135 + ], + "lines": [ + { + "bbox": [ + 105, + 110, + 506, + 126 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 506, + 126 + ], + "score": 1.0, + "content": "Judea Pearl and Elias Bareinboim. External validity: From do-calculus to transportability across", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 123, + 272, + 135 + ], + "spans": [ + { + "bbox": [ + 115, + 123, + 272, + 135 + ], + "score": 1.0, + "content": "populations. Statistical Science, 2014.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 141, + 504, + 164 + ], + "lines": [ + { + "bbox": [ + 106, + 141, + 505, + 154 + ], + "spans": [ + { + "bbox": [ + 106, + 141, + 505, + 154 + ], + "score": 1.0, + "content": "Danilo Jimenez Rezende and Shakir Mohamed. Variational inference with normalizing flows. In", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 152, + 171, + 164 + ], + "spans": [ + { + "bbox": [ + 115, + 152, + 171, + 164 + ], + "score": 1.0, + "content": "ICML, 2015.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 107, + 171, + 505, + 204 + ], + "lines": [ + { + "bbox": [ + 105, + 169, + 505, + 185 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 505, + 185 + ], + "score": 1.0, + "content": "James Robins. A new approach to causal inference in mortality studies with a sustained exposure", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 181, + 507, + 196 + ], + "spans": [ + { + "bbox": [ + 115, + 181, + 507, + 196 + ], + "score": 1.0, + "content": "period—application to control of the healthy worker survivor effect. Mathematical modelling,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 114, + 191, + 142, + 205 + ], + "spans": [ + { + "bbox": [ + 114, + 191, + 142, + 205 + ], + "score": 1.0, + "content": "1986.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 106, + 211, + 505, + 234 + ], + "lines": [ + { + "bbox": [ + 106, + 211, + 505, + 224 + ], + "spans": [ + { + "bbox": [ + 106, + 211, + 505, + 224 + ], + "score": 1.0, + "content": "Paul R Rosenbaum and Donald B Rubin. The central role of the propensity score in observational", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 116, + 222, + 294, + 234 + ], + "spans": [ + { + "bbox": [ + 116, + 222, + 294, + 234 + ], + "score": 1.0, + "content": "studies for causal effects. Biometrika, 1983.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 107, + 241, + 504, + 264 + ], + "lines": [ + { + "bbox": [ + 105, + 240, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 240, + 505, + 253 + ], + "score": 1.0, + "content": "Donald B Rubin. Causal inference using potential outcomes. Journal of the American Statistical", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 114, + 252, + 192, + 263 + ], + "spans": [ + { + "bbox": [ + 114, + 252, + 192, + 263 + ], + "score": 1.0, + "content": "Association, 2005.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 106, + 270, + 505, + 293 + ], + "lines": [ + { + "bbox": [ + 105, + 268, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 505, + 284 + ], + "score": 1.0, + "content": "Michael Schemper, Samo Wakounig, and Georg Heinze. The estimation of average hazard ratios by", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 116, + 282, + 334, + 294 + ], + "spans": [ + { + "bbox": [ + 116, + 282, + 334, + 294 + ], + "score": 1.0, + "content": "weighted cox regression. Statistics in medicine, 2009.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 105, + 300, + 505, + 323 + ], + "lines": [ + { + "bbox": [ + 106, + 299, + 506, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 299, + 506, + 312 + ], + "score": 1.0, + "content": "Uri Shalit, Fredrik D Johansson, and David Sontag. Estimating individual treatment effect: general-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 116, + 311, + 308, + 323 + ], + "spans": [ + { + "bbox": [ + 116, + 311, + 308, + 323 + ], + "score": 1.0, + "content": "ization bounds and algorithms. In ICML, 2017.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 106, + 329, + 506, + 374 + ], + "lines": [ + { + "bbox": [ + 105, + 329, + 506, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 506, + 343 + ], + "score": 1.0, + "content": "Jincheng Shen, Lu Wang, Stephanie Daignault, Daniel E Spratt, Todd M Morgan, and Jeremy MG", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 340, + 506, + 353 + ], + "spans": [ + { + "bbox": [ + 115, + 340, + 506, + 353 + ], + "score": 1.0, + "content": "Taylor. Estimating the optimal personalized treatment strategy based on selected variables to", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 116, + 353, + 506, + 364 + ], + "spans": [ + { + "bbox": [ + 116, + 353, + 506, + 364 + ], + "score": 1.0, + "content": "prolong survival via random survival forest with weighted bootstrap. Journal of biopharmaceutical", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 363, + 182, + 374 + ], + "spans": [ + { + "bbox": [ + 115, + 363, + 182, + 374 + ], + "score": 1.0, + "content": "statistics, 2018.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 105, + 380, + 505, + 404 + ], + "lines": [ + { + "bbox": [ + 106, + 381, + 505, + 394 + ], + "spans": [ + { + "bbox": [ + 106, + 381, + 505, + 394 + ], + "score": 1.0, + "content": "Rodney A Sparapani, Brent R Logan, Robert E McCulloch, and Purushottam W Laud. Nonparametric", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 392, + 488, + 404 + ], + "spans": [ + { + "bbox": [ + 115, + 392, + 488, + 404 + ], + "score": 1.0, + "content": "survival analysis using bayesian additive regression trees (bart). Statistics in medicine, 2016.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 107, + 410, + 505, + 444 + ], + "lines": [ + { + "bbox": [ + 105, + 410, + 507, + 424 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 507, + 424 + ], + "score": 1.0, + "content": "Bharath K Sriperumbudur, Kenji Fukumizu, Arthur Gretton, Bernhard Schölkopf, Gert RG Lanckriet,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 422, + 507, + 434 + ], + "spans": [ + { + "bbox": [ + 115, + 422, + 507, + 434 + ], + "score": 1.0, + "content": "et al. On the empirical estimation of integral probability metrics. Electronic Journal of Statistics,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 114, + 431, + 143, + 446 + ], + "spans": [ + { + "bbox": [ + 114, + 431, + 143, + 446 + ], + "score": 1.0, + "content": "2012.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 106, + 451, + 506, + 485 + ], + "lines": [ + { + "bbox": [ + 105, + 451, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 505, + 464 + ], + "score": 1.0, + "content": "Ludovic Trinquart, Justine Jacot, Sarah C Conner, and Raphaël Porcher. Comparison of treatment", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 461, + 506, + 476 + ], + "spans": [ + { + "bbox": [ + 115, + 461, + 506, + 476 + ], + "score": 1.0, + "content": "effects measured by the hazard ratio and by the ratio of restricted mean survival times in oncology", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 473, + 382, + 486 + ], + "spans": [ + { + "bbox": [ + 115, + 473, + 382, + 486 + ], + "score": 1.0, + "content": "randomized controlled trials. Journal of Clinical Oncology, 2016.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 106, + 491, + 504, + 514 + ], + "lines": [ + { + "bbox": [ + 106, + 491, + 506, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 491, + 506, + 505 + ], + "score": 1.0, + "content": "Anastasios Tsiatis. Semiparametric theory and missing data. Springer Science & Business Media,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 501, + 144, + 515 + ], + "spans": [ + { + "bbox": [ + 115, + 501, + 144, + 515 + ], + "score": 1.0, + "content": "2007.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 107, + 520, + 504, + 545 + ], + "lines": [ + { + "bbox": [ + 105, + 521, + 506, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 506, + 534 + ], + "score": 1.0, + "content": "Mark J van der Laan and James M Robins. Unified approach for causal inference and censored data.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 532, + 451, + 545 + ], + "spans": [ + { + "bbox": [ + 115, + 532, + 451, + 545 + ], + "score": 1.0, + "content": "In Unified Methods for Censored Longitudinal Data and Causality. Springer, 2003.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 106, + 550, + 505, + 574 + ], + "lines": [ + { + "bbox": [ + 106, + 550, + 506, + 563 + ], + "spans": [ + { + "bbox": [ + 106, + 550, + 506, + 563 + ], + "score": 1.0, + "content": "Mark J Van der Laan and Sherri Rose. Targeted learning: causal inference for observational and", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 562, + 369, + 574 + ], + "spans": [ + { + "bbox": [ + 115, + 562, + 369, + 574 + ], + "score": 1.0, + "content": "experimental data. Springer Science & Business Media, 2011.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 105, + 579, + 475, + 592 + ], + "lines": [ + { + "bbox": [ + 106, + 579, + 477, + 594 + ], + "spans": [ + { + "bbox": [ + 106, + 579, + 477, + 594 + ], + "score": 1.0, + "content": "Cédric Villani. Optimal transport: old and new. Springer Science & Business Media, 2008.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 105, + 599, + 504, + 622 + ], + "lines": [ + { + "bbox": [ + 105, + 597, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 506, + 613 + ], + "score": 1.0, + "content": "Stefan Wager and Susan Athey. Estimation and inference of heterogeneous treatment effects using", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 610, + 400, + 622 + ], + "spans": [ + { + "bbox": [ + 115, + 610, + 400, + 622 + ], + "score": 1.0, + "content": "random forests. Journal of the American Statistical Association, 2018.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 105, + 628, + 504, + 651 + ], + "lines": [ + { + "bbox": [ + 106, + 628, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 628, + 505, + 641 + ], + "score": 1.0, + "content": "Lee-Jen Wei. The accelerated failure time model: a useful alternative to the cox regression model in", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 640, + 303, + 651 + ], + "spans": [ + { + "bbox": [ + 116, + 640, + 303, + 651 + ], + "score": 1.0, + "content": "survival analysis. Statistics in medicine, 1992.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 107, + 657, + 505, + 692 + ], + "lines": [ + { + "bbox": [ + 106, + 658, + 505, + 670 + ], + "spans": [ + { + "bbox": [ + 106, + 658, + 505, + 670 + ], + "score": 1.0, + "content": "Timothy J Wilt, Hanna E Bloomfield, Roderick MacDonald, David Nelson, Indulis Rutks, Michael", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 668, + 506, + 682 + ], + "spans": [ + { + "bbox": [ + 115, + 668, + 506, + 682 + ], + "score": 1.0, + "content": "Ho, Gregory Larsen, Anthony McCall, Sandra Pineros, and Anne Sales. Effectiveness of statin", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 680, + 448, + 693 + ], + "spans": [ + { + "bbox": [ + 116, + 680, + 448, + 693 + ], + "score": 1.0, + "content": "therapy in adults with coronary heart disease. Archives of internal medicine, 2004.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 108, + 698, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "Salim Yusuf, Jackie Bosch, Gilles Dagenais, Jun Zhu, Denis Xavier, Lisheng Liu, Prem Pais, Patricio", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 116, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 116, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "López-Jaramillo, Lawrence A Leiter, Antonio Dans, et al. Cholesterol lowering in intermediate-risk", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 720, + 445, + 733 + ], + "spans": [ + { + "bbox": [ + 115, + 720, + 445, + 733 + ], + "score": 1.0, + "content": "persons without cardiovascular disease. New England Journal of Medicine, 2016.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 307, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 505, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Alfred Müller. Integral probability metrics and their generating classes of functions. Advances in", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 114, + 93, + 225, + 105 + ], + "spans": [ + { + "bbox": [ + 114, + 93, + 225, + 105 + ], + "score": 1.0, + "content": "Applied Probability, 1997.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 106, + 82, + 505, + 105 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 111, + 504, + 135 + ], + "lines": [ + { + "bbox": [ + 105, + 110, + 506, + 126 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 506, + 126 + ], + "score": 1.0, + "content": "Judea Pearl and Elias Bareinboim. External validity: From do-calculus to transportability across", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 123, + 272, + 135 + ], + "spans": [ + { + "bbox": [ + 115, + 123, + 272, + 135 + ], + "score": 1.0, + "content": "populations. Statistical Science, 2014.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5, + "bbox_fs": [ + 105, + 110, + 506, + 135 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 141, + 504, + 164 + ], + "lines": [ + { + "bbox": [ + 106, + 141, + 505, + 154 + ], + "spans": [ + { + "bbox": [ + 106, + 141, + 505, + 154 + ], + "score": 1.0, + "content": "Danilo Jimenez Rezende and Shakir Mohamed. Variational inference with normalizing flows. In", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 152, + 171, + 164 + ], + "spans": [ + { + "bbox": [ + 115, + 152, + 171, + 164 + ], + "score": 1.0, + "content": "ICML, 2015.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 106, + 141, + 505, + 164 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 171, + 505, + 204 + ], + "lines": [ + { + "bbox": [ + 105, + 169, + 505, + 185 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 505, + 185 + ], + "score": 1.0, + "content": "James Robins. A new approach to causal inference in mortality studies with a sustained exposure", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 181, + 507, + 196 + ], + "spans": [ + { + "bbox": [ + 115, + 181, + 507, + 196 + ], + "score": 1.0, + "content": "period—application to control of the healthy worker survivor effect. Mathematical modelling,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 114, + 191, + 142, + 205 + ], + "spans": [ + { + "bbox": [ + 114, + 191, + 142, + 205 + ], + "score": 1.0, + "content": "1986.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7, + "bbox_fs": [ + 105, + 169, + 507, + 205 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 211, + 505, + 234 + ], + "lines": [ + { + "bbox": [ + 106, + 211, + 505, + 224 + ], + "spans": [ + { + "bbox": [ + 106, + 211, + 505, + 224 + ], + "score": 1.0, + "content": "Paul R Rosenbaum and Donald B Rubin. The central role of the propensity score in observational", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 116, + 222, + 294, + 234 + ], + "spans": [ + { + "bbox": [ + 116, + 222, + 294, + 234 + ], + "score": 1.0, + "content": "studies for causal effects. Biometrika, 1983.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5, + "bbox_fs": [ + 106, + 211, + 505, + 234 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 241, + 504, + 264 + ], + "lines": [ + { + "bbox": [ + 105, + 240, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 240, + 505, + 253 + ], + "score": 1.0, + "content": "Donald B Rubin. Causal inference using potential outcomes. Journal of the American Statistical", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 114, + 252, + 192, + 263 + ], + "spans": [ + { + "bbox": [ + 114, + 252, + 192, + 263 + ], + "score": 1.0, + "content": "Association, 2005.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 240, + 505, + 263 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 270, + 505, + 293 + ], + "lines": [ + { + "bbox": [ + 105, + 268, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 505, + 284 + ], + "score": 1.0, + "content": "Michael Schemper, Samo Wakounig, and Georg Heinze. The estimation of average hazard ratios by", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 116, + 282, + 334, + 294 + ], + "spans": [ + { + "bbox": [ + 116, + 282, + 334, + 294 + ], + "score": 1.0, + "content": "weighted cox regression. Statistics in medicine, 2009.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5, + "bbox_fs": [ + 105, + 268, + 505, + 294 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 300, + 505, + 323 + ], + "lines": [ + { + "bbox": [ + 106, + 299, + 506, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 299, + 506, + 312 + ], + "score": 1.0, + "content": "Uri Shalit, Fredrik D Johansson, and David Sontag. Estimating individual treatment effect: general-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 116, + 311, + 308, + 323 + ], + "spans": [ + { + "bbox": [ + 116, + 311, + 308, + 323 + ], + "score": 1.0, + "content": "ization bounds and algorithms. In ICML, 2017.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 106, + 299, + 506, + 323 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 329, + 506, + 374 + ], + "lines": [ + { + "bbox": [ + 105, + 329, + 506, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 506, + 343 + ], + "score": 1.0, + "content": "Jincheng Shen, Lu Wang, Stephanie Daignault, Daniel E Spratt, Todd M Morgan, and Jeremy MG", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 340, + 506, + 353 + ], + "spans": [ + { + "bbox": [ + 115, + 340, + 506, + 353 + ], + "score": 1.0, + "content": "Taylor. Estimating the optimal personalized treatment strategy based on selected variables to", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 116, + 353, + 506, + 364 + ], + "spans": [ + { + "bbox": [ + 116, + 353, + 506, + 364 + ], + "score": 1.0, + "content": "prolong survival via random survival forest with weighted bootstrap. Journal of biopharmaceutical", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 363, + 182, + 374 + ], + "spans": [ + { + "bbox": [ + 115, + 363, + 182, + 374 + ], + "score": 1.0, + "content": "statistics, 2018.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 329, + 506, + 374 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 380, + 505, + 404 + ], + "lines": [ + { + "bbox": [ + 106, + 381, + 505, + 394 + ], + "spans": [ + { + "bbox": [ + 106, + 381, + 505, + 394 + ], + "score": 1.0, + "content": "Rodney A Sparapani, Brent R Logan, Robert E McCulloch, and Purushottam W Laud. Nonparametric", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 392, + 488, + 404 + ], + "spans": [ + { + "bbox": [ + 115, + 392, + 488, + 404 + ], + "score": 1.0, + "content": "survival analysis using bayesian additive regression trees (bart). Statistics in medicine, 2016.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5, + "bbox_fs": [ + 106, + 381, + 505, + 404 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 410, + 505, + 444 + ], + "lines": [ + { + "bbox": [ + 105, + 410, + 507, + 424 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 507, + 424 + ], + "score": 1.0, + "content": "Bharath K Sriperumbudur, Kenji Fukumizu, Arthur Gretton, Bernhard Schölkopf, Gert RG Lanckriet,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 422, + 507, + 434 + ], + "spans": [ + { + "bbox": [ + 115, + 422, + 507, + 434 + ], + "score": 1.0, + "content": "et al. On the empirical estimation of integral probability metrics. Electronic Journal of Statistics,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 114, + 431, + 143, + 446 + ], + "spans": [ + { + "bbox": [ + 114, + 431, + 143, + 446 + ], + "score": 1.0, + "content": "2012.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 410, + 507, + 446 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 451, + 506, + 485 + ], + "lines": [ + { + "bbox": [ + 105, + 451, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 505, + 464 + ], + "score": 1.0, + "content": "Ludovic Trinquart, Justine Jacot, Sarah C Conner, and Raphaël Porcher. Comparison of treatment", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 461, + 506, + 476 + ], + "spans": [ + { + "bbox": [ + 115, + 461, + 506, + 476 + ], + "score": 1.0, + "content": "effects measured by the hazard ratio and by the ratio of restricted mean survival times in oncology", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 473, + 382, + 486 + ], + "spans": [ + { + "bbox": [ + 115, + 473, + 382, + 486 + ], + "score": 1.0, + "content": "randomized controlled trials. Journal of Clinical Oncology, 2016.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27, + "bbox_fs": [ + 105, + 451, + 506, + 486 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 491, + 504, + 514 + ], + "lines": [ + { + "bbox": [ + 106, + 491, + 506, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 491, + 506, + 505 + ], + "score": 1.0, + "content": "Anastasios Tsiatis. Semiparametric theory and missing data. Springer Science & Business Media,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 501, + 144, + 515 + ], + "spans": [ + { + "bbox": [ + 115, + 501, + 144, + 515 + ], + "score": 1.0, + "content": "2007.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5, + "bbox_fs": [ + 106, + 491, + 506, + 515 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 520, + 504, + 545 + ], + "lines": [ + { + "bbox": [ + 105, + 521, + 506, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 506, + 534 + ], + "score": 1.0, + "content": "Mark J van der Laan and James M Robins. Unified approach for causal inference and censored data.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 532, + 451, + 545 + ], + "spans": [ + { + "bbox": [ + 115, + 532, + 451, + 545 + ], + "score": 1.0, + "content": "In Unified Methods for Censored Longitudinal Data and Causality. Springer, 2003.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 521, + 506, + 545 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 550, + 505, + 574 + ], + "lines": [ + { + "bbox": [ + 106, + 550, + 506, + 563 + ], + "spans": [ + { + "bbox": [ + 106, + 550, + 506, + 563 + ], + "score": 1.0, + "content": "Mark J Van der Laan and Sherri Rose. Targeted learning: causal inference for observational and", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 562, + 369, + 574 + ], + "spans": [ + { + "bbox": [ + 115, + 562, + 369, + 574 + ], + "score": 1.0, + "content": "experimental data. Springer Science & Business Media, 2011.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5, + "bbox_fs": [ + 106, + 550, + 506, + 574 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 579, + 475, + 592 + ], + "lines": [ + { + "bbox": [ + 106, + 579, + 477, + 594 + ], + "spans": [ + { + "bbox": [ + 106, + 579, + 477, + 594 + ], + "score": 1.0, + "content": "Cédric Villani. Optimal transport: old and new. Springer Science & Business Media, 2008.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35, + "bbox_fs": [ + 106, + 579, + 477, + 594 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 599, + 504, + 622 + ], + "lines": [ + { + "bbox": [ + 105, + 597, + 506, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 506, + 613 + ], + "score": 1.0, + "content": "Stefan Wager and Susan Athey. Estimation and inference of heterogeneous treatment effects using", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 610, + 400, + 622 + ], + "spans": [ + { + "bbox": [ + 115, + 610, + 400, + 622 + ], + "score": 1.0, + "content": "random forests. Journal of the American Statistical Association, 2018.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5, + "bbox_fs": [ + 105, + 597, + 506, + 622 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 628, + 504, + 651 + ], + "lines": [ + { + "bbox": [ + 106, + 628, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 628, + 505, + 641 + ], + "score": 1.0, + "content": "Lee-Jen Wei. The accelerated failure time model: a useful alternative to the cox regression model in", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 640, + 303, + 651 + ], + "spans": [ + { + "bbox": [ + 116, + 640, + 303, + 651 + ], + "score": 1.0, + "content": "survival analysis. Statistics in medicine, 1992.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 106, + 628, + 505, + 651 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 657, + 505, + 692 + ], + "lines": [ + { + "bbox": [ + 106, + 658, + 505, + 670 + ], + "spans": [ + { + "bbox": [ + 106, + 658, + 505, + 670 + ], + "score": 1.0, + "content": "Timothy J Wilt, Hanna E Bloomfield, Roderick MacDonald, David Nelson, Indulis Rutks, Michael", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 668, + 506, + 682 + ], + "spans": [ + { + "bbox": [ + 115, + 668, + 506, + 682 + ], + "score": 1.0, + "content": "Ho, Gregory Larsen, Anthony McCall, Sandra Pineros, and Anne Sales. Effectiveness of statin", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 680, + 448, + 693 + ], + "spans": [ + { + "bbox": [ + 116, + 680, + 448, + 693 + ], + "score": 1.0, + "content": "therapy in adults with coronary heart disease. Archives of internal medicine, 2004.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41, + "bbox_fs": [ + 106, + 658, + 506, + 693 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 698, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "Salim Yusuf, Jackie Bosch, Gilles Dagenais, Jun Zhu, Denis Xavier, Lisheng Liu, Prem Pais, Patricio", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 116, + 710, + 505, + 721 + ], + "spans": [ + { + "bbox": [ + 116, + 710, + 505, + 721 + ], + "score": 1.0, + "content": "López-Jaramillo, Lawrence A Leiter, Antonio Dans, et al. Cholesterol lowering in intermediate-risk", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 720, + 445, + 733 + ], + "spans": [ + { + "bbox": [ + 115, + 720, + 445, + 733 + ], + "score": 1.0, + "content": "persons without cardiovascular disease. New England Journal of Medicine, 2016.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44, + "bbox_fs": [ + 105, + 698, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Yao Zhang, Alexis Bellot, and Mihaela van der Schaar. Learning overlapping representations for the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 94, + 379, + 105 + ], + "spans": [ + { + "bbox": [ + 116, + 94, + 379, + 105 + ], + "score": 1.0, + "content": "estimation of individualized treatment effects. In AISTATS, 2020.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 109, + 112, + 504, + 146 + ], + "lines": [ + { + "bbox": [ + 108, + 112, + 505, + 124 + ], + "spans": [ + { + "bbox": [ + 108, + 112, + 505, + 124 + ], + "score": 1.0, + "content": "Lihui Zhao, Lu Tian, Hajime Uno, Scott D Solomon, Marc A Pfeffer, Jerald S Schindler, and Lee Jen", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 123, + 505, + 136 + ], + "spans": [ + { + "bbox": [ + 115, + 123, + 505, + 136 + ], + "score": 1.0, + "content": "Wei. Utilizing the integrated difference of two survival functions to quantify the treatment contrast", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 134, + 484, + 147 + ], + "spans": [ + { + "bbox": [ + 115, + 134, + 484, + 147 + ], + "score": 1.0, + "content": "for designing, monitoring, and analyzing a comparative clinical study. Clinical trials, 2012.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 307, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 308, + 38 + ], + "score": 1.0, + "content": "Under review as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Yao Zhang, Alexis Bellot, and Mihaela van der Schaar. Learning overlapping representations for the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 94, + 379, + 105 + ], + "spans": [ + { + "bbox": [ + 116, + 94, + 379, + 105 + ], + "score": 1.0, + "content": "estimation of individualized treatment effects. In AISTATS, 2020.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 106, + 82, + 505, + 105 + ] + }, + { + "type": "text", + "bbox": [ + 109, + 112, + 504, + 146 + ], + "lines": [ + { + "bbox": [ + 108, + 112, + 505, + 124 + ], + "spans": [ + { + "bbox": [ + 108, + 112, + 505, + 124 + ], + "score": 1.0, + "content": "Lihui Zhao, Lu Tian, Hajime Uno, Scott D Solomon, Marc A Pfeffer, Jerald S Schindler, and Lee Jen", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 123, + 505, + 136 + ], + "spans": [ + { + "bbox": [ + 115, + 123, + 505, + 136 + ], + "score": 1.0, + "content": "Wei. Utilizing the integrated difference of two survival functions to quantify the treatment contrast", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 134, + 484, + 147 + ], + "spans": [ + { + "bbox": [ + 115, + 134, + 484, + 147 + ], + "score": 1.0, + "content": "for designing, monitoring, and analyzing a comparative clinical study. Clinical trials, 2012.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3, + "bbox_fs": [ + 108, + 112, + 505, + 147 + ] + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/train/3ZeGLibhFo0/3ZeGLibhFo0_model.json b/parse/train/3ZeGLibhFo0/3ZeGLibhFo0_model.json new file mode 100644 index 0000000000000000000000000000000000000000..641cd3a5e6b18f47e508608e6584d79688fc8de4 --- /dev/null +++ b/parse/train/3ZeGLibhFo0/3ZeGLibhFo0_model.json @@ -0,0 +1,21370 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 398, + 589, + 1304, + 589, + 1304, + 1016, + 398, + 1016 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 298, + 1438, + 1404, + 1438, + 1404, + 1803, + 298, + 1803 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1146, + 1404, + 1146, + 1404, + 1423, + 298, + 1423 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1820, + 1402, + 1820, + 1402, + 2034, + 298, + 2034 + ], + "score": 0.978 + }, + { + "category_id": 0, + "poly": [ + 300, + 219, + 1400, + 219, + 1400, + 324, + 300, + 324 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 313, + 377, + 680, + 377, + 680, + 438, + 313, + 438 + ], + "score": 0.934 + }, + { + "category_id": 0, + "poly": [ + 303, + 1078, + 573, + 1078, + 573, + 1112, + 303, + 1112 + ], + "score": 0.897 + }, + { + "category_id": 0, + "poly": [ + 773, + 518, + 927, + 518, + 927, + 553, + 773, + 553 + ], + "score": 0.866 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 856, + 2088, + 856, + 2112, + 841, + 2112 + ], + "score": 0.727 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 854, + 75, + 854, + 104, + 298, + 104 + ], + "score": 0.697 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 854, + 75, + 854, + 105, + 298, + 105 + ], + "score": 0.195 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 218.0, + 1403.0, + 218.0, + 1403.0, + 273.0, + 295.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 275.0, + 1058.0, + 275.0, + 1058.0, + 329.0, + 294.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1074.0, + 579.0, + 1074.0, + 579.0, + 1119.0, + 294.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 515.0, + 934.0, + 515.0, + 934.0, + 558.0, + 768.0, + 558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2088.0, + 857.0, + 2088.0, + 857.0, + 2116.0, + 840.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 857.0, + 72.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 589.0, + 1307.0, + 589.0, + 1307.0, + 623.0, + 394.0, + 623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 619.0, + 1306.0, + 619.0, + 1306.0, + 654.0, + 394.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 649.0, + 1306.0, + 649.0, + 1306.0, + 685.0, + 392.0, + 685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 680.0, + 1308.0, + 680.0, + 1308.0, + 715.0, + 393.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 711.0, + 1305.0, + 711.0, + 1305.0, + 746.0, + 394.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 742.0, + 1305.0, + 742.0, + 1305.0, + 776.0, + 393.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 772.0, + 1306.0, + 772.0, + 1306.0, + 808.0, + 393.0, + 808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 803.0, + 1308.0, + 803.0, + 1308.0, + 837.0, + 393.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 833.0, + 1306.0, + 833.0, + 1306.0, + 868.0, + 393.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 864.0, + 1305.0, + 864.0, + 1305.0, + 896.0, + 392.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 893.0, + 1307.0, + 893.0, + 1307.0, + 928.0, + 394.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 924.0, + 1305.0, + 924.0, + 1305.0, + 959.0, + 394.0, + 959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 956.0, + 1305.0, + 956.0, + 1305.0, + 987.0, + 394.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 987.0, + 707.0, + 987.0, + 707.0, + 1018.0, + 395.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1436.0, + 1405.0, + 1436.0, + 1405.0, + 1474.0, + 293.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1469.0, + 1405.0, + 1469.0, + 1405.0, + 1503.0, + 296.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1498.0, + 1405.0, + 1498.0, + 1405.0, + 1535.0, + 293.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1528.0, + 1406.0, + 1528.0, + 1406.0, + 1565.0, + 292.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1562.0, + 1404.0, + 1562.0, + 1404.0, + 1593.0, + 296.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1590.0, + 1406.0, + 1590.0, + 1406.0, + 1624.0, + 293.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1623.0, + 1405.0, + 1623.0, + 1405.0, + 1653.0, + 296.0, + 1653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1651.0, + 1405.0, + 1651.0, + 1405.0, + 1685.0, + 294.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1681.0, + 1406.0, + 1681.0, + 1406.0, + 1715.0, + 294.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1713.0, + 1405.0, + 1713.0, + 1405.0, + 1745.0, + 293.0, + 1745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1742.0, + 1404.0, + 1742.0, + 1404.0, + 1775.0, + 294.0, + 1775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1774.0, + 1003.0, + 1774.0, + 1003.0, + 1805.0, + 296.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1148.0, + 1405.0, + 1148.0, + 1405.0, + 1181.0, + 297.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1180.0, + 1406.0, + 1180.0, + 1406.0, + 1213.0, + 296.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1210.0, + 1405.0, + 1210.0, + 1405.0, + 1243.0, + 294.0, + 1243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1239.0, + 1406.0, + 1239.0, + 1406.0, + 1275.0, + 293.0, + 1275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1267.0, + 1405.0, + 1267.0, + 1405.0, + 1308.0, + 292.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1298.0, + 1406.0, + 1298.0, + 1406.0, + 1336.0, + 293.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1328.0, + 1406.0, + 1328.0, + 1406.0, + 1367.0, + 292.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1360.0, + 1404.0, + 1360.0, + 1404.0, + 1397.0, + 294.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1390.0, + 789.0, + 1390.0, + 789.0, + 1429.0, + 293.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1818.0, + 1404.0, + 1818.0, + 1404.0, + 1855.0, + 293.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1847.0, + 1406.0, + 1847.0, + 1406.0, + 1888.0, + 292.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1880.0, + 1406.0, + 1880.0, + 1406.0, + 1915.0, + 294.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1910.0, + 1406.0, + 1910.0, + 1406.0, + 1945.0, + 294.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1940.0, + 1406.0, + 1940.0, + 1406.0, + 1976.0, + 293.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1970.0, + 1408.0, + 1970.0, + 1408.0, + 2008.0, + 294.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2001.0, + 1406.0, + 2001.0, + 1406.0, + 2036.0, + 293.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 315.0, + 377.0, + 560.0, + 377.0, + 560.0, + 410.0, + 315.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 407.0, + 682.0, + 407.0, + 682.0, + 440.0, + 311.0, + 440.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 400, + 1404, + 400, + 1404, + 826, + 298, + 826 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 843, + 1405, + 843, + 1405, + 1360, + 298, + 1360 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 300, + 1514, + 1406, + 1514, + 1406, + 1881, + 300, + 1881 + ], + "score": 0.969 + }, + { + "category_id": 1, + "poly": [ + 298, + 229, + 1403, + 229, + 1403, + 383, + 298, + 383 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 298, + 1376, + 1404, + 1376, + 1404, + 1499, + 298, + 1499 + ], + "score": 0.965 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 852, + 76, + 852, + 104, + 300, + 104 + ], + "score": 0.858 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.704 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.191 + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 856.0, + 73.0, + 856.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 839.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 397.0, + 1409.0, + 397.0, + 1409.0, + 435.0, + 295.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 431.0, + 1405.0, + 431.0, + 1405.0, + 463.0, + 296.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 457.0, + 1408.0, + 457.0, + 1408.0, + 497.0, + 292.0, + 497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 489.0, + 1408.0, + 489.0, + 1408.0, + 526.0, + 293.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 521.0, + 1406.0, + 521.0, + 1406.0, + 557.0, + 291.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 550.0, + 1406.0, + 550.0, + 1406.0, + 588.0, + 292.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 580.0, + 1406.0, + 580.0, + 1406.0, + 617.0, + 293.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 608.0, + 1406.0, + 608.0, + 1406.0, + 651.0, + 292.0, + 651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 638.0, + 1406.0, + 638.0, + 1406.0, + 681.0, + 292.0, + 681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 672.0, + 1406.0, + 672.0, + 1406.0, + 709.0, + 293.0, + 709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 703.0, + 1405.0, + 703.0, + 1405.0, + 738.0, + 293.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 732.0, + 1404.0, + 732.0, + 1404.0, + 770.0, + 293.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 763.0, + 1404.0, + 763.0, + 1404.0, + 799.0, + 294.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 796.0, + 1406.0, + 796.0, + 1406.0, + 828.0, + 296.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 842.0, + 1405.0, + 842.0, + 1405.0, + 877.0, + 295.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 871.0, + 1405.0, + 871.0, + 1405.0, + 907.0, + 295.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 904.0, + 1405.0, + 904.0, + 1405.0, + 938.0, + 293.0, + 938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 934.0, + 1406.0, + 934.0, + 1406.0, + 967.0, + 295.0, + 967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 961.0, + 1407.0, + 961.0, + 1407.0, + 999.0, + 293.0, + 999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 992.0, + 1406.0, + 992.0, + 1406.0, + 1030.0, + 293.0, + 1030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1026.0, + 1403.0, + 1026.0, + 1403.0, + 1058.0, + 296.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1054.0, + 1406.0, + 1054.0, + 1406.0, + 1090.0, + 295.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1085.0, + 1407.0, + 1085.0, + 1407.0, + 1121.0, + 293.0, + 1121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1116.0, + 1407.0, + 1116.0, + 1407.0, + 1152.0, + 295.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1148.0, + 1403.0, + 1148.0, + 1403.0, + 1180.0, + 296.0, + 1180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1175.0, + 1406.0, + 1175.0, + 1406.0, + 1213.0, + 293.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1207.0, + 1408.0, + 1207.0, + 1408.0, + 1244.0, + 293.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1238.0, + 1405.0, + 1238.0, + 1405.0, + 1272.0, + 293.0, + 1272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1270.0, + 1403.0, + 1270.0, + 1403.0, + 1302.0, + 296.0, + 1302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1298.0, + 1406.0, + 1298.0, + 1406.0, + 1334.0, + 295.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1332.0, + 510.0, + 1332.0, + 510.0, + 1362.0, + 294.0, + 1362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1511.0, + 1406.0, + 1511.0, + 1406.0, + 1552.0, + 297.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1547.0, + 1404.0, + 1547.0, + 1404.0, + 1581.0, + 320.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1575.0, + 1410.0, + 1575.0, + 1410.0, + 1612.0, + 318.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1603.0, + 1410.0, + 1603.0, + 1410.0, + 1643.0, + 298.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1638.0, + 1404.0, + 1638.0, + 1404.0, + 1672.0, + 322.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1670.0, + 842.0, + 1670.0, + 842.0, + 1700.0, + 322.0, + 1700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1696.0, + 1405.0, + 1696.0, + 1405.0, + 1733.0, + 298.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1728.0, + 1406.0, + 1728.0, + 1406.0, + 1762.0, + 322.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1757.0, + 1406.0, + 1757.0, + 1406.0, + 1794.0, + 320.0, + 1794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1791.0, + 591.0, + 1791.0, + 591.0, + 1820.0, + 319.0, + 1820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 1815.0, + 1406.0, + 1815.0, + 1406.0, + 1858.0, + 300.0, + 1858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1850.0, + 1158.0, + 1850.0, + 1158.0, + 1885.0, + 320.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 232.0, + 1403.0, + 232.0, + 1403.0, + 261.0, + 297.0, + 261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 262.0, + 1404.0, + 262.0, + 1404.0, + 295.0, + 296.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 286.0, + 1407.0, + 286.0, + 1407.0, + 331.0, + 292.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 321.0, + 1404.0, + 321.0, + 1404.0, + 356.0, + 294.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 351.0, + 1139.0, + 351.0, + 1139.0, + 388.0, + 294.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1376.0, + 1406.0, + 1376.0, + 1406.0, + 1412.0, + 293.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1407.0, + 1405.0, + 1407.0, + 1405.0, + 1440.0, + 293.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1439.0, + 1404.0, + 1439.0, + 1404.0, + 1471.0, + 294.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1467.0, + 1382.0, + 1467.0, + 1382.0, + 1504.0, + 292.0, + 1504.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 747, + 1407, + 747, + 1407, + 1025, + 297, + 1025 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 296, + 1441, + 1406, + 1441, + 1406, + 1726, + 296, + 1726 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 296, + 1206, + 1405, + 1206, + 1405, + 1424, + 296, + 1424 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 1038, + 1405, + 1038, + 1405, + 1194, + 297, + 1194 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1910, + 1405, + 1910, + 1405, + 2035, + 298, + 2035 + ], + "score": 0.974 + }, + { + "category_id": 3, + "poly": [ + 343, + 203, + 1342, + 203, + 1342, + 457, + 343, + 457 + ], + "score": 0.967 + }, + { + "category_id": 4, + "poly": [ + 295, + 467, + 1407, + 467, + 1407, + 652, + 295, + 652 + ], + "score": 0.964 + }, + { + "category_id": 1, + "poly": [ + 302, + 1831, + 1400, + 1831, + 1400, + 1897, + 302, + 1897 + ], + "score": 0.955 + }, + { + "category_id": 8, + "poly": [ + 447, + 1736, + 1250, + 1736, + 1250, + 1812, + 447, + 1812 + ], + "score": 0.95 + }, + { + "category_id": 0, + "poly": [ + 302, + 678, + 712, + 678, + 712, + 714, + 302, + 714 + ], + "score": 0.91 + }, + { + "category_id": 9, + "poly": [ + 1367, + 1760, + 1400, + 1760, + 1400, + 1789, + 1367, + 1789 + ], + "score": 0.876 + }, + { + "category_id": 2, + "poly": [ + 298, + 76, + 854, + 76, + 854, + 105, + 298, + 105 + ], + "score": 0.852 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.633 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.491 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 852, + 76, + 852, + 104, + 299, + 104 + ], + "score": 0.126 + }, + { + "category_id": 14, + "poly": [ + 448, + 1736, + 1253, + 1736, + 1253, + 1812, + 448, + 1812 + ], + "score": 0.95, + "latex": "\\lambda ( t | x ) = \\operatorname* { l i m } _ { d t 0 } { \\frac { P ( t < T < t + d t | X = x ) } { P ( T > t | X = x ) d t } } = - { \\frac { d \\log S ( t | x ) } { d t } } = { \\frac { f ( t | x ) } { S ( t | x ) } } ." + }, + { + "category_id": 13, + "poly": [ + 324, + 1162, + 520, + 1162, + 520, + 1195, + 324, + 1195 + ], + "score": 0.93, + "latex": "Y = \\operatorname* { m i n } ( T _ { A } , C )" + }, + { + "category_id": 13, + "poly": [ + 549, + 1836, + 1062, + 1836, + 1062, + 1869, + 549, + 1869 + ], + "score": 0.92, + "latex": "f ( t | x ) \\triangleq P ( T = t | X = x ) = \\lambda ( t | x ) S ( t | x )" + }, + { + "category_id": 13, + "poly": [ + 631, + 1574, + 704, + 1574, + 704, + 1606, + 631, + 1606 + ], + "score": 0.92, + "latex": "\\lambda ( t | x )" + }, + { + "category_id": 13, + "poly": [ + 989, + 1131, + 1106, + 1131, + 1106, + 1164, + 989, + 1164 + ], + "score": 0.92, + "latex": "\\delta \\in \\{ 0 , 1 \\}" + }, + { + "category_id": 13, + "poly": [ + 1099, + 499, + 1221, + 499, + 1221, + 530, + 1099, + 530 + ], + "score": 0.92, + "latex": "r \\ = \\ \\Phi ( x )" + }, + { + "category_id": 13, + "poly": [ + 447, + 1300, + 546, + 1300, + 546, + 1333, + 447, + 1333 + ], + "score": 0.92, + "latex": "p ( T _ { 0 } | X )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1975, + 389, + 1975, + 389, + 2006, + 298, + 2006 + ], + "score": 0.92, + "latex": "\\lambda _ { A } ( t | x )" + }, + { + "category_id": 13, + "poly": [ + 989, + 1359, + 1268, + 1359, + 1268, + 1395, + 989, + 1395 + ], + "score": 0.92, + "latex": "\\mathcal { D } = \\{ ( y _ { i } , \\delta _ { i } , x _ { i } , a _ { i } ) \\} _ { i = 1 } ^ { N }" + }, + { + "category_id": 13, + "poly": [ + 887, + 1575, + 962, + 1575, + 962, + 1606, + 887, + 1606 + ], + "score": 0.92, + "latex": "S ( t | x )" + }, + { + "category_id": 13, + "poly": [ + 1224, + 528, + 1316, + 528, + 1316, + 560, + 1224, + 560 + ], + "score": 0.92, + "latex": "h _ { A } ( r , \\tilde { \\epsilon } )" + }, + { + "category_id": 13, + "poly": [ + 617, + 527, + 835, + 527, + 835, + 561, + 617, + 561 + ], + "score": 0.92, + "latex": "\\bar { t } _ { a } \\sim p ( T _ { A } | X = x )" + }, + { + "category_id": 13, + "poly": [ + 1056, + 994, + 1121, + 994, + 1121, + 1024, + 1056, + 1024 + ], + "score": 0.92, + "latex": "T _ { 1 - A }" + }, + { + "category_id": 13, + "poly": [ + 571, + 993, + 665, + 993, + 665, + 1023, + 571, + 1023 + ], + "score": 0.91, + "latex": "T = T _ { A }" + }, + { + "category_id": 13, + "poly": [ + 1261, + 1945, + 1353, + 1945, + 1353, + 1975, + 1261, + 1975 + ], + "score": 0.91, + "latex": "S _ { A } ( t | x )" + }, + { + "category_id": 13, + "poly": [ + 1052, + 1163, + 1148, + 1163, + 1148, + 1192, + 1052, + 1192 + ], + "score": 0.91, + "latex": "T _ { A } < C" + }, + { + "category_id": 13, + "poly": [ + 1074, + 1542, + 1308, + 1542, + 1308, + 1576, + 1074, + 1576 + ], + "score": 0.91, + "latex": "S ( t | x ) \\triangleq 1 - F ( t | x )" + }, + { + "category_id": 13, + "poly": [ + 790, + 1637, + 919, + 1637, + 919, + 1667, + 790, + 1667 + ], + "score": 0.91, + "latex": "\\{ t , t + \\Delta t \\}" + }, + { + "category_id": 13, + "poly": [ + 296, + 1332, + 428, + 1332, + 428, + 1363, + 296, + 1363 + ], + "score": 0.91, + "latex": "( Y , \\delta , X , A )" + }, + { + "category_id": 13, + "poly": [ + 879, + 528, + 955, + 528, + 955, + 556, + 879, + 556 + ], + "score": 0.91, + "latex": "A = a" + }, + { + "category_id": 13, + "poly": [ + 298, + 1300, + 397, + 1300, + 397, + 1332, + 298, + 1332 + ], + "score": 0.9, + "latex": "p ( T _ { 1 } | X )" + }, + { + "category_id": 13, + "poly": [ + 1082, + 872, + 1165, + 872, + 1165, + 899, + 1082, + 899 + ], + "score": 0.9, + "latex": "A = 0" + }, + { + "category_id": 13, + "poly": [ + 577, + 1475, + 890, + 1475, + 890, + 1511, + 577, + 1511 + ], + "score": 0.9, + "latex": "F ( t | x ) \\triangleq P ( T \\leq t | X = x )" + }, + { + "category_id": 13, + "poly": [ + 1204, + 559, + 1253, + 559, + 1253, + 591, + 1204, + 591 + ], + "score": 0.9, + "latex": "p ( \\epsilon )" + }, + { + "category_id": 13, + "poly": [ + 1199, + 1162, + 1265, + 1162, + 1265, + 1190, + 1199, + 1190 + ], + "score": 0.9, + "latex": "\\delta = 1" + }, + { + "category_id": 13, + "poly": [ + 298, + 902, + 374, + 902, + 374, + 930, + 298, + 930 + ], + "score": 0.9, + "latex": "A = 1" + }, + { + "category_id": 13, + "poly": [ + 648, + 781, + 685, + 781, + 685, + 810, + 648, + 810 + ], + "score": 0.89, + "latex": "N _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1235, + 1271, + 1267, + 1271, + 1267, + 1300, + 1235, + 1300 + ], + "score": 0.89, + "latex": "T _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 947, + 781, + 984, + 781, + 984, + 810, + 947, + 810 + ], + "score": 0.89, + "latex": "N _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 1133, + 1948, + 1204, + 1948, + 1204, + 1974, + 1133, + 1974 + ], + "score": 0.88, + "latex": "T _ { 1 } , T _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 704, + 932, + 736, + 932, + 736, + 963, + 704, + 963 + ], + "score": 0.88, + "latex": "T _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1103, + 1210, + 1136, + 1210, + 1136, + 1239, + 1103, + 1239 + ], + "score": 0.88, + "latex": "T _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1217, + 780, + 1396, + 780, + 1396, + 811, + 1217, + 811 + ], + "score": 0.88, + "latex": "( N = N _ { 1 } + N _ { 0 } )" + }, + { + "category_id": 13, + "poly": [ + 622, + 933, + 654, + 933, + 654, + 963, + 622, + 963 + ], + "score": 0.88, + "latex": "T _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 1179, + 1977, + 1214, + 1977, + 1214, + 2003, + 1179, + 2003 + ], + "score": 0.88, + "latex": "T _ { A }" + }, + { + "category_id": 13, + "poly": [ + 995, + 1635, + 1068, + 1635, + 1068, + 1662, + 995, + 1662 + ], + "score": 0.87, + "latex": "T > t" + }, + { + "category_id": 13, + "poly": [ + 1367, + 1975, + 1397, + 1975, + 1397, + 2003, + 1367, + 2003 + ], + "score": 0.87, + "latex": "T _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 1280, + 1977, + 1311, + 1977, + 1311, + 2003, + 1280, + 2003 + ], + "score": 0.87, + "latex": "T _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1190, + 1209, + 1222, + 1209, + 1222, + 1240, + 1190, + 1240 + ], + "score": 0.87, + "latex": "T _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 595, + 1918, + 616, + 1918, + 616, + 1939, + 595, + 1939 + ], + "score": 0.87, + "latex": "A" + }, + { + "category_id": 13, + "poly": [ + 1318, + 1271, + 1348, + 1271, + 1348, + 1300, + 1318, + 1300 + ], + "score": 0.87, + "latex": "T _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 1122, + 1636, + 1217, + 1636, + 1217, + 1661, + 1122, + 1661 + ], + "score": 0.86, + "latex": "\\Delta t 0" + }, + { + "category_id": 13, + "poly": [ + 1342, + 469, + 1404, + 469, + 1404, + 497, + 1342, + 497 + ], + "score": 0.84, + "latex": "X =" + }, + { + "category_id": 13, + "poly": [ + 1211, + 1301, + 1239, + 1301, + 1239, + 1327, + 1211, + 1327 + ], + "score": 0.83, + "latex": "N" + }, + { + "category_id": 13, + "poly": [ + 663, + 1133, + 688, + 1133, + 688, + 1159, + 663, + 1159 + ], + "score": 0.82, + "latex": "C" + }, + { + "category_id": 13, + "poly": [ + 825, + 812, + 853, + 812, + 853, + 838, + 825, + 838 + ], + "score": 0.82, + "latex": "X" + }, + { + "category_id": 13, + "poly": [ + 298, + 1240, + 325, + 1240, + 325, + 1267, + 298, + 1267 + ], + "score": 0.82, + "latex": "X" + }, + { + "category_id": 13, + "poly": [ + 493, + 781, + 522, + 781, + 522, + 807, + 493, + 807 + ], + "score": 0.78, + "latex": "N" + }, + { + "category_id": 13, + "poly": [ + 967, + 872, + 990, + 872, + 990, + 898, + 967, + 898 + ], + "score": 0.77, + "latex": "A" + }, + { + "category_id": 13, + "poly": [ + 1024, + 903, + 1047, + 903, + 1047, + 929, + 1024, + 929 + ], + "score": 0.76, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 765, + 504, + 782, + 504, + 782, + 525, + 765, + 525 + ], + "score": 0.74, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 835, + 1611, + 847, + 1611, + 847, + 1630, + 835, + 1630 + ], + "score": 0.73, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 934, + 1515, + 952, + 1515, + 952, + 1536, + 934, + 1536 + ], + "score": 0.71, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 517, + 1512, + 530, + 1512, + 530, + 1537, + 517, + 1537 + ], + "score": 0.7, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 297, + 504, + 317, + 504, + 317, + 525, + 297, + 525 + ], + "score": 0.69, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 622, + 1872, + 643, + 1872, + 643, + 1892, + 622, + 1892 + ], + "score": 0.69, + "latex": "\\&" + }, + { + "category_id": 13, + "poly": [ + 497, + 1478, + 521, + 1478, + 521, + 1505, + 497, + 1505 + ], + "score": 0.68, + "latex": "A" + }, + { + "category_id": 13, + "poly": [ + 921, + 561, + 937, + 561, + 937, + 587, + 921, + 587 + ], + "score": 0.45, + "latex": "\\tilde { \\epsilon }" + }, + { + "category_id": 15, + "poly": [ + 975.0, + 218.0, + 994.0, + 218.0, + 994.0, + 238.0, + 975.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1235.0, + 214.0, + 1257.0, + 214.0, + 1257.0, + 238.0, + 1235.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 250.0, + 462.0, + 250.0, + 462.0, + 293.0, + 393.0, + 293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 260.0, + 772.0, + 260.0, + 772.0, + 270.0, + 764.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 292.0, + 509.0, + 292.0, + 509.0, + 323.0, + 489.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 626.0, + 265.0, + 766.0, + 265.0, + 766.0, + 300.0, + 626.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 278.0, + 916.0, + 278.0, + 916.0, + 303.0, + 886.0, + 303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 948.0, + 280.0, + 972.0, + 280.0, + 972.0, + 298.0, + 948.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1164.0, + 290.0, + 1180.0, + 290.0, + 1180.0, + 303.0, + 1164.0, + 303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1315.0, + 291.0, + 1331.0, + 291.0, + 1331.0, + 305.0, + 1315.0, + 305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 326.0, + 933.0, + 326.0, + 933.0, + 342.0, + 919.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 967.0, + 319.0, + 1000.0, + 319.0, + 1000.0, + 350.0, + 967.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 332.0, + 1197.0, + 332.0, + 1197.0, + 345.0, + 1188.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1229.0, + 327.0, + 1258.0, + 327.0, + 1258.0, + 353.0, + 1229.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1287.0, + 330.0, + 1305.0, + 330.0, + 1305.0, + 349.0, + 1287.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 627.0, + 386.0, + 765.0, + 386.0, + 765.0, + 423.0, + 627.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 967.0, + 391.0, + 992.0, + 391.0, + 992.0, + 415.0, + 967.0, + 415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 419.0, + 425.0, + 710.0, + 425.0, + 710.0, + 457.0, + 419.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 867.0, + 426.0, + 1078.0, + 426.0, + 1078.0, + 458.0, + 867.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 425.0, + 1330.0, + 425.0, + 1330.0, + 457.0, + 1169.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 214.5, + 608.0, + 214.5, + 608.0, + 256.5, + 511.0, + 256.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 521.0, + 350.0, + 591.0, + 350.0, + 591.0, + 388.5, + 521.0, + 388.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 467.0, + 1341.0, + 467.0, + 1341.0, + 502.0, + 294.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 497.0, + 296.0, + 497.0, + 296.0, + 533.0, + 293.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 497.0, + 764.0, + 497.0, + 764.0, + 533.0, + 318.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 497.0, + 1098.0, + 497.0, + 1098.0, + 533.0, + 783.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1222.0, + 497.0, + 1405.0, + 497.0, + 1405.0, + 533.0, + 1222.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 529.0, + 616.0, + 529.0, + 616.0, + 561.0, + 296.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 529.0, + 878.0, + 529.0, + 878.0, + 561.0, + 836.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 956.0, + 529.0, + 1223.0, + 529.0, + 1223.0, + 561.0, + 956.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1317.0, + 529.0, + 1404.0, + 529.0, + 1404.0, + 561.0, + 1317.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 559.0, + 920.0, + 559.0, + 920.0, + 591.0, + 296.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 559.0, + 1203.0, + 559.0, + 1203.0, + 591.0, + 938.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1254.0, + 559.0, + 1404.0, + 559.0, + 1404.0, + 591.0, + 1254.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 590.0, + 1405.0, + 590.0, + 1405.0, + 622.0, + 294.0, + 622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 621.0, + 563.0, + 621.0, + 563.0, + 656.0, + 294.0, + 656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 674.0, + 717.0, + 674.0, + 717.0, + 719.0, + 292.0, + 719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 857.0, + 72.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 856.0, + 73.0, + 856.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 748.0, + 1409.0, + 748.0, + 1409.0, + 785.0, + 295.0, + 785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 778.0, + 492.0, + 778.0, + 492.0, + 816.0, + 294.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 778.0, + 647.0, + 778.0, + 647.0, + 816.0, + 523.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 778.0, + 946.0, + 778.0, + 946.0, + 816.0, + 686.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 778.0, + 1216.0, + 778.0, + 1216.0, + 816.0, + 985.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1397.0, + 778.0, + 1409.0, + 778.0, + 1409.0, + 816.0, + 1397.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 809.0, + 824.0, + 809.0, + 824.0, + 847.0, + 294.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 854.0, + 809.0, + 1407.0, + 809.0, + 1407.0, + 847.0, + 854.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 840.0, + 1408.0, + 840.0, + 1408.0, + 875.0, + 294.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 869.0, + 966.0, + 869.0, + 966.0, + 906.0, + 294.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 991.0, + 869.0, + 1081.0, + 869.0, + 1081.0, + 906.0, + 991.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1166.0, + 869.0, + 1407.0, + 869.0, + 1407.0, + 906.0, + 1166.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 898.0, + 297.0, + 898.0, + 297.0, + 939.0, + 294.0, + 939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 898.0, + 1023.0, + 898.0, + 1023.0, + 939.0, + 375.0, + 939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1048.0, + 898.0, + 1407.0, + 898.0, + 1407.0, + 939.0, + 1048.0, + 939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 929.0, + 621.0, + 929.0, + 621.0, + 969.0, + 294.0, + 969.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 655.0, + 929.0, + 703.0, + 929.0, + 703.0, + 969.0, + 655.0, + 969.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 737.0, + 929.0, + 1405.0, + 929.0, + 1405.0, + 969.0, + 737.0, + 969.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 958.0, + 1410.0, + 958.0, + 1410.0, + 1002.0, + 292.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 992.0, + 570.0, + 992.0, + 570.0, + 1029.0, + 294.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 666.0, + 992.0, + 1055.0, + 992.0, + 1055.0, + 1029.0, + 666.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1122.0, + 992.0, + 1291.0, + 992.0, + 1291.0, + 1029.0, + 1122.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1440.0, + 1406.0, + 1440.0, + 1406.0, + 1480.0, + 293.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1475.0, + 496.0, + 1475.0, + 496.0, + 1513.0, + 294.0, + 1513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 1475.0, + 576.0, + 1475.0, + 576.0, + 1513.0, + 522.0, + 1513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 1475.0, + 1406.0, + 1475.0, + 1406.0, + 1513.0, + 891.0, + 1513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1509.0, + 516.0, + 1509.0, + 516.0, + 1543.0, + 295.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 1509.0, + 933.0, + 1509.0, + 933.0, + 1543.0, + 531.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 953.0, + 1509.0, + 1404.0, + 1509.0, + 1404.0, + 1543.0, + 953.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1540.0, + 1073.0, + 1540.0, + 1073.0, + 1577.0, + 293.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1309.0, + 1540.0, + 1406.0, + 1540.0, + 1406.0, + 1577.0, + 1309.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1571.0, + 630.0, + 1571.0, + 630.0, + 1608.0, + 294.0, + 1608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 1571.0, + 886.0, + 1571.0, + 886.0, + 1608.0, + 705.0, + 1608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 1571.0, + 1404.0, + 1571.0, + 1404.0, + 1608.0, + 963.0, + 1608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1604.0, + 834.0, + 1604.0, + 834.0, + 1638.0, + 295.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1604.0, + 1404.0, + 1604.0, + 1404.0, + 1638.0, + 848.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1633.0, + 789.0, + 1633.0, + 789.0, + 1668.0, + 293.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 1633.0, + 994.0, + 1633.0, + 994.0, + 1668.0, + 920.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1069.0, + 1633.0, + 1121.0, + 1633.0, + 1121.0, + 1668.0, + 1069.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 1633.0, + 1407.0, + 1633.0, + 1407.0, + 1668.0, + 1218.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1663.0, + 1406.0, + 1663.0, + 1406.0, + 1698.0, + 293.0, + 1698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1692.0, + 457.0, + 1692.0, + 457.0, + 1727.0, + 294.0, + 1727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1204.0, + 1102.0, + 1204.0, + 1102.0, + 1244.0, + 291.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1204.0, + 1189.0, + 1204.0, + 1189.0, + 1244.0, + 1137.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1223.0, + 1204.0, + 1406.0, + 1204.0, + 1406.0, + 1244.0, + 1223.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 1239.0, + 1406.0, + 1239.0, + 1406.0, + 1274.0, + 326.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1269.0, + 1234.0, + 1269.0, + 1234.0, + 1304.0, + 294.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1268.0, + 1269.0, + 1317.0, + 1269.0, + 1317.0, + 1304.0, + 1268.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1349.0, + 1269.0, + 1408.0, + 1269.0, + 1408.0, + 1304.0, + 1349.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1299.0, + 297.0, + 1299.0, + 297.0, + 1334.0, + 291.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 398.0, + 1299.0, + 446.0, + 1299.0, + 446.0, + 1334.0, + 398.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1299.0, + 1210.0, + 1299.0, + 1210.0, + 1334.0, + 547.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1240.0, + 1299.0, + 1407.0, + 1299.0, + 1407.0, + 1334.0, + 1240.0, + 1334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1326.0, + 295.0, + 1326.0, + 295.0, + 1367.0, + 290.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 429.0, + 1326.0, + 1408.0, + 1326.0, + 1408.0, + 1367.0, + 429.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1350.0, + 988.0, + 1350.0, + 988.0, + 1404.0, + 288.0, + 1404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1269.0, + 1350.0, + 1412.0, + 1350.0, + 1412.0, + 1404.0, + 1269.0, + 1404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1389.0, + 1036.0, + 1389.0, + 1036.0, + 1427.0, + 294.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1037.0, + 1406.0, + 1037.0, + 1406.0, + 1075.0, + 295.0, + 1075.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1069.0, + 1403.0, + 1069.0, + 1403.0, + 1103.0, + 296.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1100.0, + 1407.0, + 1100.0, + 1407.0, + 1137.0, + 294.0, + 1137.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1128.0, + 662.0, + 1128.0, + 662.0, + 1167.0, + 294.0, + 1167.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 689.0, + 1128.0, + 988.0, + 1128.0, + 988.0, + 1167.0, + 689.0, + 1167.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1107.0, + 1128.0, + 1405.0, + 1128.0, + 1405.0, + 1167.0, + 1107.0, + 1167.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1160.0, + 323.0, + 1160.0, + 323.0, + 1197.0, + 294.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 521.0, + 1160.0, + 1051.0, + 1160.0, + 1051.0, + 1197.0, + 521.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 1160.0, + 1198.0, + 1160.0, + 1198.0, + 1197.0, + 1149.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1266.0, + 1160.0, + 1278.0, + 1160.0, + 1278.0, + 1197.0, + 1266.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1912.0, + 594.0, + 1912.0, + 594.0, + 1944.0, + 296.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 617.0, + 1912.0, + 1407.0, + 1912.0, + 1407.0, + 1944.0, + 617.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1942.0, + 1132.0, + 1942.0, + 1132.0, + 1976.0, + 292.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1205.0, + 1942.0, + 1260.0, + 1942.0, + 1260.0, + 1976.0, + 1205.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 1942.0, + 1407.0, + 1942.0, + 1407.0, + 1976.0, + 1354.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 1971.0, + 1178.0, + 1971.0, + 1178.0, + 2008.0, + 390.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 1971.0, + 1279.0, + 1971.0, + 1279.0, + 2008.0, + 1215.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1312.0, + 1971.0, + 1366.0, + 1971.0, + 1366.0, + 2008.0, + 1312.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 1971.0, + 1408.0, + 1971.0, + 1408.0, + 2008.0, + 1398.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1998.0, + 1405.0, + 1998.0, + 1405.0, + 2039.0, + 292.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1828.0, + 548.0, + 1828.0, + 548.0, + 1871.0, + 292.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1063.0, + 1828.0, + 1406.0, + 1828.0, + 1406.0, + 1871.0, + 1063.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1863.0, + 621.0, + 1863.0, + 621.0, + 1899.0, + 296.0, + 1899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 644.0, + 1863.0, + 804.0, + 1863.0, + 804.0, + 1899.0, + 644.0, + 1899.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 789, + 1405, + 789, + 1405, + 1191, + 298, + 1191 + ], + "score": 0.986 + }, + { + "category_id": 1, + "poly": [ + 297, + 223, + 1406, + 223, + 1406, + 693, + 297, + 693 + ], + "score": 0.985 + }, + { + "category_id": 1, + "poly": [ + 297, + 1377, + 1405, + 1377, + 1405, + 1689, + 297, + 1689 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 297, + 1701, + 1404, + 1701, + 1404, + 1894, + 297, + 1894 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 299, + 1906, + 1403, + 1906, + 1403, + 2034, + 299, + 2034 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 296, + 1208, + 1398, + 1208, + 1398, + 1271, + 296, + 1271 + ], + "score": 0.943 + }, + { + "category_id": 0, + "poly": [ + 299, + 729, + 513, + 729, + 513, + 764, + 299, + 764 + ], + "score": 0.904 + }, + { + "category_id": 8, + "poly": [ + 686, + 1287, + 1014, + 1287, + 1014, + 1364, + 686, + 1364 + ], + "score": 0.891 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1291, + 1399, + 1291, + 1399, + 1320, + 1366, + 1320 + ], + "score": 0.877 + }, + { + "category_id": 9, + "poly": [ + 1367, + 1331, + 1399, + 1331, + 1399, + 1359, + 1367, + 1359 + ], + "score": 0.855 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 854, + 75, + 854, + 104, + 298, + 104 + ], + "score": 0.849 + }, + { + "category_id": 2, + "poly": [ + 840, + 2089, + 857, + 2089, + 857, + 2111, + 840, + 2111 + ], + "score": 0.772 + }, + { + "category_id": 8, + "poly": [ + 694, + 1288, + 1013, + 1288, + 1013, + 1323, + 694, + 1323 + ], + "score": 0.194 + }, + { + "category_id": 8, + "poly": [ + 686, + 1327, + 940, + 1327, + 940, + 1364, + 686, + 1364 + ], + "score": 0.19 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 853, + 76, + 853, + 104, + 299, + 104 + ], + "score": 0.151 + }, + { + "category_id": 13, + "poly": [ + 609, + 1004, + 739, + 1004, + 739, + 1038, + 609, + 1038 + ], + "score": 0.93, + "latex": "A = \\{ 1 , 0 \\}" + }, + { + "category_id": 13, + "poly": [ + 844, + 1593, + 991, + 1593, + 991, + 1626, + 844, + 1626 + ], + "score": 0.93, + "latex": "\\ell _ { h , \\Phi } ( t _ { a } , y , \\delta )" + }, + { + "category_id": 13, + "poly": [ + 1144, + 629, + 1247, + 629, + 1247, + 663, + 1144, + 663 + ], + "score": 0.93, + "latex": "p ( C _ { 0 } | X )" + }, + { + "category_id": 13, + "poly": [ + 1099, + 1830, + 1253, + 1830, + 1253, + 1864, + 1099, + 1864 + ], + "score": 0.93, + "latex": "\\{ h _ { A } ( \\cdot ) , \\Phi ( \\cdot ) \\}" + }, + { + "category_id": 13, + "poly": [ + 984, + 629, + 1088, + 629, + 1088, + 663, + 984, + 663 + ], + "score": 0.93, + "latex": "p ( C _ { 1 } | X )" + }, + { + "category_id": 13, + "poly": [ + 414, + 1500, + 569, + 1500, + 569, + 1534, + 414, + 1534 + ], + "score": 0.93, + "latex": "\\{ h _ { A } ( \\cdot ) , \\Phi ( \\cdot ) \\bar \\}" + }, + { + "category_id": 13, + "poly": [ + 927, + 1240, + 1080, + 1240, + 1080, + 1273, + 927, + 1273 + ], + "score": 0.93, + "latex": "\\{ h _ { A } ( \\cdot ) , \\Phi ( \\cdot ) \\}" + }, + { + "category_id": 13, + "poly": [ + 1125, + 1096, + 1353, + 1096, + 1353, + 1130, + 1125, + 1130 + ], + "score": 0.92, + "latex": "\\{ h _ { A } ( \\cdot ) , \\nu _ { A } ( \\cdot ) , \\Phi ( \\cdot ) \\}" + }, + { + "category_id": 13, + "poly": [ + 1233, + 1734, + 1403, + 1734, + 1403, + 1768, + 1233, + 1768 + ], + "score": 0.92, + "latex": "- \\ell _ { h , \\Phi } ( t _ { a } , y , \\delta )" + }, + { + "category_id": 13, + "poly": [ + 443, + 1657, + 512, + 1657, + 512, + 1689, + 443, + 1689 + ], + "score": 0.92, + "latex": "\\{ y , \\delta \\}" + }, + { + "category_id": 13, + "poly": [ + 647, + 508, + 814, + 508, + 814, + 541, + 647, + 541 + ], + "score": 0.92, + "latex": "p ( X = x ) > 0" + }, + { + "category_id": 14, + "poly": [ + 685, + 1284, + 1015, + 1284, + 1015, + 1370, + 685, + 1370 + ], + "score": 0.92, + "latex": "\\begin{array} { c } { { t \\sim p _ { h , \\Phi } ( T | X = x , A = a ) } } \\\\ { { { \\it t } _ { a } \\sim p _ { h , \\Phi } ( T _ { a } | X = x ) } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 296, + 1799, + 758, + 1799, + 758, + 1834, + 296, + 1834 + ], + "score": 0.91, + "latex": "\\log f _ { h , \\Phi } ( t _ { a } | x ) + ( 1 - \\bar { \\delta } ) \\cdot \\log S _ { h , \\Phi } ( t _ { a } | x )" + }, + { + "category_id": 13, + "poly": [ + 541, + 974, + 607, + 974, + 607, + 1007, + 541, + 1007 + ], + "score": 0.91, + "latex": "h _ { A } ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 1207, + 1004, + 1403, + 1004, + 1403, + 1039, + 1207, + 1039 + ], + "score": 0.91, + "latex": "p _ { h , \\Phi } ( T _ { A } | X = x )" + }, + { + "category_id": 13, + "poly": [ + 1011, + 568, + 1172, + 568, + 1172, + 601, + 1011, + 601 + ], + "score": 0.91, + "latex": "T \\perp \\perp C | X , A" + }, + { + "category_id": 13, + "poly": [ + 633, + 1440, + 820, + 1440, + 820, + 1473, + 633, + 1473 + ], + "score": 0.91, + "latex": "T = T _ { A } | A = a" + }, + { + "category_id": 13, + "poly": [ + 903, + 1471, + 1179, + 1471, + 1179, + 1504, + 903, + 1504 + ], + "score": 0.91, + "latex": "c _ { a } \\sim p _ { \\nu , \\Phi } ( C _ { A } | X = x )" + }, + { + "category_id": 13, + "poly": [ + 481, + 1096, + 739, + 1096, + 739, + 1131, + 481, + 1131 + ], + "score": 0.91, + "latex": "c _ { a } \\sim p _ { \\nu , \\Phi } ( C _ { A } | X = x )" + }, + { + "category_id": 13, + "poly": [ + 932, + 1410, + 1297, + 1410, + 1297, + 1443, + 932, + 1443 + ], + "score": 0.9, + "latex": "0 < P ( A = \\mathsf { \\bar { a } } | X \\doteq x ) < 1 ." + }, + { + "category_id": 13, + "poly": [ + 996, + 1065, + 1058, + 1065, + 1058, + 1099, + 996, + 1099 + ], + "score": 0.9, + "latex": "\\nu _ { A } ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 298, + 416, + 499, + 416, + 499, + 450, + 298, + 450 + ], + "score": 0.9, + "latex": "P ( A = a | X = x )" + }, + { + "category_id": 13, + "poly": [ + 525, + 225, + 1251, + 225, + 1251, + 265, + 525, + 265 + ], + "score": 0.9, + "latex": "\\begin{array} { r } { \\mathrm { { I T E } } ( t , x ) = \\int _ { 0 } ^ { t _ { \\operatorname* { m a x } } } \\{ S _ { 1 } ( t | x ) - S _ { 0 } ( t | x ) \\} \\mathrm { d } t = \\mathbb { E } \\{ T _ { 1 } - T _ { 0 } | X = x \\} . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 505, + 265, + 853, + 265, + 853, + 297, + 505, + 297 + ], + "score": 0.9, + "latex": "\\mathrm { I T E } ( t , x ) = S _ { 1 } ( t | x ) - S _ { 0 } ( t | x )" + }, + { + "category_id": 13, + "poly": [ + 993, + 1703, + 1193, + 1703, + 1193, + 1738, + 993, + 1738 + ], + "score": 0.9, + "latex": "p _ { h , \\Phi } ( T _ { A } | X = x )" + }, + { + "category_id": 13, + "poly": [ + 1256, + 1940, + 1404, + 1940, + 1404, + 1976, + 1256, + 1976 + ], + "score": 0.9, + "latex": "\\zeta _ { h , \\Phi } ( x , a ) \\triangleq" + }, + { + "category_id": 13, + "poly": [ + 775, + 1627, + 1037, + 1627, + 1037, + 1659, + 775, + 1659 + ], + "score": 0.9, + "latex": "c _ { a } \\sim p _ { \\nu , \\Phi } ( C _ { A } | X = x )" + }, + { + "category_id": 13, + "poly": [ + 297, + 1974, + 633, + 1974, + 633, + 2009, + 297, + 2009 + ], + "score": 0.9, + "latex": "\\mathbb { E } _ { ( y , \\delta , x ) \\sim p ( Y , \\delta | X ) } \\ell _ { h , \\Phi } ( t _ { a } , y , \\delta )" + }, + { + "category_id": 13, + "poly": [ + 594, + 1562, + 897, + 1562, + 897, + 1594, + 594, + 1594 + ], + "score": 0.9, + "latex": "( y , \\delta , x , a ) \\sim p ( Y , \\delta , X , A )" + }, + { + "category_id": 13, + "poly": [ + 627, + 1501, + 690, + 1501, + 690, + 1534, + 627, + 1534 + ], + "score": 0.9, + "latex": "\\nu _ { A } ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 1344, + 2003, + 1397, + 2003, + 1397, + 2034, + 1344, + 2034 + ], + "score": 0.9, + "latex": "\\mathcal { L } _ { \\mathrm { C F } }" + }, + { + "category_id": 13, + "poly": [ + 297, + 507, + 460, + 507, + 460, + 541, + 297, + 541 + ], + "score": 0.9, + "latex": "1 | X = x ) < 1" + }, + { + "category_id": 13, + "poly": [ + 297, + 1004, + 557, + 1004, + 557, + 1038, + 297, + 1038 + ], + "score": 0.9, + "latex": "\\bar { t } _ { a } \\sim p _ { h , \\Phi } ( T _ { A } | X = x )" + }, + { + "category_id": 13, + "poly": [ + 857, + 1767, + 1393, + 1767, + 1393, + 1803, + 857, + 1803 + ], + "score": 0.9, + "latex": "- \\ell _ { h , \\Phi } ( t _ { a } , y , \\delta ) \\triangleq \\log p _ { h , \\Phi } ( T _ { a } | X = x ) = \\delta" + }, + { + "category_id": 13, + "poly": [ + 855, + 914, + 969, + 914, + 969, + 944, + 855, + 944 + ], + "score": 0.9, + "latex": "r = \\Phi ( x )" + }, + { + "category_id": 13, + "poly": [ + 1256, + 2004, + 1294, + 2004, + 1294, + 2034, + 1256, + 2034 + ], + "score": 0.89, + "latex": "\\mathcal { L } _ { \\mathrm { F } }" + }, + { + "category_id": 13, + "poly": [ + 455, + 1627, + 713, + 1627, + 713, + 1659, + 455, + 1659 + ], + "score": 0.89, + "latex": "t _ { a } \\sim p _ { h , \\Phi } ( T _ { A } | X = x )" + }, + { + "category_id": 13, + "poly": [ + 1072, + 263, + 1398, + 263, + 1398, + 297, + 1072, + 297 + ], + "score": 0.89, + "latex": "\\mathrm { I T E } ( t , x ) = \\lambda _ { 1 } ( t | x ) / \\lambda _ { 0 } ( t | x )" + }, + { + "category_id": 13, + "poly": [ + 366, + 1039, + 447, + 1039, + 447, + 1064, + 366, + 1064 + ], + "score": 0.88, + "latex": "A = a" + }, + { + "category_id": 13, + "poly": [ + 1265, + 479, + 1407, + 479, + 1407, + 511, + 1265, + 511 + ], + "score": 0.88, + "latex": "0 < P ( A =" + }, + { + "category_id": 13, + "poly": [ + 686, + 630, + 721, + 630, + 721, + 661, + 686, + 661 + ], + "score": 0.88, + "latex": "C _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 594, + 631, + 630, + 631, + 630, + 660, + 594, + 660 + ], + "score": 0.88, + "latex": "C _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 875, + 1007, + 902, + 1007, + 902, + 1036, + 875, + 1036 + ], + "score": 0.87, + "latex": "t _ { a }" + }, + { + "category_id": 13, + "poly": [ + 875, + 944, + 964, + 944, + 964, + 972, + 875, + 972 + ], + "score": 0.86, + "latex": "r \\in { \\mathbb { R } } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1593, + 754, + 1593, + 754, + 1628, + 297, + 1628 + ], + "score": 0.86, + "latex": "\\mathcal { L } = \\mathbb { E } _ { ( y , \\delta , x , a ) \\sim p ( Y , \\delta , X , A ) } \\left[ \\ell _ { h , \\Phi } ( t _ { a } , y , \\delta ) \\right]" + }, + { + "category_id": 13, + "poly": [ + 1199, + 446, + 1397, + 446, + 1397, + 480, + 1199, + 480 + ], + "score": 0.85, + "latex": "\\{ T _ { 1 } , T _ { 0 } \\} \\perp \\perp A | X" + }, + { + "category_id": 13, + "poly": [ + 1094, + 1379, + 1194, + 1379, + 1194, + 1412, + 1094, + 1412 + ], + "score": 0.84, + "latex": "\\{ T _ { 0 } , T _ { 1 } \\}" + }, + { + "category_id": 13, + "poly": [ + 1248, + 1379, + 1306, + 1379, + 1306, + 1412, + 1248, + 1412 + ], + "score": 0.84, + "latex": "A | X" + }, + { + "category_id": 13, + "poly": [ + 651, + 1862, + 672, + 1862, + 672, + 1888, + 651, + 1888 + ], + "score": 0.82, + "latex": "\\mathcal { L }" + }, + { + "category_id": 13, + "poly": [ + 1143, + 1533, + 1165, + 1533, + 1165, + 1559, + 1143, + 1559 + ], + "score": 0.81, + "latex": "\\mathcal { L }" + }, + { + "category_id": 13, + "poly": [ + 1333, + 1980, + 1353, + 1980, + 1353, + 2000, + 1333, + 2000 + ], + "score": 0.8, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 1110, + 1563, + 1135, + 1563, + 1135, + 1589, + 1110, + 1589 + ], + "score": 0.8, + "latex": "\\mathcal { D }" + }, + { + "category_id": 13, + "poly": [ + 1234, + 570, + 1274, + 570, + 1274, + 599, + 1234, + 599 + ], + "score": 0.8, + "latex": "( i i )" + }, + { + "category_id": 13, + "poly": [ + 1028, + 944, + 1070, + 944, + 1070, + 976, + 1028, + 976 + ], + "score": 0.79, + "latex": "( i i )" + }, + { + "category_id": 13, + "poly": [ + 1205, + 1943, + 1222, + 1943, + 1222, + 1970, + 1205, + 1970 + ], + "score": 0.77, + "latex": "\\delta" + }, + { + "category_id": 13, + "poly": [ + 576, + 914, + 609, + 914, + 609, + 946, + 576, + 946 + ], + "score": 0.77, + "latex": "( i )" + }, + { + "category_id": 13, + "poly": [ + 536, + 2012, + 553, + 2012, + 553, + 2030, + 536, + 2030 + ], + "score": 0.77, + "latex": "a" + }, + { + "category_id": 13, + "poly": [ + 1287, + 1836, + 1305, + 1836, + 1305, + 1858, + 1287, + 1858 + ], + "score": 0.76, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 530, + 1947, + 548, + 1947, + 548, + 1974, + 530, + 1974 + ], + "score": 0.76, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 1075, + 1914, + 1094, + 1914, + 1094, + 1935, + 1075, + 1935 + ], + "score": 0.75, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 349, + 949, + 369, + 949, + 369, + 971, + 349, + 971 + ], + "score": 0.75, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 1384, + 1916, + 1401, + 1916, + 1401, + 1935, + 1384, + 1935 + ], + "score": 0.72, + "latex": "a" + }, + { + "category_id": 13, + "poly": [ + 370, + 570, + 402, + 570, + 402, + 600, + 370, + 600 + ], + "score": 0.7, + "latex": "( i )" + }, + { + "category_id": 13, + "poly": [ + 612, + 599, + 637, + 599, + 637, + 627, + 612, + 627 + ], + "score": 0.64, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 676, + 599, + 702, + 599, + 702, + 626, + 676, + 626 + ], + "score": 0.63, + "latex": "C" + }, + { + "category_id": 13, + "poly": [ + 869, + 479, + 909, + 479, + 909, + 508, + 869, + 508 + ], + "score": 0.44, + "latex": "( i i )" + }, + { + "category_id": 13, + "poly": [ + 1011, + 569, + 1037, + 569, + 1037, + 598, + 1011, + 598 + ], + "score": 0.38, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 612, + 599, + 703, + 599, + 703, + 628, + 612, + 628 + ], + "score": 0.32, + "latex": "T \\perp \\perp C" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 723.0, + 519.0, + 723.0, + 519.0, + 773.0, + 291.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 857.0, + 72.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2086.0, + 862.0, + 2086.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 856.0, + 73.0, + 856.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 789.0, + 1408.0, + 789.0, + 1408.0, + 827.0, + 293.0, + 827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 821.0, + 1407.0, + 821.0, + 1407.0, + 857.0, + 293.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 851.0, + 1407.0, + 851.0, + 1407.0, + 888.0, + 295.0, + 888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 883.0, + 1405.0, + 883.0, + 1405.0, + 919.0, + 295.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 912.0, + 575.0, + 912.0, + 575.0, + 951.0, + 292.0, + 951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 610.0, + 912.0, + 854.0, + 912.0, + 854.0, + 951.0, + 610.0, + 951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 970.0, + 912.0, + 1410.0, + 912.0, + 1410.0, + 951.0, + 970.0, + 951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 940.0, + 348.0, + 940.0, + 348.0, + 981.0, + 293.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 940.0, + 874.0, + 940.0, + 874.0, + 981.0, + 370.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 940.0, + 1027.0, + 940.0, + 1027.0, + 981.0, + 965.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1071.0, + 940.0, + 1408.0, + 940.0, + 1408.0, + 981.0, + 1071.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 973.0, + 540.0, + 973.0, + 540.0, + 1010.0, + 292.0, + 1010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 608.0, + 973.0, + 1408.0, + 973.0, + 1408.0, + 1010.0, + 608.0, + 1010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1002.0, + 296.0, + 1002.0, + 296.0, + 1042.0, + 292.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 558.0, + 1002.0, + 608.0, + 1002.0, + 608.0, + 1042.0, + 558.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 1002.0, + 874.0, + 1002.0, + 874.0, + 1042.0, + 740.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 903.0, + 1002.0, + 1206.0, + 1002.0, + 1206.0, + 1042.0, + 903.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 1002.0, + 1407.0, + 1002.0, + 1407.0, + 1042.0, + 1404.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1032.0, + 365.0, + 1032.0, + 365.0, + 1074.0, + 291.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 1032.0, + 1406.0, + 1032.0, + 1406.0, + 1074.0, + 448.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1065.0, + 995.0, + 1065.0, + 995.0, + 1101.0, + 293.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 1065.0, + 1405.0, + 1065.0, + 1405.0, + 1101.0, + 1059.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1095.0, + 480.0, + 1095.0, + 480.0, + 1135.0, + 293.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 1095.0, + 1124.0, + 1095.0, + 1124.0, + 1135.0, + 740.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 1095.0, + 1406.0, + 1095.0, + 1406.0, + 1135.0, + 1354.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1125.0, + 1407.0, + 1125.0, + 1407.0, + 1164.0, + 291.0, + 1164.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1159.0, + 706.0, + 1159.0, + 706.0, + 1191.0, + 293.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 216.0, + 524.0, + 216.0, + 524.0, + 280.0, + 286.0, + 280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 216.0, + 1414.0, + 216.0, + 1414.0, + 280.0, + 1252.0, + 280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 261.0, + 504.0, + 261.0, + 504.0, + 300.0, + 292.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 854.0, + 261.0, + 1071.0, + 261.0, + 1071.0, + 300.0, + 854.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 261.0, + 1409.0, + 261.0, + 1409.0, + 300.0, + 1399.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 295.0, + 1408.0, + 295.0, + 1408.0, + 327.0, + 295.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 326.0, + 1403.0, + 326.0, + 1403.0, + 357.0, + 296.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 354.0, + 1409.0, + 354.0, + 1409.0, + 392.0, + 292.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 383.0, + 1409.0, + 383.0, + 1409.0, + 422.0, + 292.0, + 422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 413.0, + 297.0, + 413.0, + 297.0, + 452.0, + 294.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 500.0, + 413.0, + 1407.0, + 413.0, + 1407.0, + 452.0, + 500.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 445.0, + 1198.0, + 445.0, + 1198.0, + 481.0, + 294.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 445.0, + 1406.0, + 445.0, + 1406.0, + 481.0, + 1398.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 477.0, + 868.0, + 477.0, + 868.0, + 511.0, + 291.0, + 511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 910.0, + 477.0, + 1264.0, + 477.0, + 1264.0, + 511.0, + 910.0, + 511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 461.0, + 508.0, + 646.0, + 508.0, + 646.0, + 543.0, + 461.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 815.0, + 508.0, + 1406.0, + 508.0, + 1406.0, + 543.0, + 815.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 537.0, + 1406.0, + 537.0, + 1406.0, + 574.0, + 292.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 566.0, + 369.0, + 566.0, + 369.0, + 604.0, + 291.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 566.0, + 1010.0, + 566.0, + 1010.0, + 604.0, + 403.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 566.0, + 1233.0, + 566.0, + 1233.0, + 604.0, + 1173.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1275.0, + 566.0, + 1407.0, + 566.0, + 1407.0, + 604.0, + 1275.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 598.0, + 611.0, + 598.0, + 611.0, + 634.0, + 294.0, + 634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 598.0, + 1406.0, + 598.0, + 1406.0, + 634.0, + 704.0, + 634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 630.0, + 593.0, + 630.0, + 593.0, + 665.0, + 295.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 631.0, + 630.0, + 685.0, + 630.0, + 685.0, + 665.0, + 631.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 722.0, + 630.0, + 983.0, + 630.0, + 983.0, + 665.0, + 722.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 630.0, + 1143.0, + 630.0, + 1143.0, + 665.0, + 1089.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1248.0, + 630.0, + 1408.0, + 630.0, + 1408.0, + 665.0, + 1248.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 660.0, + 1083.0, + 660.0, + 1083.0, + 696.0, + 295.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1376.0, + 1093.0, + 1376.0, + 1093.0, + 1412.0, + 295.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 1376.0, + 1247.0, + 1376.0, + 1247.0, + 1412.0, + 1195.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1307.0, + 1376.0, + 1406.0, + 1376.0, + 1406.0, + 1412.0, + 1307.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1407.0, + 931.0, + 1407.0, + 931.0, + 1443.0, + 294.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1298.0, + 1407.0, + 1405.0, + 1407.0, + 1405.0, + 1443.0, + 1298.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1438.0, + 632.0, + 1438.0, + 632.0, + 1474.0, + 295.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 821.0, + 1438.0, + 1405.0, + 1438.0, + 1405.0, + 1474.0, + 821.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1467.0, + 902.0, + 1467.0, + 902.0, + 1506.0, + 294.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 1467.0, + 1405.0, + 1467.0, + 1405.0, + 1506.0, + 1180.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1496.0, + 413.0, + 1496.0, + 413.0, + 1538.0, + 292.0, + 1538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 1496.0, + 626.0, + 1496.0, + 626.0, + 1538.0, + 570.0, + 1538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 1496.0, + 1408.0, + 1496.0, + 1408.0, + 1538.0, + 691.0, + 1538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1528.0, + 1142.0, + 1528.0, + 1142.0, + 1568.0, + 294.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1166.0, + 1528.0, + 1405.0, + 1528.0, + 1405.0, + 1568.0, + 1166.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1559.0, + 593.0, + 1559.0, + 593.0, + 1595.0, + 295.0, + 1595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 898.0, + 1559.0, + 1109.0, + 1559.0, + 1109.0, + 1595.0, + 898.0, + 1595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 1559.0, + 1403.0, + 1559.0, + 1403.0, + 1595.0, + 1136.0, + 1595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1587.0, + 296.0, + 1587.0, + 296.0, + 1636.0, + 291.0, + 1636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 1587.0, + 843.0, + 1587.0, + 843.0, + 1636.0, + 755.0, + 1636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 1587.0, + 1408.0, + 1587.0, + 1408.0, + 1636.0, + 992.0, + 1636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1624.0, + 454.0, + 1624.0, + 454.0, + 1660.0, + 294.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 1624.0, + 774.0, + 1624.0, + 774.0, + 1660.0, + 714.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 1624.0, + 1405.0, + 1624.0, + 1405.0, + 1660.0, + 1038.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1653.0, + 442.0, + 1653.0, + 442.0, + 1693.0, + 295.0, + 1693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 1653.0, + 1144.0, + 1653.0, + 1144.0, + 1693.0, + 513.0, + 1693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1701.0, + 992.0, + 1701.0, + 992.0, + 1737.0, + 295.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1194.0, + 1701.0, + 1406.0, + 1701.0, + 1406.0, + 1737.0, + 1194.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1731.0, + 1232.0, + 1731.0, + 1232.0, + 1770.0, + 294.0, + 1770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1767.0, + 856.0, + 1767.0, + 856.0, + 1804.0, + 294.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 1767.0, + 1406.0, + 1767.0, + 1406.0, + 1804.0, + 1394.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 1797.0, + 1406.0, + 1797.0, + 1406.0, + 1838.0, + 759.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1829.0, + 1098.0, + 1829.0, + 1098.0, + 1866.0, + 292.0, + 1866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1254.0, + 1829.0, + 1286.0, + 1829.0, + 1286.0, + 1866.0, + 1254.0, + 1866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1306.0, + 1829.0, + 1405.0, + 1829.0, + 1405.0, + 1866.0, + 1306.0, + 1866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1858.0, + 650.0, + 1858.0, + 650.0, + 1898.0, + 294.0, + 1898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 673.0, + 1858.0, + 1098.0, + 1858.0, + 1098.0, + 1898.0, + 673.0, + 1898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1905.0, + 1074.0, + 1905.0, + 1074.0, + 1943.0, + 294.0, + 1943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 1905.0, + 1383.0, + 1905.0, + 1383.0, + 1943.0, + 1095.0, + 1943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1905.0, + 1405.0, + 1905.0, + 1405.0, + 1943.0, + 1402.0, + 1943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1939.0, + 529.0, + 1939.0, + 529.0, + 1977.0, + 292.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 549.0, + 1939.0, + 1204.0, + 1939.0, + 1204.0, + 1977.0, + 549.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1223.0, + 1939.0, + 1255.0, + 1939.0, + 1255.0, + 1977.0, + 1223.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1405.0, + 1939.0, + 1408.0, + 1939.0, + 1408.0, + 1977.0, + 1405.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1969.0, + 296.0, + 1969.0, + 296.0, + 2013.0, + 290.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 634.0, + 1969.0, + 1332.0, + 1969.0, + 1332.0, + 2013.0, + 634.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 1969.0, + 1409.0, + 1969.0, + 1409.0, + 2013.0, + 1354.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 2000.0, + 535.0, + 2000.0, + 535.0, + 2040.0, + 292.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 2000.0, + 1255.0, + 2000.0, + 1255.0, + 2040.0, + 554.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1295.0, + 2000.0, + 1343.0, + 2000.0, + 1343.0, + 2040.0, + 1295.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 2000.0, + 1409.0, + 2000.0, + 1409.0, + 2040.0, + 1398.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1208.0, + 647.0, + 1208.0, + 647.0, + 1240.0, + 297.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 1206.0, + 1404.0, + 1206.0, + 1404.0, + 1242.0, + 660.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1236.0, + 926.0, + 1236.0, + 926.0, + 1275.0, + 294.0, + 1275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1081.0, + 1236.0, + 1141.0, + 1236.0, + 1141.0, + 1275.0, + 1081.0, + 1275.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1071, + 1407, + 1071, + 1407, + 1562, + 296, + 1562 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 296, + 1645, + 1405, + 1645, + 1405, + 1866, + 296, + 1866 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 296, + 351, + 1405, + 351, + 1405, + 542, + 296, + 542 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 296, + 794, + 1406, + 794, + 1406, + 1050, + 296, + 1050 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 567, + 1406, + 567, + 1406, + 668, + 297, + 668 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 297, + 1950, + 1406, + 1950, + 1406, + 2044, + 297, + 2044 + ], + "score": 0.973 + }, + { + "category_id": 8, + "poly": [ + 500, + 672, + 1198, + 672, + 1198, + 760, + 500, + 760 + ], + "score": 0.954 + }, + { + "category_id": 8, + "poly": [ + 453, + 286, + 1244, + 286, + 1244, + 327, + 453, + 327 + ], + "score": 0.932 + }, + { + "category_id": 1, + "poly": [ + 296, + 228, + 931, + 228, + 931, + 263, + 296, + 263 + ], + "score": 0.928 + }, + { + "category_id": 8, + "poly": [ + 386, + 1889, + 1304, + 1889, + 1304, + 1932, + 386, + 1932 + ], + "score": 0.919 + }, + { + "category_id": 8, + "poly": [ + 381, + 1584, + 1314, + 1584, + 1314, + 1623, + 381, + 1623 + ], + "score": 0.911 + }, + { + "category_id": 9, + "poly": [ + 1366, + 725, + 1400, + 725, + 1400, + 754, + 1366, + 754 + ], + "score": 0.882 + }, + { + "category_id": 9, + "poly": [ + 1366, + 290, + 1400, + 290, + 1400, + 319, + 1366, + 319 + ], + "score": 0.879 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1894, + 1400, + 1894, + 1400, + 1925, + 1366, + 1925 + ], + "score": 0.875 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1589, + 1400, + 1589, + 1400, + 1618, + 1366, + 1618 + ], + "score": 0.863 + }, + { + "category_id": 2, + "poly": [ + 297, + 74, + 854, + 74, + 854, + 106, + 297, + 106 + ], + "score": 0.76 + }, + { + "category_id": 2, + "poly": [ + 841, + 2087, + 859, + 2087, + 859, + 2112, + 841, + 2112 + ], + "score": 0.748 + }, + { + "category_id": 2, + "poly": [ + 297, + 74, + 854, + 74, + 854, + 106, + 297, + 106 + ], + "score": 0.216 + }, + { + "category_id": 13, + "poly": [ + 760, + 1164, + 886, + 1164, + 886, + 1198, + 760, + 1198 + ], + "score": 0.94, + "latex": "S _ { h , \\Phi } ( t _ { a } | x )" + }, + { + "category_id": 13, + "poly": [ + 1105, + 1351, + 1258, + 1351, + 1258, + 1381, + 1105, + 1381 + ], + "score": 0.93, + "latex": "\\{ U _ { h } , W _ { h } , b _ { h } \\}" + }, + { + "category_id": 13, + "poly": [ + 411, + 604, + 664, + 604, + 664, + 640, + 411, + 640 + ], + "score": 0.93, + "latex": "p _ { \\Phi } ^ { A = a } \\triangleq p _ { \\Phi } ( R | A = a )" + }, + { + "category_id": 13, + "poly": [ + 414, + 922, + 564, + 922, + 564, + 957, + 414, + 957 + ], + "score": 0.93, + "latex": "\\{ h _ { a } ( \\cdot ) , \\Phi ( \\cdot ) \\}" + }, + { + "category_id": 13, + "poly": [ + 898, + 569, + 1123, + 569, + 1123, + 605, + 898, + 605 + ], + "score": 0.93, + "latex": "\\alpha ^ { - 1 } \\zeta _ { h , \\Phi } ( x , a ) \\in G" + }, + { + "category_id": 13, + "poly": [ + 799, + 1711, + 907, + 1711, + 907, + 1744, + 799, + 1744 + ], + "score": 0.93, + "latex": "h _ { A } ( r , \\tilde { \\epsilon } _ { h } )" + }, + { + "category_id": 13, + "poly": [ + 1178, + 2012, + 1325, + 2012, + 1325, + 2045, + 1178, + 2045 + ], + "score": 0.92, + "latex": "p ( Y , \\delta , X , A )" + }, + { + "category_id": 13, + "poly": [ + 963, + 1712, + 1023, + 1712, + 1023, + 1744, + 963, + 1744 + ], + "score": 0.92, + "latex": "\\Phi ( x )" + }, + { + "category_id": 13, + "poly": [ + 903, + 1103, + 1173, + 1103, + 1173, + 1138, + 903, + 1138 + ], + "score": 0.91, + "latex": "t _ { a } \\sim \\stackrel { \\sim } { p } _ { h , \\Phi } ( T _ { A } | X = x )" + }, + { + "category_id": 13, + "poly": [ + 297, + 411, + 586, + 411, + 586, + 448, + 297, + 448 + ], + "score": 0.91, + "latex": "u \\cdot \\mathcal { L } _ { \\mathrm { F } } ^ { A = 1 } + \\left( 1 - u \\right) \\cdot \\mathcal { L } _ { \\mathrm { F } } ^ { A = 0 }" + }, + { + "category_id": 13, + "poly": [ + 934, + 479, + 988, + 479, + 988, + 509, + 934, + 509 + ], + "score": 0.91, + "latex": "\\mathcal { L } _ { \\mathrm { C F } }" + }, + { + "category_id": 13, + "poly": [ + 632, + 231, + 809, + 231, + 809, + 262, + 632, + 262 + ], + "score": 0.91, + "latex": "\\mathcal { L } = \\mathcal { L } _ { \\mathrm { F } } + \\mathcal { L } _ { \\mathrm { C F } }" + }, + { + "category_id": 13, + "poly": [ + 346, + 350, + 540, + 350, + 540, + 386, + 346, + 386 + ], + "score": 0.91, + "latex": "u \\triangleq P ( A = 1 )" + }, + { + "category_id": 13, + "poly": [ + 654, + 1012, + 935, + 1012, + 935, + 1049, + 654, + 1049 + ], + "score": 0.91, + "latex": "\\mathrm { I P M } _ { G } ( p _ { \\Phi } ^ { A = 1 } , p _ { \\Phi } ^ { \\tilde { A = 0 } } ) \\stackrel { . } { = } 0" + }, + { + "category_id": 13, + "poly": [ + 1349, + 1291, + 1396, + 1291, + 1396, + 1320, + 1349, + 1320 + ], + "score": 0.91, + "latex": "p ( \\epsilon )" + }, + { + "category_id": 14, + "poly": [ + 500, + 669, + 1197, + 669, + 1197, + 764, + 500, + 764 + ], + "score": 0.91, + "latex": "\\begin{array} { r l } & { \\mathcal { L } _ { \\mathrm { C F } } \\leq ( 1 - u ) \\cdot \\mathcal { L } _ { \\mathrm { F } } ^ { A = 1 } + u \\cdot \\mathcal { L } _ { \\mathrm { F } } ^ { A = 0 } + \\alpha \\cdot \\mathrm { I P M } _ { G } ( p _ { \\Phi } ^ { A = 1 } , p _ { \\Phi } ^ { A = 0 } ) } \\\\ & { \\quad \\mathcal { L } \\leq \\mathcal { L } _ { \\mathrm { F } } ^ { A = 1 } + \\mathcal { L } _ { \\mathrm { F } } ^ { A = 0 } + \\alpha \\cdot \\mathrm { I P M } _ { G } ( p _ { \\Phi } ^ { A = 1 } , p _ { \\Phi } ^ { A = 0 } ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 701, + 509, + 754, + 509, + 754, + 539, + 701, + 539 + ], + "score": 0.9, + "latex": "\\mathcal { L } _ { \\mathrm { C F } }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1135, + 500, + 1135, + 500, + 1167, + 298, + 1167 + ], + "score": 0.9, + "latex": "p _ { \\nu , \\Phi } ( C _ { A } | X = x )" + }, + { + "category_id": 13, + "poly": [ + 1093, + 412, + 1393, + 412, + 1393, + 449, + 1093, + 449 + ], + "score": 0.9, + "latex": "\\mathcal { L } _ { \\mathrm { C F } } = \\dot { ( } 1 - u ) \\cdot \\mathcal { L } _ { \\mathrm { C F } } ^ { A = 1 } + u" + }, + { + "category_id": 13, + "poly": [ + 425, + 1982, + 492, + 1982, + 492, + 2010, + 425, + 2010 + ], + "score": 0.9, + "latex": "\\delta = 1" + }, + { + "category_id": 13, + "poly": [ + 542, + 570, + 594, + 570, + 594, + 604, + 542, + 604 + ], + "score": 0.9, + "latex": "\\Phi ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 298, + 447, + 365, + 447, + 365, + 482, + 298, + 482 + ], + "score": 0.9, + "latex": "\\mathcal { L } _ { \\mathrm { C F } } ^ { A = 0 }" + }, + { + "category_id": 13, + "poly": [ + 1203, + 1134, + 1402, + 1134, + 1402, + 1168, + 1203, + 1168 + ], + "score": 0.9, + "latex": "p _ { h , \\Phi } ( T _ { A } | X = x )" + }, + { + "category_id": 13, + "poly": [ + 1116, + 863, + 1250, + 863, + 1250, + 894, + 1116, + 894 + ], + "score": 0.9, + "latex": "g : m \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1286, + 488, + 1286, + 488, + 1321, + 297, + 1321 + ], + "score": 0.9, + "latex": "p _ { h , \\Phi } ( T _ { a } | X = x )" + }, + { + "category_id": 13, + "poly": [ + 1182, + 607, + 1254, + 607, + 1254, + 635, + 1182, + 635 + ], + "score": 0.89, + "latex": "\\alpha > 0" + }, + { + "category_id": 14, + "poly": [ + 451, + 285, + 1246, + 285, + 1246, + 326, + 451, + 326 + ], + "score": 0.89, + "latex": "\\begin{array} { r } { \\mathcal { L } _ { \\mathrm { F } } = \\mathbb { E } _ { ( x , a ) \\sim p ( A , X ) } \\zeta _ { h , \\Phi } ( x , a ) , \\quad \\mathcal { L } _ { \\mathrm { C F } } = \\mathbb { E } _ { ( x , a ) \\sim p ( 1 - A , X ) } \\zeta _ { h , \\Phi } ( x , a ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1338, + 1073, + 1403, + 1073, + 1403, + 1107, + 1338, + 1107 + ], + "score": 0.89, + "latex": "h _ { A } ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 1236, + 1982, + 1303, + 1982, + 1303, + 2009, + 1236, + 2009 + ], + "score": 0.89, + "latex": "\\delta = 0" + }, + { + "category_id": 13, + "poly": [ + 1048, + 608, + 1123, + 608, + 1123, + 636, + 1048, + 636 + ], + "score": 0.89, + "latex": "A = a" + }, + { + "category_id": 13, + "poly": [ + 347, + 1104, + 411, + 1104, + 411, + 1135, + 347, + 1135 + ], + "score": 0.89, + "latex": "\\nu _ { A } ( \\cdot )" + }, + { + "category_id": 14, + "poly": [ + 392, + 1886, + 1303, + 1886, + 1303, + 1931, + 392, + 1931 + ], + "score": 0.88, + "latex": "\\mathcal { L } _ { \\mathrm { F } } ^ { \\mathrm { C S A } } \\triangleq \\mathbb { E } _ { ( y , \\delta , x , a ) \\sim p ( Y , \\delta , X , A ) , \\epsilon \\sim p ( \\epsilon ) } \\left[ \\delta \\cdot \\left( \\left| y - t _ { a } \\right| \\right) + \\left( 1 - \\delta \\right) \\cdot \\left( \\operatorname* { m a x } ( 0 , y - t _ { a } ) \\right) \\right] ," + }, + { + "category_id": 13, + "poly": [ + 376, + 1647, + 595, + 1647, + 595, + 1683, + 376, + 1683 + ], + "score": 0.88, + "latex": "\\{ U _ { h } , W _ { h } \\} \\in \\mathbb { R } ^ { d \\times d }" + }, + { + "category_id": 13, + "poly": [ + 1067, + 1681, + 1097, + 1681, + 1097, + 1711, + 1067, + 1711 + ], + "score": 0.88, + "latex": "\\tilde { \\epsilon } _ { h }" + }, + { + "category_id": 13, + "poly": [ + 871, + 1955, + 897, + 1955, + 897, + 1981, + 871, + 1981 + ], + "score": 0.87, + "latex": "t _ { a }" + }, + { + "category_id": 13, + "poly": [ + 1332, + 384, + 1404, + 384, + 1404, + 413, + 1332, + 413 + ], + "score": 0.86, + "latex": "\\mathcal { L } _ { \\mathrm { F } } =" + }, + { + "category_id": 13, + "poly": [ + 414, + 860, + 1028, + 860, + 1028, + 899, + 414, + 899 + ], + "score": 0.86, + "latex": "\\begin{array} { r } { \\mathrm { I P M } _ { G } ( p , q ) \\triangleq \\operatorname* { s u p } _ { g \\in G } \\vert \\int _ { M } g ( m ) \\left( p ( m ) - q ( m ) \\right) d m \\vert } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 690, + 1985, + 714, + 1985, + 714, + 2012, + 690, + 2012 + ], + "score": 0.86, + "latex": "t _ { a }" + }, + { + "category_id": 13, + "poly": [ + 1339, + 1109, + 1404, + 1109, + 1404, + 1134, + 1339, + 1134 + ], + "score": 0.84, + "latex": "c _ { a } \\sim" + }, + { + "category_id": 13, + "poly": [ + 1224, + 1476, + 1243, + 1476, + 1243, + 1495, + 1224, + 1495 + ], + "score": 0.83, + "latex": "_ { i i }" + }, + { + "category_id": 13, + "poly": [ + 1346, + 1259, + 1404, + 1259, + 1404, + 1287, + 1346, + 1287 + ], + "score": 0.83, + "latex": "t _ { a } \\sim" + }, + { + "category_id": 13, + "poly": [ + 805, + 510, + 827, + 510, + 827, + 536, + 805, + 536 + ], + "score": 0.82, + "latex": "\\mathcal { L }" + }, + { + "category_id": 14, + "poly": [ + 384, + 1584, + 1316, + 1584, + 1316, + 1623, + 384, + 1623 + ], + "score": 0.82, + "latex": "{ \\widetilde { \\epsilon } } _ { h } = \\epsilon + U _ { h } \\operatorname { t a n h } ( W _ { h } \\epsilon + b _ { h } ) , \\quad \\epsilon \\sim \\operatorname { U n i f o r m } ( 0 , 1 ) , t _ { a } = h _ { A } ( r , { \\widetilde { \\epsilon } } _ { h } ) , \\quad r = \\Phi ( x )" + }, + { + "category_id": 13, + "poly": [ + 1375, + 830, + 1398, + 830, + 1398, + 856, + 1375, + 856 + ], + "score": 0.82, + "latex": "R" + }, + { + "category_id": 13, + "poly": [ + 953, + 954, + 994, + 954, + 994, + 986, + 953, + 986 + ], + "score": 0.81, + "latex": "( i i )" + }, + { + "category_id": 13, + "poly": [ + 611, + 1646, + 767, + 1646, + 767, + 1683, + 611, + 1683 + ], + "score": 0.81, + "latex": "\\{ b _ { h } , \\epsilon \\} \\in \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 828, + 834, + 846, + 834, + 846, + 860, + 828, + 860 + ], + "score": 0.81, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 1052, + 1957, + 1069, + 1957, + 1069, + 1982, + 1052, + 1982 + ], + "score": 0.8, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 939, + 1680, + 1000, + 1680, + 1000, + 1712, + 939, + 1712 + ], + "score": 0.8, + "latex": "( 0 , 1 )" + }, + { + "category_id": 13, + "poly": [ + 920, + 895, + 953, + 895, + 953, + 921, + 920, + 921 + ], + "score": 0.79, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 490, + 1773, + 507, + 1773, + 507, + 1799, + 490, + 1799 + ], + "score": 0.79, + "latex": "d" + }, + { + "category_id": 13, + "poly": [ + 898, + 833, + 916, + 833, + 916, + 860, + 898, + 860 + ], + "score": 0.79, + "latex": "q" + }, + { + "category_id": 13, + "poly": [ + 1215, + 572, + 1240, + 572, + 1240, + 599, + 1215, + 599 + ], + "score": 0.78, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 468, + 1687, + 483, + 1687, + 483, + 1707, + 468, + 1707 + ], + "score": 0.74, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 364, + 1777, + 380, + 1777, + 380, + 1799, + 364, + 1799 + ], + "score": 0.74, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 1067, + 830, + 1099, + 830, + 1099, + 856, + 1067, + 856 + ], + "score": 0.74, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 298, + 1779, + 314, + 1779, + 314, + 1799, + 298, + 1799 + ], + "score": 0.73, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 1371, + 923, + 1403, + 923, + 1403, + 957, + 1371, + 957 + ], + "score": 0.65, + "latex": "( i )" + }, + { + "category_id": 13, + "poly": [ + 781, + 1650, + 800, + 1650, + 800, + 1677, + 781, + 1677 + ], + "score": 0.65, + "latex": "d" + }, + { + "category_id": 13, + "poly": [ + 669, + 1717, + 684, + 1717, + 684, + 1738, + 669, + 1738 + ], + "score": 0.64, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 745, + 1442, + 758, + 1442, + 758, + 1467, + 745, + 1467 + ], + "score": 0.63, + "latex": "\\romannumeral 1" + }, + { + "category_id": 13, + "poly": [ + 444, + 1539, + 458, + 1539, + 458, + 1557, + 444, + 1557 + ], + "score": 0.62, + "latex": "\\epsilon" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 856.0, + 72.0, + 856.0, + 109.0, + 297.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2118.0, + 839.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 72.0, + 856.0, + 72.0, + 856.0, + 109.0, + 297.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1070.0, + 1337.0, + 1070.0, + 1337.0, + 1108.0, + 295.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 1070.0, + 1407.0, + 1070.0, + 1407.0, + 1108.0, + 1404.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1101.0, + 346.0, + 1101.0, + 346.0, + 1140.0, + 293.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 412.0, + 1101.0, + 902.0, + 1101.0, + 902.0, + 1140.0, + 412.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1174.0, + 1101.0, + 1338.0, + 1101.0, + 1338.0, + 1140.0, + 1174.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1405.0, + 1101.0, + 1409.0, + 1101.0, + 1409.0, + 1140.0, + 1405.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1131.0, + 297.0, + 1131.0, + 297.0, + 1170.0, + 290.0, + 1170.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 501.0, + 1131.0, + 1202.0, + 1131.0, + 1202.0, + 1170.0, + 501.0, + 1170.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1403.0, + 1131.0, + 1409.0, + 1131.0, + 1409.0, + 1170.0, + 1403.0, + 1170.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1164.0, + 759.0, + 1164.0, + 759.0, + 1198.0, + 293.0, + 1198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 887.0, + 1164.0, + 1405.0, + 1164.0, + 1405.0, + 1198.0, + 887.0, + 1198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1196.0, + 1404.0, + 1196.0, + 1404.0, + 1227.0, + 295.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1227.0, + 1405.0, + 1227.0, + 1405.0, + 1257.0, + 296.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1255.0, + 1345.0, + 1255.0, + 1345.0, + 1289.0, + 294.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1283.0, + 296.0, + 1283.0, + 296.0, + 1324.0, + 289.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1283.0, + 1348.0, + 1283.0, + 1348.0, + 1324.0, + 489.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1397.0, + 1283.0, + 1408.0, + 1283.0, + 1408.0, + 1324.0, + 1397.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1317.0, + 1409.0, + 1317.0, + 1409.0, + 1348.0, + 293.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1344.0, + 1104.0, + 1344.0, + 1104.0, + 1382.0, + 293.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1259.0, + 1344.0, + 1408.0, + 1344.0, + 1408.0, + 1382.0, + 1259.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1373.0, + 1407.0, + 1373.0, + 1407.0, + 1413.0, + 291.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1404.0, + 1405.0, + 1404.0, + 1405.0, + 1445.0, + 291.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1436.0, + 744.0, + 1436.0, + 744.0, + 1473.0, + 293.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 1436.0, + 1405.0, + 1436.0, + 1405.0, + 1473.0, + 759.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1468.0, + 1223.0, + 1468.0, + 1223.0, + 1502.0, + 294.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1244.0, + 1468.0, + 1405.0, + 1468.0, + 1405.0, + 1502.0, + 1244.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1497.0, + 1405.0, + 1497.0, + 1405.0, + 1533.0, + 294.0, + 1533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1531.0, + 443.0, + 1531.0, + 443.0, + 1562.0, + 295.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 459.0, + 1531.0, + 955.0, + 1531.0, + 955.0, + 1562.0, + 459.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1641.0, + 375.0, + 1641.0, + 375.0, + 1688.0, + 293.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 1641.0, + 610.0, + 1641.0, + 610.0, + 1688.0, + 596.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 1641.0, + 780.0, + 1641.0, + 780.0, + 1688.0, + 768.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 801.0, + 1641.0, + 1408.0, + 1641.0, + 1408.0, + 1688.0, + 801.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1678.0, + 467.0, + 1678.0, + 467.0, + 1715.0, + 293.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 484.0, + 1678.0, + 938.0, + 1678.0, + 938.0, + 1715.0, + 484.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1678.0, + 1066.0, + 1678.0, + 1066.0, + 1715.0, + 1001.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 1678.0, + 1405.0, + 1678.0, + 1405.0, + 1715.0, + 1098.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1709.0, + 668.0, + 1709.0, + 668.0, + 1748.0, + 294.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 1709.0, + 798.0, + 1709.0, + 798.0, + 1748.0, + 685.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 1709.0, + 962.0, + 1709.0, + 962.0, + 1748.0, + 908.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 1709.0, + 1407.0, + 1709.0, + 1407.0, + 1748.0, + 1024.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1740.0, + 1407.0, + 1740.0, + 1407.0, + 1774.0, + 294.0, + 1774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1773.0, + 297.0, + 1773.0, + 297.0, + 1804.0, + 294.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 315.0, + 1773.0, + 363.0, + 1773.0, + 363.0, + 1804.0, + 315.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1773.0, + 489.0, + 1773.0, + 489.0, + 1804.0, + 381.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 1773.0, + 1405.0, + 1773.0, + 1405.0, + 1804.0, + 508.0, + 1804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1803.0, + 1405.0, + 1803.0, + 1405.0, + 1834.0, + 294.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1832.0, + 1217.0, + 1832.0, + 1217.0, + 1867.0, + 295.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 346.0, + 345.0, + 346.0, + 345.0, + 391.0, + 291.0, + 391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 346.0, + 1406.0, + 346.0, + 1406.0, + 391.0, + 541.0, + 391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 378.0, + 1331.0, + 378.0, + 1331.0, + 422.0, + 291.0, + 422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 283.0, + 397.0, + 296.0, + 397.0, + 296.0, + 466.0, + 283.0, + 466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 397.0, + 1092.0, + 397.0, + 1092.0, + 466.0, + 587.0, + 466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 397.0, + 1415.0, + 397.0, + 1415.0, + 466.0, + 1394.0, + 466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 430.0, + 297.0, + 430.0, + 297.0, + 498.0, + 285.0, + 498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 366.0, + 430.0, + 1416.0, + 430.0, + 1416.0, + 498.0, + 366.0, + 498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 478.0, + 933.0, + 478.0, + 933.0, + 513.0, + 291.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 478.0, + 1407.0, + 478.0, + 1407.0, + 513.0, + 989.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 508.0, + 700.0, + 508.0, + 700.0, + 544.0, + 294.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 508.0, + 804.0, + 508.0, + 804.0, + 544.0, + 755.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 508.0, + 1237.0, + 508.0, + 1237.0, + 544.0, + 828.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 795.0, + 1406.0, + 795.0, + 1406.0, + 833.0, + 294.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 827.0, + 827.0, + 827.0, + 827.0, + 864.0, + 294.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 827.0, + 897.0, + 827.0, + 897.0, + 864.0, + 847.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 827.0, + 1066.0, + 827.0, + 1066.0, + 864.0, + 917.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 827.0, + 1374.0, + 827.0, + 1374.0, + 864.0, + 1100.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 827.0, + 1409.0, + 827.0, + 1409.0, + 864.0, + 1399.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 856.0, + 413.0, + 856.0, + 413.0, + 906.0, + 290.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 856.0, + 1115.0, + 856.0, + 1115.0, + 906.0, + 1029.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 856.0, + 1412.0, + 856.0, + 1412.0, + 906.0, + 1251.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 890.0, + 919.0, + 890.0, + 919.0, + 928.0, + 294.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 890.0, + 1407.0, + 890.0, + 1407.0, + 928.0, + 954.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 921.0, + 413.0, + 921.0, + 413.0, + 963.0, + 293.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 565.0, + 921.0, + 1370.0, + 921.0, + 1370.0, + 963.0, + 565.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 921.0, + 1409.0, + 921.0, + 1409.0, + 963.0, + 1404.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 953.0, + 952.0, + 953.0, + 952.0, + 991.0, + 294.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 995.0, + 953.0, + 1407.0, + 953.0, + 1407.0, + 991.0, + 995.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 983.0, + 1406.0, + 983.0, + 1406.0, + 1021.0, + 293.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 1007.0, + 653.0, + 1007.0, + 653.0, + 1059.0, + 285.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 1007.0, + 949.0, + 1007.0, + 949.0, + 1059.0, + 936.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 565.0, + 541.0, + 565.0, + 541.0, + 607.0, + 295.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 565.0, + 897.0, + 565.0, + 897.0, + 607.0, + 595.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 565.0, + 1214.0, + 565.0, + 1214.0, + 607.0, + 1124.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1241.0, + 565.0, + 1409.0, + 565.0, + 1409.0, + 607.0, + 1241.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 591.0, + 410.0, + 591.0, + 410.0, + 654.0, + 285.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 665.0, + 591.0, + 1047.0, + 591.0, + 1047.0, + 654.0, + 665.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 591.0, + 1181.0, + 591.0, + 1181.0, + 654.0, + 1124.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 591.0, + 1417.0, + 591.0, + 1417.0, + 654.0, + 1255.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 632.0, + 471.0, + 632.0, + 471.0, + 671.0, + 294.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1950.0, + 870.0, + 1950.0, + 870.0, + 1987.0, + 294.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 898.0, + 1950.0, + 1051.0, + 1950.0, + 1051.0, + 1987.0, + 898.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1070.0, + 1950.0, + 1406.0, + 1950.0, + 1406.0, + 1987.0, + 1070.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1981.0, + 424.0, + 1981.0, + 424.0, + 2015.0, + 295.0, + 2015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 493.0, + 1981.0, + 689.0, + 1981.0, + 689.0, + 2015.0, + 493.0, + 2015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 1981.0, + 1235.0, + 1981.0, + 1235.0, + 2015.0, + 715.0, + 2015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1304.0, + 1981.0, + 1407.0, + 1981.0, + 1407.0, + 2015.0, + 1304.0, + 2015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 2012.0, + 1177.0, + 2012.0, + 1177.0, + 2046.0, + 296.0, + 2046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1326.0, + 2012.0, + 1336.0, + 2012.0, + 1336.0, + 2046.0, + 1326.0, + 2046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 227.0, + 631.0, + 227.0, + 631.0, + 269.0, + 293.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 810.0, + 227.0, + 934.0, + 227.0, + 934.0, + 269.0, + 810.0, + 269.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 977, + 1407, + 977, + 1407, + 1283, + 297, + 1283 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 297, + 1293, + 1406, + 1293, + 1406, + 1737, + 297, + 1737 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 296, + 695, + 1406, + 695, + 1406, + 888, + 296, + 888 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 296, + 227, + 1404, + 227, + 1404, + 386, + 296, + 386 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1908, + 1405, + 1908, + 1405, + 2036, + 298, + 2036 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 583, + 1404, + 583, + 1404, + 683, + 298, + 683 + ], + "score": 0.975 + }, + { + "category_id": 8, + "poly": [ + 474, + 1800, + 1224, + 1800, + 1224, + 1884, + 474, + 1884 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 294, + 451, + 1401, + 451, + 1401, + 515, + 294, + 515 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 294, + 1749, + 1400, + 1749, + 1400, + 1785, + 294, + 1785 + ], + "score": 0.918 + }, + { + "category_id": 8, + "poly": [ + 373, + 398, + 1314, + 398, + 1314, + 440, + 373, + 440 + ], + "score": 0.914 + }, + { + "category_id": 0, + "poly": [ + 298, + 932, + 485, + 932, + 485, + 966, + 298, + 966 + ], + "score": 0.906 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 854, + 75, + 854, + 105, + 298, + 105 + ], + "score": 0.906 + }, + { + "category_id": 8, + "poly": [ + 306, + 528, + 1333, + 528, + 1333, + 569, + 306, + 569 + ], + "score": 0.888 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1826, + 1399, + 1826, + 1399, + 1856, + 1352, + 1856 + ], + "score": 0.872 + }, + { + "category_id": 9, + "poly": [ + 1366, + 532, + 1400, + 532, + 1400, + 563, + 1366, + 563 + ], + "score": 0.861 + }, + { + "category_id": 9, + "poly": [ + 1366, + 402, + 1400, + 402, + 1400, + 432, + 1366, + 432 + ], + "score": 0.853 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 859, + 2088, + 859, + 2112, + 840, + 2112 + ], + "score": 0.796 + }, + { + "category_id": 14, + "poly": [ + 475, + 1798, + 1222, + 1798, + 1222, + 1883, + 475, + 1883 + ], + "score": 0.95, + "latex": "\\mathrm { H R } ( t ) = \\frac { \\lambda _ { 1 } ( t ) } { \\lambda _ { 0 } ( t ) } = \\frac { S _ { 0 } ( t ) } { S _ { 1 } ( t ) } \\cdot \\frac { S _ { 1 } ^ { \\prime } ( t ) } { S _ { 0 } ^ { \\prime } ( t ) } , \\quad \\mathrm { H R } ( t ) = \\frac { \\hat { S } _ { 0 } ^ { \\mathrm { P K M } } ( t ) } { \\hat { S } _ { 1 } ^ { \\mathrm { P K M } } ( t ) } \\cdot \\frac { m _ { 1 } ( t ) } { m _ { 0 } ( t ) } ," + }, + { + "category_id": 13, + "poly": [ + 817, + 824, + 980, + 824, + 980, + 859, + 817, + 859 + ], + "score": 0.93, + "latex": "\\alpha ^ { - 1 } \\zeta _ { h , \\Phi } ( \\bar { x , } a )" + }, + { + "category_id": 13, + "poly": [ + 414, + 584, + 558, + 584, + 558, + 618, + 414, + 618 + ], + "score": 0.93, + "latex": "\\ell _ { \\mathrm { T C } } ( h , \\nu , \\Phi )" + }, + { + "category_id": 13, + "poly": [ + 896, + 1567, + 1355, + 1567, + 1355, + 1614, + 896, + 1614 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\mathrm { ~ \\cdot ~ } _ { \\mathrm { b } } ( t ) \\colon \\mathrm { H R } _ { \\mathrm { C o x P H } } ( t ) = \\frac { \\lambda ( t | a = 1 ) } { \\lambda ( t | a = 0 ) } = \\exp ( \\beta ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 298, + 352, + 402, + 352, + 402, + 386, + 298, + 386 + ], + "score": 0.93, + "latex": "\\nu _ { A } ( r , \\tilde { \\epsilon } _ { \\nu } )" + }, + { + "category_id": 13, + "poly": [ + 436, + 1704, + 521, + 1704, + 521, + 1738, + 436, + 1738 + ], + "score": 0.92, + "latex": "f ( t _ { 0 } | x )" + }, + { + "category_id": 13, + "poly": [ + 573, + 1704, + 699, + 1704, + 699, + 1737, + 573, + 1737 + ], + "score": 0.92, + "latex": "t _ { a } = h _ { A } ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 940, + 726, + 1007, + 726, + 1007, + 762, + 940, + 762 + ], + "score": 0.92, + "latex": "\\mathcal { L } _ { \\mathrm { F } } ^ { \\mathrm { C S A } }" + }, + { + "category_id": 13, + "poly": [ + 634, + 696, + 863, + 696, + 863, + 732, + 634, + 732 + ], + "score": 0.92, + "latex": "\\{ h _ { A } ( \\cdot ) , \\Phi ( \\cdot ) , \\nu _ { A } ( \\cdot ) \\}" + }, + { + "category_id": 13, + "poly": [ + 753, + 1945, + 812, + 1945, + 812, + 1976, + 753, + 1976 + ], + "score": 0.92, + "latex": "S ^ { \\prime } ( t )" + }, + { + "category_id": 13, + "poly": [ + 891, + 484, + 979, + 484, + 979, + 513, + 891, + 513 + ], + "score": 0.92, + "latex": "c _ { a } < t _ { a }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1704, + 384, + 1704, + 384, + 1737, + 298, + 1737 + ], + "score": 0.92, + "latex": "f ( t _ { 1 } | x )" + }, + { + "category_id": 13, + "poly": [ + 1346, + 2003, + 1397, + 2003, + 1397, + 2036, + 1346, + 2036 + ], + "score": 0.92, + "latex": "S ( t )" + }, + { + "category_id": 13, + "poly": [ + 1124, + 484, + 1213, + 484, + 1213, + 513, + 1124, + 513 + ], + "score": 0.92, + "latex": "t \\sb a < c \\sb a" + }, + { + "category_id": 13, + "poly": [ + 1130, + 1943, + 1182, + 1943, + 1182, + 1975, + 1130, + 1975 + ], + "score": 0.92, + "latex": "S ( t )" + }, + { + "category_id": 13, + "poly": [ + 1211, + 1972, + 1333, + 1972, + 1333, + 2006, + 1211, + 2006 + ], + "score": 0.92, + "latex": "S ^ { \\prime } ( t ) \\approx m" + }, + { + "category_id": 13, + "poly": [ + 687, + 1476, + 959, + 1476, + 959, + 1510, + 687, + 1510 + ], + "score": 0.92, + "latex": "\\lambda ( t | a ) = \\lambda _ { \\mathrm { b } } \\bar { ( } t ) \\exp ( a \\beta )" + }, + { + "category_id": 13, + "poly": [ + 391, + 1943, + 443, + 1943, + 443, + 1975, + 391, + 1975 + ], + "score": 0.91, + "latex": "S ( t )" + }, + { + "category_id": 13, + "poly": [ + 843, + 321, + 995, + 321, + 995, + 354, + 843, + 354 + ], + "score": 0.91, + "latex": "\\{ U _ { \\nu } , W _ { \\nu } , b _ { \\nu } \\}" + }, + { + "category_id": 13, + "poly": [ + 849, + 2004, + 908, + 2004, + 908, + 2036, + 849, + 2036 + ], + "score": 0.91, + "latex": "S ^ { \\prime } ( t )" + }, + { + "category_id": 13, + "poly": [ + 873, + 353, + 1129, + 353, + 1129, + 387, + 873, + 387 + ], + "score": 0.91, + "latex": "c _ { a } \\sim p _ { \\nu , \\Phi } ( C _ { A } | X = x )" + }, + { + "category_id": 13, + "poly": [ + 934, + 1973, + 1120, + 1973, + 1120, + 2006, + 934, + 2006 + ], + "score": 0.91, + "latex": "S ( t ) = m \\cdot t + c ." + }, + { + "category_id": 13, + "poly": [ + 884, + 1909, + 1083, + 1909, + 1083, + 1945, + 884, + 1945 + ], + "score": 0.91, + "latex": "S ^ { \\prime } ( t ) \\triangleq d S ( t ) / d t" + }, + { + "category_id": 13, + "poly": [ + 1287, + 1747, + 1361, + 1747, + 1361, + 1784, + 1287, + 1784 + ], + "score": 0.9, + "latex": "\\mathrm { \\hat { H R } } ( t )" + }, + { + "category_id": 13, + "poly": [ + 1038, + 262, + 1115, + 262, + 1115, + 290, + 1038, + 290 + ], + "score": 0.9, + "latex": "\\delta = 0" + }, + { + "category_id": 13, + "poly": [ + 920, + 1355, + 1101, + 1355, + 1101, + 1388, + 920, + 1388 + ], + "score": 0.9, + "latex": "\\mathrm { H R } ( t ) < 1 , > 1" + }, + { + "category_id": 13, + "poly": [ + 1008, + 482, + 1075, + 482, + 1075, + 510, + 1008, + 510 + ], + "score": 0.9, + "latex": "\\delta = 0" + }, + { + "category_id": 13, + "poly": [ + 591, + 292, + 670, + 292, + 670, + 320, + 591, + 320 + ], + "score": 0.9, + "latex": "\\delta \\ : = \\ : 1" + }, + { + "category_id": 13, + "poly": [ + 1242, + 482, + 1308, + 482, + 1308, + 510, + 1242, + 510 + ], + "score": 0.9, + "latex": "\\delta = 1" + }, + { + "category_id": 13, + "poly": [ + 298, + 759, + 437, + 759, + 437, + 798, + 298, + 798 + ], + "score": 0.88, + "latex": "\\mathcal { L } _ { \\mathrm { F } } ^ { \\mathrm { C S A - I N F O } }" + }, + { + "category_id": 14, + "poly": [ + 304, + 527, + 1344, + 527, + 1344, + 569, + 304, + 569 + ], + "score": 0.88, + "latex": "\\ell _ { \\mathrm { T C } } ( h , \\nu , \\Phi ) = \\mathbb { E } _ { ( \\delta , x , a ) \\sim p ( \\delta , X , A ) , \\epsilon \\sim p ( \\epsilon ) } \\left[ \\delta \\cdot ( \\operatorname* { m a x } ( 0 , t _ { a } - c _ { a } ) ) + ( 1 - \\delta ) \\cdot ( \\operatorname* { m a x } ( 0 , c _ { a } - t _ { a } ) ) \\right] ." + }, + { + "category_id": 13, + "poly": [ + 409, + 1911, + 485, + 1911, + 485, + 1944, + 409, + 1944 + ], + "score": 0.88, + "latex": "\\mathrm { H R } ( t )" + }, + { + "category_id": 14, + "poly": [ + 375, + 398, + 1323, + 398, + 1323, + 439, + 375, + 439 + ], + "score": 0.88, + "latex": "\\ell _ { c } ( \\nu , \\Phi ) = \\mathbb { E } _ { ( y , \\delta , x , a ) \\sim p ( y , \\delta , X , A ) , \\epsilon \\sim p ( \\epsilon ) } \\left[ ( 1 - \\delta ) \\cdot ( | y - c _ { a } | ) + \\delta \\cdot ( \\operatorname* { m a x } ( 0 , y - c _ { a } ) ) \\right] ." + }, + { + "category_id": 13, + "poly": [ + 1223, + 1294, + 1298, + 1294, + 1298, + 1327, + 1223, + 1327 + ], + "score": 0.88, + "latex": "\\mathrm { H R } ( t )" + }, + { + "category_id": 13, + "poly": [ + 1143, + 1614, + 1217, + 1614, + 1217, + 1646, + 1143, + 1646 + ], + "score": 0.85, + "latex": "\\mathrm { H R } ( t )" + }, + { + "category_id": 13, + "poly": [ + 959, + 266, + 988, + 266, + 988, + 292, + 959, + 292 + ], + "score": 0.85, + "latex": "c _ { a }" + }, + { + "category_id": 13, + "poly": [ + 1017, + 1539, + 1038, + 1539, + 1038, + 1569, + 1017, + 1569 + ], + "score": 0.84, + "latex": "\\beta" + }, + { + "category_id": 13, + "poly": [ + 1329, + 266, + 1358, + 266, + 1358, + 293, + 1329, + 293 + ], + "score": 0.83, + "latex": "c _ { a }" + }, + { + "category_id": 13, + "poly": [ + 847, + 732, + 871, + 732, + 871, + 756, + 847, + 756 + ], + "score": 0.82, + "latex": "\\mathcal { D }" + }, + { + "category_id": 13, + "poly": [ + 636, + 644, + 1014, + 644, + 1014, + 683, + 636, + 683 + ], + "score": 0.82, + "latex": "\\mathcal { L } _ { \\mathrm { F } } ^ { \\mathrm { C S A - I N F O } } \\triangleq \\mathcal { L } _ { \\mathrm { F } } ^ { \\mathrm { C S A } } + \\ell _ { \\mathrm { c } } + \\ell _ { \\mathrm { T C } }" + }, + { + "category_id": 13, + "poly": [ + 1139, + 1356, + 1186, + 1356, + 1186, + 1383, + 1139, + 1383 + ], + "score": 0.81, + "latex": "\\approx 1" + }, + { + "category_id": 13, + "poly": [ + 1327, + 1415, + 1403, + 1415, + 1403, + 1448, + 1327, + 1448 + ], + "score": 0.8, + "latex": "\\mathrm { H R } ( { \\dot { t } } )" + }, + { + "category_id": 13, + "poly": [ + 522, + 297, + 542, + 297, + 542, + 323, + 522, + 323 + ], + "score": 0.79, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 344, + 860, + 366, + 860, + 366, + 883, + 344, + 883 + ], + "score": 0.78, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 1218, + 1386, + 1293, + 1386, + 1293, + 1418, + 1218, + 1418 + ], + "score": 0.76, + "latex": "\\mathrm { H R } ( t )" + }, + { + "category_id": 13, + "poly": [ + 917, + 1389, + 931, + 1389, + 931, + 1413, + 917, + 1413 + ], + "score": 0.62, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 881, + 1573, + 944, + 1573, + 944, + 1608, + 881, + 1608 + ], + "score": 0.53, + "latex": "\\lambda _ { \\mathrm { b } } ( t )" + }, + { + "category_id": 13, + "poly": [ + 637, + 643, + 776, + 643, + 776, + 683, + 637, + 683 + ], + "score": 0.35, + "latex": "{ \\mathcal { L } } _ { \\mathrm { F } } ^ { \\mathrm { C S A - I N F O } }" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 926.0, + 489.0, + 926.0, + 489.0, + 973.0, + 291.0, + 973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2086.0, + 862.0, + 2086.0, + 862.0, + 2116.0, + 840.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 978.0, + 1410.0, + 978.0, + 1410.0, + 1013.0, + 295.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1007.0, + 1407.0, + 1007.0, + 1407.0, + 1044.0, + 294.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1037.0, + 1408.0, + 1037.0, + 1408.0, + 1073.0, + 295.0, + 1073.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1069.0, + 1407.0, + 1069.0, + 1407.0, + 1105.0, + 294.0, + 1105.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1097.0, + 1405.0, + 1097.0, + 1405.0, + 1135.0, + 295.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1130.0, + 1407.0, + 1130.0, + 1407.0, + 1166.0, + 294.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1159.0, + 1409.0, + 1159.0, + 1409.0, + 1196.0, + 292.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1190.0, + 1405.0, + 1190.0, + 1405.0, + 1226.0, + 292.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1220.0, + 1407.0, + 1220.0, + 1407.0, + 1257.0, + 292.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1249.0, + 745.0, + 1249.0, + 745.0, + 1287.0, + 293.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1293.0, + 1222.0, + 1293.0, + 1222.0, + 1331.0, + 294.0, + 1331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1299.0, + 1293.0, + 1406.0, + 1293.0, + 1406.0, + 1331.0, + 1299.0, + 1331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1324.0, + 1406.0, + 1324.0, + 1406.0, + 1361.0, + 294.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1354.0, + 919.0, + 1354.0, + 919.0, + 1391.0, + 295.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1102.0, + 1354.0, + 1138.0, + 1354.0, + 1138.0, + 1391.0, + 1102.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1187.0, + 1354.0, + 1406.0, + 1354.0, + 1406.0, + 1391.0, + 1187.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1383.0, + 916.0, + 1383.0, + 916.0, + 1423.0, + 294.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 932.0, + 1383.0, + 1217.0, + 1383.0, + 1217.0, + 1423.0, + 932.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1294.0, + 1383.0, + 1406.0, + 1383.0, + 1406.0, + 1423.0, + 1294.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1416.0, + 1326.0, + 1416.0, + 1326.0, + 1452.0, + 294.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1446.0, + 1404.0, + 1446.0, + 1404.0, + 1481.0, + 292.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1476.0, + 686.0, + 1476.0, + 686.0, + 1512.0, + 294.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 960.0, + 1476.0, + 1404.0, + 1476.0, + 1404.0, + 1512.0, + 960.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1508.0, + 1404.0, + 1508.0, + 1404.0, + 1541.0, + 296.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1536.0, + 1016.0, + 1536.0, + 1016.0, + 1573.0, + 294.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1039.0, + 1536.0, + 1406.0, + 1536.0, + 1406.0, + 1573.0, + 1039.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1566.0, + 880.0, + 1566.0, + 880.0, + 1616.0, + 291.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1356.0, + 1566.0, + 1406.0, + 1566.0, + 1406.0, + 1616.0, + 1356.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1612.0, + 1142.0, + 1612.0, + 1142.0, + 1647.0, + 294.0, + 1647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 1612.0, + 1406.0, + 1612.0, + 1406.0, + 1647.0, + 1218.0, + 1647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1639.0, + 1406.0, + 1639.0, + 1406.0, + 1681.0, + 291.0, + 1681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1671.0, + 1408.0, + 1671.0, + 1408.0, + 1710.0, + 292.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 1703.0, + 435.0, + 1703.0, + 435.0, + 1740.0, + 385.0, + 1740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 1703.0, + 572.0, + 1703.0, + 572.0, + 1740.0, + 522.0, + 1740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 1703.0, + 810.0, + 1703.0, + 810.0, + 1740.0, + 700.0, + 1740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 697.0, + 633.0, + 697.0, + 633.0, + 734.0, + 294.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 864.0, + 697.0, + 1406.0, + 697.0, + 1406.0, + 734.0, + 864.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 720.0, + 846.0, + 720.0, + 846.0, + 769.0, + 290.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 720.0, + 939.0, + 720.0, + 939.0, + 769.0, + 872.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1008.0, + 720.0, + 1409.0, + 720.0, + 1409.0, + 769.0, + 1008.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 746.0, + 297.0, + 746.0, + 297.0, + 815.0, + 285.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 746.0, + 1416.0, + 746.0, + 1416.0, + 815.0, + 438.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 794.0, + 1408.0, + 794.0, + 1408.0, + 831.0, + 294.0, + 831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 824.0, + 816.0, + 824.0, + 816.0, + 861.0, + 295.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 981.0, + 824.0, + 1408.0, + 824.0, + 1408.0, + 861.0, + 981.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 854.0, + 343.0, + 854.0, + 343.0, + 891.0, + 294.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 367.0, + 854.0, + 1319.0, + 854.0, + 1319.0, + 891.0, + 367.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 226.0, + 1405.0, + 226.0, + 1405.0, + 267.0, + 294.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 258.0, + 958.0, + 258.0, + 958.0, + 298.0, + 293.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 258.0, + 1037.0, + 258.0, + 1037.0, + 298.0, + 989.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1116.0, + 258.0, + 1328.0, + 258.0, + 1328.0, + 298.0, + 1116.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1359.0, + 258.0, + 1406.0, + 258.0, + 1406.0, + 298.0, + 1359.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 290.0, + 521.0, + 290.0, + 521.0, + 328.0, + 295.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 290.0, + 590.0, + 290.0, + 590.0, + 328.0, + 543.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 671.0, + 290.0, + 1405.0, + 290.0, + 1405.0, + 328.0, + 671.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 320.0, + 842.0, + 320.0, + 842.0, + 358.0, + 294.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 320.0, + 1406.0, + 320.0, + 1406.0, + 358.0, + 996.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 352.0, + 297.0, + 352.0, + 297.0, + 390.0, + 294.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 352.0, + 872.0, + 352.0, + 872.0, + 390.0, + 403.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 352.0, + 1295.0, + 352.0, + 1295.0, + 390.0, + 1130.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1905.0, + 408.0, + 1905.0, + 408.0, + 1951.0, + 292.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 486.0, + 1905.0, + 883.0, + 1905.0, + 883.0, + 1951.0, + 486.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1084.0, + 1905.0, + 1410.0, + 1905.0, + 1410.0, + 1951.0, + 1084.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1941.0, + 390.0, + 1941.0, + 390.0, + 1978.0, + 293.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 444.0, + 1941.0, + 752.0, + 1941.0, + 752.0, + 1978.0, + 444.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 813.0, + 1941.0, + 1129.0, + 1941.0, + 1129.0, + 1978.0, + 813.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 1941.0, + 1406.0, + 1941.0, + 1406.0, + 1978.0, + 1183.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1973.0, + 933.0, + 1973.0, + 933.0, + 2006.0, + 295.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 1973.0, + 1210.0, + 1973.0, + 1210.0, + 2006.0, + 1121.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1334.0, + 1973.0, + 1405.0, + 1973.0, + 1405.0, + 2006.0, + 1334.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2002.0, + 848.0, + 2002.0, + 848.0, + 2039.0, + 293.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 909.0, + 2002.0, + 1345.0, + 2002.0, + 1345.0, + 2039.0, + 909.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 2002.0, + 1408.0, + 2002.0, + 1408.0, + 2039.0, + 1398.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 577.0, + 413.0, + 577.0, + 413.0, + 624.0, + 291.0, + 624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 577.0, + 1406.0, + 577.0, + 1406.0, + 624.0, + 559.0, + 624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 610.0, + 1404.0, + 610.0, + 1404.0, + 652.0, + 293.0, + 652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 628.0, + 635.0, + 628.0, + 635.0, + 699.0, + 284.0, + 699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1015.0, + 628.0, + 1034.0, + 628.0, + 1034.0, + 699.0, + 1015.0, + 699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 448.0, + 1404.0, + 448.0, + 1404.0, + 486.0, + 293.0, + 486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 480.0, + 890.0, + 480.0, + 890.0, + 516.0, + 296.0, + 516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 480.0, + 1007.0, + 480.0, + 1007.0, + 516.0, + 980.0, + 516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1076.0, + 480.0, + 1123.0, + 480.0, + 1123.0, + 516.0, + 1076.0, + 516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1214.0, + 480.0, + 1241.0, + 480.0, + 1241.0, + 516.0, + 1214.0, + 516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1309.0, + 480.0, + 1373.0, + 480.0, + 1373.0, + 516.0, + 1309.0, + 516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1744.0, + 1286.0, + 1744.0, + 1286.0, + 1791.0, + 293.0, + 1791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1362.0, + 1744.0, + 1405.0, + 1744.0, + 1405.0, + 1791.0, + 1362.0, + 1791.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1683, + 1406, + 1683, + 1406, + 2030, + 297, + 2030 + ], + "score": 0.985 + }, + { + "category_id": 5, + "poly": [ + 300, + 297, + 1399, + 297, + 1399, + 585, + 300, + 585 + ], + "score": 0.982, + "html": "
MethodCausalFactualC-Slope (A=0,A=1)
EPEHEEATEHR(t)C-Index (A=0,A=1)Mean COV
CoxPH-UniformNANA0.97(0.86,1.09)NANANA
CoxPH-IPWNANA0.48(0.03,7.21)NANANA
CoxPH-OWNANA0.60(0.53,0.68)NANANA
Surv-BART352.0777.890.0(0.0,0.0)(0.706,0.686)0.001(0.398,0)
AFT-Weibull367.92133.930.47(0.47,0.47)(0.21,0.267)6.209(0.707,0.729)
AFT-log-Normal377.76157.640.47(0.47,0.47)(0.675, 0.556)6.971(0.707,0.729)
SR369.4788.550.38(0.330.65)(0.791,0.744)0(0.985,1.027)
CSA (proposed)358.720.80.45(0.39,0.65)(0.787,0.767)0.131(0.985,1.026)
CSA-INFO (proposed)344.331.190.53(0.41,0.67)(0.78,0.764)0.13(0.999,1.029)
" + }, + { + "category_id": 1, + "poly": [ + 297, + 1455, + 1404, + 1455, + 1404, + 1673, + 297, + 1673 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 611, + 1406, + 611, + 1406, + 909, + 297, + 909 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 922, + 1404, + 922, + 1404, + 1078, + 298, + 1078 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 295, + 1300, + 1403, + 1300, + 1403, + 1370, + 295, + 1370 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 297, + 1095, + 1406, + 1095, + 1406, + 1283, + 297, + 1283 + ], + "score": 0.937 + }, + { + "category_id": 0, + "poly": [ + 300, + 1400, + 559, + 1400, + 559, + 1436, + 300, + 1436 + ], + "score": 0.901 + }, + { + "category_id": 6, + "poly": [ + 297, + 224, + 1401, + 224, + 1401, + 290, + 297, + 290 + ], + "score": 0.887 + }, + { + "category_id": 2, + "poly": [ + 841, + 2087, + 858, + 2087, + 858, + 2111, + 841, + 2111 + ], + "score": 0.771 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 854, + 75, + 854, + 105, + 298, + 105 + ], + "score": 0.603 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 854, + 75, + 854, + 105, + 298, + 105 + ], + "score": 0.315 + }, + { + "category_id": 13, + "poly": [ + 345, + 876, + 434, + 876, + 434, + 910, + 345, + 910 + ], + "score": 0.95, + "latex": "S ( t _ { a } | x )" + }, + { + "category_id": 13, + "poly": [ + 605, + 1248, + 843, + 1248, + 843, + 1283, + 605, + 1283 + ], + "score": 0.94, + "latex": "\\gamma ( T _ { A } ) = I \\{ T _ { A } > \\tau \\}" + }, + { + "category_id": 13, + "poly": [ + 518, + 774, + 624, + 774, + 624, + 813, + 518, + 813 + ], + "score": 0.93, + "latex": "\\hat { S } _ { a } ^ { \\mathrm { P K M } } ( t )" + }, + { + "category_id": 13, + "poly": [ + 490, + 703, + 596, + 703, + 596, + 741, + 490, + 741 + ], + "score": 0.93, + "latex": "\\hat { S } ^ { \\mathrm { P K M } } ( t )" + }, + { + "category_id": 13, + "poly": [ + 893, + 1578, + 1026, + 1578, + 1026, + 1612, + 893, + 1612 + ], + "score": 0.93, + "latex": "\\{ Y _ { \\mathrm { C F } } , \\delta _ { \\mathrm { C F } } \\}" + }, + { + "category_id": 13, + "poly": [ + 1316, + 846, + 1403, + 846, + 1403, + 880, + 1316, + 880 + ], + "score": 0.93, + "latex": "f ( t _ { a } | x )" + }, + { + "category_id": 13, + "poly": [ + 553, + 1129, + 996, + 1129, + 996, + 1185, + 553, + 1185 + ], + "score": 0.92, + "latex": "\\epsilon _ { \\mathrm { P E H E } } = \\sqrt { \\mathbb { E } _ { X } [ ( \\mathrm { I T E } ( x ) - \\mathrm { I } \\hat { \\mathrm { T E } } ( x ) ) ^ { 2 } ] }" + }, + { + "category_id": 13, + "poly": [ + 1097, + 1332, + 1334, + 1332, + 1334, + 1370, + 1097, + 1370 + ], + "score": 0.92, + "latex": "\\mathrm { A } \\hat { \\mathrm { T E } } = \\mathbb { E } _ { X } [ \\hat { \\mathrm { I T E } } ( x ) ]" + }, + { + "category_id": 13, + "poly": [ + 505, + 1640, + 683, + 1640, + 683, + 1671, + 505, + 1671 + ], + "score": 0.92, + "latex": "\\mathcal { L } = \\mathcal { L } _ { \\mathrm { F } } + \\mathcal { L } _ { \\mathrm { C F } }" + }, + { + "category_id": 13, + "poly": [ + 713, + 1014, + 812, + 1014, + 812, + 1048, + 713, + 1048 + ], + "score": 0.92, + "latex": "\\operatorname { H R } ( t | x )" + }, + { + "category_id": 13, + "poly": [ + 1197, + 775, + 1296, + 775, + 1296, + 813, + 1197, + 813 + ], + "score": 0.92, + "latex": "\\mathrm { H } \\mathrm { \\hat { R } } ( t | x )" + }, + { + "category_id": 13, + "poly": [ + 815, + 1218, + 1076, + 1218, + 1076, + 1252, + 815, + 1252 + ], + "score": 0.92, + "latex": "\\dot { t _ { a } } \\sim p _ { h , \\Phi } ( T _ { A } | X = x )" + }, + { + "category_id": 13, + "poly": [ + 413, + 612, + 473, + 612, + 473, + 646, + 413, + 646 + ], + "score": 0.92, + "latex": "S ^ { \\prime } ( t )" + }, + { + "category_id": 13, + "poly": [ + 827, + 954, + 925, + 954, + 925, + 987, + 827, + 987 + ], + "score": 0.92, + "latex": "\\mathrm { H R } ( t | x )" + }, + { + "category_id": 13, + "poly": [ + 645, + 1335, + 882, + 1335, + 882, + 1370, + 645, + 1370 + ], + "score": 0.92, + "latex": "\\mathrm { A T E } = \\mathbb { E } _ { X } [ \\mathrm { I T E } ( x ) ]" + }, + { + "category_id": 13, + "poly": [ + 297, + 1332, + 558, + 1332, + 558, + 1370, + 297, + 1370 + ], + "score": 0.92, + "latex": "\\epsilon _ { \\mathrm { A T E } } = | \\mathrm { A T E } - \\mathrm { A \\hat { T } E } |" + }, + { + "category_id": 13, + "poly": [ + 965, + 642, + 1018, + 642, + 1018, + 675, + 965, + 675 + ], + "score": 0.91, + "latex": "S ( t )" + }, + { + "category_id": 13, + "poly": [ + 1079, + 811, + 1177, + 811, + 1177, + 849, + 1079, + 849 + ], + "score": 0.91, + "latex": "\\mathrm { H } \\mathrm { \\hat { R } } ( t | x )" + }, + { + "category_id": 13, + "poly": [ + 296, + 1183, + 750, + 1183, + 750, + 1221, + 296, + 1221 + ], + "score": 0.91, + "latex": "\\mathrm { I } \\hat { \\mathrm { T E } } ( x ) = \\mathbb { E } _ { T } \\left[ \\gamma \\left( T _ { 1 } \\right) - \\gamma \\left( T _ { 0 } \\right) \\vert X = x \\right]" + }, + { + "category_id": 13, + "poly": [ + 527, + 1969, + 684, + 1969, + 684, + 2000, + 527, + 2000 + ], + "score": 0.91, + "latex": "\\hat { e } _ { i } = \\sigma ( x _ { i } ; w )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1997, + 499, + 1997, + 499, + 2031, + 298, + 2031 + ], + "score": 0.91, + "latex": "P ( A = { \\bar { 1 } } | X = x )" + }, + { + "category_id": 13, + "poly": [ + 1292, + 612, + 1344, + 612, + 1344, + 646, + 1292, + 646 + ], + "score": 0.9, + "latex": "S ( t )" + }, + { + "category_id": 13, + "poly": [ + 836, + 846, + 934, + 846, + 934, + 880, + 836, + 880 + ], + "score": 0.9, + "latex": "\\mathrm { H R } ( t | x )" + }, + { + "category_id": 13, + "poly": [ + 802, + 1188, + 851, + 1188, + 851, + 1220, + 802, + 1220 + ], + "score": 0.9, + "latex": "\\gamma ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 952, + 811, + 1028, + 811, + 1028, + 848, + 952, + 848 + ], + "score": 0.9, + "latex": "\\mathrm { \\hat { H R } } ( t )" + }, + { + "category_id": 13, + "poly": [ + 447, + 1898, + 710, + 1898, + 710, + 1938, + 447, + 1938 + ], + "score": 0.9, + "latex": "\\begin{array} { r } { \\mathrm { I P W } _ { i } = \\frac { a _ { i } } { \\hat { e } _ { i } } + \\frac { 1 - a _ { i } } { 1 - \\hat { e } _ { i } } ; i i ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 984, + 739, + 1059, + 739, + 1059, + 777, + 984, + 777 + ], + "score": 0.9, + "latex": "\\mathrm { \\hat { H R } } ( t )" + }, + { + "category_id": 13, + "poly": [ + 380, + 1220, + 428, + 1220, + 428, + 1251, + 380, + 1251 + ], + "score": 0.89, + "latex": "\\gamma ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 675, + 781, + 715, + 781, + 715, + 811, + 675, + 811 + ], + "score": 0.87, + "latex": "m _ { a }" + }, + { + "category_id": 13, + "poly": [ + 1009, + 1969, + 1034, + 1969, + 1034, + 1998, + 1009, + 1998 + ], + "score": 0.87, + "latex": "\\boldsymbol { \\hat { e } } _ { i }" + }, + { + "category_id": 13, + "poly": [ + 656, + 923, + 732, + 923, + 732, + 957, + 656, + 957 + ], + "score": 0.87, + "latex": "\\mathrm { H R } ( t )" + }, + { + "category_id": 13, + "poly": [ + 1086, + 1144, + 1175, + 1144, + 1175, + 1178, + 1086, + 1178 + ], + "score": 0.87, + "latex": "\\operatorname { I T E } ( x )" + }, + { + "category_id": 13, + "poly": [ + 369, + 1938, + 765, + 1938, + 765, + 1969, + 369, + 1969 + ], + "score": 0.86, + "latex": "\\mathrm { O W } _ { i } = a _ { i } \\cdot \\left( 1 - \\hat { e } _ { i } \\right) + \\left( 1 - a _ { i } \\right) \\cdot \\hat { e } _ { i }" + }, + { + "category_id": 13, + "poly": [ + 1049, + 1807, + 1096, + 1807, + 1096, + 1839, + 1049, + 1839 + ], + "score": 0.83, + "latex": "( i v )" + }, + { + "category_id": 13, + "poly": [ + 829, + 1639, + 1032, + 1639, + 1032, + 1673, + 829, + 1673 + ], + "score": 0.81, + "latex": "( 0 , 0 . 1 , 1 , 1 0 , 1 0 0 )" + }, + { + "category_id": 13, + "poly": [ + 1212, + 1613, + 1234, + 1613, + 1234, + 1636, + 1212, + 1636 + ], + "score": 0.79, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 523, + 1867, + 555, + 1867, + 555, + 1899, + 523, + 1899 + ], + "score": 0.76, + "latex": "( i )" + }, + { + "category_id": 13, + "poly": [ + 752, + 255, + 1017, + 255, + 1017, + 291, + 752, + 291 + ], + "score": 0.76, + "latex": "\\mathrm { H R ( t ) } = 0 . 5 2 _ { ( 0 . 3 9 , 0 . 7 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 1051, + 224, + 1105, + 224, + 1105, + 255, + 1051, + 255 + ], + "score": 0.75, + "latex": "9 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 1281, + 1554, + 1302, + 1554, + 1302, + 1576, + 1281, + 1576 + ], + "score": 0.74, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 606, + 749, + 625, + 749, + 625, + 771, + 606, + 771 + ], + "score": 0.73, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 873, + 1746, + 924, + 1746, + 924, + 1778, + 873, + 1778 + ], + "score": 0.71, + "latex": "( i i i )" + }, + { + "category_id": 13, + "poly": [ + 297, + 1547, + 373, + 1547, + 373, + 1580, + 297, + 1580 + ], + "score": 0.7, + "latex": "\\mathrm { H R } ( t )" + }, + { + "category_id": 13, + "poly": [ + 822, + 1938, + 854, + 1938, + 854, + 1965, + 822, + 1965 + ], + "score": 0.52, + "latex": "i i i" + }, + { + "category_id": 13, + "poly": [ + 1107, + 224, + 1183, + 224, + 1183, + 258, + 1107, + 258 + ], + "score": 0.45, + "latex": "\\mathrm { H R } ( t )" + }, + { + "category_id": 13, + "poly": [ + 1258, + 1716, + 1300, + 1716, + 1300, + 1749, + 1258, + 1749 + ], + "score": 0.4, + "latex": "( i i )" + }, + { + "category_id": 13, + "poly": [ + 1130, + 1684, + 1163, + 1684, + 1163, + 1718, + 1130, + 1718 + ], + "score": 0.32, + "latex": "( i )" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1396.0, + 562.0, + 1396.0, + 562.0, + 1441.0, + 291.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 223.0, + 1050.0, + 223.0, + 1050.0, + 260.0, + 295.0, + 260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1184.0, + 223.0, + 1405.0, + 223.0, + 1405.0, + 260.0, + 1184.0, + 260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 243.0, + 751.0, + 243.0, + 751.0, + 300.0, + 290.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1018.0, + 243.0, + 1032.0, + 243.0, + 1032.0, + 300.0, + 1018.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2118.0, + 839.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1684.0, + 1129.0, + 1684.0, + 1129.0, + 1719.0, + 295.0, + 1719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1164.0, + 1684.0, + 1404.0, + 1684.0, + 1404.0, + 1719.0, + 1164.0, + 1719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1713.0, + 1257.0, + 1713.0, + 1257.0, + 1751.0, + 294.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 1713.0, + 1406.0, + 1713.0, + 1406.0, + 1751.0, + 1301.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1744.0, + 872.0, + 1744.0, + 872.0, + 1780.0, + 291.0, + 1780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 1744.0, + 1409.0, + 1744.0, + 1409.0, + 1780.0, + 925.0, + 1780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1775.0, + 1407.0, + 1775.0, + 1407.0, + 1811.0, + 294.0, + 1811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1804.0, + 1048.0, + 1804.0, + 1048.0, + 1844.0, + 294.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 1804.0, + 1407.0, + 1804.0, + 1407.0, + 1844.0, + 1097.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1837.0, + 1406.0, + 1837.0, + 1406.0, + 1869.0, + 294.0, + 1869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1867.0, + 368.0, + 1867.0, + 368.0, + 1966.0, + 289.0, + 1966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 1867.0, + 821.0, + 1867.0, + 821.0, + 1966.0, + 766.0, + 1966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 855.0, + 1867.0, + 1422.0, + 1867.0, + 1422.0, + 1966.0, + 855.0, + 1966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1931.0, + 368.0, + 1931.0, + 368.0, + 1976.0, + 292.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 1931.0, + 821.0, + 1931.0, + 821.0, + 1976.0, + 766.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 855.0, + 1931.0, + 1406.0, + 1931.0, + 1406.0, + 1976.0, + 855.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1964.0, + 526.0, + 1964.0, + 526.0, + 2006.0, + 292.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 1964.0, + 1008.0, + 1964.0, + 1008.0, + 2006.0, + 685.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 1964.0, + 1406.0, + 1964.0, + 1406.0, + 2006.0, + 1035.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 500.0, + 1999.0, + 970.0, + 1999.0, + 970.0, + 2030.0, + 500.0, + 2030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1456.0, + 1404.0, + 1456.0, + 1404.0, + 1491.0, + 295.0, + 1491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1486.0, + 1402.0, + 1486.0, + 1402.0, + 1521.0, + 295.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1517.0, + 1406.0, + 1517.0, + 1406.0, + 1552.0, + 294.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1547.0, + 1280.0, + 1547.0, + 1280.0, + 1583.0, + 374.0, + 1583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 1547.0, + 1406.0, + 1547.0, + 1406.0, + 1583.0, + 1303.0, + 1583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1578.0, + 892.0, + 1578.0, + 892.0, + 1615.0, + 294.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1027.0, + 1578.0, + 1405.0, + 1578.0, + 1405.0, + 1615.0, + 1027.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1609.0, + 1211.0, + 1609.0, + 1211.0, + 1643.0, + 295.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1235.0, + 1609.0, + 1406.0, + 1609.0, + 1406.0, + 1643.0, + 1235.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1640.0, + 504.0, + 1640.0, + 504.0, + 1675.0, + 295.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 684.0, + 1640.0, + 828.0, + 1640.0, + 828.0, + 1675.0, + 684.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 1640.0, + 1044.0, + 1640.0, + 1044.0, + 1675.0, + 1033.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 608.0, + 412.0, + 608.0, + 412.0, + 651.0, + 292.0, + 651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 474.0, + 608.0, + 1291.0, + 608.0, + 1291.0, + 651.0, + 474.0, + 651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1345.0, + 608.0, + 1409.0, + 608.0, + 1409.0, + 651.0, + 1345.0, + 651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 641.0, + 964.0, + 641.0, + 964.0, + 680.0, + 294.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1019.0, + 641.0, + 1408.0, + 641.0, + 1408.0, + 680.0, + 1019.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 675.0, + 1404.0, + 675.0, + 1404.0, + 707.0, + 296.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 699.0, + 489.0, + 699.0, + 489.0, + 746.0, + 291.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 699.0, + 1409.0, + 699.0, + 1409.0, + 746.0, + 597.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 739.0, + 605.0, + 739.0, + 605.0, + 784.0, + 291.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 626.0, + 739.0, + 983.0, + 739.0, + 983.0, + 784.0, + 626.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 739.0, + 1407.0, + 739.0, + 1407.0, + 784.0, + 1060.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 282.0, + 764.0, + 517.0, + 764.0, + 517.0, + 823.0, + 282.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 625.0, + 764.0, + 674.0, + 764.0, + 674.0, + 823.0, + 625.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 764.0, + 1196.0, + 764.0, + 1196.0, + 823.0, + 716.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1297.0, + 764.0, + 1415.0, + 764.0, + 1415.0, + 823.0, + 1297.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 813.0, + 951.0, + 813.0, + 951.0, + 850.0, + 292.0, + 850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 813.0, + 1078.0, + 813.0, + 1078.0, + 850.0, + 1029.0, + 850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1178.0, + 813.0, + 1404.0, + 813.0, + 1404.0, + 850.0, + 1178.0, + 850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 843.0, + 835.0, + 843.0, + 835.0, + 884.0, + 292.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 843.0, + 1315.0, + 843.0, + 1315.0, + 884.0, + 935.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 843.0, + 1407.0, + 843.0, + 1407.0, + 884.0, + 1404.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 878.0, + 344.0, + 878.0, + 344.0, + 909.0, + 296.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 878.0, + 757.0, + 878.0, + 757.0, + 909.0, + 435.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 924.0, + 655.0, + 924.0, + 655.0, + 957.0, + 296.0, + 957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 733.0, + 924.0, + 1402.0, + 924.0, + 1402.0, + 957.0, + 733.0, + 957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 953.0, + 826.0, + 953.0, + 826.0, + 991.0, + 294.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 953.0, + 1405.0, + 953.0, + 1405.0, + 991.0, + 926.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 983.0, + 1406.0, + 983.0, + 1406.0, + 1020.0, + 293.0, + 1020.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1012.0, + 712.0, + 1012.0, + 712.0, + 1052.0, + 292.0, + 1052.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 813.0, + 1012.0, + 1406.0, + 1012.0, + 1406.0, + 1052.0, + 813.0, + 1052.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1047.0, + 864.0, + 1047.0, + 864.0, + 1080.0, + 296.0, + 1080.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1302.0, + 1405.0, + 1302.0, + 1405.0, + 1335.0, + 298.0, + 1335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1332.0, + 296.0, + 1332.0, + 296.0, + 1376.0, + 292.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 1332.0, + 644.0, + 1332.0, + 644.0, + 1376.0, + 559.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 1332.0, + 1096.0, + 1332.0, + 1096.0, + 1376.0, + 883.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1335.0, + 1332.0, + 1345.0, + 1332.0, + 1345.0, + 1376.0, + 1335.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1095.0, + 1403.0, + 1095.0, + 1403.0, + 1134.0, + 295.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1141.0, + 552.0, + 1141.0, + 552.0, + 1181.0, + 294.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 997.0, + 1141.0, + 1085.0, + 1141.0, + 1085.0, + 1181.0, + 997.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1176.0, + 1141.0, + 1407.0, + 1141.0, + 1407.0, + 1181.0, + 1176.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 1181.0, + 801.0, + 1181.0, + 801.0, + 1228.0, + 751.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1181.0, + 1409.0, + 1181.0, + 1409.0, + 1228.0, + 852.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1213.0, + 379.0, + 1213.0, + 379.0, + 1257.0, + 292.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 429.0, + 1213.0, + 814.0, + 1213.0, + 814.0, + 1257.0, + 429.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 1213.0, + 1411.0, + 1213.0, + 1411.0, + 1257.0, + 1077.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1246.0, + 604.0, + 1246.0, + 604.0, + 1284.0, + 295.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 844.0, + 1246.0, + 1335.0, + 1246.0, + 1335.0, + 1284.0, + 844.0, + 1284.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1110, + 1405, + 1110, + 1405, + 1720, + 297, + 1720 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 297, + 1758, + 1405, + 1758, + 1405, + 2035, + 297, + 2035 + ], + "score": 0.979 + }, + { + "category_id": 5, + "poly": [ + 380, + 329, + 1315, + 329, + 1315, + 611, + 380, + 611 + ], + "score": 0.978, + "html": "
MethodCausal HR(t)C-Index (A=0, A=1)Factual Mean COVC-Slope (A=0,A=1)
CoxPH-Uniform1.69(1.38,2.07)NANANA
CoxPH-IPW1.09(0.76,1.57)NANANA
CoxPH-OW0.88(0.73,1.08)NANANA
Surv-BART14.99(14.9,14.9e8)(0.629,0.630)0.003(0.232,0.084)
AFT-Weibull1.09(1.09,1.09)(0.734,0.395)8.609(0.857,0.89)
AFT-log-Normal1.55(1.46,1.55)(0.68,0.56)10.415(0.979,0.732)
SR0.58(0.53,0.71)(0.601, 0.57)0(0.491,0.63)
CSA(proposed)1.04(1.00,1.09)(0.763,0.728)0.161(0.891,0.81)
CSA-INFO (proposed)0.81(0.77,0.83)(0.752,0.651)0.156(0.907, 0.881)
" + }, + { + "category_id": 1, + "poly": [ + 298, + 883, + 1404, + 883, + 1404, + 1096, + 298, + 1096 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 298, + 638, + 883, + 638, + 883, + 882, + 298, + 882 + ], + "score": 0.97 + }, + { + "category_id": 5, + "poly": [ + 908, + 668, + 1397, + 668, + 1397, + 825, + 908, + 825 + ], + "score": 0.968, + "html": "
FRAMINGHAMACTGACTG-SYNTHETIC
Events (%)26.026.948.9
Treatment (%)10.449.555.9
N3,4351,0542,139
p322323
Missing (%)0.231.411.38
tmax (days)7,2791,2311,313
" + }, + { + "category_id": 6, + "poly": [ + 297, + 224, + 1405, + 224, + 1405, + 322, + 297, + 322 + ], + "score": 0.932 + }, + { + "category_id": 6, + "poly": [ + 912, + 627, + 1391, + 627, + 1391, + 658, + 912, + 658 + ], + "score": 0.91 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 854, + 75, + 854, + 105, + 298, + 105 + ], + "score": 0.897 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2112, + 840, + 2112 + ], + "score": 0.791 + }, + { + "category_id": 13, + "poly": [ + 559, + 2002, + 765, + 2002, + 765, + 2036, + 559, + 2036 + ], + "score": 0.92, + "latex": "\\mathrm { H R } ( \\bar { t } | x ) > 1 . 9 1 6" + }, + { + "category_id": 13, + "poly": [ + 298, + 2002, + 506, + 2002, + 506, + 2036, + 298, + 2036 + ], + "score": 0.91, + "latex": "\\mathrm { H R } ( t | x ) < 0 . 0 2 4" + }, + { + "category_id": 13, + "poly": [ + 503, + 942, + 705, + 942, + 705, + 975, + 503, + 975 + ], + "score": 0.91, + "latex": "\\bar { P ( A = 1 | X = x ) }" + }, + { + "category_id": 13, + "poly": [ + 1132, + 1005, + 1209, + 1005, + 1209, + 1035, + 1132, + 1035 + ], + "score": 0.91, + "latex": "N \\times p" + }, + { + "category_id": 13, + "poly": [ + 1096, + 2002, + 1235, + 2002, + 1235, + 2036, + 1096, + 2036 + ], + "score": 0.91, + "latex": "\\log \\mathrm { \\bar { H R } } ( t | x )" + }, + { + "category_id": 13, + "poly": [ + 528, + 1035, + 584, + 1035, + 584, + 1065, + 528, + 1065 + ], + "score": 0.9, + "latex": "t _ { \\mathrm { m a x } }" + }, + { + "category_id": 13, + "poly": [ + 1206, + 1943, + 1397, + 1943, + 1397, + 1976, + 1206, + 1976 + ], + "score": 0.9, + "latex": "\\log \\mathrm { H R } ( t | x ) < 0" + }, + { + "category_id": 13, + "poly": [ + 1073, + 1850, + 1212, + 1850, + 1212, + 1884, + 1073, + 1884 + ], + "score": 0.89, + "latex": "\\log \\mathrm { H R } ( t | x )" + }, + { + "category_id": 13, + "poly": [ + 693, + 255, + 741, + 255, + 741, + 286, + 693, + 286 + ], + "score": 0.89, + "latex": "\\delta _ { \\mathrm { C F } }" + }, + { + "category_id": 13, + "poly": [ + 1117, + 1476, + 1201, + 1476, + 1201, + 1509, + 1117, + 1509 + ], + "score": 0.89, + "latex": "\\in [ 0 , 1 ]" + }, + { + "category_id": 13, + "poly": [ + 763, + 1942, + 818, + 1942, + 818, + 1972, + 763, + 1972 + ], + "score": 0.88, + "latex": "7 0 \\%" + }, + { + "category_id": 13, + "poly": [ + 1265, + 1911, + 1403, + 1911, + 1403, + 1944, + 1265, + 1944 + ], + "score": 0.88, + "latex": "\\log \\mathrm { H } \\bar { \\mathrm { R } } ( t | x )" + }, + { + "category_id": 13, + "poly": [ + 758, + 288, + 920, + 288, + 920, + 322, + 758, + 322 + ], + "score": 0.86, + "latex": "\\mathrm { H R } ( \\mathrm { t } ) = 0 . 7 5" + }, + { + "category_id": 13, + "poly": [ + 595, + 259, + 643, + 259, + 643, + 287, + 595, + 287 + ], + "score": 0.86, + "latex": "y _ { \\mathrm { C F } }" + }, + { + "category_id": 13, + "poly": [ + 346, + 1034, + 397, + 1034, + 397, + 1066, + 346, + 1066 + ], + "score": 0.84, + "latex": "( i i i )" + }, + { + "category_id": 13, + "poly": [ + 813, + 1385, + 888, + 1385, + 888, + 1417, + 813, + 1417 + ], + "score": 0.82, + "latex": "\\mathrm { H R } ( t )" + }, + { + "category_id": 13, + "poly": [ + 935, + 1327, + 1014, + 1327, + 1014, + 1355, + 935, + 1355 + ], + "score": 0.81, + "latex": "\\epsilon _ { \\mathrm { P E H E } }" + }, + { + "category_id": 13, + "poly": [ + 1039, + 224, + 1095, + 224, + 1095, + 254, + 1039, + 254 + ], + "score": 0.78, + "latex": "9 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 1321, + 1628, + 1396, + 1628, + 1396, + 1662, + 1321, + 1662 + ], + "score": 0.78, + "latex": "\\mathrm { H R } ( t )" + }, + { + "category_id": 13, + "poly": [ + 801, + 1324, + 877, + 1324, + 877, + 1356, + 801, + 1356 + ], + "score": 0.74, + "latex": "\\mathrm { H R } ( t )" + }, + { + "category_id": 13, + "poly": [ + 895, + 255, + 1160, + 255, + 1160, + 288, + 895, + 288 + ], + "score": 0.74, + "latex": "\\mathrm { H R ( t ) } = 1 . 2 3 _ { ( 1 . 1 7 , 1 . 2 5 ) }" + }, + { + "category_id": 13, + "poly": [ + 386, + 1789, + 462, + 1789, + 462, + 1823, + 386, + 1823 + ], + "score": 0.73, + "latex": "\\mathrm { H R } ( t )" + }, + { + "category_id": 13, + "poly": [ + 750, + 852, + 801, + 852, + 801, + 884, + 750, + 884 + ], + "score": 0.65, + "latex": "( i i i )" + }, + { + "category_id": 13, + "poly": [ + 1328, + 977, + 1347, + 977, + 1347, + 1006, + 1328, + 1006 + ], + "score": 0.62, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 1083, + 973, + 1116, + 973, + 1116, + 1006, + 1083, + 1006 + ], + "score": 0.59, + "latex": "( i )" + }, + { + "category_id": 13, + "poly": [ + 1098, + 224, + 1175, + 224, + 1175, + 258, + 1098, + 258 + ], + "score": 0.56, + "latex": "\\mathrm { H R } ( t )" + }, + { + "category_id": 13, + "poly": [ + 645, + 760, + 687, + 760, + 687, + 794, + 645, + 794 + ], + "score": 0.4, + "latex": "( i i )" + }, + { + "category_id": 13, + "poly": [ + 1362, + 973, + 1403, + 973, + 1403, + 1006, + 1362, + 1006 + ], + "score": 0.34, + "latex": "( i i )" + }, + { + "category_id": 13, + "poly": [ + 484, + 1944, + 516, + 1944, + 516, + 1971, + 484, + 1971 + ], + "score": 0.25, + "latex": "_ { 2 \\mathrm { c } }" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 222.0, + 1038.0, + 222.0, + 1038.0, + 261.0, + 292.0, + 261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1176.0, + 222.0, + 1408.0, + 222.0, + 1408.0, + 261.0, + 1176.0, + 261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 249.0, + 594.0, + 249.0, + 594.0, + 294.0, + 291.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 644.0, + 249.0, + 692.0, + 249.0, + 692.0, + 294.0, + 644.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 249.0, + 894.0, + 249.0, + 894.0, + 294.0, + 742.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1161.0, + 249.0, + 1406.0, + 249.0, + 1406.0, + 294.0, + 1161.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 287.0, + 757.0, + 287.0, + 757.0, + 329.0, + 294.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 287.0, + 1257.0, + 287.0, + 1257.0, + 329.0, + 921.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 625.0, + 1396.0, + 625.0, + 1396.0, + 661.0, + 908.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 2085.0, + 859.0, + 2085.0, + 859.0, + 2116.0, + 836.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1111.0, + 1408.0, + 1111.0, + 1408.0, + 1143.0, + 296.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1143.0, + 1403.0, + 1143.0, + 1403.0, + 1173.0, + 296.0, + 1173.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1171.0, + 1406.0, + 1171.0, + 1406.0, + 1204.0, + 294.0, + 1204.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1202.0, + 1403.0, + 1202.0, + 1403.0, + 1236.0, + 296.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1229.0, + 1406.0, + 1229.0, + 1406.0, + 1270.0, + 291.0, + 1270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1260.0, + 1406.0, + 1260.0, + 1406.0, + 1302.0, + 290.0, + 1302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1295.0, + 1403.0, + 1295.0, + 1403.0, + 1326.0, + 296.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1323.0, + 800.0, + 1323.0, + 800.0, + 1360.0, + 291.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 1323.0, + 934.0, + 1323.0, + 934.0, + 1360.0, + 878.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1015.0, + 1323.0, + 1406.0, + 1323.0, + 1406.0, + 1360.0, + 1015.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1355.0, + 1403.0, + 1355.0, + 1403.0, + 1389.0, + 294.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1385.0, + 812.0, + 1385.0, + 812.0, + 1420.0, + 294.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 1385.0, + 1405.0, + 1385.0, + 1405.0, + 1420.0, + 889.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1417.0, + 1403.0, + 1417.0, + 1403.0, + 1448.0, + 296.0, + 1448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1445.0, + 1405.0, + 1445.0, + 1405.0, + 1479.0, + 292.0, + 1479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1475.0, + 1116.0, + 1475.0, + 1116.0, + 1514.0, + 292.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1202.0, + 1475.0, + 1406.0, + 1475.0, + 1406.0, + 1514.0, + 1202.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1505.0, + 1407.0, + 1505.0, + 1407.0, + 1543.0, + 292.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1538.0, + 1407.0, + 1538.0, + 1407.0, + 1573.0, + 294.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1563.0, + 1410.0, + 1563.0, + 1410.0, + 1605.0, + 292.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1599.0, + 1405.0, + 1599.0, + 1405.0, + 1629.0, + 296.0, + 1629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1627.0, + 1320.0, + 1627.0, + 1320.0, + 1664.0, + 292.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1397.0, + 1627.0, + 1407.0, + 1627.0, + 1407.0, + 1664.0, + 1397.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1658.0, + 1410.0, + 1658.0, + 1410.0, + 1695.0, + 292.0, + 1695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1690.0, + 1216.0, + 1690.0, + 1216.0, + 1724.0, + 295.0, + 1724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1756.0, + 1405.0, + 1756.0, + 1405.0, + 1794.0, + 295.0, + 1794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1788.0, + 385.0, + 1788.0, + 385.0, + 1825.0, + 294.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 463.0, + 1788.0, + 1405.0, + 1788.0, + 1405.0, + 1825.0, + 463.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1819.0, + 1406.0, + 1819.0, + 1406.0, + 1856.0, + 294.0, + 1856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1849.0, + 1072.0, + 1849.0, + 1072.0, + 1886.0, + 294.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 1849.0, + 1406.0, + 1849.0, + 1406.0, + 1886.0, + 1213.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1879.0, + 1407.0, + 1879.0, + 1407.0, + 1917.0, + 292.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1911.0, + 1264.0, + 1911.0, + 1264.0, + 1948.0, + 294.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1942.0, + 483.0, + 1942.0, + 483.0, + 1979.0, + 294.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1942.0, + 762.0, + 1942.0, + 762.0, + 1979.0, + 517.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 1942.0, + 1205.0, + 1942.0, + 1205.0, + 1979.0, + 819.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 1942.0, + 1408.0, + 1942.0, + 1408.0, + 1979.0, + 1398.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1971.0, + 1408.0, + 1971.0, + 1408.0, + 2009.0, + 294.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1999.0, + 297.0, + 1999.0, + 297.0, + 2041.0, + 292.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 507.0, + 1999.0, + 558.0, + 1999.0, + 558.0, + 2041.0, + 507.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 1999.0, + 1095.0, + 1999.0, + 1095.0, + 2041.0, + 766.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 1999.0, + 1406.0, + 1999.0, + 1406.0, + 2041.0, + 1236.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 881.0, + 1405.0, + 881.0, + 1405.0, + 916.0, + 295.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 911.0, + 1402.0, + 911.0, + 1402.0, + 945.0, + 294.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 939.0, + 502.0, + 939.0, + 502.0, + 980.0, + 292.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.0, + 939.0, + 1408.0, + 939.0, + 1408.0, + 980.0, + 706.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 971.0, + 1082.0, + 971.0, + 1082.0, + 1009.0, + 292.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1117.0, + 971.0, + 1327.0, + 971.0, + 1327.0, + 1009.0, + 1117.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1348.0, + 971.0, + 1361.0, + 971.0, + 1361.0, + 1009.0, + 1348.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1003.0, + 1131.0, + 1003.0, + 1131.0, + 1040.0, + 293.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 1003.0, + 1407.0, + 1003.0, + 1407.0, + 1040.0, + 1210.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1031.0, + 345.0, + 1031.0, + 345.0, + 1070.0, + 292.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 398.0, + 1031.0, + 527.0, + 1031.0, + 527.0, + 1070.0, + 398.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 1031.0, + 1406.0, + 1031.0, + 1406.0, + 1070.0, + 585.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1062.0, + 999.0, + 1062.0, + 999.0, + 1101.0, + 293.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 633.0, + 883.0, + 633.0, + 883.0, + 674.0, + 294.0, + 674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 667.0, + 884.0, + 667.0, + 884.0, + 700.0, + 296.0, + 700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 698.0, + 881.0, + 698.0, + 881.0, + 731.0, + 296.0, + 731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 729.0, + 885.0, + 729.0, + 885.0, + 761.0, + 296.0, + 761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 760.0, + 644.0, + 760.0, + 644.0, + 793.0, + 295.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 688.0, + 760.0, + 886.0, + 760.0, + 886.0, + 793.0, + 688.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 789.0, + 885.0, + 789.0, + 885.0, + 824.0, + 295.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 818.0, + 882.0, + 818.0, + 882.0, + 854.0, + 295.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 851.0, + 749.0, + 851.0, + 749.0, + 884.0, + 296.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 802.0, + 851.0, + 885.0, + 851.0, + 885.0, + 884.0, + 802.0, + 884.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 883, + 1406, + 883, + 1406, + 1098, + 297, + 1098 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 661, + 1406, + 661, + 1406, + 816, + 298, + 816 + ], + "score": 0.975 + }, + { + "category_id": 3, + "poly": [ + 321, + 210, + 1359, + 210, + 1359, + 493, + 321, + 493 + ], + "score": 0.969 + }, + { + "category_id": 4, + "poly": [ + 297, + 503, + 1404, + 503, + 1404, + 627, + 297, + 627 + ], + "score": 0.962 + }, + { + "category_id": 0, + "poly": [ + 299, + 838, + 562, + 838, + 562, + 872, + 299, + 872 + ], + "score": 0.899 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2111, + 840, + 2111 + ], + "score": 0.784 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 852, + 76, + 852, + 104, + 299, + 104 + ], + "score": 0.613 + }, + { + "category_id": 2, + "poly": [ + 298, + 76, + 854, + 76, + 854, + 104, + 298, + 104 + ], + "score": 0.541 + }, + { + "category_id": 13, + "poly": [ + 387, + 564, + 526, + 564, + 526, + 597, + 387, + 597 + ], + "score": 0.9, + "latex": "\\log \\mathrm { H R } ( t | x )" + }, + { + "category_id": 13, + "poly": [ + 888, + 594, + 1028, + 594, + 1028, + 627, + 888, + 627 + ], + "score": 0.89, + "latex": "\\log \\mathrm { H R } ( t | x )" + }, + { + "category_id": 13, + "poly": [ + 886, + 534, + 1008, + 534, + 1008, + 567, + 886, + 567 + ], + "score": 0.85, + "latex": "\\operatorname { \\Pi } ^ { } \\operatorname { \\Pi } ^ { } \\operatorname { \\mathrm { ) g } } \\operatorname { \\mathrm { { H R } } } ( t | x )" + }, + { + "category_id": 13, + "poly": [ + 873, + 463, + 1001, + 463, + 1001, + 494, + 873, + 494 + ], + "score": 0.76, + "latex": "\\log \\mathrm { H R } ( t | x )" + }, + { + "category_id": 13, + "poly": [ + 1216, + 460, + 1339, + 460, + 1339, + 492, + 1216, + 492 + ], + "score": 0.66, + "latex": "\\log \\mathrm { H R } ( t | x )" + }, + { + "category_id": 13, + "poly": [ + 657, + 502, + 733, + 502, + 733, + 536, + 657, + 536 + ], + "score": 0.66, + "latex": "\\mathrm { H R } ( t )" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 211.0, + 719.0, + 211.0, + 719.0, + 231.0, + 699.0, + 231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 228.0, + 372.0, + 228.0, + 372.0, + 273.0, + 341.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 225.0, + 612.0, + 225.0, + 612.0, + 273.0, + 552.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 245.0, + 718.0, + 245.0, + 718.0, + 261.0, + 701.0, + 261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 221.0, + 820.0, + 221.0, + 820.0, + 269.0, + 762.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 234.0, + 1069.0, + 234.0, + 1069.0, + 256.0, + 1037.0, + 256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1298.0, + 250.0, + 1312.0, + 250.0, + 1312.0, + 259.0, + 1298.0, + 259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1314.0, + 227.0, + 1337.0, + 227.0, + 1337.0, + 265.0, + 1314.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 275.0, + 372.0, + 275.0, + 372.0, + 346.0, + 323.0, + 346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 294.0, + 534.0, + 294.0, + 534.0, + 305.0, + 525.0, + 305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 551.0, + 269.0, + 624.0, + 269.0, + 624.0, + 316.0, + 551.0, + 316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 275.0, + 719.0, + 275.0, + 719.0, + 364.0, + 674.0, + 364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 262.0, + 860.0, + 262.0, + 860.0, + 325.0, + 762.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 265.0, + 1066.0, + 265.0, + 1066.0, + 312.0, + 1038.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1314.0, + 265.0, + 1338.0, + 265.0, + 1338.0, + 284.0, + 1314.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 322.0, + 372.0, + 322.0, + 372.0, + 343.0, + 340.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 319.0, + 804.0, + 319.0, + 804.0, + 344.0, + 763.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 323.0, + 1066.0, + 323.0, + 1066.0, + 341.0, + 1038.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.0, + 340.0, + 719.0, + 340.0, + 719.0, + 355.0, + 706.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 346.0, + 372.0, + 346.0, + 372.0, + 367.0, + 341.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 352.0, + 1069.0, + 352.0, + 1069.0, + 373.0, + 1037.0, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 369.0, + 372.0, + 369.0, + 372.0, + 391.0, + 341.0, + 391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 373.0, + 718.0, + 373.0, + 718.0, + 385.0, + 707.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 380.0, + 1069.0, + 380.0, + 1069.0, + 402.0, + 1037.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 395.0, + 371.0, + 395.0, + 371.0, + 413.0, + 342.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 405.0, + 717.0, + 405.0, + 717.0, + 416.0, + 705.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 411.0, + 1068.0, + 411.0, + 1068.0, + 429.0, + 1038.0, + 429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 378.0, + 427.0, + 398.0, + 427.0, + 398.0, + 441.0, + 378.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 426.0, + 438.0, + 426.0, + 438.0, + 442.0, + 410.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 426.0, + 478.0, + 426.0, + 478.0, + 442.0, + 449.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 483.0, + 426.0, + 514.0, + 426.0, + 514.0, + 442.0, + 483.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 426.0, + 551.0, + 426.0, + 551.0, + 442.0, + 522.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 425.0, + 629.0, + 425.0, + 629.0, + 444.0, + 555.0, + 444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 727.0, + 427.0, + 987.0, + 427.0, + 987.0, + 446.0, + 727.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 421.0, + 1154.0, + 421.0, + 1154.0, + 441.0, + 1058.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1158.0, + 422.0, + 1185.0, + 422.0, + 1185.0, + 440.0, + 1158.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1202.0, + 425.0, + 1216.0, + 425.0, + 1216.0, + 439.0, + 1202.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 426.0, + 1248.0, + 426.0, + 1248.0, + 436.0, + 1236.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1268.0, + 425.0, + 1286.0, + 425.0, + 1286.0, + 439.0, + 1268.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 422.0, + 1321.0, + 422.0, + 1321.0, + 440.0, + 1300.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1333.0, + 423.0, + 1354.0, + 423.0, + 1354.0, + 441.0, + 1333.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 439.0, + 541.0, + 439.0, + 541.0, + 462.0, + 464.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 439.0, + 898.0, + 439.0, + 898.0, + 464.0, + 819.0, + 464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.0, + 436.0, + 1271.0, + 436.0, + 1271.0, + 459.0, + 1142.0, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 460.0, + 640.0, + 460.0, + 640.0, + 490.0, + 321.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 454.0, + 872.0, + 454.0, + 872.0, + 499.0, + 672.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 454.0, + 1215.0, + 454.0, + 1215.0, + 499.0, + 1002.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1340.0, + 454.0, + 1366.0, + 454.0, + 1366.0, + 499.0, + 1340.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 485.0, + 270.0, + 525.0, + 270.0, + 525.0, + 305.0, + 485.0, + 305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 503.0, + 656.0, + 503.0, + 656.0, + 539.0, + 294.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 734.0, + 503.0, + 1405.0, + 503.0, + 1405.0, + 539.0, + 734.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 533.0, + 885.0, + 533.0, + 885.0, + 570.0, + 294.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1009.0, + 533.0, + 1406.0, + 533.0, + 1406.0, + 570.0, + 1009.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 566.0, + 386.0, + 566.0, + 386.0, + 599.0, + 295.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 566.0, + 1404.0, + 566.0, + 1404.0, + 599.0, + 527.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 592.0, + 887.0, + 592.0, + 887.0, + 632.0, + 292.0, + 632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 592.0, + 1111.0, + 592.0, + 1111.0, + 632.0, + 1029.0, + 632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 831.0, + 566.0, + 831.0, + 566.0, + 880.0, + 292.0, + 880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2087.0, + 860.0, + 2087.0, + 860.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 856.0, + 73.0, + 856.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 883.0, + 1404.0, + 883.0, + 1404.0, + 918.0, + 295.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 913.0, + 1404.0, + 913.0, + 1404.0, + 947.0, + 295.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 945.0, + 1408.0, + 945.0, + 1408.0, + 979.0, + 295.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 976.0, + 1408.0, + 976.0, + 1408.0, + 1009.0, + 294.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1004.0, + 1408.0, + 1004.0, + 1408.0, + 1041.0, + 294.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1036.0, + 1406.0, + 1036.0, + 1406.0, + 1070.0, + 294.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1069.0, + 1141.0, + 1069.0, + 1141.0, + 1100.0, + 296.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 661.0, + 1403.0, + 661.0, + 1403.0, + 695.0, + 295.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 691.0, + 1408.0, + 691.0, + 1408.0, + 729.0, + 293.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 720.0, + 1409.0, + 720.0, + 1409.0, + 762.0, + 292.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 754.0, + 1404.0, + 754.0, + 1404.0, + 788.0, + 295.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 784.0, + 1391.0, + 784.0, + 1391.0, + 820.0, + 293.0, + 820.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 280, + 1406, + 280, + 1406, + 372, + 298, + 372 + ], + "score": 0.928 + }, + { + "category_id": 1, + "poly": [ + 297, + 395, + 1400, + 395, + 1400, + 488, + 297, + 488 + ], + "score": 0.924 + }, + { + "category_id": 1, + "poly": [ + 292, + 1659, + 1402, + 1659, + 1402, + 1724, + 292, + 1724 + ], + "score": 0.895 + }, + { + "category_id": 1, + "poly": [ + 296, + 1542, + 1407, + 1542, + 1407, + 1637, + 296, + 1637 + ], + "score": 0.891 + }, + { + "category_id": 1, + "poly": [ + 297, + 1458, + 1403, + 1458, + 1403, + 1521, + 297, + 1521 + ], + "score": 0.887 + }, + { + "category_id": 1, + "poly": [ + 300, + 565, + 1399, + 565, + 1399, + 631, + 300, + 631 + ], + "score": 0.885 + }, + { + "category_id": 1, + "poly": [ + 297, + 1371, + 1403, + 1371, + 1403, + 1437, + 297, + 1437 + ], + "score": 0.882 + }, + { + "category_id": 1, + "poly": [ + 298, + 651, + 1405, + 651, + 1405, + 746, + 298, + 746 + ], + "score": 0.88 + }, + { + "category_id": 1, + "poly": [ + 292, + 1800, + 1404, + 1800, + 1404, + 1863, + 292, + 1863 + ], + "score": 0.879 + }, + { + "category_id": 1, + "poly": [ + 295, + 510, + 1402, + 510, + 1402, + 545, + 295, + 545 + ], + "score": 0.878 + }, + { + "category_id": 1, + "poly": [ + 298, + 1170, + 1405, + 1170, + 1405, + 1266, + 298, + 1266 + ], + "score": 0.876 + }, + { + "category_id": 1, + "poly": [ + 296, + 1745, + 1403, + 1745, + 1403, + 1780, + 296, + 1780 + ], + "score": 0.876 + }, + { + "category_id": 1, + "poly": [ + 306, + 969, + 1404, + 969, + 1404, + 1063, + 306, + 1063 + ], + "score": 0.873 + }, + { + "category_id": 1, + "poly": [ + 294, + 1084, + 1402, + 1084, + 1402, + 1150, + 294, + 1150 + ], + "score": 0.87 + }, + { + "category_id": 1, + "poly": [ + 297, + 883, + 1404, + 883, + 1404, + 948, + 297, + 948 + ], + "score": 0.863 + }, + { + "category_id": 1, + "poly": [ + 291, + 1885, + 1402, + 1885, + 1402, + 1950, + 291, + 1950 + ], + "score": 0.859 + }, + { + "category_id": 1, + "poly": [ + 299, + 1971, + 1398, + 1971, + 1398, + 2035, + 299, + 2035 + ], + "score": 0.844 + }, + { + "category_id": 1, + "poly": [ + 298, + 767, + 1406, + 767, + 1406, + 863, + 298, + 863 + ], + "score": 0.842 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 864, + 2088, + 864, + 2113, + 836, + 2113 + ], + "score": 0.82 + }, + { + "category_id": 1, + "poly": [ + 297, + 1286, + 1401, + 1286, + 1401, + 1352, + 297, + 1352 + ], + "score": 0.816 + }, + { + "category_id": 0, + "poly": [ + 299, + 227, + 489, + 227, + 489, + 261, + 299, + 261 + ], + "score": 0.75 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 854, + 75, + 854, + 105, + 298, + 105 + ], + "score": 0.492 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 854, + 75, + 854, + 105, + 298, + 105 + ], + "score": 0.404 + }, + { + "category_id": 1, + "poly": [ + 299, + 227, + 489, + 227, + 489, + 261, + 299, + 261 + ], + "score": 0.132 + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2123.0, + 831.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 228.0, + 491.0, + 228.0, + 491.0, + 264.0, + 297.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 280.0, + 1408.0, + 280.0, + 1408.0, + 314.0, + 295.0, + 314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 312.0, + 1404.0, + 312.0, + 1404.0, + 345.0, + 321.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 339.0, + 496.0, + 339.0, + 496.0, + 378.0, + 321.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 393.0, + 1406.0, + 393.0, + 1406.0, + 431.0, + 293.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 427.0, + 1405.0, + 427.0, + 1405.0, + 461.0, + 322.0, + 461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 457.0, + 511.0, + 457.0, + 511.0, + 489.0, + 322.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1654.0, + 1405.0, + 1654.0, + 1405.0, + 1699.0, + 295.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1690.0, + 1128.0, + 1690.0, + 1128.0, + 1726.0, + 324.0, + 1726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1541.0, + 1408.0, + 1541.0, + 1408.0, + 1581.0, + 293.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1576.0, + 1408.0, + 1576.0, + 1408.0, + 1610.0, + 323.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1605.0, + 1041.0, + 1605.0, + 1041.0, + 1639.0, + 323.0, + 1639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1458.0, + 1405.0, + 1458.0, + 1405.0, + 1493.0, + 295.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1488.0, + 601.0, + 1488.0, + 601.0, + 1524.0, + 321.0, + 1524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 564.0, + 1403.0, + 564.0, + 1403.0, + 601.0, + 295.0, + 601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 598.0, + 977.0, + 598.0, + 977.0, + 632.0, + 321.0, + 632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1369.0, + 1408.0, + 1369.0, + 1408.0, + 1411.0, + 294.0, + 1411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1403.0, + 949.0, + 1403.0, + 949.0, + 1438.0, + 322.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 651.0, + 1405.0, + 651.0, + 1405.0, + 689.0, + 295.0, + 689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 681.0, + 1402.0, + 681.0, + 1402.0, + 716.0, + 323.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 714.0, + 746.0, + 714.0, + 746.0, + 748.0, + 323.0, + 748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1798.0, + 1409.0, + 1798.0, + 1409.0, + 1837.0, + 294.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1832.0, + 645.0, + 1832.0, + 645.0, + 1863.0, + 323.0, + 1863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 509.0, + 1405.0, + 509.0, + 1405.0, + 544.0, + 297.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1168.0, + 1405.0, + 1168.0, + 1405.0, + 1209.0, + 293.0, + 1209.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1202.0, + 1405.0, + 1202.0, + 1405.0, + 1236.0, + 322.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1234.0, + 775.0, + 1234.0, + 775.0, + 1267.0, + 320.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1743.0, + 1408.0, + 1743.0, + 1408.0, + 1782.0, + 295.0, + 1782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 967.0, + 1408.0, + 967.0, + 1408.0, + 1006.0, + 297.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 998.0, + 1405.0, + 998.0, + 1405.0, + 1037.0, + 322.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1031.0, + 763.0, + 1031.0, + 763.0, + 1064.0, + 321.0, + 1064.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1085.0, + 1405.0, + 1085.0, + 1405.0, + 1122.0, + 294.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1116.0, + 1223.0, + 1116.0, + 1223.0, + 1151.0, + 323.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 882.0, + 1405.0, + 882.0, + 1405.0, + 919.0, + 295.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 914.0, + 1301.0, + 914.0, + 1301.0, + 950.0, + 324.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1883.0, + 1407.0, + 1883.0, + 1407.0, + 1923.0, + 294.0, + 1923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1916.0, + 703.0, + 1916.0, + 703.0, + 1951.0, + 323.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1968.0, + 1403.0, + 1968.0, + 1403.0, + 2008.0, + 295.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 2003.0, + 1006.0, + 2003.0, + 1006.0, + 2035.0, + 323.0, + 2035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 768.0, + 1404.0, + 768.0, + 1404.0, + 803.0, + 296.0, + 803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 798.0, + 1404.0, + 798.0, + 1404.0, + 835.0, + 324.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 830.0, + 1115.0, + 830.0, + 1115.0, + 864.0, + 324.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1284.0, + 1405.0, + 1284.0, + 1405.0, + 1326.0, + 292.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1318.0, + 831.0, + 1318.0, + 831.0, + 1351.0, + 322.0, + 1351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 228.0, + 491.0, + 228.0, + 491.0, + 264.0, + 297.0, + 264.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 300, + 229, + 1404, + 229, + 1404, + 353, + 300, + 353 + ], + "score": 0.945 + }, + { + "category_id": 1, + "poly": [ + 298, + 460, + 1407, + 460, + 1407, + 553, + 298, + 553 + ], + "score": 0.916 + }, + { + "category_id": 1, + "poly": [ + 298, + 375, + 1402, + 375, + 1402, + 438, + 298, + 438 + ], + "score": 0.912 + }, + { + "category_id": 1, + "poly": [ + 299, + 775, + 1401, + 775, + 1401, + 870, + 299, + 870 + ], + "score": 0.907 + }, + { + "category_id": 1, + "poly": [ + 297, + 574, + 1406, + 574, + 1406, + 668, + 297, + 668 + ], + "score": 0.902 + }, + { + "category_id": 1, + "poly": [ + 296, + 1231, + 1402, + 1231, + 1402, + 1296, + 296, + 1296 + ], + "score": 0.9 + }, + { + "category_id": 1, + "poly": [ + 294, + 893, + 1403, + 893, + 1403, + 957, + 294, + 957 + ], + "score": 0.896 + }, + { + "category_id": 1, + "poly": [ + 295, + 689, + 1400, + 689, + 1400, + 753, + 295, + 753 + ], + "score": 0.895 + }, + { + "category_id": 1, + "poly": [ + 292, + 977, + 1402, + 977, + 1402, + 1041, + 292, + 1041 + ], + "score": 0.894 + }, + { + "category_id": 1, + "poly": [ + 302, + 1971, + 1401, + 1971, + 1401, + 2034, + 302, + 2034 + ], + "score": 0.893 + }, + { + "category_id": 1, + "poly": [ + 298, + 1062, + 1401, + 1062, + 1401, + 1127, + 298, + 1127 + ], + "score": 0.878 + }, + { + "category_id": 1, + "poly": [ + 296, + 1146, + 1400, + 1146, + 1400, + 1212, + 296, + 1212 + ], + "score": 0.875 + }, + { + "category_id": 1, + "poly": [ + 295, + 1431, + 1403, + 1431, + 1403, + 1496, + 295, + 1496 + ], + "score": 0.87 + }, + { + "category_id": 1, + "poly": [ + 299, + 1517, + 1404, + 1517, + 1404, + 1612, + 299, + 1612 + ], + "score": 0.869 + }, + { + "category_id": 1, + "poly": [ + 298, + 1631, + 1128, + 1631, + 1128, + 1667, + 298, + 1667 + ], + "score": 0.855 + }, + { + "category_id": 1, + "poly": [ + 297, + 1686, + 1400, + 1686, + 1400, + 1751, + 297, + 1751 + ], + "score": 0.854 + }, + { + "category_id": 1, + "poly": [ + 298, + 1315, + 1405, + 1315, + 1405, + 1411, + 298, + 1411 + ], + "score": 0.848 + }, + { + "category_id": 1, + "poly": [ + 304, + 1855, + 1402, + 1855, + 1402, + 1950, + 304, + 1950 + ], + "score": 0.845 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 854, + 75, + 854, + 105, + 298, + 105 + ], + "score": 0.843 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 861, + 2088, + 861, + 2113, + 835, + 2113 + ], + "score": 0.824 + }, + { + "category_id": 1, + "poly": [ + 295, + 1770, + 1403, + 1770, + 1403, + 1835, + 295, + 1835 + ], + "score": 0.813 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 852, + 76, + 852, + 104, + 299, + 104 + ], + "score": 0.103 + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 867.0, + 2085.0, + 867.0, + 2125.0, + 831.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 856.0, + 73.0, + 856.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 227.0, + 1409.0, + 227.0, + 1409.0, + 267.0, + 293.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 262.0, + 1406.0, + 262.0, + 1406.0, + 295.0, + 323.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 292.0, + 1405.0, + 292.0, + 1405.0, + 325.0, + 321.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 323.0, + 1337.0, + 323.0, + 1337.0, + 355.0, + 323.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 461.0, + 1405.0, + 461.0, + 1405.0, + 494.0, + 296.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 491.0, + 1407.0, + 491.0, + 1407.0, + 526.0, + 321.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 521.0, + 631.0, + 521.0, + 631.0, + 555.0, + 320.0, + 555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 376.0, + 1404.0, + 376.0, + 1404.0, + 412.0, + 294.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 405.0, + 1276.0, + 405.0, + 1276.0, + 441.0, + 323.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 776.0, + 1405.0, + 776.0, + 1405.0, + 810.0, + 297.0, + 810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 807.0, + 1406.0, + 807.0, + 1406.0, + 841.0, + 322.0, + 841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 836.0, + 860.0, + 836.0, + 860.0, + 874.0, + 322.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 576.0, + 1404.0, + 576.0, + 1404.0, + 610.0, + 295.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 605.0, + 1404.0, + 605.0, + 1404.0, + 640.0, + 323.0, + 640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 638.0, + 639.0, + 638.0, + 639.0, + 668.0, + 324.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1231.0, + 1404.0, + 1231.0, + 1404.0, + 1267.0, + 295.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1264.0, + 840.0, + 1264.0, + 840.0, + 1296.0, + 323.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 890.0, + 1405.0, + 890.0, + 1405.0, + 929.0, + 294.0, + 929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 924.0, + 1406.0, + 924.0, + 1406.0, + 960.0, + 323.0, + 960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 685.0, + 1406.0, + 685.0, + 1406.0, + 727.0, + 293.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 720.0, + 511.0, + 720.0, + 511.0, + 753.0, + 323.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 978.0, + 1404.0, + 978.0, + 1404.0, + 1014.0, + 295.0, + 1014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1009.0, + 678.0, + 1009.0, + 678.0, + 1042.0, + 323.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1969.0, + 1407.0, + 1969.0, + 1407.0, + 2008.0, + 296.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 2000.0, + 1327.0, + 2000.0, + 1327.0, + 2039.0, + 322.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1060.0, + 1403.0, + 1060.0, + 1403.0, + 1100.0, + 294.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1093.0, + 1119.0, + 1093.0, + 1119.0, + 1127.0, + 322.0, + 1127.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1149.0, + 1404.0, + 1149.0, + 1404.0, + 1182.0, + 296.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1178.0, + 866.0, + 1178.0, + 866.0, + 1213.0, + 321.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1430.0, + 1405.0, + 1430.0, + 1405.0, + 1467.0, + 294.0, + 1467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1464.0, + 831.0, + 1464.0, + 831.0, + 1496.0, + 323.0, + 1496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1518.0, + 1404.0, + 1518.0, + 1404.0, + 1552.0, + 295.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1546.0, + 1405.0, + 1546.0, + 1405.0, + 1585.0, + 321.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1578.0, + 1055.0, + 1578.0, + 1055.0, + 1616.0, + 323.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1629.0, + 1129.0, + 1629.0, + 1129.0, + 1670.0, + 294.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1682.0, + 1404.0, + 1682.0, + 1404.0, + 1726.0, + 292.0, + 1726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1717.0, + 1058.0, + 1717.0, + 1058.0, + 1752.0, + 322.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1315.0, + 1406.0, + 1315.0, + 1406.0, + 1354.0, + 293.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1350.0, + 1403.0, + 1350.0, + 1403.0, + 1380.0, + 323.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1379.0, + 1051.0, + 1379.0, + 1051.0, + 1414.0, + 322.0, + 1414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1853.0, + 1407.0, + 1853.0, + 1407.0, + 1894.0, + 295.0, + 1894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1888.0, + 1406.0, + 1888.0, + 1406.0, + 1922.0, + 324.0, + 1922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1918.0, + 1256.0, + 1918.0, + 1256.0, + 1951.0, + 323.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1766.0, + 1406.0, + 1766.0, + 1406.0, + 1812.0, + 294.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1801.0, + 1038.0, + 1801.0, + 1038.0, + 1837.0, + 320.0, + 1837.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 915, + 1406, + 915, + 1406, + 1039, + 297, + 1039 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 299, + 475, + 1403, + 475, + 1403, + 569, + 299, + 569 + ], + "score": 0.921 + }, + { + "category_id": 1, + "poly": [ + 299, + 1141, + 1405, + 1141, + 1405, + 1234, + 299, + 1234 + ], + "score": 0.916 + }, + { + "category_id": 1, + "poly": [ + 294, + 229, + 1403, + 229, + 1403, + 293, + 294, + 293 + ], + "score": 0.915 + }, + { + "category_id": 1, + "poly": [ + 295, + 588, + 1403, + 588, + 1403, + 652, + 295, + 652 + ], + "score": 0.914 + }, + { + "category_id": 1, + "poly": [ + 298, + 670, + 1402, + 670, + 1402, + 734, + 298, + 734 + ], + "score": 0.907 + }, + { + "category_id": 1, + "poly": [ + 294, + 834, + 1404, + 834, + 1404, + 898, + 294, + 898 + ], + "score": 0.904 + }, + { + "category_id": 1, + "poly": [ + 296, + 392, + 1402, + 392, + 1402, + 458, + 296, + 458 + ], + "score": 0.903 + }, + { + "category_id": 1, + "poly": [ + 296, + 311, + 1402, + 311, + 1402, + 376, + 296, + 376 + ], + "score": 0.903 + }, + { + "category_id": 1, + "poly": [ + 292, + 1746, + 1402, + 1746, + 1402, + 1810, + 292, + 1810 + ], + "score": 0.9 + }, + { + "category_id": 1, + "poly": [ + 297, + 1253, + 1406, + 1253, + 1406, + 1348, + 297, + 1348 + ], + "score": 0.899 + }, + { + "category_id": 1, + "poly": [ + 296, + 751, + 1403, + 751, + 1403, + 816, + 296, + 816 + ], + "score": 0.896 + }, + { + "category_id": 1, + "poly": [ + 294, + 1058, + 1404, + 1058, + 1404, + 1123, + 294, + 1123 + ], + "score": 0.89 + }, + { + "category_id": 1, + "poly": [ + 294, + 1664, + 1401, + 1664, + 1401, + 1728, + 294, + 1728 + ], + "score": 0.887 + }, + { + "category_id": 1, + "poly": [ + 297, + 1529, + 1404, + 1529, + 1404, + 1596, + 297, + 1596 + ], + "score": 0.887 + }, + { + "category_id": 1, + "poly": [ + 298, + 1447, + 1401, + 1447, + 1401, + 1514, + 298, + 1514 + ], + "score": 0.886 + }, + { + "category_id": 1, + "poly": [ + 297, + 1365, + 1402, + 1365, + 1402, + 1429, + 297, + 1429 + ], + "score": 0.879 + }, + { + "category_id": 1, + "poly": [ + 299, + 1827, + 1403, + 1827, + 1403, + 1923, + 299, + 1923 + ], + "score": 0.877 + }, + { + "category_id": 1, + "poly": [ + 294, + 1611, + 1322, + 1611, + 1322, + 1647, + 294, + 1647 + ], + "score": 0.877 + }, + { + "category_id": 1, + "poly": [ + 300, + 1940, + 1402, + 1940, + 1402, + 2035, + 300, + 2035 + ], + "score": 0.868 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 864, + 2088, + 864, + 2113, + 836, + 2113 + ], + "score": 0.831 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 854, + 75, + 854, + 105, + 298, + 105 + ], + "score": 0.81 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 854, + 75, + 854, + 105, + 298, + 105 + ], + "score": 0.095 + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 916.0, + 1406.0, + 916.0, + 1406.0, + 953.0, + 294.0, + 953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 947.0, + 1406.0, + 947.0, + 1406.0, + 983.0, + 321.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 981.0, + 1406.0, + 981.0, + 1406.0, + 1013.0, + 324.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1009.0, + 506.0, + 1009.0, + 506.0, + 1040.0, + 321.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 472.0, + 1404.0, + 472.0, + 1404.0, + 514.0, + 292.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 505.0, + 1409.0, + 505.0, + 1409.0, + 545.0, + 321.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 531.0, + 396.0, + 531.0, + 396.0, + 572.0, + 319.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1139.0, + 1409.0, + 1139.0, + 1409.0, + 1178.0, + 293.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1173.0, + 1409.0, + 1173.0, + 1409.0, + 1207.0, + 322.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1198.0, + 398.0, + 1198.0, + 398.0, + 1239.0, + 319.0, + 1239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 229.0, + 1405.0, + 229.0, + 1405.0, + 265.0, + 296.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 260.0, + 625.0, + 260.0, + 625.0, + 294.0, + 318.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 588.0, + 1405.0, + 588.0, + 1405.0, + 624.0, + 295.0, + 624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 619.0, + 819.0, + 619.0, + 819.0, + 651.0, + 324.0, + 651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 669.0, + 1404.0, + 669.0, + 1404.0, + 705.0, + 294.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 702.0, + 535.0, + 702.0, + 535.0, + 733.0, + 319.0, + 733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 833.0, + 1406.0, + 833.0, + 1406.0, + 869.0, + 297.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 866.0, + 857.0, + 866.0, + 857.0, + 898.0, + 323.0, + 898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 393.0, + 1404.0, + 393.0, + 1404.0, + 429.0, + 295.0, + 429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 424.0, + 475.0, + 424.0, + 475.0, + 457.0, + 321.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 306.0, + 1406.0, + 306.0, + 1406.0, + 351.0, + 293.0, + 351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 344.0, + 757.0, + 344.0, + 757.0, + 377.0, + 321.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1746.0, + 1404.0, + 1746.0, + 1404.0, + 1782.0, + 295.0, + 1782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1779.0, + 843.0, + 1779.0, + 843.0, + 1809.0, + 323.0, + 1809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1253.0, + 1404.0, + 1253.0, + 1404.0, + 1291.0, + 294.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1281.0, + 1406.0, + 1281.0, + 1406.0, + 1324.0, + 321.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1314.0, + 1062.0, + 1314.0, + 1062.0, + 1350.0, + 321.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 747.0, + 1405.0, + 747.0, + 1405.0, + 791.0, + 293.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 784.0, + 929.0, + 784.0, + 929.0, + 817.0, + 323.0, + 817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1060.0, + 1405.0, + 1060.0, + 1405.0, + 1096.0, + 296.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1091.0, + 1357.0, + 1091.0, + 1357.0, + 1124.0, + 321.0, + 1124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1660.0, + 1407.0, + 1660.0, + 1407.0, + 1704.0, + 293.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1695.0, + 1113.0, + 1695.0, + 1113.0, + 1728.0, + 321.0, + 1728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1530.0, + 1406.0, + 1530.0, + 1406.0, + 1566.0, + 295.0, + 1566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1563.0, + 1025.0, + 1563.0, + 1025.0, + 1595.0, + 322.0, + 1595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1448.0, + 1406.0, + 1448.0, + 1406.0, + 1484.0, + 294.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1479.0, + 1254.0, + 1479.0, + 1254.0, + 1514.0, + 322.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1364.0, + 1408.0, + 1364.0, + 1408.0, + 1403.0, + 295.0, + 1403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1393.0, + 400.0, + 1393.0, + 400.0, + 1432.0, + 320.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1828.0, + 1404.0, + 1828.0, + 1404.0, + 1863.0, + 297.0, + 1863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1858.0, + 1407.0, + 1858.0, + 1407.0, + 1896.0, + 322.0, + 1896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1891.0, + 1245.0, + 1891.0, + 1245.0, + 1925.0, + 324.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1610.0, + 1325.0, + 1610.0, + 1325.0, + 1650.0, + 295.0, + 1650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1939.0, + 1406.0, + 1939.0, + 1406.0, + 1978.0, + 293.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1974.0, + 1404.0, + 1974.0, + 1404.0, + 2005.0, + 324.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 2002.0, + 1237.0, + 2002.0, + 1237.0, + 2037.0, + 320.0, + 2037.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 305, + 312, + 1400, + 312, + 1400, + 406, + 305, + 406 + ], + "score": 0.928 + }, + { + "category_id": 1, + "poly": [ + 295, + 229, + 1402, + 229, + 1402, + 293, + 295, + 293 + ], + "score": 0.896 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 854, + 75, + 854, + 105, + 298, + 105 + ], + "score": 0.867 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 863, + 2088, + 863, + 2112, + 836, + 2112 + ], + "score": 0.83 + }, + { + "category_id": 15, + "poly": [ + 297.0, + 73.0, + 857.0, + 73.0, + 857.0, + 108.0, + 297.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2084.0, + 867.0, + 2084.0, + 867.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 313.0, + 1404.0, + 313.0, + 1404.0, + 347.0, + 301.0, + 347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 343.0, + 1405.0, + 343.0, + 1405.0, + 380.0, + 322.0, + 380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 374.0, + 1347.0, + 374.0, + 1347.0, + 411.0, + 322.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 228.0, + 1404.0, + 228.0, + 1404.0, + 265.0, + 295.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 262.0, + 1053.0, + 262.0, + 1053.0, + 292.0, + 323.0, + 292.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/train/6NFBvWlRXaG/6NFBvWlRXaG.md b/parse/train/6NFBvWlRXaG/6NFBvWlRXaG.md new file mode 100644 index 0000000000000000000000000000000000000000..53ed5b030020c3dba3b5dd61fd6a7bf4f8146488 --- /dev/null +++ b/parse/train/6NFBvWlRXaG/6NFBvWlRXaG.md @@ -0,0 +1,844 @@ +# ON THE UNIVERSALITY OF ROTATION EQUIVARIANT POINT CLOUD NETWORKS + +Nadav Dym +Duke University +nadavdym@gmail.com + +Haggai Maron NVIDIA Research hmaron@nvidia.com + +# ABSTRACT + +Learning functions on point clouds has applications in many fields, including computer vision, computer graphics, physics, and chemistry. Recently, there has been a growing interest in neural architectures that are invariant or equivariant to all three shape-preserving transformations of point clouds: translation, rotation, and permutation. In this paper, we present a first study of the approximation power of these architectures. We first derive two sufficient conditions for an equivariant architecture to have the universal approximation property, based on a novel characterization of the space of equivariant polynomials. We then use these conditions to show that two recently suggested models (Thomas et al., 2018; Fuchs et al., 2020) are universal, and for devising two other novel universal architectures. + +# 1 INTRODUCTION + +Designing neural networks that respect data symmetry is a powerful approach for obtaining efficient deep models. Prominent examples being convolutional networks which respect the translational invariance of images, graph neural networks which respect the permutation invariance of graphs (Gilmer et al., 2017; Maron et al., 2019b), networks such as (Zaheer et al., 2017; Qi et al., 2017a) which respect the permutation invariance of sets, and networks which respect 3D rotational symmetries (Cohen et al., 2018; Weiler et al., 2018; Esteves et al., 2018; Worrall & Brostow, 2018; Kondor et al., 2018a). + +While the expressive power of equivariant models is reduced by design to include only equivariant functions, a desirable property of equivariant networks is universality: the ability to approximate any continuous equivariant function. This is not always the case: while convolutional networks and networks for sets are universal (Yarotsky, 2018; Segol & Lipman, 2019), popular graph neural networks are not (Xu et al., 2019; Morris et al., 2018). + +![](images/deeacb27baa383de014f051156e964f0944d6459137a61e1a3840aa1470e5f7e.jpg) + +In this paper, we consider the universality of networks that respect the symmetries of 3D point clouds: translations, rotations, and permutations. Designing such networks is a popular paradigm in recent years (Thomas et al., 2018; Fuchs et al., 2020; Poulenard et al., 2019; Zhao et al., 2019). While there have been many works on the universality of permutation invariant networks (Zaheer et al., 2017; Maron et al., $2 0 1 9 \mathrm { c }$ ; Keriven & Peyre, 2019), ´ and a recent work discussing the universality of rotation equivariant networks (Bogatskiy et al., 2020), this is a first paper which discusses the universality of networks which combine rotations, permutations and translations. + +We start the paper with a general, architecture-agnostic, discussion, and derive two sufficient conditions for universality. These conditions are a result of a novel characterization of equivariant polynomials for the symmetry group of interest. We use these conditions in order to prove universality of the prominent Tensor Field Networks (TFN) architecture (Thomas et al., 2018; Fuchs et al., 2020). The following is a weakened and simplified statement of Theorem 2 stated later on in the paper: + +Theorem (Simplification of Theorem 2). Any continuous equivariant function on point clouds can be approximated uniformly on compact sets by a composition of TFN layers. + +We use our general discussion to prove the universality of two additional equivariant models: the first is a simple modification of the TFN architecture which allows for universality using only low dimensional filters. The second is a minimal architecture which is based on tensor product representations, rather than the more commonly used irreducible representations of SO(3). We discuss the advantages and disadvantages of both approaches. + +To summarize, the contributions of this paper are: (1) A general approach for proving the universality of rotation equivariant models for point clouds; (2) A proof that two recent equivariant models (Thomas et al., 2018; Fuchs et al., 2020) are universal; (3) Two additional simple and novel universal architectures. + +# 2 PREVIOUS WORK + +Deep learning on point clouds. (Qi et al., 2017a; Zaheer et al., 2017) were the first to apply neural networks directly to the raw point cloud data, by using pointwise functions and pooling operations. Many subsequent works used local neighborhood information (Qi et al., 2017b; Wang et al., 2019b; Atzmon et al., 2018). We refer the reader to a recent survey for more details (Guo et al., 2020). In contrast with the aforementioned works which focused solely on permutation invariance, more related to this paper are works that additionally incorporated invariance to rigid motions. (Thomas et al., 2018) proposed Tensor Field Networks (TFN) and showed their efficacy on physics and chemistry tasks.(Kondor et al., 2018b) also suggested an equivariant model for continuous rotations. (Li et al., 2019) suggested models that are equivariant to discrete subgroups of SO(3). (Poulenard et al., 2019) suggested an invariant model based on spherical harmonics. (Fuchs et al., 2020) followed TFN and added an attention mechanism. Recently, (Zhao et al., 2019) proposed a quaternion equivariant point capsule network that also achieves rotation and translation invariance. + +Universal approximation for invariant networks. Understanding the approximation power of invariant models is a popular research goal. Most of the current results assume that the symmetry group is a permutation group. (Zaheer et al., 2017; Qi et al., $2 0 1 7 \mathrm { a }$ ; Segol & Lipman, 2019; Maron et al., 2020; Serviansky et al., 2020) proved universality for several $S _ { n }$ -invariant and equivariant models. (Maron et al., 2019b;a; Keriven & Peyre, 2019; Maehara & NT, 2019) studied the approx- ´ imation power of high-order graph neural networks. (Maron et al., 2019c; Ravanbakhsh, 2020) targeted universality of networks that use high-order representations for permutation groups(Yarotsky, 2018) provided several theoretical constructions of universal equivariant neural network models based on polynomial invariants, including an $S E ( 2 )$ equivariant model. In a recent work (Bogatskiy et al., 2020) presented a universal approximation theorem for networks that are equivariant to several Lie groups including $S O ( 3 )$ . The main difference from our paper is that we prove a universality theorem for a more complex group that besides rotations also includes translations and permutations. + +# 3 A FRAMEWORK FOR PROVING UNIVERSALITY + +In this section, we describe a framework for proving the universality of equivariant networks. We begin with some mathematical preliminaries: + +# 3.1 MATHEMATICAL SETUP + +An action of a group $G$ on a real vector space $W$ is a collection of maps $\rho ( g ) : W \to W$ defined for any $g \in G$ , such that $\rho ( g _ { 1 } ) \circ \rho ( g _ { 2 } ) = \rho ( g _ { 1 } g _ { 2 } )$ for all $g _ { 1 } , g _ { 2 } \in G$ , and the identity element of $G$ is mapped to the identity mapping on $W$ . We say $\rho$ is a representation of $G$ if $\rho ( g )$ is a linear map for every $g \in G$ . As is customary, when it does not cause confusion we often say that $W$ itself is a representation of $G$ . + +In this paper, we are interested in functions on point clouds. Point clouds are sets of vectors in $\mathbb { R } ^ { 3 }$ arranged as matrices: + +$$ +X = ( x _ { 1 } , \ldots , x _ { n } ) \in \mathbb { R } ^ { 3 \times n } . +$$ + +Many machine learning tasks on point clouds, such as classification, aim to learn a function which is invariant to rigid motions and relabeling of the points. Put differently, such functions are required + +to be invariant to the action of $G = \mathbb { R } ^ { 3 } \rtimes \mathrm { S O ( 3 ) } \times S _ { n }$ on $\mathbb { R } ^ { 3 \times n }$ via + +$$ +\rho _ { G } ( t , R , P ) ( X ) = R ( X - t 1 _ { n } ^ { T } ) P ^ { T } , +$$ + +where $t \in \mathbb { R } ^ { 3 }$ defines a translation, $R$ is a rotation and $P$ is a permutation matrix. + +Equivariant functions are generalizations of invariant functions: If $G$ acts on $W _ { 1 }$ via some action $\bar { \rho _ { 1 } ( g ) }$ , and on $W _ { 2 }$ via some other group action $\rho _ { 2 } ( g )$ , we say that a function $f : W _ { 1 } \to W _ { 2 }$ is equivariant if + +Invariant functions correspond to the special case where $\rho _ { 2 } ( g )$ is the identity mapping for all $g \in G$ + +In some machine learning tasks on point clouds, the functions learned are not invariant but rather equivariant. For example, segmentation tasks assign a discrete label to each point. They are invariant to translations and rotations but equivariant to permutations – in the sense that permuting the input causes a corresponding permutation of the output. Another example is predicting a normal for each point of a point cloud. This task is invariant to translations but equivariant to both rotations and permutations. + +In this paper, we are interested in learning equivariant functions from point clouds into $W _ { T } ^ { n }$ , where $W _ { T }$ is some representation of SO(3). The equivariance of these functions is with respect to the action $\rho _ { G }$ on point clouds defined in equation 1, and the action of $G$ on $W _ { T } ^ { n }$ defined by applying the rotation action from the left and permutation action from the right as in 1, but ‘ignoring’ the translation component. Thus, $G$ -equivariant functions will be translation invariant. This formulation of equivariance includes the normals prediction example by taking $W _ { T } = \mathbb { R } ^ { 3 }$ , as well as the segmentation case by setting $W _ { T } = \mathbb { R }$ with the trivial identity representation. We focus on the harder case of functions into $W _ { T } ^ { n }$ which are equivariant to permutations, since it easily implies the easier case of permutation invariant functions to $W _ { T }$ . + +Notation. We use the notation $ { \mathbb { N } } _ { + } = { \mathbb { N } } \cup \lbrace 0 \rbrace$ and $\begin{array} { r } { \mathbb { N } _ { + } ^ { * } = \bigcup _ { r \in \mathbb { N } } \mathbb { N } _ { + } ^ { r } } \end{array}$ . We set $[ D ] = \{ 1 , \dots , D \}$ and $[ D ] _ { 0 } = \{ 0 , \dots , D \}$ . + +Proofs. Proofs appear in the appendices, arranged according to sections. + +# 3.2 CONDITIONS FOR UNIVERSALITY + +The semi-lifted approach In general, highly expressive equivariant neural networks can be achieved by using a ‘lifted approach’, where intermediate features in the network belong to high dimensional representations of the group. In the context of point clouds where typically $n \gg 3$ , many papers, e.g., (Thomas et al., 2018; Kondor, 2018; Bogatskiy et al., 2020) use a ‘semi-lifted’ approach, where hidden layers hold only higher dimensional representations of SO(3), but not high order permutation representations. In this subsection, we propose a strategy for achieving universality with the semi-lifted approach. + +We begin by an axiomatic formulation of the semi-lifted approach (see illustration in inset): we assume that our neural networks are composed of two main components: the first component is a family $\mathcal { F } _ { \mathrm { f e a t } }$ of parametric continuous $G$ -equivariant functions $f _ { \mathrm { f e a t } }$ which map the original point cloud $\mathbb { R } ^ { 3 \times n }$ to a semi-lifted point cloud $\begin{array} { c c l } { W _ { \mathrm { f e a t } } ^ { n } } & { = } & { \bigoplus _ { i = 1 } ^ { n } W _ { \mathrm { f e a t } } } \end{array}$ , where $W _ { \mathrm { f e a t } }$ is a lifted (i.e., high-order) representation of SO(3). + +The second component is a family of parametric linear SO(3)-equivariant functions $\mathcal { F } _ { \mathrm { p o o l } }$ , which map from the high order representation $W _ { \mathrm { f e a t } }$ down to the target representation $W _ { T }$ . Each such $\mathrm { S O ( 3 ) }$ –equivariant function $\Lambda$ : $W _ { \mathrm { f e a t } } \to W _ { T }$ can be extended to a $\mathrm { S O } ( 3 ) \times S _ { n }$ equivariant function $\hat { \Lambda } : W _ { \mathrm { f e a t } } ^ { n } \to W _ { T } ^ { n }$ by applying $\Lambda$ elementwise. For every positive integer $C$ , these two families of functions induce a family of functions $\mathcal { F } _ { C }$ obtained by summing $C$ different compositions of these functions: + +![](images/b54a82d14eb02b10dbee355a8abcb92b8088b25aadb6409a1bedc597d43bf20d.jpg) + +$$ +\mathcal { F } _ { C } ( \mathcal { F } _ { \mathrm { f e a t } } , \mathcal { F } _ { \mathrm { p o o l } } ) = \{ f | f ( X ) = \sum _ { c = 1 } ^ { C } \hat { \Lambda } _ { c } ( g _ { c } ( X ) ) , \ : ( \Lambda _ { c } , g _ { c } ) \in \mathcal { F } _ { \mathrm { p o o l } } \times \mathcal { F } _ { \mathrm { f e a t } } \} . +$$ + +Conditions for universality We now describe two conditions that guarantee universality using the semi-lifted approach. The first step is showing, as in (Yarotsky, 2018), that continuous $G$ -equivariant functions $\mathcal { C } _ { G } \mathrm { ( } \mathbb { R } ^ { 3 \times n } , W _ { T } ^ { n } \mathrm { ) }$ can be approximated by $G$ -equivariant polynomials $\mathcal { P } _ { G } ( \mathbb { R } ^ { 3 \times n } , \bar { W } _ { T } ^ { n } )$ . + +Lemma 1. Any continuous $G$ -equivariant function in $\mathcal { C } _ { G } ( \mathbb { R } ^ { 3 \times n } , W _ { T } ^ { n } )$ can be approximated uniformly on compact sets by $G$ -equivariant polynomials in $\mathcal { P } _ { G } ( \mathbb { R } ^ { 3 \times n } , \bar { W } _ { T } ^ { n } )$ . + +Universality is now reduced to the approximation of $G$ -equivariant polynomials. Next, we provide two conditions which guarantee that $G$ -equivariant polynomials of degree $D$ can be expressed by function spaces $\mathcal { F } _ { C } ( \mathcal { F } _ { \mathrm { f e a t } } , \mathcal { F } _ { \mathrm { p o o l } } )$ as defined in equation 2. The idea behind these conditions is that explicit characterizations of polynomials equivariant to the joint action of translations, rotations and permutations is challenging. However, it is possible to explicitly characterize polynomials equivariant to translations and permutations (but not rotations). The key observation is then that this characterization can be rewritten as a sum of functions to $W _ { \mathrm { f e a t } } ^ { n }$ , a high dimensional representations of SO(3) which is equivariant to translations, permutations and rotations, composed with a linear map which is permutation equivariant (but does not respect rotations). Accordingly, our first condition is that $\mathcal { F } _ { \mathrm { f e a t } }$ contains a spanning set of such functions to $W _ { \mathrm { f e a t } } ^ { n }$ . We call this condition $D$ -spanning: + +Definition 1 ( $D$ -spanning). For $D \in { \mathbb { N } } _ { + }$ , let $\mathcal { F } _ { \mathrm { f e a t } }$ be a subset of $\mathcal { C } _ { G } ( \mathbb { R } ^ { 3 \times n } , W _ { \mathrm { f e a t } } ^ { n } )$ . We say that $\mathcal { F } _ { \mathrm { f e a t } }$ is $D$ -spanning, if there exist $f _ { 1 } , \dots , f _ { K } \in \mathcal { F } _ { \mathrm { f e a t } }$ , such that every polynomial $p : \mathbb { R } ^ { 3 \times n } \mathbb { R } ^ { n }$ of degree $D$ which is invariant to translations and equivariant to permutations, can be written as + +$$ +p ( X ) = \sum _ { k = 1 } ^ { K } \hat { \Lambda } _ { k } ( f _ { k } ( X ) ) , +$$ + +where $\Lambda _ { k } : W _ { \mathrm { f e a t } } \to \mathbb { R }$ are all linear functionals, and ${ \hat { \boldsymbol { \Lambda } } } _ { k } : { \cal W } _ { \mathrm { f e a t } } ^ { n } \mathbb { R } ^ { n }$ are the functions defined by elementwise applications of $\Lambda _ { k }$ . + +In Lemma 4 we explicitly construct a $D$ -spanning family of functions. This provides us with a concrete condition which implies $D$ -spanning for other function families as well. + +The second condition is that $\mathcal { F } _ { \mathrm { p o o l } }$ contains all SO(3) linear equivariant layers. We call this condition Linear universality. Intuitively, taking linear rotation equivariant $\Lambda _ { k }$ in equation 3 ensures that the resulting function $p$ will be rotation equivariant and thus fully $G$ -equivariant, and linear universality guarantees the ability to express all such $G$ invariant functions. + +Definition 2 (Linear universality). We say that a collection $\mathcal { F } _ { \mathrm { p o o l } }$ of equivariant linear functionals between two representations $W _ { \mathrm { f e a t } }$ and $W _ { T }$ of SO(3) is linearly universal, if it contains all linear SO(3)-equivariant mappings between the two representations. + +When these two conditions apply, a rather simple symmetrization arguments leads to the following theorem: + +Theorem 1. If $\mathcal { F } _ { \mathrm { f e a t } }$ is $D$ -spanning and $\mathcal { F } _ { \mathrm { p o o l } }$ is linearly universal, then there exists some $C ( D ) \in$ $\mathbb { N }$ such that for all $C \geq C ( D )$ the function space $\mathcal { F } _ { C } ( \mathcal { F } _ { \mathrm { f e a t } } , \mathcal { F } _ { \mathrm { p o o l } } )$ contains all $G$ -equivariant polynomials of degree $\le D$ . + +Proof idea. By the $D$ -spanning assumption, there exist $f _ { 1 } , \dots , f _ { K } \in { \mathcal { F } } _ { \mathrm { f e a t } }$ such that any vector valued polynomial invariant to translations and equivariant to permutations is of the form + +$$ +p ( X ) = \sum _ { k = 1 } ^ { K } \hat { \Lambda } _ { k } ( f _ { k } ( X ) ) , +$$ + +While by definition this holds for functions $p$ whose image is $\mathbb { R } ^ { n }$ , this is easily extended to functions to $W _ { T } ^ { n }$ as well. + +It remains to show that when $p$ is also SO(3)-equivariant, we can choose $\Lambda _ { k }$ to be SO(3) equivariant. +This is accomplished by averaging over SO(3). + +As a result of Theorem 1 and Lemma 1 we obtain our universality result (see inset for illustration) + +Corollary 1. For all ${ \cal C } , { \cal D } \in \mathbb { N } _ { + }$ , let $\mathcal { F } _ { C , D }$ denote function spaces generated by a pair of functions spaces which are $D$ -spanning and linearly universal as in equation 2. Then any continuous $G$ -equivariant function in $\mathcal { C } _ { G } ( \mathbb { R } ^ { 3 \times n } , W _ { T } ^ { n } )$ can be approximated uniformly on compact sets by equivariant functions in + +![](images/e9ffa6a3d2fc7c2c788f5f90cefc79b1dd49ce1bf98aa7f80e26cd6b3b4cedca.jpg) + +$$ +{ \mathcal { F } } = \bigcup _ { D \in \mathbb { N } } { \mathcal { F } } _ { C ( D ) , D } . +$$ + +# 3.3 UNIVERSALITY CONDITIONS IN ACTION + +In the remainder of the paper, we prove the universality of several $G$ -equivariant architectures, based on the framework we discussed in the previous subsection. We discuss two different strategies for achieving universality, which differ mainly in the type of lifted representations of SO(3) they use: (i) The first strategy uses (direct sums of) tensor-product representations; (ii) The second uses (direct sums of) irreducible representations. The main advantage of the first strategy from the perspective of our methodology is that achieving the $D$ -spanning property is more straightforward. The advantage of irreducible representations is that they almost automatically guarantees the linear universality property. + +In Section 4 we discuss universality through tensor product representations, and give an example of a minimal tensor representation network architecture that would satisfy universality. In section 5 we discuss universality through irreducible representations, which is currently the more common strategy. We show that the TFN architecture (Thomas et al., 2018; Fuchs et al., 2020) which follows this strategy is universal, and describe a simple tweak that achieves universality using only low order filters, though the representations throughout the network are high dimensional. + +# 4 UNIVERSALITY WITH TENSOR REPRESENTATIONS + +In this section, we prove universality for models that are based on tensor product representations, as defined below. The main advantage of this approach is that $D$ -spanning is achieved rather easily. The main drawbacks are that its data representation is somewhat redundant and that characterizing the linear equivariant layers is more laborious. + +Tensor representations We begin by defining tensor representations. For $k \in \mathbb { N } _ { + }$ denote $\mathcal { T } _ { k } = \mathbb { R } ^ { 3 ^ { k } }$ . SO(3) acts on $\mathcal { T } _ { k }$ by the tensor product representation, i.e., by applying the matrix Kronecker product $k$ times: $\rho _ { k } ( R ) : =$ $\bar { R } ^ { \otimes k }$ . The inset illustrates the vector spaces and action for $k = 1 , 2 , 3$ . With this action, for any $\bar { i } _ { 1 } , \ldots , i _ { k } \in [ n ]$ , the map from $\mathbb { R } ^ { 3 \times n }$ to $\mathcal { T } _ { k }$ defined by + +![](images/50d9d12099cc8cbcdfa045b7f9ad7b985e39b62e026d157b80b6dee7accd6a00.jpg) + +$$ +( x _ { 1 } , \ldots , x _ { n } ) \mapsto x _ { i _ { 1 } } \otimes x _ { i _ { 2 } } \ldots \otimes x _ { i _ { k } } +$$ + +is SO(3) equivariant. + +A $D$ -spanning family We now show that tensor representations can be used to define a finite set of $D$ -spanning functions. The lifted representation $W _ { \mathrm { f e a t } }$ will be given by + +$$ +W _ { \mathrm { f e a t } } ^ { \mathcal { T } } = \bigoplus _ { T = 0 } ^ { D } \mathcal { T } _ { T } . +$$ + +The $D$ -spanning functions are indexed by vectors $\vec { r } = \left( r _ { 1 } , \ldots , r _ { K } \right)$ , where each $r _ { k }$ is a non-negative integer. Denoting $T = \| r \| _ { 1 }$ , the functions $Q ^ { ( \vec { r } ) } : \mathbb { R } ^ { 3 \times n } \mathcal { T } _ { T } ^ { n }$ , $Q ^ { ( \vec { r } ) } = ( Q _ { j } ^ { ( \vec { r } ) } ) _ { j = 1 } ^ { n }$ ))nj=1 are defined for fixed $j \in [ n ]$ by + +$$ +Q _ { j } ^ { ( \vec { r } ) } ( X ) = \sum _ { i _ { 2 } , \ldots , i _ { K } = 1 } ^ { n } x _ { j } ^ { \otimes r _ { 1 } } \otimes x _ { i _ { 2 } } ^ { \otimes r _ { 2 } } \otimes x _ { i _ { 3 } } ^ { \otimes r _ { 3 } } \otimes \ldots \otimes x _ { i _ { K } } ^ { \otimes r _ { K } } . +$$ + +The functions $Q _ { j } ^ { ( \vec { r } ) }$ are SO(3) equivariant as they are a sum of equivariant functions from equation 5. Thus $Q ^ { ( \vec { r } ) }$ is $\mathrm { S O } ( 3 ) \times S _ { n }$ equivariant. The motivation behind the definition of these functions is that known characterizations of permutation equivariant polynomials Segol & Lipman (2019) tell us that the entries of these tensor valued functions span all permutation equivariant polynomials (see the proof of Lemma 2 for more details). + +To account for translation invariance, we compose the functions $Q ^ { ( \vec { r } ) }$ with the centralization operation and define the set of functions + +$$ +\mathcal { Q } _ { D } = \{ \iota \circ Q ^ { ( \vec { r } ) } ( X - \frac { 1 } { n } X 1 _ { n } \boldsymbol { 1 } _ { n } ^ { T } ) | \| \vec { r } \| _ { 1 } \leq D \} , +$$ + +where $\iota$ is the natural embedding that takes each $\mathcal { T } _ { T }$ into $W _ { \mathrm { f e a t } } ^ { \mathcal { T } } = \oplus _ { T = 0 } ^ { D } \mathcal { T } _ { T }$ . In the following lemma, we prove that this set is $D$ -spanning. + +Lemma 2. For every $D \in { \mathbb { N } } _ { + }$ , the set $\mathcal { Q } _ { D }$ is $D$ -spanning. + +Proof idea. It is known (Segol & Lipman, 2019) (Theorem 2) that polynomials $p : \mathbb { R } ^ { 3 \times n } \mathbb { R } ^ { n }$ which are $S _ { n }$ -equivariant, are spanned by polynomials of the form $p _ { \vec { \alpha } } = ( p _ { \vec { \alpha } } ^ { j } ) _ { j = 1 } ^ { n }$ , defined as + +$$ +p _ { \vec { \alpha } } ^ { j } ( X ) = \sum _ { i _ { 2 } , \ldots , i _ { K } = 1 } ^ { n } x _ { j } ^ { \alpha _ { 1 } } x _ { i _ { 2 } } ^ { \alpha _ { 2 } } \cdot \cdot \cdot x _ { i _ { k } } ^ { \alpha _ { k } } +$$ + +where $\vec { \alpha } = ( \alpha _ { 1 } , \ldots , \alpha _ { K } )$ and each $\alpha _ { k } \in \mathbb { N } _ { + } ^ { 3 }$ is a multi-index. We first show that these polynomials can be extracted from $Q ^ { ( \vec { r } ) }$ and then use them to represent $p$ . □ + +A minimal universal architecture Once we have shown that $\mathcal { Q } _ { D }$ is $D$ -spanning, we can design $D$ -spanning architectures, by devising architectures that are able to span all elements of $\mathcal { Q } _ { D }$ . As we will now show, the compositional nature of neural networks allows us to do this in a very clean manner. + +We define a parametric function $f ( X , V | \theta _ { 1 } , \theta _ { 2 } )$ which maps $\mathbb { R } ^ { 3 \times n } \oplus \mathcal { T } _ { k } ^ { n }$ to $\mathbb { R } ^ { 3 \times n } \oplus \mathcal { T } _ { k + 1 } ^ { n }$ as follows: For all $j \in [ n ]$ , we have $f _ { j } ( X , V ) = ( x _ { j } , \tilde { V } _ { j } ( X , V ) )$ , where + +$$ +\tilde { V } _ { j } ( X , V | \theta _ { 1 } , \theta _ { 2 } ) = \theta _ { 1 } \left( x _ { j } \otimes V _ { j } \right) + \theta _ { 2 } \sum _ { i } \left( x _ { i } \otimes V _ { i } \right) +$$ + +We denote the set of functions $( X , V ) \mapsto f ( X , V | \theta _ { 1 } , \theta _ { 2 } )$ obtained by choosing the parameters $\theta _ { 1 } , \theta _ { 2 } \in \mathbb { R }$ , by $\mathcal { F } _ { m i n }$ . While in the hidden layers of our network the data is represented using both coordinates $( X , V )$ , the input to the network only contains an $X$ coordinate and the output only contains a $V$ coordinate. To this end, we define the functions + +$$ +\operatorname { e x t } ( X ) = ( X , 1 _ { n } ) \operatorname { a n d } \pi _ { V } ( X , V ) = V . +$$ + +We can achieve $D$ -spanning by composition of functions in $\mathcal { F } _ { m i n }$ with these functions and centralizing: + +Lemma 3. The function set $\mathcal { Q } _ { D }$ is contained in + +$$ +\mathcal { F } _ { \mathrm { f e a t } } = \big \{ \iota \circ \pi _ { V } \circ f ^ { 1 } \circ f ^ { 2 } \circ \ldots \circ f ^ { T } \circ \mathrm { e x t } ( X - \frac { 1 } { n } X \boldsymbol { 1 } _ { n } \boldsymbol { 1 } _ { n } ^ { T } ) | f ^ { j } \in \mathcal { F } _ { m i n } , T \leq D \big \} . +$$ + +Thus $\mathcal { F } _ { \mathrm { f e a t } }$ is $D$ -spanning. + +Proof idea. The proof is technical and follows by induction on $D$ . + +To complete the construction of a universal network, we now need to characterize all linear equivariant functions from $W _ { \mathrm { f e a t } } ^ { \mathcal { T } }$ to the target representation $W _ { T }$ . In Appendix $\mathbf { G }$ we show how this can be done for the trivial representation $W _ { T } = \mathbb { R }$ . This characterization gives us a set of linear functions $\mathcal { F } _ { \mathrm { p o o l } }$ , which combined with $\mathcal { F } _ { \mathrm { f e a t } }$ defined in equation 11 (corresponds to SO(3) invariant functions) + +gives us a universal architecture as in Theorem 1. However, the disadvantage of this approach is that implementation of the linear functions in $\mathcal { F } _ { \mathrm { p o o l } }$ is somewhat cumbersome. + +In the next section we discuss irreducible representations, which give us a systematic way to address linear equivariant mappings into any $W _ { T }$ . Proving $D$ -spanning for these networks is accomplished via the $D$ -spanning property of tensor representations, through the following lemma + +Lemma 4. If all functions in $\mathcal { Q } _ { D }$ can be written as + +$$ +\iota \circ Q ^ { ( \vec { r } ) } ( X - \frac { 1 } { n } X 1 _ { n } 1 _ { n } ^ { T } ) = \sum _ { k = 1 } ^ { K } \hat { A } _ { k } f _ { k } ( X ) , +$$ + +where $f _ { k } \in \mathcal { F } _ { \mathrm { f e a t } }$ , $A _ { k } : W _ { \mathrm { f e a t } } \to W _ { \mathrm { f e a t } } ^ { \mathcal { T } }$ and ${ \hat { A } } _ { k } : W _ { \mathrm { f e a t } } ^ { n } \to ( W _ { \mathrm { f e a t } } ^ { \mathcal { T } } ) ^ { n }$ is defined by elementwise application of $A _ { k }$ , then $\mathcal { F } _ { \mathrm { f e a t } }$ is $D$ -spanning. + +We note that as before, $A _ { k }$ are not necessarily SO(3)- equivariant. + +Proof idea. The lemma follows directly from the assumptions. + +# 5 UNIVERSALITY WITH IRREDUCIBLE REPRESENTATIONS + +In this section, we discuss how to achieve universality when using irreducible representations of SO(3). We will begin by defining irreducible representations, and explaining how linear universality is easily achieved by them, while the $D$ -spanning properties of tensor representations can be preserved. This discussion can be seen as an interpretation of the choices made in the construction of TFN and similar networks in the literature. We then show that these architectures are indeed universal. + +# 5.1 IRREDUCIBLE REPRESENTATIONS OF $S O ( 3 )$ + +In general, any finite-dimensional representation $W$ of a compact group $H$ can be decomposed into irreducible representations: a subspace $W _ { 0 } \subset W$ is $H$ -invariant if $h w \in W _ { 0 }$ for all $h \in H , w \in W _ { 0 }$ . A representation $W$ is irreducible if it has no non-trivial invariant subspaces. In the case of SO(3), all irreducible real representations are defined by matrices $D ^ { ( \ell ) } ( R )$ , called the real Wigner D-matrices, acting on $W _ { \ell } : = \overset { \bullet } { \mathbb { R } } ^ { 2 \ell + 1 }$ by matrix multiplication. In particular, the representation for $\ell = 0 , 1$ are $D ^ { ( 0 ) } ( R ) = 1$ and $D ^ { ( 1 ) } ( R ) = R$ . + +Linear maps between irreducible representations As mentioned above, one of the main advantages of using irreducible representations is that there is a very simple characterization of all linear equivariant maps between two direct sums of irreducible representations. We use the notation $W _ { l }$ for direct sums of irreducible representations, where $\boldsymbol { l } = ( \ell _ { 1 } , \dots , \ell _ { K } ) \in \mathbb { N } _ { + } ^ { K }$ and $W _ { l } = \bigoplus _ { k = 1 } ^ { K } W _ { \ell _ { k } }$ + +Lemma 5. Let $l ^ { ( 1 ) } = ( \ell _ { 1 } ^ { ( 1 ) } , \dots , \ell _ { K _ { 1 } } ^ { ( 1 ) } )$ and $l ^ { ( 2 ) } = ( \ell _ { 1 } ^ { ( 2 ) } , \dots , \ell _ { K _ { 2 } } ^ { ( 2 ) } )$ . A function $\boldsymbol { \Lambda } = \left( \Lambda _ { 1 } , \ldots , \Lambda _ { K _ { 2 } } \right)$ is a linear equivariant mapping between $W _ { l ^ { ( 1 ) } }$ and $W _ { l ^ { ( 2 ) } }$ , if and only if there exists a $K _ { 1 } \times K _ { 2 }$ matrix $M$ with $M _ { i j } = 0$ whenever $\bar { \ell } _ { i } ^ { ( 1 ) } \neq \ell _ { j } ^ { ( 2 ) }$ , such that + +$$ +\Lambda _ { j } ( V ) = \sum _ { i = 1 } ^ { K _ { 1 } } M _ { i j } V _ { i } +$$ + +where $V = ( V _ { i } ) _ { i = 1 } ^ { K _ { 1 } }$ and $V _ { i } \in W _ { \ell _ { i } ^ { ( 1 ) } }$ for all $i = 1 , \ldots , K _ { 1 }$ + +Proof idea. This lemma is a simple generalization of Schur’s lemma, a classical tool in representation theory, which asserts that a non-zero linear map between irreducible representations is a scaled multiply of the identity mapping. Lemma 5 was stated in the complex setting in Kondor (2018). While Schur’s lemma, and thus Lemma 5, does not always hold for representations over the reals, we observe here that it holds for real irreducible representations of SO(3) since their dimension is always odd. □ + +Clebsch-Gordan decomposition of tensor products As any finite-dimensional representation of SO(3) can be decomposed into a direct sum of irreducible representations, this is true for tensor representations as well. In particular, the Clebsch-Gordan coefficients provide an explicit formula for decomposing the tensor product of two irreducible representations $W _ { \ell _ { 1 } }$ and $W _ { \ell _ { 2 } }$ into a direct sum of irreducible representations. This decomposition can be easily extended to decompose the tensor product $W _ { l _ { 1 } } \otimes W _ { l _ { 2 } }$ into a direct sum of irreducible representations, where $l _ { 1 } , l _ { 2 }$ are now vectors. In matrix notation, this means there is a unitary linear equivariant $U ( l _ { 1 } , l _ { 2 } )$ mapping of $W _ { l _ { 1 } } \otimes W _ { l _ { 2 } }$ onto $W _ { l }$ , where the explicit values of $l = l ( l _ { 1 } , l _ { 2 } )$ and the matrix $U ( l _ { 1 } , l _ { 2 } )$ can be inferred directly from the case where $\ell _ { 1 }$ and $\ell _ { 2 }$ are scalars. + +By repeatedly taking tensor products and applying Clebsch-Gordan decompositions to the result, TFN and similar architectures can achieve the $D$ -spanning property in a manner analogous to tensor representations, and also enjoy linear universality since they maintain irreducible representations throughout the network. + +# 5.2 TENSOR FIELD NETWORKS + +We now describe the basic layers of the TFN architecture (Thomas et al., 2018), which are based on irreducible representations, and suggest an architecture based on these layers which can approximate $G$ -equivariant maps into any representation $W _ { l _ { T } } ^ { n } , l _ { T } \in \mathbb { N } _ { + } ^ { * }$ . There are some superficial differences Tbetween our description of TFN and the description in the original paper, for more details see Appendix F. + +We note that the universality of TFN also implies the universality of Fuchs et al. (2020), which is a generalization of TFN that enables adding an attention mechanism. Assuming the attention mechanism is not restricted to local neighborhoods, this method is at least as expressive as TFN. + +TFNs are composed of three types of layers: (i) Convolution (ii) Self-interaction and (iii) Nonlinearities. In our architecture, we only use the first two layers types, which we will now describe:1. + +Convolution. Convolutional layers involve taking tensor products of a filter and a feature vector to create a new feature vector, and then decomposing into irreducible representations. Unlike in standard CNN, a filter here depends on the input, and is a function $F : \bar { \mathbb { R } ^ { 3 } } W _ { l _ { D } }$ , where $\boldsymbol { l } _ { D } =$ $[ 0 , 1 , \ldots , D ] ^ { T }$ . The $\ell$ -th component of the filter $F ( \boldsymbol { x } ) = \left[ F ^ { ( 0 ) } ( x ) , \ldots , F ^ { ( D ) } ( x ) \right]$ will be given by + +$$ +F _ { m } ^ { \ell } ( x ) = R ^ { ( \ell ) } ( \| x \| ) Y _ { m } ^ { \ell } ( \hat { x } ) , m = - \ell , \ldots , \ell +$$ + +where $\hat { x } = x / \| x \|$ if $x \neq 0$ and $\hat { x } ~ = ~ 0$ otherwise, $Y _ { m } ^ { \ell }$ are spherical harmonics, and $R ^ { ( \ell ) }$ any polynomial of degree $\leq D$ . In Appendix F we show that these polynomial functions can be replaced by fully connected networks, since the latter can approximate all polynomials uniformly. + +The convolution of an input feature $V \in W _ { l _ { i } } ^ { n }$ and a filter $F$ as defined above, will give an output feature $\tilde { V } = ( \tilde { V } _ { a } ) _ { a = 1 } ^ { n } \in W _ { l _ { 0 } } ^ { n }$ , where $\boldsymbol { l _ { o } } = \boldsymbol { l } ( \boldsymbol { l _ { f } } , \boldsymbol { l _ { i } } )$ , which is given by + +$$ +\tilde { V } _ { a } ( X , V ) = U ( l _ { f } , l _ { i } ) \left( \theta _ { 0 } V _ { a } + \sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \otimes V _ { b } \right) . +$$ + +More formally we will think of convolutional layer as functions of the form $f ( X , V ) \quad = \quad$ $( X , { \tilde { V } } ( X , V ) )$ . These functions are defined by a choice of $D$ , a choice of a scalar polynomial $R ^ { ( \ell ) } , \ell = 0 , \dots , D$ , and a choice of the parameter $\theta _ { 0 } \in \mathbb { R }$ in equation 14. We denote the set of all such functions $f$ by $\mathcal { F } _ { D }$ . + +Self Interaction layers. Self interaction layers are linear functions from $\hat { \Lambda } : W _ { l } ^ { n } \to W _ { l , r } ^ { n }$ , which are obtained from elementwise application of equivariant linear functions $\Lambda : W _ { l } \stackrel { \cdot } { } W _ { l _ { T } }$ . These linear functions can be specified by a choice of matrix $M$ with the sparsity pattern described in Lemma 5. + +Activation functions. TFN, as well as other papers, proposed several activation functions. We find that these layers are not necessary for universality and thus we do not define them here. + +Network architecture. For our universality proof, we suggest a simple architecture which depends on two positive integer parameters $( C , D )$ : For given $D$ , we will define $\mathcal { F } _ { \mathrm { f e a t } } ( D )$ as the set of + +function obtained by $2 D$ recursive convolutions + +$$ +{ \mathcal { F } } _ { \mathrm { f e a t } } ( D ) = \{ \pi _ { V } \circ f ^ { 2 D } \circ \ldots f ^ { 2 } \circ f ^ { 1 } \circ \operatorname { e x t } ( X ) | f ^ { j } \in { \mathcal { F } } _ { D } \} , +$$ + +where ext and πV are defined as in equation 10. The output of a function in Ffeat(D) is in W nl(D), for some $\boldsymbol { l } ( D )$ which depends on $D$ . We then define $\mathcal { F } _ { \mathrm { p o o l } } ( D )$ to be the self-interaction layers which map $W _ { l ( D ) } ^ { n }$ to $W _ { l _ { T } } ^ { n }$ . This choice of $\mathcal { F } _ { \mathrm { f e a t } } ( D )$ and $\mathcal { F } _ { \mathrm { p o o l } } ( D )$ , together with a choice of the number of channels $C$ T, defines the final network architecture $\mathcal { F } _ { C , D } ^ { \mathrm { T F N } } = \mathcal { F } _ { C } ( \mathcal { F } _ { \mathrm { f e a t } } ( D ) , \mathcal { F } _ { \mathrm { p o o l } } ( D ) )$ as in equation 2. In the appendix we prove the universality of TFN: + +Theorem 2. For all $n \in \mathbb { N } , l _ { T } \in \mathbb { N } _ { + } ^ { * }$ + +1. For $D \in { \mathbb { N } } _ { + }$ , every $G$ -equivariant polynomial $p : \mathbb { R } ^ { 3 \times n } W _ { T } ^ { n }$ of degree $D$ is in $\mathcal { F } _ { C ( D ) , D } ^ { \mathrm { T F N } }$ 2. Every continuoby functions in $G$ t function can be approximated uniformly on compact sets $\cup _ { D \in \mathbb { N } _ { + } } \mathcal { F } _ { C ( D ) , D } ^ { \mathrm { T F N } }$ + +As discussed previously, the linear universality of $\mathcal { F } _ { \mathrm { p o o l } }$ is guaranteed. Thus proving Theorem 2 amounts to showing that $\mathcal { F } _ { \mathrm { f e a t } } ( D )$ is $D$ -spanning. This is done using the sufficient condition for $D$ -spanning defined in Lemma 4. + +Proof idea. The proof is rather technical and involved. A useful observation (see Dai & Xu (2013)) used in the proof is that the filters of orders $\ell = 0 , 1 , \ldots , D$ , defined in equation 13, span all polynomial functions of degree $D$ on $\mathbb { R } ^ { 3 }$ . This observation is used to show that all functions in $\mathcal { Q } _ { D }$ can be expressed by $\mathcal { F } _ { \mathrm { f e a t } } ( D )$ and so $\mathcal { F } _ { \mathrm { f e a t } }$ is $D$ -spanning, as stated in Lemma 2. □ + +Alternative architecture The complexity of the TFN network used to construct $G$ -equivariant polynomials of degree $D$ , can be reduced using a simple modifications of the convolutional layer in equation 14: We add two parameters $\theta _ { 1 } , \theta _ { 2 } \in \mathbb { R }$ to the convolutional layer, which is now defined as: + +$$ +\tilde { V } _ { a } ( X , V ) = U ( l _ { f } , l _ { i } ) \left( \theta _ { 1 } \sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \otimes V _ { b } + \theta _ { 2 } \sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \otimes V _ { a } \right) . +$$ + +With this simple change, we can show that $\mathcal { F } _ { \mathrm { f e a t } } ( D )$ is $D$ -spanning even if we only take filters of order 0 and 1 throughout the network. This is shown in Appendix E. + +# 6 CONCLUSION + +In this paper, we have presented a new framework for proving the universality of $G$ -equivariant point cloud networks. We used this framework for proving the universality of the TFN model Thomas et al. (2018); Fuchs et al. (2020), and for devising two additional novel simple universal architectures. In the future we hope to extend these simple constructions to operational $G$ -equivariant networks with universality guarantees and competitive practical performance. + +Our universal architectures do not require activation functions, and use a single self-interaction layer. In Appendix H we present an experiment indicating that the performance of TFN is not significantly altered by these simplifications. Our architectures also require high order representations, and our experiments show that using increasingly high order representations does indeed improve performance. To date, practical TFN implementation included a relatively small amount of layers, and did not use very high order representations. We believe our theoretical results will inspire interest in stable implementation of larger architectures. On the other hand, an interesting open problem is understanding whether universality can be achieved using only low-dimensional representations. + +Finally, we believe that the framework we developed here will be useful for proving the universality of other $G$ -equivariant models for point cloud networks, and other related equivariant models. We note that large parts of our discussion can be easily generalized to symmetry groups of the form $G = \mathbb { R } ^ { d } \rtimes \breve { H } \times \bar { S } _ { n }$ acting on $\mathbb { R } ^ { d \times n }$ , where $H$ can be any compact topological group. + +Acknowledgments The authors would like to thank Fabian B. Fuchs for making code available and Taco Cohen for helpful discussion. N.D. is supported by THEORINET Simons award 814643. + +# REFERENCES + +Matan Atzmon, Haggai Maron, and Yaron Lipman. Point convolutional neural networks by extension operators. arXiv preprint arXiv:1803.10091, 2018. + +Alexander Bogatskiy, Brandon Anderson, Jan T Offermann, Marwah Roussi, David W Miller, and Risi Kondor. Lorentz group equivariant neural network for particle physics. arXiv preprint arXiv:2006.04780, 2020. + +Taco S Cohen, Mario Geiger, Jonas Kohler, and Max Welling. Spherical cnns. ¨ arXiv preprint arXiv:1801.10130, 2018. + +Feng Dai and Yuan Xu. Approximation theory and harmonic analysis on spheres and balls, volume 23. Springer, 2013. + +Carlos Esteves, Christine Allen-Blanchette, Ameesh Makadia, and Kostas Daniilidis. Learning so (3) equivariant representations with spherical cnns. In Proceedings of the European Conference on Computer Vision (ECCV), pp. 52–68, 2018. + +Fabian B Fuchs, Daniel E Worrall, Volker Fischer, and Max Welling. Se (3)-transformers: 3d rototranslation equivariant attention networks. arXiv preprint arXiv:2006.10503, 2020. + +William Fulton and Joe Harris. Representation theory: a first course, volume 129. Springer Science & Business Media, 2013. + +Justin Gilmer, Samuel S Schoenholz, Patrick F Riley, Oriol Vinyals, and George E Dahl. Neural message passing for quantum chemistry. arXiv preprint arXiv:1704.01212, 2017. + +Yulan Guo, Hanyun Wang, Qingyong Hu, Hao Liu, Li Liu, and Mohammed Bennamoun. Deep learning for 3d point clouds: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020. + +Nicolas Keriven and Gabriel Peyre. Universal invariant and equivariant graph neural networks. ´ CoRR, abs/1905.04943, 2019. URL http://arxiv.org/abs/1905.04943. + +Risi Kondor. N-body networks: a covariant hierarchical neural network architecture for learning atomic potentials. arXiv preprint arXiv:1803.01588, 2018. + +Risi Kondor, Zhen Lin, and Shubhendu Trivedi. Clebsch–gordan nets: a fully fourier space spherical convolutional neural network. In Advances in Neural Information Processing Systems, pp. 10117– 10126, 2018a. + +Risi Kondor, Hy Truong Son, Horace Pan, Brandon Anderson, and Shubhendu Trivedi. Covariant compositional networks for learning graphs. arXiv preprint arXiv:1801.02144, 2018b. + +Hanspeter Kraft and Claudio Procesi. Classical invariant theory, a primer. Lecture Notes, Version, 2000. + +Jiaxin Li, Yingcai Bi, and Gim Hee Lee. Discrete rotation equivariance for point cloud recognition. In 2019 International Conference on Robotics and Automation (ICRA), pp. 7269–7275. IEEE, 2019. + +Takanori Maehara and Hoang NT. A simple proof of the universality of invariant/equivariant graph neural networks, 2019. + +Haggai Maron, Heli Ben-Hamu, Hadar Serviansky, and Yaron Lipman. Provably powerful graph networks. arXiv preprint arXiv:1905.11136, 2019a. + +Haggai Maron, Heli Ben-Hamu, Nadav Shamir, and Yaron Lipman. Invariant and equivariant graph networks. In International Conference on Learning Representations, 2019b. URL https: //openreview.net/forum?id=Syx72jC9tm. + +Haggai Maron, Ethan Fetaya, Nimrod Segol, and Yaron Lipman. On the universality of invariant networks. In International conference on machine learning, 2019c. + +Haggai Maron, Or Litany, Gal Chechik, and Ethan Fetaya. On learning sets of symmetric elements. arXiv preprint arXiv:2002.08599, 2020. + +Christopher Morris, Martin Ritzert, Matthias Fey, William L Hamilton, Jan Eric Lenssen, Gaurav Rattan, and Martin Grohe. Weisfeiler and leman go neural: Higher-order graph neural networks. arXiv preprint arXiv:1810.02244, 2018. + +Adam Paszke, Sam Gross, Soumith Chintala, Gregory Chanan, Edward Yang, Zachary DeVito, Zeming Lin, Alban Desmaison, Luca Antiga, and Adam Lerer. Automatic differentiation in pytorch. 2017. + +Adrien Poulenard, Marie-Julie Rakotosaona, Yann Ponty, and Maks Ovsjanikov. Effective rotationinvariant point cnn with spherical harmonics kernels. In 2019 International Conference on 3D Vision (3DV), pp. 47–56. IEEE, 2019. + +Charles R Qi, Hao Su, Kaichun Mo, and Leonidas J Guibas. Pointnet: Deep learning on point sets for 3d classification and segmentation. Proc. Computer Vision and Pattern Recognition (CVPR), IEEE, 1(2):4, 2017a. + +Charles Ruizhongtai Qi, Li Yi, Hao Su, and Leonidas J Guibas. Pointne $^ { + + }$ : Deep hierarchical feature learning on point sets in a metric space. In Advances in neural information processing systems, pp. 5099–5108, 2017b. + +Raghunathan Ramakrishnan, Pavlo O Dral, Matthias Rupp, and O Anatole Von Lilienfeld. Quantum chemistry structures and properties of 134 kilo molecules. Scientific data, 1(1):1–7, 2014. + +Siamak Ravanbakhsh. Universal equivariant multilayer perceptrons. arXiv preprint arXiv:2002.02912, 2020. + +Nimrod Segol and Yaron Lipman. On universal equivariant set networks. arXiv preprint arXiv:1910.02421, 2019. + +Hadar Serviansky, Nimrod Segol, Jonathan Shlomi, Kyle Cranmer, Eilam Gross, Haggai Maron, and Yaron Lipman. Set2graph: Learning graphs from sets. arXiv preprint arXiv:2002.08772, 2020. + +Nathaniel Thomas, Tess Smidt, Steven Kearnes, Lusann Yang, Li Li, Kai Kohlhoff, and Patrick Riley. Tensor field networks: Rotation-and translation-equivariant neural networks for 3d point clouds. arXiv preprint arXiv:1802.08219, 2018. + +Minjie Wang, Lingfan Yu, Da Zheng, Quan Gan, Yu Gai, Zihao Ye, Mufei Li, Jinjing Zhou, Qi Huang, Chao Ma, et al. Deep graph library: Towards efficient and scalable deep learning on graphs. arXiv preprint arXiv:1909.01315, 2019a. + +Yue Wang, Yongbin Sun, Ziwei Liu, Sanjay E Sarma, Michael M Bronstein, and Justin M Solomon. Dynamic graph cnn for learning on point clouds. Acm Transactions On Graphics (tog), 38(5): 1–12, 2019b. + +Maurice Weiler, Mario Geiger, Max Welling, Wouter Boomsma, and Taco Cohen. 3D Steerable CNNs: Learning Rotationally Equivariant Features in Volumetric Data. 2018. URL http: //arxiv.org/abs/1807.02547. + +Daniel Worrall and Gabriel Brostow. Cubenet: Equivariance to 3d rotation and translation. In Proceedings of the European Conference on Computer Vision (ECCV), pp. 567–584, 2018. + +Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural networks? In International Conference on Learning Representations, 2019. URL https: //openreview.net/forum?id $=$ ryGs6iA5Km. + +Dmitry Yarotsky. Universal approximations of invariant maps by neural networks. arXiv preprint arXiv:1804.10306, 2018. + +Manzil Zaheer, Satwik Kottur, Siamak Ravanbakhsh, Barnabas Poczos, Ruslan R Salakhutdinov, and Alexander J Smola. Deep sets. In Advances in neural information processing systems, pp. 3391–3401, 2017. + +Yongheng Zhao, Tolga Birdal, Jan Eric Lenssen, Emanuele Menegatti, Leonidas Guibas, and Federico Tombari. Quaternion equivariant capsule networks for 3d point clouds. arXiv preprint arXiv:1912.12098, 2019. + +# A NOTATION + +We introduce some notation for the proofs in the appendices. We use the shortened notation $\bar { X } =$ $\begin{array} { r } { X - \frac { 1 } { n } X \mathbb { 1 } _ { n } \mathbb { 1 } _ { n } ^ { T } } \end{array}$ and denote the columns of $\bar { X }$ by $\left( \hat { x } _ { 1 } , \ldots , \hat { x } _ { n } \right)$ . We denote + +$$ +\Sigma _ { T } = \{ \vec { r } \in \mathbb { N } _ { + } ^ { * } | \| \vec { r } \| _ { 1 } = T \} +$$ + +# B PROOFS FOR SECTION 3 + +# B.1 $G$ -EQUIVARIANT POLYNOMIALS ARE DENSE + +A first step in proving denseness of $G$ -equivariance polynomials, and in the proof used in the next subsection is the following simple lemma, which shows that translation invariance can be dealt with simply by centralizing the point cloud. + +In the following, $\rho _ { W _ { T } }$ is some representation of $\mathrm { S O ( 3 ) }$ on a finite dimensional real vector space $W _ { T }$ this induces an action $\rho _ { W _ { T } \times S _ { n } }$ of $\mathrm { S O } ( 3 ) \times S _ { n }$ on $W _ { T } ^ { n }$ by + +$$ +\rho _ { W _ { T } \times S _ { n } } ( R , P ) ( Y ) = \rho _ { W _ { T } } ( R ) Y P ^ { T } +$$ + +This is also the action of $G$ which we consider, $\rho _ { G } ~ = ~ \rho _ { \underline { { W } } _ { T } \times S _ { n } }$ , where we have invariance with respect to the translation coordinate. The action of $G$ on $\mathbb { R } ^ { 3 \times n }$ is defined in equation 1. + +Lemma B.1. A function $f : \mathbb { R } ^ { 3 \times n } W _ { T } ^ { n }$ is $G$ -equivariant, if and only if there exists a function $h$ which is equivariant with respect to the action of $\mathrm { S O } ( 3 ) \times S _ { n }$ on $\mathbb { R } ^ { 3 \times n }$ , and + +$$ +f ( X ) = h ( X - \frac 1 n X 1 _ { n } 1 _ { n } ^ { T } ) +$$ + +Proof. Recall that $G$ -equivariance means $\mathrm { S O } ( 3 ) \times S _ { n }$ equivariance and translation invariance. Thus if $f$ is $G$ -equivariant then equation 16 holds with $h = f$ . + +On the other hand, if $f$ satisfies equation 16 then we claim it is $G$ -equivariant. Indeed, for all $( t , R , P ) \in \mathbb { R } ^ { d } \rtimes \mathrm { S O ( 3 ) } \times S _ { n }$ , since $P ^ { T } 1 _ { n } 1 _ { n } ^ { T } = 1 _ { n } 1 _ { n } ^ { T } = 1 _ { n } 1 _ { n } ^ { T } P ^ { T }$ , + +$$ +\begin{array} { l } { f \left( \rho _ { G } ( t , R , P ) ( X ) \right) = f ( R ( X + t \boldsymbol { 1 } _ { n } ) P ^ { T } ) = h ( R ( X + t \boldsymbol { 1 } _ { n } ) P ^ { T } - \displaystyle \frac { 1 } { n } R ( X + t \boldsymbol { 1 } ) P ^ { T } \boldsymbol { 1 } _ { n } \boldsymbol { 1 } _ { n } ^ { T } ) } \\ { \displaystyle \quad \quad = h ( R ( X - \frac { 1 } { n } X \boldsymbol { 1 } _ { n } \boldsymbol { 1 } _ { n } ^ { T } ) P ^ { T } ) = h \left( \rho _ { \mathbb { R } ^ { 3 } \times S _ { n } } ( R , P ) ( X - \frac { 1 } { n } X \boldsymbol { 1 } _ { n } \boldsymbol { 1 } _ { n } ^ { T } ) \right) } \\ { \displaystyle \quad = \rho _ { W _ { T } \times S _ { n } } ( R , P ) h \left( X - \frac { 1 } { n } X \boldsymbol { 1 } _ { n } \boldsymbol { 1 } _ { n } ^ { T } \right) } \\ { \displaystyle \quad = \rho _ { G } ( t , R , P ) f ( X ) . } \end{array} +$$ + +We now prove denseness of $G$ -equivariant polynomials in the space of $G$ -invariant continuous functions (Lemma 1). + +Lemma 1. Any continuous $G$ -equivariant function in $\mathcal { C } _ { G } ( \mathbb { R } ^ { 3 \times n } , W _ { T } ^ { n } )$ can be approximated uniformly on compact sets by $G$ -equivariant polynomials in $\mathcal { P } _ { G } ( \mathbb { R } ^ { 3 \times n } , \dot { W } _ { T } ^ { n } )$ . + +Proof of Lemma 1. Let $K ~ \subseteq ~ \mathbb { R } ^ { 3 \times n }$ be a compact set. We need to show that continuous $G$ - +equivariant functions can be approximated undenote the compact set which is the image of By Lemma B.1, it is sufficient to show that eve $K$ y ene $G$ -equivarianlizing map ivariant con $K _ { 0 }$ $K$ $\begin{array} { r } { \dot { X } \stackrel { \bullet } { \mapsto } X - \frac { 1 } { n } X 1 _ { n } 1 _ { n } ^ { T } } \end{array}$ $\mathrm { S O } ( 3 ) \times S _ { n }$ $f$ +be approximated uniformly on $K _ { 0 }$ by a sequence of $\mathrm { S O } ( 3 ) \times S _ { n }$ equivariant polynomials $p _ { k }$ . The +argument is now concluded by the following general lemma: + +Lemma B.2. Let $G$ be a compact group, Let $\rho _ { 1 }$ and $\rho _ { 2 }$ be continuous2representations of $G$ on the Euclidean spaces $W _ { 1 }$ and $W _ { 2 }$ . Let $K \subseteq W _ { 1 }$ be a compact set. Then every equivariant function $f : W _ { 1 } \to W _ { 2 }$ can be approximated uniformly on $K$ by a sequence of equivariant polynomials $p _ { k } : W _ { 1 } \mapsto W _ { 2 }$ . + +Let $\mu$ be the Haar probability measure associated with the compact group $G$ . Let $K _ { 1 }$ denote the compact set obtained as an image of the compact set $G \times K$ under the continuous mapping + +$$ +( g , X ) \mapsto \rho _ { 1 } ( g ) X . +$$ + +Using the Stone-Weierstrass theorem, let $p _ { k }$ be a sequence of (not necessarily equivariant) polynomials which approximate $f$ uniformly on $K _ { 1 }$ . Every degree $D$ polynomial $p : W _ { 1 } \to W _ { 2 }$ induces a $G$ -equivariant function + +$$ +\langle p \rangle ( X ) = \int _ { G } \rho _ { 2 } ( g ^ { - 1 } ) p ( \rho _ { 1 } ( g ) X ) d \mu ( g ) . +$$ + +This functformly on $\langle p \rangle$ is a degree “Riemann $D$ polynomial as ms” of the form $\langle p \rangle$ e approximated uwhich are degree $K _ { 1 }$ $\begin{array} { r } { \sum _ { j = 1 } ^ { N } w _ { j } \rho _ { 2 } ( g _ { j } ^ { - 1 } ) p ( \rho _ { 1 } ( g _ { j } ) X ) } \end{array}$ $D$ polynomials, and because degree polynomials are closed in $C ( K _ { 1 } )$ . + +Now for all $X \ \in \ K _ { 1 }$ , continuity of the function $g \mapsto \rho _ { 2 } ( g ^ { - 1 } )$ implies that the operator norm of $\rho _ { 2 } ( g ^ { - 1 } )$ is bounded uniformly by some constant $N > 0$ , and so + +$$ +\begin{array} { l } { \displaystyle { | \langle p _ { k } \rangle ( X ) - f ( X ) | = \left| \int _ { G } \rho _ { 2 } ( g ^ { - 1 } ) p _ { k } ( \rho _ { 1 } ( g ) X ) - \rho _ { 2 } ( g ^ { - 1 } ) f ( \rho _ { 1 } ( g ) X ) d \mu ( g ) \right| } } \\ { \displaystyle { \phantom { \rho _ { k } ( \rho _ { 1 } ) ( } = \left| \int _ { G } \rho _ { 2 } ( g ^ { - 1 } ) \left[ p _ { k } ( \rho _ { 1 } ( g ) X ) - f ( \rho _ { 1 } ( g ) X ) \right] d \mu ( g ) \right| \leq N \| f - p _ { k } \| _ { C ( K _ { 1 } ) } \to 0 } } \end{array} +$$ + +# B.2 PROOF OF THEOREM 1 + +Theorem 1. If $\mathcal { F } _ { \mathrm { f e a t } }$ is $D$ -spanning and $\mathcal { F } _ { \mathrm { p o o l } }$ is linearly universal, then there exists some $C ( D ) \in$ $\mathbb { N }$ such that for all $C \geq C ( D )$ the function space $\mathcal { F } _ { C } ( \mathcal { F } _ { \mathrm { f e a t } } , \mathcal { F } _ { \mathrm { p o o l } } )$ contains all $G$ -equivariant polynomials of degree $\le D$ . + +Proof. By the $D$ -spanning assumption, there exist $f _ { 1 } , \dots , f _ { K } \in \mathcal { F } _ { \mathrm { f e a t } }$ such that any vector valued polynomial $p : \mathbb { R } ^ { 3 \bar { \times } n } \bar { \mathbb { R } } ^ { n }$ invariant to translations and equivariant to permutations is of the form + +$$ +p ( X ) = \sum _ { k = 1 } ^ { K } \hat { \Lambda } _ { k } ( f _ { k } ( X ) ) , +$$ + +where $\Lambda _ { k }$ are linear functions to $\mathbb { R }$ . If $p$ is a matrix valued polynomial mapping $\mathbb { R } ^ { 3 \times n }$ to $W _ { T } ^ { n } = \mathbb { R } ^ { t \times n }$ , which is invariant to translations and equivariant to permutations, then it is of the form $p = ( p _ { i j } ) _ { i \in [ t ] , j \in [ n ] }$ , and each $p _ { i } = ( p _ { i j } ) _ { j \in [ n ] }$ is itself invariant to translations and permutation equivariant. It follows that matrix valued $p$ can also be written in the form equation 17, the only difference being that the image of the linear functions $\Lambda _ { k }$ is now $\mathbb { R } ^ { t }$ . + +Now let $p : \mathbb { R } ^ { 3 \times n } W _ { T } ^ { n }$ be a $G$ -equivariant polynomial of degree $\le D$ . It remains to show that we can choose $\Lambda _ { k }$ to be SO(3) equivariant. We do this by a symmetrization argument: denote the Haar probability measure on $\mathrm { S O ( 3 ) }$ by $\nu$ , and the action of SO(3) on $W _ { \mathrm { f e a t } }$ and $W _ { T }$ by $\rho _ { 1 }$ and $\rho _ { 2 }$ respectively Denote $p = ( p _ { j } ) _ { j = 1 } ^ { n }$ and $f _ { k } = ( f _ { k } ^ { j } ) _ { j = 1 } ^ { n }$ . For every $j = 1 , \dotsc , n$ , we use the SO(3) equivariance of $p _ { j }$ and $f _ { k } ^ { j }$ to obtain + +$$ +\begin{array} { l } { { \displaystyle p _ { j } ( X ) = \int _ { \mathrm { S O } ( 3 ) } \rho _ { 2 } ( R ^ { - 1 } ) \circ p _ { j } ( R X ) d \nu ( R ) = \sum _ { k = 1 } ^ { K } \int _ { \mathrm { S O } ( 3 ) } \rho _ { 2 } ( R ^ { - 1 } ) \circ \Lambda _ { k } \circ f _ { j } ^ { k } ( R X ) d \nu ( R ) } } \\ { { \displaystyle \qquad = \sum _ { k = 1 } ^ { K } \int _ { \mathrm { S O } ( 3 ) } \rho _ { 2 } ( R ^ { - 1 } ) \circ \Lambda _ { k } \left( \rho _ { 1 } ( R ) \circ f _ { k } ^ { j } ( X ) \right) d \nu ( R ) = \sum _ { k = 1 } ^ { K } \tilde { \Lambda } _ { k } \circ f _ { k } ^ { j } ( X ) , } } \end{array} +$$ + +2By this we mean that the maps $( g , X ) \mapsto \rho _ { j } ( g ) X , j = 1 , 2$ are jointly continuous + +where $\tilde { \Lambda } _ { k }$ stands for the equivariant linear functional from $W _ { \mathrm { f e a t } }$ to $W _ { T }$ , defined for $w \in W _ { \mathrm { f e a t } }$ by + +$$ +\tilde { \Lambda } _ { k } ( w ) = \int _ { \mathrm { S O ( 3 ) } } \rho _ { 2 } ( R ^ { - 1 } ) \circ \Lambda _ { k } \left( \rho _ { 1 } ( R ) w \right) d \nu ( R ) . +$$ + +Thus we have shown that $p$ is in $\mathcal { F } _ { C } ( \mathcal { F } _ { \mathrm { f e a t } } , \mathcal { F } _ { \mathrm { p o o l } } )$ for $C = K$ , as required. + +# C PROOFS FOR SECTION 4 + +We prove Lemma 2 + +Lemma 2. For every $D \in { \mathbb { N } } _ { + }$ , the set $\mathcal { Q } _ { D }$ is $D$ -spanning. + +Proof. It is known (Segol & Lipman, 2019) (Theorem 2) that polynomials $p : \mathbb { R } ^ { 3 \times n } \mathbb { R } ^ { n }$ which are $S _ { n }$ -equivariant, are spanned by polynomials of the form $p _ { \vec { \alpha } } = \stackrel { \cdot } { ( } p _ { \vec { \alpha } } ^ { j } ) _ { j = 1 } ^ { n }$ , defined as + +$$ +p _ { \vec { \alpha } } ^ { j } ( X ) = \sum _ { i _ { 2 } , \ldots , i _ { K } = 1 } ^ { n } x _ { j } ^ { \alpha _ { 1 } } x _ { i _ { 2 } } ^ { \alpha _ { 2 } } \cdot \cdot \cdot x _ { i _ { k } } ^ { \alpha _ { k } } +$$ + +where $\vec { \alpha } \ : = \ : ( \alpha _ { 1 } , \ldots , \alpha _ { K } )$ and each $\alpha _ { k } ~ \in ~ \mathbb { N } _ { + } ^ { 3 }$ is a multi-index. It follows that $S _ { n }$ -equivariant polynomials of degree $\leq D$ are spanned by polynomials of the form $p _ { \vec { \alpha } } ^ { j }$ where $\begin{array} { r } { \sum _ { k = 1 } ^ { K } \left| \alpha _ { k } \right| \leq D } \end{array}$ . Denoting $r _ { k } = | \alpha _ { k } | , k = 1 , \ldots K$ , the sum of all $r _ { k }$ by $T$ , and $\vec { r } = ( r _ { k } ) _ { k = 1 } ^ { K }$ , we see that there exists a linear functional $\Lambda _ { \vec { \alpha } , \vec { r } } : \mathcal { T } _ { T } \to \mathbb { R }$ such that + +$$ +p _ { \vec { \alpha } } ^ { j } ( X ) = \Lambda _ { \vec { \alpha } , \vec { r } ^ { } ^ { \mathrm { ~ O ~ } } } Q _ { j } ^ { \vec { r } } ( X ) +$$ + +where we recall that $Q ^ { \vec { r } } = \left( Q _ { j } ^ { ( \vec { r } ) } ( X ) \right) _ { j = 1 } ^ { n }$ is defined in equation 6 as + +$$ +Q _ { j } ^ { ( \vec { r } ) } ( X ) = \sum _ { i _ { 2 } , \ldots , i _ { K } = 1 } ^ { n } x _ { j } ^ { \otimes r _ { 1 } } \otimes x _ { i _ { 2 } } ^ { \otimes r _ { 2 } } \otimes x _ { i _ { 3 } } ^ { \otimes r _ { 3 } } \otimes \ldots \otimes x _ { i _ { K } } ^ { \otimes r _ { K } } . +$$ + +Thus polynomials $p = ( p _ { j } ) _ { j = 1 } ^ { n }$ which are of degree $\leq D$ , and are $S _ { n }$ equivariant, can be written as + +$$ +p _ { j } ( X ) = \sum _ { T \leq D } \sum _ { \vec { r } \in \Sigma _ { T } } \sum _ { \vec { \alpha } | | \alpha _ { k } | = r _ { k } } \Lambda _ { \vec { \alpha } , \vec { r } } \left( Q _ { j } ^ { ( \vec { r } ) } ( X ) \right) = \sum _ { T \leq D } \sum _ { \vec { r } \in \Sigma _ { T } } \Lambda _ { \vec { r } } \left( \iota \circ Q _ { j } ^ { ( \vec { r } ) } ( X ) \right) , j = 1 , \dots , n , +$$ + +where $\begin{array} { r } { \Lambda _ { \vec { r } } = \sum _ { \vec { \alpha } | | \alpha _ { k } | = r _ { k } } \Lambda _ { \vec { \alpha } , \vec { r } } \circ \iota _ { T } ^ { - 1 } } \end{array}$ , and $\iota _ { T } ^ { - 1 }$ is the left inverse of the embedding $\iota$ . If $p$ is also translation invariant, then + +$$ +p ( X ) = p ( X - \frac { 1 } { n } X 1 _ { n } \mathbf { 1 } _ { n } ^ { T } ) = \sum _ { T \leq D } \sum _ { \vec { r } \in \Sigma _ { T } } \hat { \Lambda } _ { \vec { r } } \left( \iota \circ Q ^ { ( \vec { r } ) } ( X - \frac { 1 } { n } X 1 _ { n } \mathbf { 1 } _ { n } ^ { T } ) \right) . +$$ + +Thus $\mathcal { Q } _ { D }$ is $D$ -spanning. + +We prove Lemma 3 + +Lemma 3. The function set $\mathcal { Q } _ { D }$ is contained in + +$$ +\mathcal { F } _ { \mathrm { f e a t } } = \big \{ \iota \circ \pi _ { V } \circ f ^ { 1 } \circ f ^ { 2 } \circ \ldots \circ f ^ { T } \circ \mathrm { e x t } ( X - \frac { 1 } { n } X \boldsymbol { 1 } _ { n } \boldsymbol { 1 } _ { n } ^ { T } ) | f ^ { j } \in \mathcal { F } _ { m i n } , T \leq D \big \} . +$$ + +Thus $\mathcal { F } _ { \mathrm { f e a t } }$ is $D$ -spanning. + +Proof. In this proof we make the dependence of $\mathcal { F } _ { \mathrm { f e a t } }$ on $D$ explicit and denote $\mathcal { F } _ { \mathrm { f e a t } } ( D )$ . + +We prove the claim by induction on $D$ . Assume $D = 0$ . Then $\mathcal { Q } _ { 0 }$ contains only the constant function $X \mapsto 1 _ { n } \in \mathcal { T } _ { 0 } ^ { n }$ , and this is precisely the function $\pi _ { V } \circ \mathrm { e x t } \in \mathcal { F } _ { \mathrm { f e a t } } ( 0 )$ . + +Now assume the claim holds for all $D ^ { \prime }$ with $D - 1 \geq D ^ { \prime } \geq 0$ , and prove the claim for $D$ . Choose $\vec { r } = ( r _ { 1 } , \ldots , r _ { k } ) \in \Sigma _ { T }$ for some $T \leq D$ , we need to show that the function $Q ^ { ( \vec { r } ) }$ is in $\mathcal { F } _ { \mathrm { f e a t } } ( D )$ . Since ${ \mathcal { F } } _ { \mathrm { f e a t } } ( D - 1 ) \subseteq { \mathcal { F } } _ { \mathrm { f e a t } } ( D )$ we know from the induction hypothesis that this is true if $T < D$ . Now assume $T = D$ . We consider two cases: + +1. If $r _ { 1 } > 0$ , we set $\tilde { r } = ( r _ { 1 } - 1 , r _ { 2 } , \dots , r _ { K } )$ . We know that $\iota \circ Q ^ { ( \tilde { r } ) } ( \bar { X } ) \in \mathcal { F } _ { \mathrm { f e a t } } ( D - 1 )$ by the induction hypothesis. So there exist $f _ { 2 } , \ldots , f _ { D }$ such that + +$$ +\iota \circ \pi _ { V } \circ f _ { 2 } \circ . . . \circ f _ { D } \circ \mathrm { e x t } ( \bar { X } ) = \iota \circ Q ^ { ( \tilde { r } ) } ( \bar { X } ) . +$$ + +Now choose $f _ { 1 } \in \mathcal { F } _ { m i n }$ to be the function whose $V$ coordinate $\tilde { V } = ( \tilde { V } _ { j } ) _ { j = 1 } ^ { n }$ , is given by $\tilde { V } _ { j } ( X , V ) = x _ { j } \otimes V _ { j }$ , obtained by setting $\theta _ { 1 } = 1 , \theta _ { 2 } = 0$ in equation 9. Then , we have + +$$ +\begin{array} { r l } & { \tilde { V } _ { j } ( \bar { X } , Q ^ { ( \tilde { r } ) } ( \bar { X } ) ) = \displaystyle \sum _ { i _ { 2 } , \ldots , i _ { K } = 1 } ^ { n } \bar { x } _ { j } \otimes \bar { x } _ { j } ^ { \otimes ( r _ { 1 } - 1 ) } \otimes \bar { x } _ { i _ { 2 } } ^ { \otimes r _ { 2 } } \otimes \ldots \otimes \bar { x } _ { i _ { K } } ^ { \otimes r _ { K } } } \\ & { \qquad = Q _ { j } ^ { ( \tilde { r } ) } ( \bar { X } ) . } \end{array} +$$ + +and so + +$$ +\begin{array} { c } { \displaystyle { \imath \circ \pi _ { V } \circ f _ { 1 } \circ f _ { 2 } \circ . . . \circ f _ { D } \circ \mathrm { e x t } ( X - \frac 1 n X 1 _ { n } \mathbf { 1 } _ { n } ^ { T } ) = \imath \circ Q ^ { ( \vec { r } ) } ( \bar { X } ) . } } \\ { \displaystyle { \mathrm { } } } \\ { \displaystyle { \mathrm { } \mathrm { } } \iota \circ Q ^ { ( \vec { r } ) } ( X - \frac 1 n X 1 _ { n } \mathbf { 1 } _ { n } ^ { T } ) \in \mathcal { F } _ { \mathrm { f e a t } } ( D ) . } \end{array} +$$ + +2. If $r _ { 1 } = 0$ . We assume without loss of generality that $r _ { 2 } > 0$ . Set $\tilde { r } = ( r _ { 2 } - 1 , r _ { 3 } , \dots , r _ { K } )$ . As before by the induction hypothesis there exist $f _ { 2 } , \ldots , f _ { D }$ which satisfy equation 19. This time we choose $f _ { 1 } \in \mathcal { F } _ { m i n }$ to be the function whose $V$ coordinate $\tilde { V } = \tilde { ( } \tilde { V } _ { j } ) _ { j = 1 } ^ { n }$ , is given by $\begin{array} { r } { \tilde { V } _ { j } ( X , V ) = \sum _ { j } x _ { j } \otimes V _ { j } } \end{array}$ , obtained by setting $\theta _ { 1 } = 0 , \theta _ { 2 } = 1$ in equation 9. Then we have + +$$ +\begin{array} { r l } & { \tilde { V } _ { j } ( \bar { X } , Q ^ { ( \tilde { r } ) } ( \bar { X } ) ) = \displaystyle \sum _ { j = 1 } ^ { n } \displaystyle \sum _ { i _ { 3 } , \ldots , i _ { K } = 1 } ^ { n } \bar { x } _ { j } \otimes \bar { x } _ { j } ^ { \otimes ( r _ { 2 } - 1 ) } \otimes \bar { x } _ { i _ { 3 } } ^ { \otimes r _ { 2 } } \otimes \ldots \otimes \bar { x } _ { i _ { K } } ^ { \otimes r _ { K } } } \\ & { \qquad = \displaystyle \sum _ { i _ { 2 } , i _ { 3 } , \ldots , i _ { K } = 1 } ^ { n } \bar { x } _ { i _ { 2 } } ^ { \otimes r _ { 2 } } \otimes \bar { x } _ { i _ { 3 } } ^ { \otimes r _ { 2 } } \otimes \ldots \otimes \bar { x } _ { i _ { K } } ^ { \otimes r _ { K } } } \\ & { \qquad = Q _ { j } ^ { ( \tilde { r } ) } ( \bar { X } ) . } \end{array} +$$ + +Thus equation 20 holds, and so again we have that $\begin{array} { r } { \iota \circ Q ^ { ( \vec { r } ) } ( X - \frac { 1 } { n } X 1 _ { n } \mathbf { 1 } _ { n } ^ { T } ) \in \mathcal { F } _ { \mathrm { f e a t } } ( D ) . } \end{array}$ + +Finally we prove Lemma 4 + +Lemma 4. If all functions in $\mathcal { Q } _ { D }$ can be written as + +$$ +\iota \circ Q ^ { ( \vec { r } ) } ( X - \frac { 1 } { n } X 1 _ { n } 1 _ { n } ^ { T } ) = \sum _ { k = 1 } ^ { K } \hat { A } _ { k } f _ { k } ( X ) , +$$ + +where $f _ { k } \in \mathcal { F } _ { \mathrm { f e a t } }$ , $A _ { k } : W _ { \mathrm { f e a t } } \to W _ { \mathrm { f e a t } } ^ { \mathcal { T } }$ and ${ \hat { A } } _ { k } : W _ { \mathrm { f e a t } } ^ { n } \to ( W _ { \mathrm { f e a t } } ^ { \mathcal { T } } ) ^ { n }$ is defined by elementwise application of $A _ { k }$ , then $\mathcal { F } _ { \mathrm { f e a t } }$ is $D$ -spanning. + +Proof. If the conditions in Lemma 4 hold, then since $\mathcal { Q } _ { D }$ is $D$ -spanning, every translation invariant and permutation equivariant polynomials $p$ of degree $D$ can be written as + +$$ +\begin{array} { l } { \displaystyle p ( X ) = \sum _ { \vec { r } ^ { \prime } | | \vec { r } | | _ { 1 } \le D } \hat { \Lambda } _ { \vec { r } } \bigg ( \iota \circ Q ^ { ( \vec { r } ) } ( X - \frac 1 n X \mathbf { 1 } _ { n } \mathbf { 1 } _ { n } ^ { T } ) \bigg ) = \sum _ { \vec { r } ^ { \prime } | | \vec { r } ^ { \prime } | | _ { 1 } \le D } \hat { \Lambda } _ { \vec { r } } \bigg ( \sum _ { k = 1 } ^ { K _ { \vec { r } } } \iota \circ \hat { A } _ { k , \vec { r } } f _ { k , \vec { r } } ( X ) \bigg ) } \\ { = \sum _ { \vec { r } ^ { \prime } | | \vec { r } | | _ { 1 } \le D } \sum _ { k = 1 } ^ { K _ { \vec { r } } } \hat { \Lambda } _ { k , \vec { r } } ( f _ { k , \vec { r } } ( X ) ) } \end{array} +$$ + +where we denote $\Lambda _ { k , \vec { r } } = \Lambda _ { \vec { r } } \circ \iota \circ A _ { k , \vec { r } } .$ Thus we proved $\mathcal { F } _ { \mathrm { f e a t } }$ is $D$ -spanning. + +# D PROOFS FOR SECTION 5 + +We prove Lemma 5 + +Lemma 5. Let $l ^ { ( 1 ) } = ( \ell _ { 1 } ^ { ( 1 ) } , \dots , \ell _ { K _ { 1 } } ^ { ( 1 ) } )$ $l ^ { ( 2 ) } = ( \ell _ { 1 } ^ { ( 2 ) } , \dots , \ell _ { K _ { 2 } } ^ { ( 2 ) } )$ . A function $\boldsymbol { \Lambda } = \left( \Lambda _ { 1 } , \ldots , \Lambda _ { K _ { 2 } } \right)$ a linear equivariant mapping between $W _ { l ^ { ( 1 ) } }$ and $W _ { l ^ { ( 2 ) } }$ , if and only if there exists a $K _ { 1 } \times K _ { 2 }$ matrix $M$ with $M _ { i j } = 0$ whenever $\ell _ { i } ^ { ( 1 ) } \neq \ell _ { j } ^ { ( 2 ) }$ , such that + +$$ +\Lambda _ { j } ( V ) = \sum _ { i = 1 } ^ { K _ { 1 } } M _ { i j } V _ { i } +$$ + +where $V = ( V _ { i } ) _ { i = 1 } ^ { K _ { 1 } }$ and $V _ { i } \in W _ { \ell _ { i } ^ { ( 1 ) } }$ for all $i = 1 , \ldots , K _ { 1 }$ + +Proof. As mentioned in the main text, this lemma is based on Schur’s lemma. This lemma is typically stated for complex representations, but holds for odd dimensional real representation as well. We recount the lemma and its proof here for completeness (see also (Fulton & Harris, 2013)). + +Lemma D.1 (Schur’s Lemma for SO(3)). Let $\Lambda : W _ { \ell _ { 1 } } \to W _ { \ell _ { 2 } }$ be a linear equivariant map. If $\ell _ { 1 } \neq \ell _ { 2 }$ then $\Lambda = 0$ . Otherwise $\Lambda$ is a scalar multiply of the identity. + +Proof. Let $\Lambda : W _ { \ell _ { 1 } } \to W _ { \ell _ { 2 } }$ be a linear equivariant map. The image and kernel of $\Lambda$ are invariant subspaces of $W _ { \ell _ { 1 } }$ and $W _ { \ell _ { 2 } }$ , respectively. It follows that if $\Lambda \neq 0$ then $\Lambda$ is a linear isomorphism so necessarily $\ell _ { 1 } = \ell _ { 2 }$ . Now assume $\ell _ { 1 } = \ell _ { 2 }$ . Since the dimension of $W _ { \ell _ { 1 } }$ is odd, $\Lambda$ has a real eigenvalue $\lambda$ . The linear function $\Lambda - \lambda I$ is equivariant and has a non-trivial kernel, so $\Lambda - \lambda I =$ 0. □ + +We now return to the proof of Lemma 5. Note that each $\Lambda _ { j } : W _ { l ^ { ( 1 ) } } \to W _ { \ell _ { \vec { \cdot } } ^ { ( 2 ) } }$ is linear and SO(3) equivariant. Next denote the restrictions of each $\Lambda _ { j }$ to $W _ { \ell _ { i } ^ { ( 1 ) } } , i = 1 , \dots , K _ { 2 }$ by $\Lambda _ { i j }$ , and note that + +$$ +\Lambda _ { j } ( V _ { 1 } , \dots , V _ { K _ { 1 } } ) = \sum _ { i = 1 } ^ { K _ { 1 } } \Lambda _ { i j } ( V _ { i } ) . +$$ + +By considering vectors in $W _ { l ^ { ( 1 ) } }$ of the form $( 0 , \ldots , 0 , V _ { i } , 0 \ldots , 0 )$ we see that each $\Lambda _ { i j } : W _ { \ell _ { i } ^ { ( 1 ) } } \to$ $W _ { \ell _ { j } ^ { ( 2 ) } }$ is linear and SO(3)-equivariant. Thus by Schur’s lemma, if $\ell _ { i } ^ { ( 1 ) } = \ell _ { j } ^ { ( 2 ) }$ then $\Lambda _ { i j } ( V _ { i } ) = M _ { i j } V _ { i }$ for some real $M _ { i j }$ , and otherwise $M _ { i j } = 0$ . Plugging this into equation 21 we obtain equation 12. + +We prove Theorem 2 which shows that the TFN network described in the main text is universal: + +Theorem 2. For all $n \in \mathbb { N } , l _ { T } \in \mathbb { N } _ { + } ^ { * }$ , + +1. For $D \in { \mathbb { N } } _ { + }$ , every $G$ -equivariant polynomial $p : \mathbb { R } ^ { 3 \times n } W _ { T } ^ { n }$ of degree $D$ is in $\mathcal { F } _ { C ( D ) , D } ^ { \mathrm { T F N } }$ 2. Every continuous $G$ -equivariant function can be approximated uniformly on compact sets by functions in ∪D∈N+ FTFNC(D),D + +Proof. As mentioned in the main text, we only need to show that the function space $\mathcal { F } _ { \mathrm { f e a t } } ( D )$ is $D$ -spanning. Recall that $\mathcal { F } _ { \mathrm { f e a t } } ( D )$ is obtained by $2 D$ consecutive convolutions with $D$ -filters. In general, we denote the space of functions defined by applying $J$ consecutive convolutions by $\mathcal { G } _ { J , D }$ . + +If $\mathcal { V }$ is a space of functions from $\mathbb { R } ^ { 3 \times n } Y ^ { n }$ , we denote by $\langle \mathcal { V } , \mathcal { T } _ { T } \rangle$ the space of all functions $p : \mathbb { R } ^ { 3 \times n } \overset { \cdot } { } \mathcal { T } _ { T } ^ { n }$ of the form + +$$ +p ( X ) = \sum _ { k = 1 } ^ { K } \hat { A } _ { k } f _ { k } ( X ) , +$$ + +where $A _ { k } : Y \mathcal { T } _ { T }$ are linear functions, $\hat { A } _ { k } : Y ^ { n } \mathcal { T } _ { T } ^ { n }$ are induced by elementwise application, and $f _ { k } \in \mathcal { V }$ . This notation is useful because: (i) by Lemma 4 it is sufficient to show that $Q ^ { ( \vec { r } ) } ( \bar { X } )$ is in $\langle \mathcal { G } _ { 2 D , D } , \mathcal { T } _ { T } \rangle$ for all $\vec { r } \in \Sigma _ { T }$ and all $T \leq D$ , and because (ii) it enables comparison of the expressive power of function spaces $\mathcal { \scriptsize { N } } _ { 1 } , \mathcal { \scriptsize { N } } _ { 2 }$ whose elements map to different spaces $Y _ { 1 } ^ { n } , Y _ { 2 } ^ { n }$ , since the elements in $\langle y _ { i } , \mathcal { T } _ { T } \rangle , i = \bar { 1 } , 2$ both map to the same space. In particular, note that if for every $f \in \mathcal { V } _ { 2 }$ there is a $g \in \mathcal { V } 1$ and a linear map $A : Y _ { 1 } Y _ { 2 }$ such that $f ( X ) \ = \ \hat { A } \circ g ( X )$ , then $\langle { \mathcal { V } } _ { 2 } , { \mathcal { T } } _ { T } \rangle \subseteq \langle { \mathcal { V } } _ { 1 } , { \mathcal { T } } _ { T } \rangle$ . + +We now use this abstract discussion to prove some useful results: the first is that for the purpose of this lemma, we can ‘forget about’ the multiplication by a unitary matrix in equation 14, used for decomposition into irreducible representations: To see this, denote by $\tilde { \mathcal { G } } _ { J , D }$ the function space obtained by taking $J$ consecutive convolutions with $D$ -filters without multiplying by a unitary matrix in equation 14. Since Kronecker products of unitary matrices are unitary matrices, we obtain that the elements in $\mathcal { G } _ { J , D }$ and $\tilde { \mathcal { G } } _ { J , D }$ differ only by multiplication by a unitary matrix, and thus $\langle \tilde { \mathcal { G } } _ { J , D } , \mathcal { T } _ { T } \rangle \subseteq$ $\langle \mathcal { G } _ { J , D } , \mathcal { T } _ { T } \rangle$ and $\langle \mathcal { G } _ { J , D } , \mathcal { T } _ { T } \rangle \subseteq \langle \tilde { \mathcal { G } } _ { J , D } , \mathcal { T } _ { T } \rangle$ , so both sets are equal. + +Next, we prove that adding convolutional layers (enlarging $J$ ) or taking higher order filters (enlarging $D$ ) can only increase the expressive power of a network. + +Lemma D.2. For all $J , D , T \in \mathbb { N } _ { + }$ , $l . \langle \mathcal { G } _ { J , D } , \mathcal { T } _ { T } \rangle \subseteq \langle \mathcal { G } _ { J + 1 , D } , \mathcal { T } _ { T } \rangle .$ 2 $\langle \mathcal { G } _ { J , D } , \mathcal { T } _ { T } \rangle \subseteq \langle \mathcal { G } _ { J , D + 1 } , \mathcal { T } _ { T } \rangle .$ . + +Proof. The first claim follows from the fact that every function $f$ in $\langle \mathcal { G } _ { J , D } , \mathcal { T } _ { T } \rangle$ can be identified with a function in $\langle \mathcal { G } _ { J + 1 , D } , \mathcal { T } _ { T } \rangle$ by taking the $J + 1$ convolutional layer in equation 14 with $\theta _ { 0 } =$ $1 , F = 0$ . + +The second claim follows from the fact that $D$ -filters can be identified with $D + 1$ -filters whose $D + 1$ -th entry is $0$ . □ + +The last preliminary lemma we will need is + +Lemma D.3. For every $J , D \in \mathbb { N } _ { + }$ , and every $t , s \in \mathbb { N } _ { + }$ , if $\dot { \boldsymbol { p } } \in \langle \mathcal { G } _ { J , D } , \mathcal { T } _ { t } \rangle$ , then the function $q$ defined by + +$$ +q _ { a } ( X ) = \sum _ { b = 1 } ^ { n } ( { \bar { x } } _ { a } - { \bar { x } } _ { b } ) ^ { \otimes s } \otimes p _ { b } ( X ) +$$ + +is in $\langle \mathcal { G } _ { J + 1 , D } , \mathcal { T } _ { t + s } \rangle$ + +Proof. This lemma is based on the fact that the space of $s$ homogeneous polynomial on $\mathbb { R } ^ { 3 }$ is spanned by polynomials of the form $\| x \| ^ { s - \ell } Y _ { m } ^ { \ell } ( \bar { x } )$ for $\ell = s , s - 2 , s - 4 \ldots$ (Dai & $\mathrm { X u }$ , 2013). For each such $\ell$ , and $s \leq D$ , these polynomials can be realized by filters $F ^ { ( \ell ) }$ by setting $R ^ { ( \ell ) } ( \| x \| ) = \| x \| ^ { s }$ so that + +$$ +F _ { m } ^ { ( \ell ) } ( x ) = \| x \| ^ { s } Y _ { m } ^ { \ell } ( \hat { x } ) = \| x \| ^ { s - \ell } Y _ { m } ^ { \ell } ( x ) . +$$ + +For every $D \ \in \ \mathbb { N }$ and $s \leq D$ , we can construct a $D$ -filter $F ^ { s , D } = ( F ^ { ( 0 ) } , \ldots , F ^ { ( D ) } )$ where $F ^ { ( s ) } , F ^ { ( s - 2 ) } , \ldots .$ are as defined above and the other filters are zero. Since both the entries of $F ^ { s , D } ( x )$ , and the entries of $x ^ { \otimes s }$ , span the space of $s$ -homogeneous polynomials on $\mathbb { R } ^ { 3 }$ , it follows that there exists a linear mapping $B _ { s } : W _ { l _ { D } } T _ { s }$ so that + +$$ +x ^ { \otimes s } = B _ { s } ( F ^ { s , D } ( { \boldsymbol { x } } ) ) , \forall { \boldsymbol { x } } \in \mathbb { R } ^ { 3 } . +$$ + +Thus, since $p$ can be written as a sum of compositions of linear mappings with functions in $\mathcal { G } _ { J , D }$ as in equation 22, and similarly $x ^ { \otimes s }$ is obtained as a linear image of functions in $\mathcal { G } _ { 1 , D }$ as in equation 23, we deduce that + +$$ +\sum _ { b = 1 } ^ { n } ( x _ { a } - x _ { b } ) \otimes p _ { b } ( X ) = \sum _ { b = 1 } ^ { n } ( \bar { x } _ { a } - \bar { x } _ { b } ) \otimes p _ { b } ( X ) +$$ + +is in $\langle \mathcal { G } _ { J + 1 , D } , \mathcal { T } _ { t + s } \rangle$ + +As a final preliminary, we note that $D$ -filters can perform an averaging operation by setting $R ^ { ( 0 ) } = 1$ and $\theta _ { 0 } , R ^ { ( \mathrm { 1 } ) } , \ldots , R ^ { ( \dot { D } ) } = 0$ in equation 13 and equation 14 . We call this $D$ -filter an averaging filter. + +We are now ready to prove our claim: we need to show that for every $D , T \in \mathbb { N } _ { + }$ where $T \leq D$ , for every $\vec { r } \in \Sigma _ { T }$ , the function $Q ^ { ( \vec { r } ) }$ is in $\langle \mathcal { G } _ { 2 D , D } , \mathcal { T } _ { T } \rangle$ . Note that due to the inclusion relations in Lemma D.2 it is sufficient to prove this for the case $T = D$ . We prove this by induction on $D$ . For $D = 0$ , vectors $\vec { r } \in \Sigma _ { 0 }$ contains only zeros and so + +$$ +Q ^ { ( \vec { r } ) } ( \bar { X } ) = 1 _ { n } = \pi _ { V } \circ \mathrm { e x t } ( X ) \in \langle \mathcal { G } _ { 0 , 0 } , \mathcal { T } _ { 0 } \rangle . +$$ + +We now assume the claim is true for all $D ^ { \prime }$ with $D > D ^ { \prime } \geq 0$ and prove the claim is true for $D$ . We need to show that for every $\vec { r } \in \Sigma _ { D }$ the function $Q ^ { ( \vec { r } ) }$ is in $\langle \mathcal { G } _ { 2 D , D } , \mathcal { T } _ { D } \rangle$ . We prove this yet again by induction, this time on the value of $r _ { 1 }$ : assume that $\vec { r } \in \Sigma _ { D }$ and $r _ { 1 } = 0 .$ . Denote by $\tilde { r }$ the vector in $\Sigma _ { D - 1 }$ defined by + +$$ +\tilde { r } = ( r _ { 2 } - 1 , r _ { 3 } , \ldots , r _ { K } ) . +$$ + +By the induction assumption on $D$ , we know that $Q ^ { ( \tilde { r } ) } ( \bar { X } ) \in \mathcal { G } _ { 2 ( D - 1 ) , D - 1 , D - 1 }$ and so + +$$ +\begin{array} { l } { \displaystyle q _ { a } ( X ) = \sum _ { b = 1 } ^ { n } ( \bar { x } _ { a } - \bar { x } _ { b } ) \otimes Q _ { b } ^ { ( \bar { r } ) } ( \bar { X } ) = \sum _ { b = 1 } ^ { n } ( \bar { x } _ { a } - \bar { x } _ { b } ) \otimes \bar { x } _ { b } ^ { \otimes r _ { 2 } - 1 } \otimes \sum _ { i _ { 3 } , \dots , i _ { K } = 1 } ^ { n } \bar { x } _ { i _ { 3 } } ^ { \otimes r _ { 3 } } \otimes \dots \otimes \bar { x } _ { i _ { K } } ^ { \otimes r _ { K } } } \\ { \displaystyle \quad = \left( \bar { x } _ { a } \otimes \sum _ { b = 1 } ^ { n } Q _ { b } ^ { ( \bar { r } ) } ( \bar { X } ) \right) - Q ^ { ( \bar { r } ) } ( \bar { X } ) } \end{array} +$$ + +is in $\langle \mathcal { G } _ { 2 D - 1 , D - 1 } , \mathcal { T } _ { D } \rangle$ by Lemma D.3, which is contained in $\langle \mathcal { G } _ { 2 D - 1 , D } , \mathcal { T } _ { D } \rangle$ by Lemma D.2. Since $\textstyle { \bar { x } } _ { a }$ has zero mean, while $Q _ { a } ^ { ( \vec { r } ) } ( \bar { X } )$ does not depend on $a$ since $r _ { 1 } ~ = ~ 0$ , applying an averaging filter to $q _ { a }$ gives us a constant value $- Q _ { a } ^ { ( \vec { r } ) } ( \bar { X } )$ in each coordinate $a \in [ n ]$ , and so $Q ^ { ( \vec { r } ) } ( \bar { X } )$ is in $\langle \mathcal { G } _ { 2 D , D } , \mathcal { T } _ { D } \rangle$ . + +Now assume the claim is true for all $\vec { r } \in \Sigma _ { D }$ which sum to $D$ , and whose first coordinate is smaller than some $r _ { 1 } ^ { \prime } \geq 1$ , we now prove the claim is true when the first coordinate of $\vec { r }$ is equal to $r _ { 1 } ^ { \prime }$ . The vector $\tilde { \boldsymbol { r } } = \left( r _ { 2 } , \ldots , r _ { K } \right)$ obtained from $\vec { r }$ by removing the first coordinate, sums to $D ^ { \prime } = D - r _ { 1 } ^ { \prime } <$ $D$ , and so by the induction hypothesis on $D$ we know that $Q ^ { ( \tilde { r } ) } \in \langle { \mathcal { G } _ { 2 D ^ { \prime } , D ^ { \prime } } } , { \mathcal { T } _ { D ^ { \prime } } } \rangle$ . By Lemma D.3 we obtain a function $q _ { a } \in \langle \mathcal { G } _ { 2 D ^ { \prime } + 1 , D ^ { \prime } } , \mathcal { T } _ { D } \rangle \subseteq \langle \mathcal { G } _ { 2 D , D } , \mathcal { T } _ { D } \rangle$ defined by + +$$ +\begin{array} { r l } & { q _ { a } ( X ) = \displaystyle \sum _ { b = 1 } ^ { n } ( \bar { x } _ { a } - \bar { x } _ { b } ) ^ { \otimes r _ { 1 } } \otimes Q _ { b } ^ { ( \bar { r } ) } ( \bar { X } ) } \\ & { \qquad = \displaystyle \sum _ { b = 1 } ^ { n } ( \bar { x } _ { a } - \bar { x } _ { b } ) ^ { \otimes r _ { 1 } } \otimes \bar { x } _ { b } ^ { \otimes r _ { 2 } } \otimes \displaystyle \sum _ { i _ { 3 } , \ldots , i _ { K } = 1 } ^ { n } \bar { x } _ { i _ { 3 } } ^ { \otimes r _ { 3 } } \otimes \ldots \otimes \bar { x } _ { i _ { K } } ^ { \otimes r _ { K } } } \\ & { \qquad = Q _ { a } ^ { ( \bar { r } ) } ( \bar { X } ) + \mathrm { a d d i t i o n a l ~ t e r m s } } \end{array} +$$ + +where the additional terms are linear combinations of functions of the form $P _ { D } Q _ { a } ^ { ( r ^ { \prime } ) } ( \bar { X } )$ where $r ^ { \prime } \in \Sigma _ { D }$ and their first coordinate $r _ { 1 }$ is smaller than $r _ { 1 } ^ { \prime }$ , and $P _ { D } : \mathcal { T } _ { D } \mathcal { T } _ { D }$ is a permutation. By the induction hypothesis on $r _ { 1 }$ , each such $Q ^ { ( r ^ { \prime } ) }$ is in $\langle \mathcal { G } _ { 2 D , D } , \mathcal { T } _ { D } \rangle$ . It follows that $P _ { D } Q _ { a } ^ { ( r ^ { \prime } ) } ( \bar { X } ) , a =$ $1 , \ldots , n$ , and thus $Q ^ { ( \vec { r } ) } ( \bar { X } )$ , are in $\langle \mathcal { G } _ { 2 D , D } , \mathcal { T } _ { D } \rangle$ as well. This concludes the proof of Theorem 2. + +# E ALTERNATIVE TFN ARCHITECTURE + +In this appendix we show that replacing the standard TFN convolutional layer with the layer defined in equation 15: + +$$ +\tilde { V } _ { a } ( X , V ) = U ( l _ { f } , l _ { i } ) \left( \theta _ { 1 } \sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \otimes V _ { b } + \theta _ { 2 } \sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \otimes V _ { a } \right) , +$$ + +we can obtain $D$ -spanning networks using $2 D$ consecutive convolutions with 1-filters (that is, filters in $W _ { l _ { 1 } }$ , where $l _ { 1 } \stackrel { \cdot } { = } [ 0 , 1 ] ^ { \frac { \cdot } { T } }$ ). Our discussion here is somewhat informal, meant to provide the general + +ideas without delving into the details as we have done for the standard TFN architecture in the proof of Theorem 2. In the end of our discussion we will explain what is necessary to make this argument completely rigorous. + +We will only need two fixed filters for our argument here: The first is the 1-filter $F _ { I d } = ( F ^ { ( 0 ) } , F ^ { ( 1 ) } )$ defined by setting $R ^ { ( 0 ) } ( \| x \| ) = 0$ and $R ^ { ( 1 ) } ( \| x \| ) = \| x \|$ to obtain + +$$ +F _ { I d } ( x ) = \| x \| Y ^ { 1 } ( \hat { x } ) = \| x \| \hat { x } = x . +$$ + +The second is the filter $F _ { 1 }$ defined by setting $R ^ { ( 0 ) } ( \| x \| ) = 1$ and $R ^ { ( 1 ) } ( \| x \| ) = 0$ , so that + +$$ +F _ { 1 } ( x ) = 1 . +$$ + +We prove our claim by showing that a pair of convolutions with 1-filters can construct any convolutional layer defined in equation 9 for the $D$ -spanning architecture using tensor representations. The claim then follows from the fact that $D$ convolutions of the latter architecture suffice for achieving $D$ -spanning, as shown in Lemma 3. + +Convolutions for tensor representations, defined in equation 9, are composed of two terms: + +$$ +\tilde { V } _ { a } ^ { \mathrm { t e n s o r } , 1 } ( \bar { X } , V ) = \bar { x } _ { a } \otimes V _ { a } \mathrm { a n d } \tilde { V } _ { a } ^ { \mathrm { t e n s o r } , 2 } ( \bar { X } , V ) = \sum _ { b = 1 } ^ { n } \bar { x } _ { b } \otimes V _ { b } . +$$ + +To obtain the first term $\tilde { V } _ { a } ^ { \mathrm { t e n s o r , 1 } }$ , we set $\theta _ { 1 } = 0 , \theta _ { 2 } = 1 / n , F = F _ { I d }$ in equation 15 we obtain (the decomposition into irreducibles of) $\tilde { V } _ { a } ^ { \mathrm { t e n s o r , 1 } } ( \bar { X } , V ) = \bar { x } _ { a } \otimes V _ { a }$ . Thus this term can in fact be expressed by a single convolution. We can leave this outcome unchanged by a second convolution, defined by setting $\bar { \theta } _ { 1 } = 0 , \theta _ { 2 } = 1 / n , F = F _ { 1 }$ . + +To obtain the second term $\tilde { V } _ { a } ^ { \mathrm { t e n s o r } , 2 }$ , we apply a first convolution with $\theta _ { 1 } = - 1 , F = F _ { I d } , \theta _ { 2 } = 0$ , to obtain + +$$ +\sum _ { b = 1 } ^ { n } ( x _ { b } - x _ { a } ) \otimes V _ { b } = \sum _ { b = 1 } ^ { n } ( { \bar { x } } _ { b } - { \bar { x } } _ { a } ) \otimes V _ { b } = { \tilde { V } } _ { a } ^ { { \mathrm { t e n s o r } } , 2 } ( V , { \bar { X } } ) - { \bar { x } } _ { a } \otimes \sum _ { b = 1 } ^ { n } V _ { b } +$$ + +By applying an additional averaging filter, defined by setting $\textstyle \theta _ { 1 } = { \frac { 1 } { n } } , F = F _ { 1 } , \theta _ { 2 } = 0$ , we obtain $\tilde { V } _ { a } ^ { \mathrm { t e n s o r } , 2 } ( V , \bar { X } )$ . This concludes our ‘informal proof’. + +Our discussion here has been somewhat inaccurate, since in practice $F _ { I d } ( x ) = ( 0 , x ) \in W _ { 0 } \oplus W _ { 1 }$ and $F _ { 1 } ( x ) = ( 1 , 0 ) \in W _ { 0 } \oplus W _ { 1 }$ . Moreover, in our proof we have glossed over the multiplication by the unitary matrix used to obtain decomposition into irreducible representations. However the ideas discussed here can be used to show that $2 D$ convolutions with 1-filters can satisfy the sufficient condition for $D$ -spanning defined in Lemma 4. See our treatment of Theorem 2 for more details. + +# F COMPARISON WITH ORIGINAL TFN PAPER + +In this Appendix we discuss three superficial differences between the presentation of the TFN architecture in Thomas et al. (2018) and our presentation here: + +1. We define convolutional layers between features residing in direct sums of irreducible representations, while (Thomas et al., 2018) focuses on features which inhabit a single irreducible representation. This difference is non-essential, as direct sums of irreducible representations can be represented as multiple channels where each feature inhabits a single irreducible representation. 2. The term $\theta _ { 0 } V _ { a }$ in equation 14 appears in (Fuchs et al., 2020), but does not appear explicitly in (Thomas et al., 2018). However it can be obtained by concatenation of the input of a self-interaction layer to the output, and then applying a self-interaction layer. 3. We take the scalar functions $R ^ { ( \ell ) }$ to be polynomials, while (Thomas et al., 2018) take them to be fully connected networks composed with radial basis functions. Using polynomial scalar bases is convenient for our presentation here since it enables exact expression of equivariant polynomials. Replacing polynomial bases with fully connected networks, we obtain approximation of equivariant polynomials instead of exact expression. It can be shown that if $p$ is a $G$ -equivariant polynomial which can be expressed by some network $\mathcal { F } _ { C , D }$ defined with filters coming from a polynomial scalar basis, then $p$ can be approximated on a compact set $K$ , up to an arbitrary $\epsilon$ error, by a similar network with scalar functions coming from a sufficiently large fully connected network. + +# G TENSOR UNIVERSALITY + +In this section we show how to construct the complete set $\mathcal { F } _ { \mathrm { p o o l } }$ of linear $\mathrm { S O ( 3 ) }$ invariant functionals from $W _ { \mathrm { f e a t } } ^ { \mathcal { T } } = \bigoplus _ { T = 0 } ^ { D } \mathcal { T } _ { T }$ to $\mathbb { R }$ . Since each such functional $\Lambda$ is of the form + +$$ +\Lambda ( w _ { 0 } , . . . , w _ { D } ) = \sum _ { T = 0 } ^ { D } \Lambda _ { T } ( w _ { T } ) , +$$ + +where each $\Lambda _ { T }$ is $\mathrm { S O ( 3 ) }$ -invariant, it is sufficient to characterize all linear SO(3)-invariant functionals $\Lambda : \mathcal { T } _ { D } \to \mathbb { R }$ . + +It will be convenient to denote + +$$ +W = \mathbb { R } ^ { 3 } \operatorname { a n d } W ^ { \otimes D } \cong \mathbb { R } ^ { 3 ^ { D } } = \mathcal { T } _ { D } . +$$ + +We achieve our characterization using the bijective correspondence between linear functional $\Lambda :$ $W ^ { \otimes D } \to \mathbb { R }$ and multi-linear functions $\tilde { \Lambda } : W ^ { D } \to \mathbb { R }$ : each such $\Lambda$ corresponds to a unique $\hat { \Lambda }$ , such that + +$$ +\tilde { \Lambda } ( e _ { i _ { 1 } } , \ldots , e _ { i _ { D } } ) = \Lambda ( e _ { i _ { 1 } } \otimes \ldots \otimes e _ { i _ { D } } ) , \forall ( i _ { 1 } , \ldots , i _ { D } ) \in [ 3 ] ^ { D } , +$$ + +where $e _ { 1 } , e _ { 2 } , e _ { 3 }$ denote the standard basis elements of $\mathbb { R } ^ { 3 }$ . We define a spanning set of equivariant linear functionals on $W ^ { \otimes D }$ via a corresponding characterization for multi-linear functionals on $W ^ { D }$ . Specifically, set + +$$ +K _ { D } = \{ k \in \mathbb { N } _ { + } | D - 3 k { \mathrm { i s ~ e v e n ~ a n d ~ n o n } } { \mathrm { - n e g a t i v e . ~ } } \} +$$ + +For $k \in K _ { D }$ we define a multi-linear functional: + +$$ +\begin{array} { r } { \tilde { \Lambda } _ { k } ( w _ { 1 } , \dots , w _ { D } ) = \operatorname* { d e t } ( w _ { 1 } , w _ { 2 } , w _ { 3 } ) \times \dots \times \operatorname* { d e t } ( w _ { 3 k - 2 } , w _ { 3 k - 1 } , w _ { 3 k } ) \times \langle w _ { 3 k + 1 } , w _ { 3 k + 2 } \rangle \times \dots } \\ { \times \langle w _ { D - 1 } , w _ { D } \rangle , \qquad ( 2 \tilde { \lambda } _ { 1 } - \tilde { \lambda } _ { 1 } ) \times \dots , } \end{array} +$$ + +and for $( k , \sigma ) \in K _ { D } \times S _ { D }$ we define + +$$ +\tilde { \Lambda } _ { k , \sigma } ( w _ { 1 } , \ldots , w _ { D } ) = \tilde { \Lambda } _ { k } ( w _ { \sigma ( 1 ) } , \ldots , w _ { \sigma ( D ) } ) +$$ + +Proposition 1. The space of linear invariant functions from $\mathcal { T } _ { D }$ to $\mathbb { R }$ is spanned by the set of linear invariant functionals $\lambda _ { D } = \{ \Lambda _ { k , \sigma } | \left( k , \sigma \right) \in K _ { D } \times S _ { D } \}$ induced by the multi-linear functional $\tilde { \Lambda } _ { k , \sigma }$ described in equation 25 and equation $2 6$ + +We note that (i) equation 24 provides a (cumbersome) way to compute all linear invariant functionals $\Lambda _ { k , \sigma }$ explicitly by evaluating the corresponding $\tilde { \Lambda } _ { k , \sigma }$ on the $3 ^ { D }$ elements of the standard basis and (ii) the set $\lambda _ { D }$ is spanning, but is not linearly independent. For example, since $\langle w _ { 1 } , w _ { 2 } \rangle = \langle w _ { 2 } , w _ { 1 } \rangle$ , the space of $\mathrm { S O ( 3 ) }$ invariant functionals on ${ \mathcal { T } } _ { 2 } = { \overline { { W } } } ^ { \otimes 2 }$ is one dimensional while $| \lambda _ { 2 } | = 2$ . + +Proof of Proposition $^ { l }$ . We first show that the bijective correspondence between linear functional $\Lambda : W ^ { \otimes D } \to \mathbb { R }$ and multi-linear functions $\tilde { \Lambda } : \dot { W } ^ { D } \to \mathbb { R }$ , extends to a bijective correspondence between SO(3)-invariant linear/multi-linear functionals. The action of SO(3) on $W ^ { D }$ is defined by + +$$ +\tilde { \rho } ( R ) ( w _ { 1 } , \ldots , w _ { D } ) = ( R w _ { 1 } , \ldots , R w _ { D } ) . +$$ + +The action $\rho ( R ) = R ^ { \otimes D }$ of SO(3) on $W ^ { \otimes D }$ is such that the map + +$$ +( w _ { 1 } , \dots , w _ { D } ) \mapsto w _ { 1 } \otimes w _ { 2 } \dots w _ { D } +$$ + +is SO(3)- equivariant. It follows that if $\tilde { \Lambda }$ and $\Lambda$ satisfy equation 24, then for all $R \in \mathrm { S O ( 3 ) }$ , the same equation holds for the pair $\tilde { \Lambda } \circ \tilde { \rho } ( R )$ and $\Lambda \circ \rho ( R )$ . Thus SO(3)-invariance of $\tilde { \Lambda }$ is equivalent to $\mathrm { S O ( 3 ) }$ -invariance of $\Lambda$ . + +Multi-linear functionals on $W ^ { D }$ invariant to $\tilde { \rho }$ are a subset of the set of polynomials on $W ^ { D }$ invariant to $\tilde { \rho }$ . It is known (see (Kraft & Procesi, 2000), page 114), that all such polynomials are algebraically generated by functions of the form + +$$ +\begin{array} { r } { \operatorname* { d e t } ( w _ { i _ { 1 } } , w _ { i _ { 2 } } , w _ { i _ { 3 } } ) \mathrm { ~ a n d ~ } \langle w _ { j _ { 1 } } , w _ { j _ { 2 } } \rangle , \mathrm { ~ w h e r e ~ } i _ { 1 } , i _ { 2 } , i _ { 3 } , j _ { 1 } , j _ { 2 } \in [ D ] . } \end{array} +$$ + +Equivalently, $\mathrm { S O ( 3 ) }$ -invariant polynomials are spanned by linear combinations of polynomials of the form + +$$ +\operatorname * { d e t } \bigl ( w _ { i _ { 1 } } , w _ { i _ { 2 } } , w _ { i _ { 3 } } \bigr ) \operatorname * { d e t } \bigl ( w _ { i _ { 4 } } , w _ { i _ { 5 } } , w _ { i _ { 6 } } \bigr ) \ldots \langle w _ { j _ { 1 } } , w _ { j _ { 2 } } \rangle \langle w _ { j _ { 3 } } , w _ { j _ { 4 } } \rangle \ldots . +$$ + +When considering the subset of multi-linear invariant polynomials, we see that they must be spanned by polynomials as in equation 27, where each $w _ { 1 } , \ldots , w _ { D }$ appears exactly once in each polynomial in the spanning set. These precisely correspond to the functions in $\lambda _ { D }$ . + +# H EXPERIMENTS + +Table 1: Results obtained on the QM9 dataset Ramakrishnan et al. (2014) for different design choices in the TFN architecture. Results are reported for the $\epsilon _ { h o m o }$ target variable, and are multiplied by $1 0 ^ { 3 }$ . + +
Model variant Mean l1error
Order 0irreps 111.2 ± 0.3
Order 0-1 ireps 82.1 ± 0.4
Order 0-2 irreps 62.1 ± 1.5
Order 0-3 irreps 53.2 ± 1.9
Order 0-4 irreps 51.2 ± 0.4
Order O-3 irreps without non-linearity 53.1 ± 0.8
Order O-3 irreps + self-interaction only in the final layer 55.2 ± 0.3
+ +This section provides an experimental evaluation of different design choices of the TFN architecture, inspired by our theoretical analysis. We study the following questions: + +1. The importance of non-linear activation. Our proof shows that using non-linear activation functions is not necessary for proving universality. Here, we empirically test the possibility of removing these layers. +2. The importance of high-dimensional irreducible representations. Our theoretical analysis shows that in order to represent/approximate high degree polynomials, high-order representations should be used. Here, we check whether using high-order representations has practical benefits. +3. The effect of self-interaction layers. Our proof suggests that it is enough to use self interaction linear layers at the end of the model. We empirically compare this approach with the more common approach of using self-interaction layers after each convolutional layer. + +Dataset. We use the QM9 (Ramakrishnan et al., 2014) dataset for our experiments. The dataset contains 134K molecules, with node 3D positions, 5 categorical node features and 4 categorical edge features. The task is a molecule property prediction regression task. + +Framework. We used pytorch (Paszke et al., 2017)as the deep learning framework and the Deep Graph Library (DGL) (Wang et al., 2019a) as the graph learning framework. All experiments ran on NVIDIA GV100 GPUs. + +Experimental setup. We use the the TFN implementation from Fuchs et al. (2020). We trained each model variant for 50 epochs on the $\epsilon _ { h o m o }$ target variable using an $\ell _ { 1 }$ loss function and the ADAM optimizer with learning rate $1 0 ^ { - 3 }$ and report results on the test set on the final epoch averaged over two runs. We used the default parameters and data splits from Fuchs et al. (2020). + +Architecture. The architecture consists of 4 TFN convolutional layers, each followed by a linear self-interaction layer. We used 16 copies of each irreducible representation used. We used normbased non-linearity as in the original TFN paper (Thomas et al., 2018). These convolutional layers are followed by a max-pooling layer and two fully connected layers with $1 6 d$ features in the hidden layer, where $d$ is the maximal degree of irreducible representations used. + +Results. Table 1 and Figure 1 present the results. The main conclusions are: (1) The experiments show that, at least for this task, using non-linear activations does not improve performance. This result fits our theoretical analysis which shows that these layers are not needed for universality. (2) Figure 1 presents a plot of error vs representation degrees used. The plot clearly shows that using high-dimensional representations (up to order 3) improves performance, which also fits our analysis. Using representation orders higher than 3 is significantly more time consuming, and was found to have little effect on the results (as in in Fuchs et al. (2020)), though we believe this to be application-dependent. (3) Using self interaction layers only at the end of the model is shown to have marginal negative effect on the results. + +![](images/0a3168963674044c22256599e104a6aa2cefc267903dcecc69ad907cfab638f0.jpg) +Figure 1: $\ell _ { 1 }$ error versus maximal irreducible representation used. It is clear that the error is reduced as higher order representation are used. Semi-transparent color represents standard deviation. \ No newline at end of file diff --git a/parse/train/6NFBvWlRXaG/6NFBvWlRXaG_content_list.json b/parse/train/6NFBvWlRXaG/6NFBvWlRXaG_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..2be80e9d5610b0e31cf47e5866f0d2f03a06de20 --- /dev/null +++ b/parse/train/6NFBvWlRXaG/6NFBvWlRXaG_content_list.json @@ -0,0 +1,4073 @@ +[ + { + "type": "text", + "text": "ON THE UNIVERSALITY OF ROTATION EQUIVARIANT POINT CLOUD NETWORKS ", + "text_level": 1, + "bbox": [ + 176, + 98, + 820, + 146 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Nadav Dym \nDuke University \nnadavdym@gmail.com ", + "bbox": [ + 183, + 170, + 362, + 212 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Haggai Maron NVIDIA Research hmaron@nvidia.com ", + "bbox": [ + 511, + 170, + 681, + 212 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "ABSTRACT ", + "text_level": 1, + "bbox": [ + 454, + 250, + 544, + 263 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Learning functions on point clouds has applications in many fields, including computer vision, computer graphics, physics, and chemistry. Recently, there has been a growing interest in neural architectures that are invariant or equivariant to all three shape-preserving transformations of point clouds: translation, rotation, and permutation. In this paper, we present a first study of the approximation power of these architectures. We first derive two sufficient conditions for an equivariant architecture to have the universal approximation property, based on a novel characterization of the space of equivariant polynomials. We then use these conditions to show that two recently suggested models (Thomas et al., 2018; Fuchs et al., 2020) are universal, and for devising two other novel universal architectures. ", + "bbox": [ + 233, + 280, + 764, + 417 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 INTRODUCTION ", + "text_level": 1, + "bbox": [ + 176, + 444, + 339, + 459 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Designing neural networks that respect data symmetry is a powerful approach for obtaining efficient deep models. Prominent examples being convolutional networks which respect the translational invariance of images, graph neural networks which respect the permutation invariance of graphs (Gilmer et al., 2017; Maron et al., 2019b), networks such as (Zaheer et al., 2017; Qi et al., 2017a) which respect the permutation invariance of sets, and networks which respect 3D rotational symmetries (Cohen et al., 2018; Weiler et al., 2018; Esteves et al., 2018; Worrall & Brostow, 2018; Kondor et al., 2018a). ", + "bbox": [ + 173, + 476, + 825, + 571 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "While the expressive power of equivariant models is reduced by design to include only equivariant functions, a desirable property of equivariant networks is universality: the ability to approximate any continuous equivariant function. This is not always the case: while convolutional networks and networks for sets are universal (Yarotsky, 2018; Segol & Lipman, 2019), popular graph neural networks are not (Xu et al., 2019; Morris et al., 2018). ", + "bbox": [ + 174, + 573, + 549, + 683 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/deeacb27baa383de014f051156e964f0944d6459137a61e1a3840aa1470e5f7e.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 583, + 590, + 781, + 698 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "In this paper, we consider the universality of networks that respect the symmetries of 3D point clouds: translations, rotations, and permutations. Designing such networks is a popular paradigm in recent years (Thomas et al., 2018; Fuchs et al., 2020; Poulenard et al., 2019; Zhao et al., 2019). While there have been many works on the universality of permutation invariant networks (Zaheer et al., 2017; Maron et al., $2 0 1 9 \\mathrm { c }$ ; Keriven & Peyre, 2019), ´ and a recent work discussing the universality of rotation equivariant networks (Bogatskiy et al., 2020), this is a first paper which discusses the universality of networks which combine rotations, permutations and translations. ", + "bbox": [ + 174, + 690, + 549, + 704 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 704, + 825, + 801 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "We start the paper with a general, architecture-agnostic, discussion, and derive two sufficient conditions for universality. These conditions are a result of a novel characterization of equivariant polynomials for the symmetry group of interest. We use these conditions in order to prove universality of the prominent Tensor Field Networks (TFN) architecture (Thomas et al., 2018; Fuchs et al., 2020). The following is a weakened and simplified statement of Theorem 2 stated later on in the paper: ", + "bbox": [ + 174, + 808, + 825, + 892 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Theorem (Simplification of Theorem 2). Any continuous equivariant function on point clouds can be approximated uniformly on compact sets by a composition of TFN layers. ", + "bbox": [ + 174, + 895, + 823, + 924 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "We use our general discussion to prove the universality of two additional equivariant models: the first is a simple modification of the TFN architecture which allows for universality using only low dimensional filters. The second is a minimal architecture which is based on tensor product representations, rather than the more commonly used irreducible representations of SO(3). We discuss the advantages and disadvantages of both approaches. ", + "bbox": [ + 174, + 103, + 823, + 174 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "To summarize, the contributions of this paper are: (1) A general approach for proving the universality of rotation equivariant models for point clouds; (2) A proof that two recent equivariant models (Thomas et al., 2018; Fuchs et al., 2020) are universal; (3) Two additional simple and novel universal architectures. ", + "bbox": [ + 174, + 180, + 823, + 236 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2 PREVIOUS WORK ", + "text_level": 1, + "bbox": [ + 176, + 258, + 348, + 275 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Deep learning on point clouds. (Qi et al., 2017a; Zaheer et al., 2017) were the first to apply neural networks directly to the raw point cloud data, by using pointwise functions and pooling operations. Many subsequent works used local neighborhood information (Qi et al., 2017b; Wang et al., 2019b; Atzmon et al., 2018). We refer the reader to a recent survey for more details (Guo et al., 2020). In contrast with the aforementioned works which focused solely on permutation invariance, more related to this paper are works that additionally incorporated invariance to rigid motions. (Thomas et al., 2018) proposed Tensor Field Networks (TFN) and showed their efficacy on physics and chemistry tasks.(Kondor et al., 2018b) also suggested an equivariant model for continuous rotations. (Li et al., 2019) suggested models that are equivariant to discrete subgroups of SO(3). (Poulenard et al., 2019) suggested an invariant model based on spherical harmonics. (Fuchs et al., 2020) followed TFN and added an attention mechanism. Recently, (Zhao et al., 2019) proposed a quaternion equivariant point capsule network that also achieves rotation and translation invariance. ", + "bbox": [ + 173, + 292, + 825, + 459 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Universal approximation for invariant networks. Understanding the approximation power of invariant models is a popular research goal. Most of the current results assume that the symmetry group is a permutation group. (Zaheer et al., 2017; Qi et al., $2 0 1 7 \\mathrm { a }$ ; Segol & Lipman, 2019; Maron et al., 2020; Serviansky et al., 2020) proved universality for several $S _ { n }$ -invariant and equivariant models. (Maron et al., 2019b;a; Keriven & Peyre, 2019; Maehara & NT, 2019) studied the approx- ´ imation power of high-order graph neural networks. (Maron et al., 2019c; Ravanbakhsh, 2020) targeted universality of networks that use high-order representations for permutation groups(Yarotsky, 2018) provided several theoretical constructions of universal equivariant neural network models based on polynomial invariants, including an $S E ( 2 )$ equivariant model. In a recent work (Bogatskiy et al., 2020) presented a universal approximation theorem for networks that are equivariant to several Lie groups including $S O ( 3 )$ . The main difference from our paper is that we prove a universality theorem for a more complex group that besides rotations also includes translations and permutations. ", + "bbox": [ + 174, + 467, + 825, + 633 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "3 A FRAMEWORK FOR PROVING UNIVERSALITY ", + "text_level": 1, + "bbox": [ + 176, + 657, + 583, + 671 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "In this section, we describe a framework for proving the universality of equivariant networks. We begin with some mathematical preliminaries: ", + "bbox": [ + 174, + 689, + 823, + 718 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "3.1 MATHEMATICAL SETUP ", + "text_level": 1, + "bbox": [ + 176, + 737, + 379, + 752 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "An action of a group $G$ on a real vector space $W$ is a collection of maps $\\rho ( g ) : W \\to W$ defined for any $g \\in G$ , such that $\\rho ( g _ { 1 } ) \\circ \\rho ( g _ { 2 } ) = \\rho ( g _ { 1 } g _ { 2 } )$ for all $g _ { 1 } , g _ { 2 } \\in G$ , and the identity element of $G$ is mapped to the identity mapping on $W$ . We say $\\rho$ is a representation of $G$ if $\\rho ( g )$ is a linear map for every $g \\in G$ . As is customary, when it does not cause confusion we often say that $W$ itself is a representation of $G$ . ", + "bbox": [ + 174, + 763, + 825, + 834 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "In this paper, we are interested in functions on point clouds. Point clouds are sets of vectors in $\\mathbb { R } ^ { 3 }$ arranged as matrices: ", + "bbox": [ + 171, + 840, + 823, + 869 + ], + "page_idx": 1 + }, + { + "type": "equation", + "img_path": "images/2b2080272f92a716470478cabee758a436313e049083470d0833964e2e318b44.jpg", + "text": "$$\nX = ( x _ { 1 } , \\ldots , x _ { n } ) \\in \\mathbb { R } ^ { 3 \\times n } .\n$$", + "text_format": "latex", + "bbox": [ + 406, + 869, + 591, + 888 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Many machine learning tasks on point clouds, such as classification, aim to learn a function which is invariant to rigid motions and relabeling of the points. Put differently, such functions are required ", + "bbox": [ + 173, + 895, + 825, + 924 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "to be invariant to the action of $G = \\mathbb { R } ^ { 3 } \\rtimes \\mathrm { S O ( 3 ) } \\times S _ { n }$ on $\\mathbb { R } ^ { 3 \\times n }$ via ", + "bbox": [ + 173, + 102, + 617, + 118 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/80e403e3b0b33933be7a0061a3d3d29a6dfa55bc07f442f7a89f4af8dbd9fd6f.jpg", + "text": "$$\n\\rho _ { G } ( t , R , P ) ( X ) = R ( X - t 1 _ { n } ^ { T } ) P ^ { T } ,\n$$", + "text_format": "latex", + "bbox": [ + 377, + 121, + 617, + 138 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "where $t \\in \\mathbb { R } ^ { 3 }$ defines a translation, $R$ is a rotation and $P$ is a permutation matrix. ", + "bbox": [ + 176, + 140, + 705, + 155 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Equivariant functions are generalizations of invariant functions: If $G$ acts on $W _ { 1 }$ via some action $\\bar { \\rho _ { 1 } ( g ) }$ , and on $W _ { 2 }$ via some other group action $\\rho _ { 2 } ( g )$ , we say that a function $f : W _ { 1 } \\to W _ { 2 }$ is equivariant if ", + "bbox": [ + 174, + 160, + 825, + 202 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Invariant functions correspond to the special case where $\\rho _ { 2 } ( g )$ is the identity mapping for all $g \\in G$ ", + "bbox": [ + 176, + 218, + 821, + 233 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "In some machine learning tasks on point clouds, the functions learned are not invariant but rather equivariant. For example, segmentation tasks assign a discrete label to each point. They are invariant to translations and rotations but equivariant to permutations – in the sense that permuting the input causes a corresponding permutation of the output. Another example is predicting a normal for each point of a point cloud. This task is invariant to translations but equivariant to both rotations and permutations. ", + "bbox": [ + 173, + 239, + 825, + 323 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "In this paper, we are interested in learning equivariant functions from point clouds into $W _ { T } ^ { n }$ , where $W _ { T }$ is some representation of SO(3). The equivariance of these functions is with respect to the action $\\rho _ { G }$ on point clouds defined in equation 1, and the action of $G$ on $W _ { T } ^ { n }$ defined by applying the rotation action from the left and permutation action from the right as in 1, but ‘ignoring’ the translation component. Thus, $G$ -equivariant functions will be translation invariant. This formulation of equivariance includes the normals prediction example by taking $W _ { T } = \\mathbb { R } ^ { 3 }$ , as well as the segmentation case by setting $W _ { T } = \\mathbb { R }$ with the trivial identity representation. We focus on the harder case of functions into $W _ { T } ^ { n }$ which are equivariant to permutations, since it easily implies the easier case of permutation invariant functions to $W _ { T }$ . ", + "bbox": [ + 173, + 329, + 825, + 455 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Notation. We use the notation $ { \\mathbb { N } } _ { + } = { \\mathbb { N } } \\cup \\lbrace 0 \\rbrace$ and $\\begin{array} { r } { \\mathbb { N } _ { + } ^ { * } = \\bigcup _ { r \\in \\mathbb { N } } \\mathbb { N } _ { + } ^ { r } } \\end{array}$ . We set $[ D ] = \\{ 1 , \\dots , D \\}$ and $[ D ] _ { 0 } = \\{ 0 , \\dots , D \\}$ . ", + "bbox": [ + 173, + 460, + 821, + 493 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Proofs. Proofs appear in the appendices, arranged according to sections. ", + "bbox": [ + 174, + 498, + 648, + 513 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3.2 CONDITIONS FOR UNIVERSALITY ", + "text_level": 1, + "bbox": [ + 178, + 529, + 444, + 542 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "The semi-lifted approach In general, highly expressive equivariant neural networks can be achieved by using a ‘lifted approach’, where intermediate features in the network belong to high dimensional representations of the group. In the context of point clouds where typically $n \\gg 3$ , many papers, e.g., (Thomas et al., 2018; Kondor, 2018; Bogatskiy et al., 2020) use a ‘semi-lifted’ approach, where hidden layers hold only higher dimensional representations of SO(3), but not high order permutation representations. In this subsection, we propose a strategy for achieving universality with the semi-lifted approach. ", + "bbox": [ + 173, + 554, + 825, + 652 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "We begin by an axiomatic formulation of the semi-lifted approach (see illustration in inset): we assume that our neural networks are composed of two main components: the first component is a family $\\mathcal { F } _ { \\mathrm { f e a t } }$ of parametric continuous $G$ -equivariant functions $f _ { \\mathrm { f e a t } }$ which map the original point cloud $\\mathbb { R } ^ { 3 \\times n }$ to a semi-lifted point cloud $\\begin{array} { c c l } { W _ { \\mathrm { f e a t } } ^ { n } } & { = } & { \\bigoplus _ { i = 1 } ^ { n } W _ { \\mathrm { f e a t } } } \\end{array}$ , where $W _ { \\mathrm { f e a t } }$ is a lifted (i.e., high-order) representation of SO(3). ", + "bbox": [ + 173, + 660, + 825, + 729 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "The second component is a family of parametric linear SO(3)-equivariant functions $\\mathcal { F } _ { \\mathrm { p o o l } }$ , which map from the high order representation $W _ { \\mathrm { f e a t } }$ down to the target representation $W _ { T }$ . Each such $\\mathrm { S O ( 3 ) }$ –equivariant function $\\Lambda$ : $W _ { \\mathrm { f e a t } } \\to W _ { T }$ can be extended to a $\\mathrm { S O } ( 3 ) \\times S _ { n }$ equivariant function $\\hat { \\Lambda } : W _ { \\mathrm { f e a t } } ^ { n } \\to W _ { T } ^ { n }$ by applying $\\Lambda$ elementwise. For every positive integer $C$ , these two families of functions induce a family of functions $\\mathcal { F } _ { C }$ obtained by summing $C$ different compositions of these functions: ", + "bbox": [ + 174, + 731, + 485, + 883 + ], + "page_idx": 2 + }, + { + "type": "image", + "img_path": "images/b54a82d14eb02b10dbee355a8abcb92b8088b25aadb6409a1bedc597d43bf20d.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 521, + 733, + 766, + 862 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/b06c45d6a0b3b6d8ed3f7c06f8d25672199ece3ea9ef5d00ecf9d267757b82bb.jpg", + "text": "$$\n\\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } , \\mathcal { F } _ { \\mathrm { p o o l } } ) = \\{ f | f ( X ) = \\sum _ { c = 1 } ^ { C } \\hat { \\Lambda } _ { c } ( g _ { c } ( X ) ) , \\ : ( \\Lambda _ { c } , g _ { c } ) \\in \\mathcal { F } _ { \\mathrm { p o o l } } \\times \\mathcal { F } _ { \\mathrm { f e a t } } \\} .\n$$", + "text_format": "latex", + "bbox": [ + 245, + 885, + 751, + 928 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Conditions for universality We now describe two conditions that guarantee universality using the semi-lifted approach. The first step is showing, as in (Yarotsky, 2018), that continuous $G$ -equivariant functions $\\mathcal { C } _ { G } \\mathrm { ( } \\mathbb { R } ^ { 3 \\times n } , W _ { T } ^ { n } \\mathrm { ) }$ can be approximated by $G$ -equivariant polynomials $\\mathcal { P } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , \\bar { W } _ { T } ^ { n } )$ . ", + "bbox": [ + 173, + 103, + 823, + 147 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Lemma 1. Any continuous $G$ -equivariant function in $\\mathcal { C } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , W _ { T } ^ { n } )$ can be approximated uniformly on compact sets by $G$ -equivariant polynomials in $\\mathcal { P } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , \\bar { W } _ { T } ^ { n } )$ . ", + "bbox": [ + 173, + 150, + 823, + 180 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Universality is now reduced to the approximation of $G$ -equivariant polynomials. Next, we provide two conditions which guarantee that $G$ -equivariant polynomials of degree $D$ can be expressed by function spaces $\\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } , \\mathcal { F } _ { \\mathrm { p o o l } } )$ as defined in equation 2. The idea behind these conditions is that explicit characterizations of polynomials equivariant to the joint action of translations, rotations and permutations is challenging. However, it is possible to explicitly characterize polynomials equivariant to translations and permutations (but not rotations). The key observation is then that this characterization can be rewritten as a sum of functions to $W _ { \\mathrm { f e a t } } ^ { n }$ , a high dimensional representations of SO(3) which is equivariant to translations, permutations and rotations, composed with a linear map which is permutation equivariant (but does not respect rotations). Accordingly, our first condition is that $\\mathcal { F } _ { \\mathrm { f e a t } }$ contains a spanning set of such functions to $W _ { \\mathrm { f e a t } } ^ { n }$ . We call this condition $D$ -spanning: ", + "bbox": [ + 174, + 191, + 825, + 345 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Definition 1 ( $D$ -spanning). For $D \\in { \\mathbb { N } } _ { + }$ , let $\\mathcal { F } _ { \\mathrm { f e a t } }$ be a subset of $\\mathcal { C } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , W _ { \\mathrm { f e a t } } ^ { n } )$ . We say that $\\mathcal { F } _ { \\mathrm { f e a t } }$ is $D$ -spanning, if there exist $f _ { 1 } , \\dots , f _ { K } \\in \\mathcal { F } _ { \\mathrm { f e a t } }$ , such that every polynomial $p : \\mathbb { R } ^ { 3 \\times n } \\mathbb { R } ^ { n }$ of degree $D$ which is invariant to translations and equivariant to permutations, can be written as ", + "bbox": [ + 173, + 349, + 825, + 393 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/43e6309bd685ec48eb69a7abb94cee27f9f294cbd921a8df42aa6dc02eebdbe9.jpg", + "text": "$$\np ( X ) = \\sum _ { k = 1 } ^ { K } \\hat { \\Lambda } _ { k } ( f _ { k } ( X ) ) ,\n$$", + "text_format": "latex", + "bbox": [ + 415, + 401, + 581, + 445 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "where $\\Lambda _ { k } : W _ { \\mathrm { f e a t } } \\to \\mathbb { R }$ are all linear functionals, and ${ \\hat { \\boldsymbol { \\Lambda } } } _ { k } : { \\cal W } _ { \\mathrm { f e a t } } ^ { n } \\mathbb { R } ^ { n }$ are the functions defined by elementwise applications of $\\Lambda _ { k }$ . ", + "bbox": [ + 171, + 454, + 825, + 486 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "In Lemma 4 we explicitly construct a $D$ -spanning family of functions. This provides us with a concrete condition which implies $D$ -spanning for other function families as well. ", + "bbox": [ + 174, + 497, + 823, + 526 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "The second condition is that $\\mathcal { F } _ { \\mathrm { p o o l } }$ contains all SO(3) linear equivariant layers. We call this condition Linear universality. Intuitively, taking linear rotation equivariant $\\Lambda _ { k }$ in equation 3 ensures that the resulting function $p$ will be rotation equivariant and thus fully $G$ -equivariant, and linear universality guarantees the ability to express all such $G$ invariant functions. ", + "bbox": [ + 174, + 531, + 825, + 588 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Definition 2 (Linear universality). We say that a collection $\\mathcal { F } _ { \\mathrm { p o o l } }$ of equivariant linear functionals between two representations $W _ { \\mathrm { f e a t } }$ and $W _ { T }$ of SO(3) is linearly universal, if it contains all linear SO(3)-equivariant mappings between the two representations. ", + "bbox": [ + 174, + 593, + 825, + 637 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "When these two conditions apply, a rather simple symmetrization arguments leads to the following theorem: ", + "bbox": [ + 174, + 648, + 823, + 678 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Theorem 1. If $\\mathcal { F } _ { \\mathrm { f e a t } }$ is $D$ -spanning and $\\mathcal { F } _ { \\mathrm { p o o l } }$ is linearly universal, then there exists some $C ( D ) \\in$ $\\mathbb { N }$ such that for all $C \\geq C ( D )$ the function space $\\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } , \\mathcal { F } _ { \\mathrm { p o o l } } )$ contains all $G$ -equivariant polynomials of degree $\\le D$ . ", + "bbox": [ + 173, + 681, + 825, + 726 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Proof idea. By the $D$ -spanning assumption, there exist $f _ { 1 } , \\dots , f _ { K } \\in { \\mathcal { F } } _ { \\mathrm { f e a t } }$ such that any vector valued polynomial invariant to translations and equivariant to permutations is of the form ", + "bbox": [ + 173, + 752, + 823, + 781 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/65688a9ed2ba7a86da01fc534b78beb1d3c5369d1355607144fa4b5d092569e0.jpg", + "text": "$$\np ( X ) = \\sum _ { k = 1 } ^ { K } \\hat { \\Lambda } _ { k } ( f _ { k } ( X ) ) ,\n$$", + "text_format": "latex", + "bbox": [ + 415, + 790, + 581, + 834 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "While by definition this holds for functions $p$ whose image is $\\mathbb { R } ^ { n }$ , this is easily extended to functions to $W _ { T } ^ { n }$ as well. ", + "bbox": [ + 171, + 840, + 825, + 871 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "It remains to show that when $p$ is also SO(3)-equivariant, we can choose $\\Lambda _ { k }$ to be SO(3) equivariant. \nThis is accomplished by averaging over SO(3). ", + "bbox": [ + 171, + 876, + 823, + 906 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "As a result of Theorem 1 and Lemma 1 we obtain our universality result (see inset for illustration) ", + "bbox": [ + 174, + 103, + 549, + 132 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Corollary 1. For all ${ \\cal C } , { \\cal D } \\in \\mathbb { N } _ { + }$ , let $\\mathcal { F } _ { C , D }$ denote function spaces generated by a pair of functions spaces which are $D$ -spanning and linearly universal as in equation 2. Then any continuous $G$ -equivariant function in $\\mathcal { C } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , W _ { T } ^ { n } )$ can be approximated uniformly on compact sets by equivariant functions in ", + "bbox": [ + 173, + 135, + 549, + 191 + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/e9ffa6a3d2fc7c2c788f5f90cefc79b1dd49ce1bf98aa7f80e26cd6b3b4cedca.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 575, + 85, + 787, + 166 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "", + "bbox": [ + 168, + 191, + 776, + 207 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/1c264ee43605c8770f7fcfe0fb711e379d5b74553fd50d1b112fdb2da713aff9.jpg", + "text": "$$\n{ \\mathcal { F } } = \\bigcup _ { D \\in \\mathbb { N } } { \\mathcal { F } } _ { C ( D ) , D } .\n$$", + "text_format": "latex", + "bbox": [ + 431, + 212, + 565, + 246 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "3.3 UNIVERSALITY CONDITIONS IN ACTION ", + "text_level": 1, + "bbox": [ + 176, + 260, + 488, + 273 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "In the remainder of the paper, we prove the universality of several $G$ -equivariant architectures, based on the framework we discussed in the previous subsection. We discuss two different strategies for achieving universality, which differ mainly in the type of lifted representations of SO(3) they use: (i) The first strategy uses (direct sums of) tensor-product representations; (ii) The second uses (direct sums of) irreducible representations. The main advantage of the first strategy from the perspective of our methodology is that achieving the $D$ -spanning property is more straightforward. The advantage of irreducible representations is that they almost automatically guarantees the linear universality property. ", + "bbox": [ + 173, + 285, + 825, + 398 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "In Section 4 we discuss universality through tensor product representations, and give an example of a minimal tensor representation network architecture that would satisfy universality. In section 5 we discuss universality through irreducible representations, which is currently the more common strategy. We show that the TFN architecture (Thomas et al., 2018; Fuchs et al., 2020) which follows this strategy is universal, and describe a simple tweak that achieves universality using only low order filters, though the representations throughout the network are high dimensional. ", + "bbox": [ + 173, + 404, + 825, + 488 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "4 UNIVERSALITY WITH TENSOR REPRESENTATIONS ", + "text_level": 1, + "bbox": [ + 174, + 508, + 617, + 523 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "In this section, we prove universality for models that are based on tensor product representations, as defined below. The main advantage of this approach is that $D$ -spanning is achieved rather easily. The main drawbacks are that its data representation is somewhat redundant and that characterizing the linear equivariant layers is more laborious. ", + "bbox": [ + 173, + 539, + 823, + 594 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Tensor representations We begin by defining tensor representations. For $k \\in \\mathbb { N } _ { + }$ denote $\\mathcal { T } _ { k } = \\mathbb { R } ^ { 3 ^ { k } }$ . SO(3) acts on $\\mathcal { T } _ { k }$ by the tensor product representation, i.e., by applying the matrix Kronecker product $k$ times: $\\rho _ { k } ( R ) : =$ $\\bar { R } ^ { \\otimes k }$ . The inset illustrates the vector spaces and action for $k = 1 , 2 , 3$ . With this action, for any $\\bar { i } _ { 1 } , \\ldots , i _ { k } \\in [ n ]$ , the map from $\\mathbb { R } ^ { 3 \\times n }$ to $\\mathcal { T } _ { k }$ defined by ", + "bbox": [ + 173, + 616, + 550, + 688 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 688, + 766, + 703 + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/50d9d12099cc8cbcdfa045b7f9ad7b985e39b62e026d157b80b6dee7accd6a00.jpg", + "image_caption": [], + "image_footnote": [], + "bbox": [ + 570, + 613, + 792, + 685 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/3a715a86537245fcd6b8c4ec6286409899ac3e9af73d7db3cea35849934105da.jpg", + "text": "$$\n( x _ { 1 } , \\ldots , x _ { n } ) \\mapsto x _ { i _ { 1 } } \\otimes x _ { i _ { 2 } } \\ldots \\otimes x _ { i _ { k } }\n$$", + "text_format": "latex", + "bbox": [ + 379, + 708, + 617, + 724 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "is SO(3) equivariant. ", + "bbox": [ + 173, + 728, + 313, + 744 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "A $D$ -spanning family We now show that tensor representations can be used to define a finite set of $D$ -spanning functions. The lifted representation $W _ { \\mathrm { f e a t } }$ will be given by ", + "bbox": [ + 173, + 751, + 826, + 780 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/61830aefac19c1277d057845830e6bfae00c53974c5a2f9063ff38a8ceec272e.jpg", + "text": "$$\nW _ { \\mathrm { f e a t } } ^ { \\mathcal { T } } = \\bigoplus _ { T = 0 } ^ { D } \\mathcal { T } _ { T } .\n$$", + "text_format": "latex", + "bbox": [ + 441, + 784, + 557, + 827 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "The $D$ -spanning functions are indexed by vectors $\\vec { r } = \\left( r _ { 1 } , \\ldots , r _ { K } \\right)$ , where each $r _ { k }$ is a non-negative integer. Denoting $T = \\| r \\| _ { 1 }$ , the functions $Q ^ { ( \\vec { r } ) } : \\mathbb { R } ^ { 3 \\times n } \\mathcal { T } _ { T } ^ { n }$ , $Q ^ { ( \\vec { r } ) } = ( Q _ { j } ^ { ( \\vec { r } ) } ) _ { j = 1 } ^ { n }$ ))nj=1 are defined for fixed $j \\in [ n ]$ by ", + "bbox": [ + 174, + 832, + 825, + 881 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/e6ca06060048c51bdbe91ba75c3f61ec795f6819e3895e4f638adf7baba11d0e.jpg", + "text": "$$\nQ _ { j } ^ { ( \\vec { r } ) } ( X ) = \\sum _ { i _ { 2 } , \\ldots , i _ { K } = 1 } ^ { n } x _ { j } ^ { \\otimes r _ { 1 } } \\otimes x _ { i _ { 2 } } ^ { \\otimes r _ { 2 } } \\otimes x _ { i _ { 3 } } ^ { \\otimes r _ { 3 } } \\otimes \\ldots \\otimes x _ { i _ { K } } ^ { \\otimes r _ { K } } .\n$$", + "text_format": "latex", + "bbox": [ + 305, + 885, + 692, + 928 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "The functions $Q _ { j } ^ { ( \\vec { r } ) }$ are SO(3) equivariant as they are a sum of equivariant functions from equation 5. Thus $Q ^ { ( \\vec { r } ) }$ is $\\mathrm { S O } ( 3 ) \\times S _ { n }$ equivariant. The motivation behind the definition of these functions is that known characterizations of permutation equivariant polynomials Segol & Lipman (2019) tell us that the entries of these tensor valued functions span all permutation equivariant polynomials (see the proof of Lemma 2 for more details). ", + "bbox": [ + 173, + 102, + 825, + 179 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "To account for translation invariance, we compose the functions $Q ^ { ( \\vec { r } ) }$ with the centralization operation and define the set of functions ", + "bbox": [ + 174, + 184, + 821, + 214 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/577822362a69d996a5b8f585aa8be41f18d57409dd20b244abea1af4e17b8356.jpg", + "text": "$$\n\\mathcal { Q } _ { D } = \\{ \\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac { 1 } { n } X 1 _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) | \\| \\vec { r } \\| _ { 1 } \\leq D \\} ,\n$$", + "text_format": "latex", + "bbox": [ + 344, + 218, + 651, + 250 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "where $\\iota$ is the natural embedding that takes each $\\mathcal { T } _ { T }$ into $W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } } = \\oplus _ { T = 0 } ^ { D } \\mathcal { T } _ { T }$ . In the following lemma, we prove that this set is $D$ -spanning. ", + "bbox": [ + 174, + 257, + 823, + 287 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Lemma 2. For every $D \\in { \\mathbb { N } } _ { + }$ , the set $\\mathcal { Q } _ { D }$ is $D$ -spanning. ", + "bbox": [ + 173, + 290, + 558, + 306 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Proof idea. It is known (Segol & Lipman, 2019) (Theorem 2) that polynomials $p : \\mathbb { R } ^ { 3 \\times n } \\mathbb { R } ^ { n }$ which are $S _ { n }$ -equivariant, are spanned by polynomials of the form $p _ { \\vec { \\alpha } } = ( p _ { \\vec { \\alpha } } ^ { j } ) _ { j = 1 } ^ { n }$ , defined as ", + "bbox": [ + 171, + 327, + 823, + 359 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/99ab3f48ed1763969604fac776352a2c075cf44d5e166c7a767f93ed4702223c.jpg", + "text": "$$\np _ { \\vec { \\alpha } } ^ { j } ( X ) = \\sum _ { i _ { 2 } , \\ldots , i _ { K } = 1 } ^ { n } x _ { j } ^ { \\alpha _ { 1 } } x _ { i _ { 2 } } ^ { \\alpha _ { 2 } } \\cdot \\cdot \\cdot x _ { i _ { k } } ^ { \\alpha _ { k } }\n$$", + "text_format": "latex", + "bbox": [ + 379, + 367, + 616, + 410 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "where $\\vec { \\alpha } = ( \\alpha _ { 1 } , \\ldots , \\alpha _ { K } )$ and each $\\alpha _ { k } \\in \\mathbb { N } _ { + } ^ { 3 }$ is a multi-index. We first show that these polynomials can be extracted from $Q ^ { ( \\vec { r } ) }$ and then use them to represent $p$ . □ ", + "bbox": [ + 174, + 417, + 825, + 450 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "A minimal universal architecture Once we have shown that $\\mathcal { Q } _ { D }$ is $D$ -spanning, we can design $D$ -spanning architectures, by devising architectures that are able to span all elements of $\\mathcal { Q } _ { D }$ . As we will now show, the compositional nature of neural networks allows us to do this in a very clean manner. ", + "bbox": [ + 173, + 465, + 826, + 522 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "We define a parametric function $f ( X , V | \\theta _ { 1 } , \\theta _ { 2 } )$ which maps $\\mathbb { R } ^ { 3 \\times n } \\oplus \\mathcal { T } _ { k } ^ { n }$ to $\\mathbb { R } ^ { 3 \\times n } \\oplus \\mathcal { T } _ { k + 1 } ^ { n }$ as follows: For all $j \\in [ n ]$ , we have $f _ { j } ( X , V ) = ( x _ { j } , \\tilde { V } _ { j } ( X , V ) )$ , where ", + "bbox": [ + 173, + 527, + 820, + 563 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/d38e84166719d7c03679c4c996ad32c73150b74c54182a5d36ccac62eb944303.jpg", + "text": "$$\n\\tilde { V } _ { j } ( X , V | \\theta _ { 1 } , \\theta _ { 2 } ) = \\theta _ { 1 } \\left( x _ { j } \\otimes V _ { j } \\right) + \\theta _ { 2 } \\sum _ { i } \\left( x _ { i } \\otimes V _ { i } \\right)\n$$", + "text_format": "latex", + "bbox": [ + 328, + 577, + 669, + 611 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "We denote the set of functions $( X , V ) \\mapsto f ( X , V | \\theta _ { 1 } , \\theta _ { 2 } )$ obtained by choosing the parameters $\\theta _ { 1 } , \\theta _ { 2 } \\in \\mathbb { R }$ , by $\\mathcal { F } _ { m i n }$ . While in the hidden layers of our network the data is represented using both coordinates $( X , V )$ , the input to the network only contains an $X$ coordinate and the output only contains a $V$ coordinate. To this end, we define the functions ", + "bbox": [ + 173, + 621, + 825, + 678 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/a5eede119b22fdd477db38131758e79a5298c33c1d635ab71ae68ec67c6b3b40.jpg", + "text": "$$\n\\operatorname { e x t } ( X ) = ( X , 1 _ { n } ) \\operatorname { a n d } \\pi _ { V } ( X , V ) = V .\n$$", + "text_format": "latex", + "bbox": [ + 366, + 685, + 630, + 703 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "We can achieve $D$ -spanning by composition of functions in $\\mathcal { F } _ { m i n }$ with these functions and centralizing: ", + "bbox": [ + 173, + 709, + 826, + 739 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Lemma 3. The function set $\\mathcal { Q } _ { D }$ is contained in ", + "bbox": [ + 171, + 742, + 488, + 757 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/93ede582cc29e75094f8d3a9e1daafa85c8f9c1049c9e766450977511ecf598a.jpg", + "text": "$$\n\\mathcal { F } _ { \\mathrm { f e a t } } = \\big \\{ \\iota \\circ \\pi _ { V } \\circ f ^ { 1 } \\circ f ^ { 2 } \\circ \\ldots \\circ f ^ { T } \\circ \\mathrm { e x t } ( X - \\frac { 1 } { n } X \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) | f ^ { j } \\in \\mathcal { F } _ { m i n } , T \\leq D \\big \\} .\n$$", + "text_format": "latex", + "bbox": [ + 227, + 763, + 769, + 794 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Thus $\\mathcal { F } _ { \\mathrm { f e a t } }$ is $D$ -spanning. ", + "bbox": [ + 174, + 799, + 346, + 815 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Proof idea. The proof is technical and follows by induction on $D$ . ", + "bbox": [ + 173, + 835, + 607, + 852 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "To complete the construction of a universal network, we now need to characterize all linear equivariant functions from $W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } }$ to the target representation $W _ { T }$ . In Appendix $\\mathbf { G }$ we show how this can be done for the trivial representation $W _ { T } = \\mathbb { R }$ . This characterization gives us a set of linear functions $\\mathcal { F } _ { \\mathrm { p o o l } }$ , which combined with $\\mathcal { F } _ { \\mathrm { f e a t } }$ defined in equation 11 (corresponds to SO(3) invariant functions) ", + "bbox": [ + 174, + 867, + 825, + 924 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "gives us a universal architecture as in Theorem 1. However, the disadvantage of this approach is that implementation of the linear functions in $\\mathcal { F } _ { \\mathrm { p o o l } }$ is somewhat cumbersome. ", + "bbox": [ + 169, + 103, + 823, + 133 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "In the next section we discuss irreducible representations, which give us a systematic way to address linear equivariant mappings into any $W _ { T }$ . Proving $D$ -spanning for these networks is accomplished via the $D$ -spanning property of tensor representations, through the following lemma ", + "bbox": [ + 174, + 138, + 823, + 181 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Lemma 4. If all functions in $\\mathcal { Q } _ { D }$ can be written as ", + "bbox": [ + 173, + 184, + 513, + 200 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/e30e8f7ccf0c360604f9bbc4ab4bfc3b7f392383f90a98dc50473b858cbedd5b.jpg", + "text": "$$\n\\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac { 1 } { n } X 1 _ { n } 1 _ { n } ^ { T } ) = \\sum _ { k = 1 } ^ { K } \\hat { A } _ { k } f _ { k } ( X ) ,\n$$", + "text_format": "latex", + "bbox": [ + 356, + 204, + 638, + 248 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "where $f _ { k } \\in \\mathcal { F } _ { \\mathrm { f e a t } }$ , $A _ { k } : W _ { \\mathrm { f e a t } } \\to W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } }$ and ${ \\hat { A } } _ { k } : W _ { \\mathrm { f e a t } } ^ { n } \\to ( W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } } ) ^ { n }$ is defined by elementwise application of $A _ { k }$ , then $\\mathcal { F } _ { \\mathrm { f e a t } }$ is $D$ -spanning. ", + "bbox": [ + 173, + 253, + 825, + 285 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "We note that as before, $A _ { k }$ are not necessarily SO(3)- equivariant. ", + "bbox": [ + 174, + 295, + 607, + 310 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Proof idea. The lemma follows directly from the assumptions. ", + "bbox": [ + 173, + 325, + 584, + 340 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "5 UNIVERSALITY WITH IRREDUCIBLE REPRESENTATIONS ", + "text_level": 1, + "bbox": [ + 176, + 369, + 665, + 386 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "In this section, we discuss how to achieve universality when using irreducible representations of SO(3). We will begin by defining irreducible representations, and explaining how linear universality is easily achieved by them, while the $D$ -spanning properties of tensor representations can be preserved. This discussion can be seen as an interpretation of the choices made in the construction of TFN and similar networks in the literature. We then show that these architectures are indeed universal. ", + "bbox": [ + 173, + 400, + 825, + 484 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "5.1 IRREDUCIBLE REPRESENTATIONS OF $S O ( 3 )$ ", + "text_level": 1, + "bbox": [ + 174, + 501, + 516, + 516 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "In general, any finite-dimensional representation $W$ of a compact group $H$ can be decomposed into irreducible representations: a subspace $W _ { 0 } \\subset W$ is $H$ -invariant if $h w \\in W _ { 0 }$ for all $h \\in H , w \\in W _ { 0 }$ . A representation $W$ is irreducible if it has no non-trivial invariant subspaces. In the case of SO(3), all irreducible real representations are defined by matrices $D ^ { ( \\ell ) } ( R )$ , called the real Wigner D-matrices, acting on $W _ { \\ell } : = \\overset { \\bullet } { \\mathbb { R } } ^ { 2 \\ell + 1 }$ by matrix multiplication. In particular, the representation for $\\ell = 0 , 1$ are $D ^ { ( 0 ) } ( R ) = 1$ and $D ^ { ( 1 ) } ( R ) = R$ . ", + "bbox": [ + 173, + 526, + 825, + 614 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Linear maps between irreducible representations As mentioned above, one of the main advantages of using irreducible representations is that there is a very simple characterization of all linear equivariant maps between two direct sums of irreducible representations. We use the notation $W _ { l }$ for direct sums of irreducible representations, where $\\boldsymbol { l } = ( \\ell _ { 1 } , \\dots , \\ell _ { K } ) \\in \\mathbb { N } _ { + } ^ { K }$ and $W _ { l } = \\bigoplus _ { k = 1 } ^ { K } W _ { \\ell _ { k } }$ ", + "bbox": [ + 173, + 621, + 825, + 680 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Lemma 5. Let $l ^ { ( 1 ) } = ( \\ell _ { 1 } ^ { ( 1 ) } , \\dots , \\ell _ { K _ { 1 } } ^ { ( 1 ) } )$ and $l ^ { ( 2 ) } = ( \\ell _ { 1 } ^ { ( 2 ) } , \\dots , \\ell _ { K _ { 2 } } ^ { ( 2 ) } )$ . A function $\\boldsymbol { \\Lambda } = \\left( \\Lambda _ { 1 } , \\ldots , \\Lambda _ { K _ { 2 } } \\right)$ is a linear equivariant mapping between $W _ { l ^ { ( 1 ) } }$ and $W _ { l ^ { ( 2 ) } }$ , if and only if there exists a $K _ { 1 } \\times K _ { 2 }$ matrix $M$ with $M _ { i j } = 0$ whenever $\\bar { \\ell } _ { i } ^ { ( 1 ) } \\neq \\ell _ { j } ^ { ( 2 ) }$ , such that ", + "bbox": [ + 173, + 693, + 825, + 744 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/c582f18bbf70a46c752752da5f8838c3a628cadb9e49ebbc6e5a8d85c122efdf.jpg", + "text": "$$\n\\Lambda _ { j } ( V ) = \\sum _ { i = 1 } ^ { K _ { 1 } } M _ { i j } V _ { i }\n$$", + "text_format": "latex", + "bbox": [ + 429, + 750, + 568, + 794 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "where $V = ( V _ { i } ) _ { i = 1 } ^ { K _ { 1 } }$ and $V _ { i } \\in W _ { \\ell _ { i } ^ { ( 1 ) } }$ for all $i = 1 , \\ldots , K _ { 1 }$ ", + "bbox": [ + 173, + 799, + 552, + 820 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Proof idea. This lemma is a simple generalization of Schur’s lemma, a classical tool in representation theory, which asserts that a non-zero linear map between irreducible representations is a scaled multiply of the identity mapping. Lemma 5 was stated in the complex setting in Kondor (2018). While Schur’s lemma, and thus Lemma 5, does not always hold for representations over the reals, we observe here that it holds for real irreducible representations of SO(3) since their dimension is always odd. □ ", + "bbox": [ + 173, + 839, + 825, + 924 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Clebsch-Gordan decomposition of tensor products As any finite-dimensional representation of SO(3) can be decomposed into a direct sum of irreducible representations, this is true for tensor representations as well. In particular, the Clebsch-Gordan coefficients provide an explicit formula for decomposing the tensor product of two irreducible representations $W _ { \\ell _ { 1 } }$ and $W _ { \\ell _ { 2 } }$ into a direct sum of irreducible representations. This decomposition can be easily extended to decompose the tensor product $W _ { l _ { 1 } } \\otimes W _ { l _ { 2 } }$ into a direct sum of irreducible representations, where $l _ { 1 } , l _ { 2 }$ are now vectors. In matrix notation, this means there is a unitary linear equivariant $U ( l _ { 1 } , l _ { 2 } )$ mapping of $W _ { l _ { 1 } } \\otimes W _ { l _ { 2 } }$ onto $W _ { l }$ , where the explicit values of $l = l ( l _ { 1 } , l _ { 2 } )$ and the matrix $U ( l _ { 1 } , l _ { 2 } )$ can be inferred directly from the case where $\\ell _ { 1 }$ and $\\ell _ { 2 }$ are scalars. ", + "bbox": [ + 173, + 103, + 825, + 229 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "By repeatedly taking tensor products and applying Clebsch-Gordan decompositions to the result, TFN and similar architectures can achieve the $D$ -spanning property in a manner analogous to tensor representations, and also enjoy linear universality since they maintain irreducible representations throughout the network. ", + "bbox": [ + 174, + 236, + 825, + 291 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "5.2 TENSOR FIELD NETWORKS ", + "text_level": 1, + "bbox": [ + 176, + 308, + 400, + 321 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "We now describe the basic layers of the TFN architecture (Thomas et al., 2018), which are based on irreducible representations, and suggest an architecture based on these layers which can approximate $G$ -equivariant maps into any representation $W _ { l _ { T } } ^ { n } , l _ { T } \\in \\mathbb { N } _ { + } ^ { * }$ . There are some superficial differences Tbetween our description of TFN and the description in the original paper, for more details see Appendix F. ", + "bbox": [ + 174, + 333, + 825, + 405 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "We note that the universality of TFN also implies the universality of Fuchs et al. (2020), which is a generalization of TFN that enables adding an attention mechanism. Assuming the attention mechanism is not restricted to local neighborhoods, this method is at least as expressive as TFN. ", + "bbox": [ + 174, + 410, + 825, + 453 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "TFNs are composed of three types of layers: (i) Convolution (ii) Self-interaction and (iii) Nonlinearities. In our architecture, we only use the first two layers types, which we will now describe:1. ", + "bbox": [ + 176, + 459, + 823, + 489 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Convolution. Convolutional layers involve taking tensor products of a filter and a feature vector to create a new feature vector, and then decomposing into irreducible representations. Unlike in standard CNN, a filter here depends on the input, and is a function $F : \\bar { \\mathbb { R } ^ { 3 } } W _ { l _ { D } }$ , where $\\boldsymbol { l } _ { D } =$ $[ 0 , 1 , \\ldots , D ] ^ { T }$ . The $\\ell$ -th component of the filter $F ( \\boldsymbol { x } ) = \\left[ F ^ { ( 0 ) } ( x ) , \\ldots , F ^ { ( D ) } ( x ) \\right]$ will be given by ", + "bbox": [ + 173, + 494, + 825, + 554 + ], + "page_idx": 7 + }, + { + "type": "equation", + "img_path": "images/ab43c4063a462588d15cfce3a32fb29492e677c37687cb4a82ed38c439545931.jpg", + "text": "$$\nF _ { m } ^ { \\ell } ( x ) = R ^ { ( \\ell ) } ( \\| x \\| ) Y _ { m } ^ { \\ell } ( \\hat { x } ) , m = - \\ell , \\ldots , \\ell\n$$", + "text_format": "latex", + "bbox": [ + 349, + 559, + 647, + 578 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "where $\\hat { x } = x / \\| x \\|$ if $x \\neq 0$ and $\\hat { x } ~ = ~ 0$ otherwise, $Y _ { m } ^ { \\ell }$ are spherical harmonics, and $R ^ { ( \\ell ) }$ any polynomial of degree $\\leq D$ . In Appendix F we show that these polynomial functions can be replaced by fully connected networks, since the latter can approximate all polynomials uniformly. ", + "bbox": [ + 174, + 583, + 825, + 627 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "The convolution of an input feature $V \\in W _ { l _ { i } } ^ { n }$ and a filter $F$ as defined above, will give an output feature $\\tilde { V } = ( \\tilde { V } _ { a } ) _ { a = 1 } ^ { n } \\in W _ { l _ { 0 } } ^ { n }$ , where $\\boldsymbol { l _ { o } } = \\boldsymbol { l } ( \\boldsymbol { l _ { f } } , \\boldsymbol { l _ { i } } )$ , which is given by ", + "bbox": [ + 174, + 632, + 823, + 666 + ], + "page_idx": 7 + }, + { + "type": "equation", + "img_path": "images/965407aa486d2589a2966a08fbbe90a36b2b8b549c1c3a671e196f7ef48eb5f8.jpg", + "text": "$$\n\\tilde { V } _ { a } ( X , V ) = U ( l _ { f } , l _ { i } ) \\left( \\theta _ { 0 } V _ { a } + \\sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \\otimes V _ { b } \\right) .\n$$", + "text_format": "latex", + "bbox": [ + 310, + 671, + 687, + 714 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "More formally we will think of convolutional layer as functions of the form $f ( X , V ) \\quad = \\quad$ $( X , { \\tilde { V } } ( X , V ) )$ . These functions are defined by a choice of $D$ , a choice of a scalar polynomial $R ^ { ( \\ell ) } , \\ell = 0 , \\dots , D$ , and a choice of the parameter $\\theta _ { 0 } \\in \\mathbb { R }$ in equation 14. We denote the set of all such functions $f$ by $\\mathcal { F } _ { D }$ . ", + "bbox": [ + 174, + 718, + 825, + 780 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Self Interaction layers. Self interaction layers are linear functions from $\\hat { \\Lambda } : W _ { l } ^ { n } \\to W _ { l , r } ^ { n }$ , which are obtained from elementwise application of equivariant linear functions $\\Lambda : W _ { l } \\stackrel { \\cdot } { } W _ { l _ { T } }$ . These linear functions can be specified by a choice of matrix $M$ with the sparsity pattern described in Lemma 5. ", + "bbox": [ + 173, + 786, + 825, + 830 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Activation functions. TFN, as well as other papers, proposed several activation functions. We find that these layers are not necessary for universality and thus we do not define them here. ", + "bbox": [ + 174, + 837, + 823, + 866 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Network architecture. For our universality proof, we suggest a simple architecture which depends on two positive integer parameters $( C , D )$ : For given $D$ , we will define $\\mathcal { F } _ { \\mathrm { f e a t } } ( D )$ as the set of ", + "bbox": [ + 174, + 871, + 823, + 901 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "function obtained by $2 D$ recursive convolutions ", + "bbox": [ + 176, + 103, + 486, + 118 + ], + "page_idx": 8 + }, + { + "type": "equation", + "img_path": "images/adc6540c8611808e70ae0a52bb304fe48558b096f6639f463c82fbedde064dd0.jpg", + "text": "$$\n{ \\mathcal { F } } _ { \\mathrm { f e a t } } ( D ) = \\{ \\pi _ { V } \\circ f ^ { 2 D } \\circ \\ldots f ^ { 2 } \\circ f ^ { 1 } \\circ \\operatorname { e x t } ( X ) | f ^ { j } \\in { \\mathcal { F } } _ { D } \\} ,\n$$", + "text_format": "latex", + "bbox": [ + 303, + 122, + 692, + 141 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "where ext and πV are defined as in equation 10. The output of a function in Ffeat(D) is in W nl(D), for some $\\boldsymbol { l } ( D )$ which depends on $D$ . We then define $\\mathcal { F } _ { \\mathrm { p o o l } } ( D )$ to be the self-interaction layers which map $W _ { l ( D ) } ^ { n }$ to $W _ { l _ { T } } ^ { n }$ . This choice of $\\mathcal { F } _ { \\mathrm { f e a t } } ( D )$ and $\\mathcal { F } _ { \\mathrm { p o o l } } ( D )$ , together with a choice of the number of channels $C$ T, defines the final network architecture $\\mathcal { F } _ { C , D } ^ { \\mathrm { T F N } } = \\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } ( D ) , \\mathcal { F } _ { \\mathrm { p o o l } } ( D ) )$ as in equation 2. In the appendix we prove the universality of TFN: ", + "bbox": [ + 173, + 143, + 825, + 223 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Theorem 2. For all $n \\in \\mathbb { N } , l _ { T } \\in \\mathbb { N } _ { + } ^ { * }$ ", + "bbox": [ + 174, + 224, + 421, + 241 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "1. For $D \\in { \\mathbb { N } } _ { + }$ , every $G$ -equivariant polynomial $p : \\mathbb { R } ^ { 3 \\times n } W _ { T } ^ { n }$ of degree $D$ is in $\\mathcal { F } _ { C ( D ) , D } ^ { \\mathrm { T F N } }$ 2. Every continuoby functions in $G$ t function can be approximated uniformly on compact sets $\\cup _ { D \\in \\mathbb { N } _ { + } } \\mathcal { F } _ { C ( D ) , D } ^ { \\mathrm { T F N } }$ ", + "bbox": [ + 207, + 251, + 825, + 311 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "As discussed previously, the linear universality of $\\mathcal { F } _ { \\mathrm { p o o l } }$ is guaranteed. Thus proving Theorem 2 amounts to showing that $\\mathcal { F } _ { \\mathrm { f e a t } } ( D )$ is $D$ -spanning. This is done using the sufficient condition for $D$ -spanning defined in Lemma 4. ", + "bbox": [ + 178, + 320, + 825, + 363 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Proof idea. The proof is rather technical and involved. A useful observation (see Dai & Xu (2013)) used in the proof is that the filters of orders $\\ell = 0 , 1 , \\ldots , D$ , defined in equation 13, span all polynomial functions of degree $D$ on $\\mathbb { R } ^ { 3 }$ . This observation is used to show that all functions in $\\mathcal { Q } _ { D }$ can be expressed by $\\mathcal { F } _ { \\mathrm { f e a t } } ( D )$ and so $\\mathcal { F } _ { \\mathrm { f e a t } }$ is $D$ -spanning, as stated in Lemma 2. □ ", + "bbox": [ + 174, + 377, + 825, + 435 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Alternative architecture The complexity of the TFN network used to construct $G$ -equivariant polynomials of degree $D$ , can be reduced using a simple modifications of the convolutional layer in equation 14: We add two parameters $\\theta _ { 1 } , \\theta _ { 2 } \\in \\mathbb { R }$ to the convolutional layer, which is now defined as: ", + "bbox": [ + 174, + 449, + 825, + 492 + ], + "page_idx": 8 + }, + { + "type": "equation", + "img_path": "images/a62cee15c1d813f1fad5e4a74cd43f39c9d1db3184a1e47f48c5df6dc91c07d9.jpg", + "text": "$$\n\\tilde { V } _ { a } ( X , V ) = U ( l _ { f } , l _ { i } ) \\left( \\theta _ { 1 } \\sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \\otimes V _ { b } + \\theta _ { 2 } \\sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \\otimes V _ { a } \\right) .\n$$", + "text_format": "latex", + "bbox": [ + 238, + 496, + 759, + 539 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "With this simple change, we can show that $\\mathcal { F } _ { \\mathrm { f e a t } } ( D )$ is $D$ -spanning even if we only take filters of order 0 and 1 throughout the network. This is shown in Appendix E. ", + "bbox": [ + 173, + 549, + 826, + 579 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "6 CONCLUSION ", + "text_level": 1, + "bbox": [ + 174, + 598, + 320, + 614 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "In this paper, we have presented a new framework for proving the universality of $G$ -equivariant point cloud networks. We used this framework for proving the universality of the TFN model Thomas et al. (2018); Fuchs et al. (2020), and for devising two additional novel simple universal architectures. In the future we hope to extend these simple constructions to operational $G$ -equivariant networks with universality guarantees and competitive practical performance. ", + "bbox": [ + 173, + 628, + 825, + 700 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Our universal architectures do not require activation functions, and use a single self-interaction layer. In Appendix H we present an experiment indicating that the performance of TFN is not significantly altered by these simplifications. Our architectures also require high order representations, and our experiments show that using increasingly high order representations does indeed improve performance. To date, practical TFN implementation included a relatively small amount of layers, and did not use very high order representations. We believe our theoretical results will inspire interest in stable implementation of larger architectures. On the other hand, an interesting open problem is understanding whether universality can be achieved using only low-dimensional representations. ", + "bbox": [ + 174, + 705, + 825, + 818 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Finally, we believe that the framework we developed here will be useful for proving the universality of other $G$ -equivariant models for point cloud networks, and other related equivariant models. We note that large parts of our discussion can be easily generalized to symmetry groups of the form $G = \\mathbb { R } ^ { d } \\rtimes \\breve { H } \\times \\bar { S } _ { n }$ acting on $\\mathbb { R } ^ { d \\times n }$ , where $H$ can be any compact topological group. ", + "bbox": [ + 173, + 824, + 825, + 881 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Acknowledgments The authors would like to thank Fabian B. Fuchs for making code available and Taco Cohen for helpful discussion. N.D. is supported by THEORINET Simons award 814643. ", + "bbox": [ + 173, + 895, + 823, + 924 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "REFERENCES ", + "text_level": 1, + "bbox": [ + 174, + 102, + 287, + 118 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Matan Atzmon, Haggai Maron, and Yaron Lipman. Point convolutional neural networks by extension operators. arXiv preprint arXiv:1803.10091, 2018. ", + "bbox": [ + 173, + 126, + 823, + 155 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Alexander Bogatskiy, Brandon Anderson, Jan T Offermann, Marwah Roussi, David W Miller, and Risi Kondor. Lorentz group equivariant neural network for particle physics. arXiv preprint arXiv:2006.04780, 2020. ", + "bbox": [ + 173, + 164, + 821, + 207 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Taco S Cohen, Mario Geiger, Jonas Kohler, and Max Welling. Spherical cnns. ¨ arXiv preprint arXiv:1801.10130, 2018. ", + "bbox": [ + 171, + 215, + 823, + 244 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Feng Dai and Yuan Xu. Approximation theory and harmonic analysis on spheres and balls, volume 23. Springer, 2013. ", + "bbox": [ + 171, + 253, + 823, + 284 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Carlos Esteves, Christine Allen-Blanchette, Ameesh Makadia, and Kostas Daniilidis. Learning so (3) equivariant representations with spherical cnns. In Proceedings of the European Conference on Computer Vision (ECCV), pp. 52–68, 2018. ", + "bbox": [ + 176, + 291, + 823, + 335 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Fabian B Fuchs, Daniel E Worrall, Volker Fischer, and Max Welling. Se (3)-transformers: 3d rototranslation equivariant attention networks. arXiv preprint arXiv:2006.10503, 2020. ", + "bbox": [ + 173, + 343, + 823, + 373 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "William Fulton and Joe Harris. Representation theory: a first course, volume 129. Springer Science & Business Media, 2013. ", + "bbox": [ + 173, + 382, + 821, + 411 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Justin Gilmer, Samuel S Schoenholz, Patrick F Riley, Oriol Vinyals, and George E Dahl. Neural message passing for quantum chemistry. arXiv preprint arXiv:1704.01212, 2017. ", + "bbox": [ + 173, + 420, + 820, + 450 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Yulan Guo, Hanyun Wang, Qingyong Hu, Hao Liu, Li Liu, and Mohammed Bennamoun. Deep learning for 3d point clouds: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020. ", + "bbox": [ + 173, + 458, + 823, + 501 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Nicolas Keriven and Gabriel Peyre. Universal invariant and equivariant graph neural networks. ´ CoRR, abs/1905.04943, 2019. URL http://arxiv.org/abs/1905.04943. ", + "bbox": [ + 173, + 510, + 821, + 540 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Risi Kondor. N-body networks: a covariant hierarchical neural network architecture for learning atomic potentials. arXiv preprint arXiv:1803.01588, 2018. ", + "bbox": [ + 173, + 549, + 823, + 578 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Risi Kondor, Zhen Lin, and Shubhendu Trivedi. Clebsch–gordan nets: a fully fourier space spherical convolutional neural network. In Advances in Neural Information Processing Systems, pp. 10117– 10126, 2018a. ", + "bbox": [ + 173, + 585, + 823, + 628 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Risi Kondor, Hy Truong Son, Horace Pan, Brandon Anderson, and Shubhendu Trivedi. Covariant compositional networks for learning graphs. arXiv preprint arXiv:1801.02144, 2018b. ", + "bbox": [ + 171, + 638, + 823, + 667 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Hanspeter Kraft and Claudio Procesi. Classical invariant theory, a primer. Lecture Notes, Version, 2000. ", + "bbox": [ + 171, + 676, + 823, + 707 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Jiaxin Li, Yingcai Bi, and Gim Hee Lee. Discrete rotation equivariance for point cloud recognition. In 2019 International Conference on Robotics and Automation (ICRA), pp. 7269–7275. IEEE, 2019. ", + "bbox": [ + 174, + 714, + 823, + 757 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Takanori Maehara and Hoang NT. A simple proof of the universality of invariant/equivariant graph neural networks, 2019. ", + "bbox": [ + 173, + 766, + 823, + 796 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Haggai Maron, Heli Ben-Hamu, Hadar Serviansky, and Yaron Lipman. Provably powerful graph networks. arXiv preprint arXiv:1905.11136, 2019a. ", + "bbox": [ + 173, + 805, + 823, + 834 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Haggai Maron, Heli Ben-Hamu, Nadav Shamir, and Yaron Lipman. Invariant and equivariant graph networks. In International Conference on Learning Representations, 2019b. URL https: //openreview.net/forum?id=Syx72jC9tm. ", + "bbox": [ + 176, + 843, + 821, + 887 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Haggai Maron, Ethan Fetaya, Nimrod Segol, and Yaron Lipman. On the universality of invariant networks. In International conference on machine learning, 2019c. ", + "bbox": [ + 173, + 895, + 820, + 924 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Haggai Maron, Or Litany, Gal Chechik, and Ethan Fetaya. On learning sets of symmetric elements. arXiv preprint arXiv:2002.08599, 2020. ", + "bbox": [ + 171, + 103, + 821, + 132 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Christopher Morris, Martin Ritzert, Matthias Fey, William L Hamilton, Jan Eric Lenssen, Gaurav Rattan, and Martin Grohe. Weisfeiler and leman go neural: Higher-order graph neural networks. arXiv preprint arXiv:1810.02244, 2018. ", + "bbox": [ + 174, + 140, + 823, + 184 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Adam Paszke, Sam Gross, Soumith Chintala, Gregory Chanan, Edward Yang, Zachary DeVito, Zeming Lin, Alban Desmaison, Luca Antiga, and Adam Lerer. Automatic differentiation in pytorch. 2017. ", + "bbox": [ + 174, + 191, + 823, + 234 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Adrien Poulenard, Marie-Julie Rakotosaona, Yann Ponty, and Maks Ovsjanikov. Effective rotationinvariant point cnn with spherical harmonics kernels. In 2019 International Conference on 3D Vision (3DV), pp. 47–56. IEEE, 2019. ", + "bbox": [ + 174, + 243, + 825, + 286 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Charles R Qi, Hao Su, Kaichun Mo, and Leonidas J Guibas. Pointnet: Deep learning on point sets for 3d classification and segmentation. Proc. Computer Vision and Pattern Recognition (CVPR), IEEE, 1(2):4, 2017a. ", + "bbox": [ + 173, + 295, + 823, + 338 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Charles Ruizhongtai Qi, Li Yi, Hao Su, and Leonidas J Guibas. Pointne $^ { + + }$ : Deep hierarchical feature learning on point sets in a metric space. In Advances in neural information processing systems, pp. 5099–5108, 2017b. ", + "bbox": [ + 173, + 347, + 823, + 390 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Raghunathan Ramakrishnan, Pavlo O Dral, Matthias Rupp, and O Anatole Von Lilienfeld. Quantum chemistry structures and properties of 134 kilo molecules. Scientific data, 1(1):1–7, 2014. ", + "bbox": [ + 173, + 398, + 823, + 428 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Siamak Ravanbakhsh. Universal equivariant multilayer perceptrons. arXiv preprint arXiv:2002.02912, 2020. ", + "bbox": [ + 171, + 435, + 825, + 465 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Nimrod Segol and Yaron Lipman. On universal equivariant set networks. arXiv preprint arXiv:1910.02421, 2019. ", + "bbox": [ + 171, + 473, + 825, + 502 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Hadar Serviansky, Nimrod Segol, Jonathan Shlomi, Kyle Cranmer, Eilam Gross, Haggai Maron, and Yaron Lipman. Set2graph: Learning graphs from sets. arXiv preprint arXiv:2002.08772, 2020. ", + "bbox": [ + 173, + 511, + 823, + 541 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Nathaniel Thomas, Tess Smidt, Steven Kearnes, Lusann Yang, Li Li, Kai Kohlhoff, and Patrick Riley. Tensor field networks: Rotation-and translation-equivariant neural networks for 3d point clouds. arXiv preprint arXiv:1802.08219, 2018. ", + "bbox": [ + 174, + 547, + 823, + 592 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Minjie Wang, Lingfan Yu, Da Zheng, Quan Gan, Yu Gai, Zihao Ye, Mufei Li, Jinjing Zhou, Qi Huang, Chao Ma, et al. Deep graph library: Towards efficient and scalable deep learning on graphs. arXiv preprint arXiv:1909.01315, 2019a. ", + "bbox": [ + 176, + 599, + 823, + 643 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Yue Wang, Yongbin Sun, Ziwei Liu, Sanjay E Sarma, Michael M Bronstein, and Justin M Solomon. Dynamic graph cnn for learning on point clouds. Acm Transactions On Graphics (tog), 38(5): 1–12, 2019b. ", + "bbox": [ + 174, + 651, + 823, + 694 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Maurice Weiler, Mario Geiger, Max Welling, Wouter Boomsma, and Taco Cohen. 3D Steerable CNNs: Learning Rotationally Equivariant Features in Volumetric Data. 2018. URL http: //arxiv.org/abs/1807.02547. ", + "bbox": [ + 174, + 703, + 823, + 746 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Daniel Worrall and Gabriel Brostow. Cubenet: Equivariance to 3d rotation and translation. In Proceedings of the European Conference on Computer Vision (ECCV), pp. 567–584, 2018. ", + "bbox": [ + 171, + 753, + 823, + 784 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural networks? In International Conference on Learning Representations, 2019. URL https: //openreview.net/forum?id $=$ ryGs6iA5Km. ", + "bbox": [ + 176, + 791, + 823, + 835 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Dmitry Yarotsky. Universal approximations of invariant maps by neural networks. arXiv preprint arXiv:1804.10306, 2018. ", + "bbox": [ + 169, + 843, + 823, + 872 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Manzil Zaheer, Satwik Kottur, Siamak Ravanbakhsh, Barnabas Poczos, Ruslan R Salakhutdinov, and Alexander J Smola. Deep sets. In Advances in neural information processing systems, pp. 3391–3401, 2017. ", + "bbox": [ + 174, + 881, + 825, + 924 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Yongheng Zhao, Tolga Birdal, Jan Eric Lenssen, Emanuele Menegatti, Leonidas Guibas, and Federico Tombari. Quaternion equivariant capsule networks for 3d point clouds. arXiv preprint arXiv:1912.12098, 2019. ", + "bbox": [ + 174, + 103, + 823, + 146 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "A NOTATION ", + "text_level": 1, + "bbox": [ + 176, + 170, + 299, + 186 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "We introduce some notation for the proofs in the appendices. We use the shortened notation $\\bar { X } =$ $\\begin{array} { r } { X - \\frac { 1 } { n } X \\mathbb { 1 } _ { n } \\mathbb { 1 } _ { n } ^ { T } } \\end{array}$ and denote the columns of $\\bar { X }$ by $\\left( \\hat { x } _ { 1 } , \\ldots , \\hat { x } _ { n } \\right)$ . We denote ", + "bbox": [ + 173, + 200, + 825, + 231 + ], + "page_idx": 11 + }, + { + "type": "equation", + "img_path": "images/ea97350b1cbf196b3aa1c4c70d02d5268f172baf7f348c41df90846546deb63e.jpg", + "text": "$$\n\\Sigma _ { T } = \\{ \\vec { r } \\in \\mathbb { N } _ { + } ^ { * } | \\| \\vec { r } \\| _ { 1 } = T \\}\n$$", + "text_format": "latex", + "bbox": [ + 405, + 233, + 593, + 252 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "B PROOFS FOR SECTION 3 ", + "text_level": 1, + "bbox": [ + 176, + 268, + 410, + 285 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "B.1 $G$ -EQUIVARIANT POLYNOMIALS ARE DENSE ", + "text_level": 1, + "bbox": [ + 174, + 300, + 522, + 314 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "A first step in proving denseness of $G$ -equivariance polynomials, and in the proof used in the next subsection is the following simple lemma, which shows that translation invariance can be dealt with simply by centralizing the point cloud. ", + "bbox": [ + 174, + 325, + 826, + 368 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "In the following, $\\rho _ { W _ { T } }$ is some representation of $\\mathrm { S O ( 3 ) }$ on a finite dimensional real vector space $W _ { T }$ this induces an action $\\rho _ { W _ { T } \\times S _ { n } }$ of $\\mathrm { S O } ( 3 ) \\times S _ { n }$ on $W _ { T } ^ { n }$ by ", + "bbox": [ + 173, + 375, + 821, + 405 + ], + "page_idx": 11 + }, + { + "type": "equation", + "img_path": "images/ed30f3a425d843a9f90649371c9135255a4219739b1c0e989ac811c1a575c53e.jpg", + "text": "$$\n\\rho _ { W _ { T } \\times S _ { n } } ( R , P ) ( Y ) = \\rho _ { W _ { T } } ( R ) Y P ^ { T }\n$$", + "text_format": "latex", + "bbox": [ + 377, + 409, + 620, + 428 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "This is also the action of $G$ which we consider, $\\rho _ { G } ~ = ~ \\rho _ { \\underline { { W } } _ { T } \\times S _ { n } }$ , where we have invariance with respect to the translation coordinate. The action of $G$ on $\\mathbb { R } ^ { 3 \\times n }$ is defined in equation 1. ", + "bbox": [ + 173, + 431, + 823, + 458 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Lemma B.1. A function $f : \\mathbb { R } ^ { 3 \\times n } W _ { T } ^ { n }$ is $G$ -equivariant, if and only if there exists a function $h$ which is equivariant with respect to the action of $\\mathrm { S O } ( 3 ) \\times S _ { n }$ on $\\mathbb { R } ^ { 3 \\times n }$ , and ", + "bbox": [ + 173, + 460, + 823, + 491 + ], + "page_idx": 11 + }, + { + "type": "equation", + "img_path": "images/24acaf8abdd52513d14c27352e968eb1385247ccf18f19540ce3a97abb1b7030.jpg", + "text": "$$\nf ( X ) = h ( X - \\frac 1 n X 1 _ { n } 1 _ { n } ^ { T } )\n$$", + "text_format": "latex", + "bbox": [ + 406, + 494, + 591, + 525 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Proof. Recall that $G$ -equivariance means $\\mathrm { S O } ( 3 ) \\times S _ { n }$ equivariance and translation invariance. Thus if $f$ is $G$ -equivariant then equation 16 holds with $h = f$ . ", + "bbox": [ + 173, + 537, + 825, + 566 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "On the other hand, if $f$ satisfies equation 16 then we claim it is $G$ -equivariant. Indeed, for all $( t , R , P ) \\in \\mathbb { R } ^ { d } \\rtimes \\mathrm { S O ( 3 ) } \\times S _ { n }$ , since $P ^ { T } 1 _ { n } 1 _ { n } ^ { T } = 1 _ { n } 1 _ { n } ^ { T } = 1 _ { n } 1 _ { n } ^ { T } P ^ { T }$ , ", + "bbox": [ + 176, + 573, + 823, + 603 + ], + "page_idx": 11 + }, + { + "type": "equation", + "img_path": "images/bfe1ba1553323a3c64485a603dd3bea449a778962e3611934c34d05529273d92.jpg", + "text": "$$\n\\begin{array} { l } { f \\left( \\rho _ { G } ( t , R , P ) ( X ) \\right) = f ( R ( X + t \\boldsymbol { 1 } _ { n } ) P ^ { T } ) = h ( R ( X + t \\boldsymbol { 1 } _ { n } ) P ^ { T } - \\displaystyle \\frac { 1 } { n } R ( X + t \\boldsymbol { 1 } ) P ^ { T } \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) } \\\\ { \\displaystyle \\quad \\quad = h ( R ( X - \\frac { 1 } { n } X \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) P ^ { T } ) = h \\left( \\rho _ { \\mathbb { R } ^ { 3 } \\times S _ { n } } ( R , P ) ( X - \\frac { 1 } { n } X \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) \\right) } \\\\ { \\displaystyle \\quad = \\rho _ { W _ { T } \\times S _ { n } } ( R , P ) h \\left( X - \\frac { 1 } { n } X \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } \\right) } \\\\ { \\displaystyle \\quad = \\rho _ { G } ( t , R , P ) f ( X ) . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 194, + 607, + 803, + 726 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "We now prove denseness of $G$ -equivariant polynomials in the space of $G$ -invariant continuous functions (Lemma 1). ", + "bbox": [ + 174, + 765, + 823, + 792 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Lemma 1. Any continuous $G$ -equivariant function in $\\mathcal { C } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , W _ { T } ^ { n } )$ can be approximated uniformly on compact sets by $G$ -equivariant polynomials in $\\mathcal { P } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , \\dot { W } _ { T } ^ { n } )$ . ", + "bbox": [ + 173, + 795, + 820, + 825 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Proof of Lemma 1. Let $K ~ \\subseteq ~ \\mathbb { R } ^ { 3 \\times n }$ be a compact set. We need to show that continuous $G$ - \nequivariant functions can be approximated undenote the compact set which is the image of By Lemma B.1, it is sufficient to show that eve $K$ y ene $G$ -equivarianlizing map ivariant con $K _ { 0 }$ $K$ $\\begin{array} { r } { \\dot { X } \\stackrel { \\bullet } { \\mapsto } X - \\frac { 1 } { n } X 1 _ { n } 1 _ { n } ^ { T } } \\end{array}$ $\\mathrm { S O } ( 3 ) \\times S _ { n }$ $f$ \nbe approximated uniformly on $K _ { 0 }$ by a sequence of $\\mathrm { S O } ( 3 ) \\times S _ { n }$ equivariant polynomials $p _ { k }$ . The \nargument is now concluded by the following general lemma: ", + "bbox": [ + 173, + 838, + 825, + 924 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Lemma B.2. Let $G$ be a compact group, Let $\\rho _ { 1 }$ and $\\rho _ { 2 }$ be continuous2representations of $G$ on the Euclidean spaces $W _ { 1 }$ and $W _ { 2 }$ . Let $K \\subseteq W _ { 1 }$ be a compact set. Then every equivariant function $f : W _ { 1 } \\to W _ { 2 }$ can be approximated uniformly on $K$ by a sequence of equivariant polynomials $p _ { k } : W _ { 1 } \\mapsto W _ { 2 }$ . ", + "bbox": [ + 173, + 102, + 825, + 160 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Let $\\mu$ be the Haar probability measure associated with the compact group $G$ . Let $K _ { 1 }$ denote the compact set obtained as an image of the compact set $G \\times K$ under the continuous mapping ", + "bbox": [ + 173, + 174, + 825, + 203 + ], + "page_idx": 12 + }, + { + "type": "equation", + "img_path": "images/5f4461c64095fc77cdd890372dae6fd44c4eb438541773c5e7fd11abefadc925.jpg", + "text": "$$\n( g , X ) \\mapsto \\rho _ { 1 } ( g ) X .\n$$", + "text_format": "latex", + "bbox": [ + 436, + 207, + 562, + 223 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Using the Stone-Weierstrass theorem, let $p _ { k }$ be a sequence of (not necessarily equivariant) polynomials which approximate $f$ uniformly on $K _ { 1 }$ . Every degree $D$ polynomial $p : W _ { 1 } \\to W _ { 2 }$ induces a $G$ -equivariant function ", + "bbox": [ + 173, + 227, + 823, + 268 + ], + "page_idx": 12 + }, + { + "type": "equation", + "img_path": "images/15eb986de5f78bc842430871066793bfc7ffb96f9d9b47216630c87a00f73dce.jpg", + "text": "$$\n\\langle p \\rangle ( X ) = \\int _ { G } \\rho _ { 2 } ( g ^ { - 1 } ) p ( \\rho _ { 1 } ( g ) X ) d \\mu ( g ) .\n$$", + "text_format": "latex", + "bbox": [ + 366, + 267, + 632, + 300 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "This functformly on $\\langle p \\rangle$ is a degree “Riemann $D$ polynomial as ms” of the form $\\langle p \\rangle$ e approximated uwhich are degree $K _ { 1 }$ $\\begin{array} { r } { \\sum _ { j = 1 } ^ { N } w _ { j } \\rho _ { 2 } ( g _ { j } ^ { - 1 } ) p ( \\rho _ { 1 } ( g _ { j } ) X ) } \\end{array}$ $D$ polynomials, and because degree polynomials are closed in $C ( K _ { 1 } )$ . ", + "bbox": [ + 173, + 301, + 825, + 349 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Now for all $X \\ \\in \\ K _ { 1 }$ , continuity of the function $g \\mapsto \\rho _ { 2 } ( g ^ { - 1 } )$ implies that the operator norm of $\\rho _ { 2 } ( g ^ { - 1 } )$ is bounded uniformly by some constant $N > 0$ , and so ", + "bbox": [ + 171, + 354, + 823, + 385 + ], + "page_idx": 12 + }, + { + "type": "equation", + "img_path": "images/667bbdb38be9394d312e35d3e43bdc61e9e0b663f47435897f5af36d3106aead.jpg", + "text": "$$\n\\begin{array} { l } { \\displaystyle { | \\langle p _ { k } \\rangle ( X ) - f ( X ) | = \\left| \\int _ { G } \\rho _ { 2 } ( g ^ { - 1 } ) p _ { k } ( \\rho _ { 1 } ( g ) X ) - \\rho _ { 2 } ( g ^ { - 1 } ) f ( \\rho _ { 1 } ( g ) X ) d \\mu ( g ) \\right| } } \\\\ { \\displaystyle { \\phantom { \\rho _ { k } ( \\rho _ { 1 } ) ( } = \\left| \\int _ { G } \\rho _ { 2 } ( g ^ { - 1 } ) \\left[ p _ { k } ( \\rho _ { 1 } ( g ) X ) - f ( \\rho _ { 1 } ( g ) X ) \\right] d \\mu ( g ) \\right| \\leq N \\| f - p _ { k } \\| _ { C ( K _ { 1 } ) } \\to 0 } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 181, + 388, + 818, + 459 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "B.2 PROOF OF THEOREM 1 ", + "text_level": 1, + "bbox": [ + 174, + 491, + 374, + 506 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Theorem 1. If $\\mathcal { F } _ { \\mathrm { f e a t } }$ is $D$ -spanning and $\\mathcal { F } _ { \\mathrm { p o o l } }$ is linearly universal, then there exists some $C ( D ) \\in$ $\\mathbb { N }$ such that for all $C \\geq C ( D )$ the function space $\\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } , \\mathcal { F } _ { \\mathrm { p o o l } } )$ contains all $G$ -equivariant polynomials of degree $\\le D$ . ", + "bbox": [ + 173, + 517, + 825, + 560 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Proof. By the $D$ -spanning assumption, there exist $f _ { 1 } , \\dots , f _ { K } \\in \\mathcal { F } _ { \\mathrm { f e a t } }$ such that any vector valued polynomial $p : \\mathbb { R } ^ { 3 \\bar { \\times } n } \\bar { \\mathbb { R } } ^ { n }$ invariant to translations and equivariant to permutations is of the form ", + "bbox": [ + 173, + 575, + 823, + 604 + ], + "page_idx": 12 + }, + { + "type": "equation", + "img_path": "images/2463d691fa6fda41539a973d733e2689860ca5ec0c6ddd0d49de84349dd24050.jpg", + "text": "$$\np ( X ) = \\sum _ { k = 1 } ^ { K } \\hat { \\Lambda } _ { k } ( f _ { k } ( X ) ) ,\n$$", + "text_format": "latex", + "bbox": [ + 415, + 608, + 581, + 651 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "where $\\Lambda _ { k }$ are linear functions to $\\mathbb { R }$ . If $p$ is a matrix valued polynomial mapping $\\mathbb { R } ^ { 3 \\times n }$ to $W _ { T } ^ { n } = \\mathbb { R } ^ { t \\times n }$ , which is invariant to translations and equivariant to permutations, then it is of the form $p = ( p _ { i j } ) _ { i \\in [ t ] , j \\in [ n ] }$ , and each $p _ { i } = ( p _ { i j } ) _ { j \\in [ n ] }$ is itself invariant to translations and permutation equivariant. It follows that matrix valued $p$ can also be written in the form equation 17, the only difference being that the image of the linear functions $\\Lambda _ { k }$ is now $\\mathbb { R } ^ { t }$ . ", + "bbox": [ + 173, + 655, + 825, + 726 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Now let $p : \\mathbb { R } ^ { 3 \\times n } W _ { T } ^ { n }$ be a $G$ -equivariant polynomial of degree $\\le D$ . It remains to show that we can choose $\\Lambda _ { k }$ to be SO(3) equivariant. We do this by a symmetrization argument: denote the Haar probability measure on $\\mathrm { S O ( 3 ) }$ by $\\nu$ , and the action of SO(3) on $W _ { \\mathrm { f e a t } }$ and $W _ { T }$ by $\\rho _ { 1 }$ and $\\rho _ { 2 }$ respectively Denote $p = ( p _ { j } ) _ { j = 1 } ^ { n }$ and $f _ { k } = ( f _ { k } ^ { j } ) _ { j = 1 } ^ { n }$ . For every $j = 1 , \\dotsc , n$ , we use the SO(3) equivariance of $p _ { j }$ and $f _ { k } ^ { j }$ to obtain ", + "bbox": [ + 173, + 731, + 823, + 810 + ], + "page_idx": 12 + }, + { + "type": "equation", + "img_path": "images/4bc18a56fd3a2e5947357c447088fa0a0015cc7008f44aeacfb7112cf513128c.jpg", + "text": "$$\n\\begin{array} { l } { { \\displaystyle p _ { j } ( X ) = \\int _ { \\mathrm { S O } ( 3 ) } \\rho _ { 2 } ( R ^ { - 1 } ) \\circ p _ { j } ( R X ) d \\nu ( R ) = \\sum _ { k = 1 } ^ { K } \\int _ { \\mathrm { S O } ( 3 ) } \\rho _ { 2 } ( R ^ { - 1 } ) \\circ \\Lambda _ { k } \\circ f _ { j } ^ { k } ( R X ) d \\nu ( R ) } } \\\\ { { \\displaystyle \\qquad = \\sum _ { k = 1 } ^ { K } \\int _ { \\mathrm { S O } ( 3 ) } \\rho _ { 2 } ( R ^ { - 1 } ) \\circ \\Lambda _ { k } \\left( \\rho _ { 1 } ( R ) \\circ f _ { k } ^ { j } ( X ) \\right) d \\nu ( R ) = \\sum _ { k = 1 } ^ { K } \\tilde { \\Lambda } _ { k } \\circ f _ { k } ^ { j } ( X ) , } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 202, + 814, + 795, + 902 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "2By this we mean that the maps $( g , X ) \\mapsto \\rho _ { j } ( g ) X , j = 1 , 2$ are jointly continuous ", + "bbox": [ + 191, + 909, + 684, + 925 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "where $\\tilde { \\Lambda } _ { k }$ stands for the equivariant linear functional from $W _ { \\mathrm { f e a t } }$ to $W _ { T }$ , defined for $w \\in W _ { \\mathrm { f e a t } }$ by ", + "bbox": [ + 176, + 102, + 815, + 119 + ], + "page_idx": 13 + }, + { + "type": "equation", + "img_path": "images/7ed28a552a728ceae7ebf55ba5d40395d91da3c7d42661bf2dedb6f9d6431448.jpg", + "text": "$$\n\\tilde { \\Lambda } _ { k } ( w ) = \\int _ { \\mathrm { S O ( 3 ) } } \\rho _ { 2 } ( R ^ { - 1 } ) \\circ \\Lambda _ { k } \\left( \\rho _ { 1 } ( R ) w \\right) d \\nu ( R ) .\n$$", + "text_format": "latex", + "bbox": [ + 334, + 122, + 663, + 156 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Thus we have shown that $p$ is in $\\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } , \\mathcal { F } _ { \\mathrm { p o o l } } )$ for $C = K$ , as required. ", + "bbox": [ + 176, + 161, + 663, + 178 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "C PROOFS FOR SECTION 4 ", + "text_level": 1, + "bbox": [ + 174, + 195, + 411, + 213 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "We prove Lemma 2 ", + "bbox": [ + 174, + 227, + 305, + 241 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Lemma 2. For every $D \\in { \\mathbb { N } } _ { + }$ , the set $\\mathcal { Q } _ { D }$ is $D$ -spanning. ", + "bbox": [ + 173, + 244, + 557, + 261 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Proof. It is known (Segol & Lipman, 2019) (Theorem 2) that polynomials $p : \\mathbb { R } ^ { 3 \\times n } \\mathbb { R } ^ { n }$ which are $S _ { n }$ -equivariant, are spanned by polynomials of the form $p _ { \\vec { \\alpha } } = \\stackrel { \\cdot } { ( } p _ { \\vec { \\alpha } } ^ { j } ) _ { j = 1 } ^ { n }$ , defined as ", + "bbox": [ + 171, + 273, + 823, + 306 + ], + "page_idx": 13 + }, + { + "type": "equation", + "img_path": "images/71ac45911fef7a6e9f0f1f15d080c1024f7fd811c2059571fc556dfbcd4b1c9b.jpg", + "text": "$$\np _ { \\vec { \\alpha } } ^ { j } ( X ) = \\sum _ { i _ { 2 } , \\ldots , i _ { K } = 1 } ^ { n } x _ { j } ^ { \\alpha _ { 1 } } x _ { i _ { 2 } } ^ { \\alpha _ { 2 } } \\cdot \\cdot \\cdot x _ { i _ { k } } ^ { \\alpha _ { k } }\n$$", + "text_format": "latex", + "bbox": [ + 380, + 310, + 616, + 354 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "where $\\vec { \\alpha } \\ : = \\ : ( \\alpha _ { 1 } , \\ldots , \\alpha _ { K } )$ and each $\\alpha _ { k } ~ \\in ~ \\mathbb { N } _ { + } ^ { 3 }$ is a multi-index. It follows that $S _ { n }$ -equivariant polynomials of degree $\\leq D$ are spanned by polynomials of the form $p _ { \\vec { \\alpha } } ^ { j }$ where $\\begin{array} { r } { \\sum _ { k = 1 } ^ { K } \\left| \\alpha _ { k } \\right| \\leq D } \\end{array}$ . Denoting $r _ { k } = | \\alpha _ { k } | , k = 1 , \\ldots K$ , the sum of all $r _ { k }$ by $T$ , and $\\vec { r } = ( r _ { k } ) _ { k = 1 } ^ { K }$ , we see that there exists a linear functional $\\Lambda _ { \\vec { \\alpha } , \\vec { r } } : \\mathcal { T } _ { T } \\to \\mathbb { R }$ such that ", + "bbox": [ + 173, + 358, + 825, + 422 + ], + "page_idx": 13 + }, + { + "type": "equation", + "img_path": "images/20b00c92a9be1f81a246671a464bcc308502fd9e032dbed0308d3ee0ca025373.jpg", + "text": "$$\np _ { \\vec { \\alpha } } ^ { j } ( X ) = \\Lambda _ { \\vec { \\alpha } , \\vec { r } ^ { } ^ { \\mathrm { ~ O ~ } } } Q _ { j } ^ { \\vec { r } } ( X )\n$$", + "text_format": "latex", + "bbox": [ + 415, + 426, + 581, + 448 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "where we recall that $Q ^ { \\vec { r } } = \\left( Q _ { j } ^ { ( \\vec { r } ) } ( X ) \\right) _ { j = 1 } ^ { n }$ is defined in equation 6 as ", + "bbox": [ + 174, + 452, + 629, + 479 + ], + "page_idx": 13 + }, + { + "type": "equation", + "img_path": "images/c77b67b34053601544f6bcf7516644f3b241b0b16af382236e465577f65a6f52.jpg", + "text": "$$\nQ _ { j } ^ { ( \\vec { r } ) } ( X ) = \\sum _ { i _ { 2 } , \\ldots , i _ { K } = 1 } ^ { n } x _ { j } ^ { \\otimes r _ { 1 } } \\otimes x _ { i _ { 2 } } ^ { \\otimes r _ { 2 } } \\otimes x _ { i _ { 3 } } ^ { \\otimes r _ { 3 } } \\otimes \\ldots \\otimes x _ { i _ { K } } ^ { \\otimes r _ { K } } .\n$$", + "text_format": "latex", + "bbox": [ + 303, + 483, + 692, + 527 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Thus polynomials $p = ( p _ { j } ) _ { j = 1 } ^ { n }$ which are of degree $\\leq D$ , and are $S _ { n }$ equivariant, can be written as ", + "bbox": [ + 176, + 537, + 818, + 554 + ], + "page_idx": 13 + }, + { + "type": "equation", + "img_path": "images/a850ab1e1e90d06a22cc3b7a3205b0605b223fbfd89301a4c88f8a588642481c.jpg", + "text": "$$\np _ { j } ( X ) = \\sum _ { T \\leq D } \\sum _ { \\vec { r } \\in \\Sigma _ { T } } \\sum _ { \\vec { \\alpha } | | \\alpha _ { k } | = r _ { k } } \\Lambda _ { \\vec { \\alpha } , \\vec { r } } \\left( Q _ { j } ^ { ( \\vec { r } ) } ( X ) \\right) = \\sum _ { T \\leq D } \\sum _ { \\vec { r } \\in \\Sigma _ { T } } \\Lambda _ { \\vec { r } } \\left( \\iota \\circ Q _ { j } ^ { ( \\vec { r } ) } ( X ) \\right) , j = 1 , \\dots , n ,\n$$", + "text_format": "latex", + "bbox": [ + 187, + 559, + 805, + 597 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "where $\\begin{array} { r } { \\Lambda _ { \\vec { r } } = \\sum _ { \\vec { \\alpha } | | \\alpha _ { k } | = r _ { k } } \\Lambda _ { \\vec { \\alpha } , \\vec { r } } \\circ \\iota _ { T } ^ { - 1 } } \\end{array}$ , and $\\iota _ { T } ^ { - 1 }$ is the left inverse of the embedding $\\iota$ . If $p$ is also translation invariant, then ", + "bbox": [ + 169, + 602, + 825, + 633 + ], + "page_idx": 13 + }, + { + "type": "equation", + "img_path": "images/72fdaf5398588aaec2a0b484faf92aa54c621a5dada65c8b759b1ebe5787c18e.jpg", + "text": "$$\np ( X ) = p ( X - \\frac { 1 } { n } X 1 _ { n } \\mathbf { 1 } _ { n } ^ { T } ) = \\sum _ { T \\leq D } \\sum _ { \\vec { r } \\in \\Sigma _ { T } } \\hat { \\Lambda } _ { \\vec { r } } \\left( \\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac { 1 } { n } X 1 _ { n } \\mathbf { 1 } _ { n } ^ { T } ) \\right) .\n$$", + "text_format": "latex", + "bbox": [ + 254, + 636, + 743, + 678 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Thus $\\mathcal { Q } _ { D }$ is $D$ -spanning. ", + "bbox": [ + 174, + 679, + 339, + 695 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "We prove Lemma 3 ", + "bbox": [ + 174, + 710, + 305, + 724 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Lemma 3. The function set $\\mathcal { Q } _ { D }$ is contained in ", + "bbox": [ + 171, + 728, + 488, + 743 + ], + "page_idx": 13 + }, + { + "type": "equation", + "img_path": "images/d9975ca5b0f8b179121a196b20c5c59424456b8b001824e241861043d7bbc3aa.jpg", + "text": "$$\n\\mathcal { F } _ { \\mathrm { f e a t } } = \\big \\{ \\iota \\circ \\pi _ { V } \\circ f ^ { 1 } \\circ f ^ { 2 } \\circ \\ldots \\circ f ^ { T } \\circ \\mathrm { e x t } ( X - \\frac { 1 } { n } X \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) | f ^ { j } \\in \\mathcal { F } _ { m i n } , T \\leq D \\big \\} .\n$$", + "text_format": "latex", + "bbox": [ + 228, + 747, + 769, + 776 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Thus $\\mathcal { F } _ { \\mathrm { f e a t } }$ is $D$ -spanning. ", + "bbox": [ + 174, + 780, + 346, + 795 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Proof. In this proof we make the dependence of $\\mathcal { F } _ { \\mathrm { f e a t } }$ on $D$ explicit and denote $\\mathcal { F } _ { \\mathrm { f e a t } } ( D )$ . ", + "bbox": [ + 171, + 809, + 764, + 825 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "We prove the claim by induction on $D$ . Assume $D = 0$ . Then $\\mathcal { Q } _ { 0 }$ contains only the constant function $X \\mapsto 1 _ { n } \\in \\mathcal { T } _ { 0 } ^ { n }$ , and this is precisely the function $\\pi _ { V } \\circ \\mathrm { e x t } \\in \\mathcal { F } _ { \\mathrm { f e a t } } ( 0 )$ . ", + "bbox": [ + 174, + 830, + 821, + 861 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Now assume the claim holds for all $D ^ { \\prime }$ with $D - 1 \\geq D ^ { \\prime } \\geq 0$ , and prove the claim for $D$ . Choose $\\vec { r } = ( r _ { 1 } , \\ldots , r _ { k } ) \\in \\Sigma _ { T }$ for some $T \\leq D$ , we need to show that the function $Q ^ { ( \\vec { r } ) }$ is in $\\mathcal { F } _ { \\mathrm { f e a t } } ( D )$ . Since ${ \\mathcal { F } } _ { \\mathrm { f e a t } } ( D - 1 ) \\subseteq { \\mathcal { F } } _ { \\mathrm { f e a t } } ( D )$ we know from the induction hypothesis that this is true if $T < D$ . Now assume $T = D$ . We consider two cases: ", + "bbox": [ + 173, + 864, + 825, + 924 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "1. If $r _ { 1 } > 0$ , we set $\\tilde { r } = ( r _ { 1 } - 1 , r _ { 2 } , \\dots , r _ { K } )$ . We know that $\\iota \\circ Q ^ { ( \\tilde { r } ) } ( \\bar { X } ) \\in \\mathcal { F } _ { \\mathrm { f e a t } } ( D - 1 )$ by the induction hypothesis. So there exist $f _ { 2 } , \\ldots , f _ { D }$ such that ", + "bbox": [ + 212, + 102, + 823, + 133 + ], + "page_idx": 14 + }, + { + "type": "equation", + "img_path": "images/40b995011f66678151cd2807461802b1c8bf5061fa4a256c935fa6cd4368a147.jpg", + "text": "$$\n\\iota \\circ \\pi _ { V } \\circ f _ { 2 } \\circ . . . \\circ f _ { D } \\circ \\mathrm { e x t } ( \\bar { X } ) = \\iota \\circ Q ^ { ( \\tilde { r } ) } ( \\bar { X } ) .\n$$", + "text_format": "latex", + "bbox": [ + 369, + 140, + 686, + 160 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "Now choose $f _ { 1 } \\in \\mathcal { F } _ { m i n }$ to be the function whose $V$ coordinate $\\tilde { V } = ( \\tilde { V } _ { j } ) _ { j = 1 } ^ { n }$ , is given by $\\tilde { V } _ { j } ( X , V ) = x _ { j } \\otimes V _ { j }$ , obtained by setting $\\theta _ { 1 } = 1 , \\theta _ { 2 } = 0$ in equation 9. Then , we have ", + "bbox": [ + 232, + 167, + 826, + 204 + ], + "page_idx": 14 + }, + { + "type": "equation", + "img_path": "images/ba9a7d56acf3f3ea11547843107770808066270253a13811a1572470db75093e.jpg", + "text": "$$\n\\begin{array} { r l } & { \\tilde { V } _ { j } ( \\bar { X } , Q ^ { ( \\tilde { r } ) } ( \\bar { X } ) ) = \\displaystyle \\sum _ { i _ { 2 } , \\ldots , i _ { K } = 1 } ^ { n } \\bar { x } _ { j } \\otimes \\bar { x } _ { j } ^ { \\otimes ( r _ { 1 } - 1 ) } \\otimes \\bar { x } _ { i _ { 2 } } ^ { \\otimes r _ { 2 } } \\otimes \\ldots \\otimes \\bar { x } _ { i _ { K } } ^ { \\otimes r _ { K } } } \\\\ & { \\qquad = Q _ { j } ^ { ( \\tilde { r } ) } ( \\bar { X } ) . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 305, + 210, + 746, + 280 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "and so ", + "bbox": [ + 232, + 284, + 279, + 297 + ], + "page_idx": 14 + }, + { + "type": "equation", + "img_path": "images/15d4df6e6d3c363ae960037b26f79b35c7236d0bad9887e341c2287dac072080.jpg", + "text": "$$\n\\begin{array} { c } { \\displaystyle { \\imath \\circ \\pi _ { V } \\circ f _ { 1 } \\circ f _ { 2 } \\circ . . . \\circ f _ { D } \\circ \\mathrm { e x t } ( X - \\frac 1 n X 1 _ { n } \\mathbf { 1 } _ { n } ^ { T } ) = \\imath \\circ Q ^ { ( \\vec { r } ) } ( \\bar { X } ) . } } \\\\ { \\displaystyle { \\mathrm { } } } \\\\ { \\displaystyle { \\mathrm { } \\mathrm { } } \\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac 1 n X 1 _ { n } \\mathbf { 1 } _ { n } ^ { T } ) \\in \\mathcal { F } _ { \\mathrm { f e a t } } ( D ) . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 232, + 295, + 743, + 351 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "2. If $r _ { 1 } = 0$ . We assume without loss of generality that $r _ { 2 } > 0$ . Set $\\tilde { r } = ( r _ { 2 } - 1 , r _ { 3 } , \\dots , r _ { K } )$ . As before by the induction hypothesis there exist $f _ { 2 } , \\ldots , f _ { D }$ which satisfy equation 19. This time we choose $f _ { 1 } \\in \\mathcal { F } _ { m i n }$ to be the function whose $V$ coordinate $\\tilde { V } = \\tilde { ( } \\tilde { V } _ { j } ) _ { j = 1 } ^ { n }$ , is given by $\\begin{array} { r } { \\tilde { V } _ { j } ( X , V ) = \\sum _ { j } x _ { j } \\otimes V _ { j } } \\end{array}$ , obtained by setting $\\theta _ { 1 } = 0 , \\theta _ { 2 } = 1$ in equation 9. Then we have ", + "bbox": [ + 214, + 356, + 825, + 434 + ], + "page_idx": 14 + }, + { + "type": "equation", + "img_path": "images/06ce0b180180e15501631eabed7866ecd38a9b64670f91d2718ba54d91ad1b0d.jpg", + "text": "$$\n\\begin{array} { r l } & { \\tilde { V } _ { j } ( \\bar { X } , Q ^ { ( \\tilde { r } ) } ( \\bar { X } ) ) = \\displaystyle \\sum _ { j = 1 } ^ { n } \\displaystyle \\sum _ { i _ { 3 } , \\ldots , i _ { K } = 1 } ^ { n } \\bar { x } _ { j } \\otimes \\bar { x } _ { j } ^ { \\otimes ( r _ { 2 } - 1 ) } \\otimes \\bar { x } _ { i _ { 3 } } ^ { \\otimes r _ { 2 } } \\otimes \\ldots \\otimes \\bar { x } _ { i _ { K } } ^ { \\otimes r _ { K } } } \\\\ & { \\qquad = \\displaystyle \\sum _ { i _ { 2 } , i _ { 3 } , \\ldots , i _ { K } = 1 } ^ { n } \\bar { x } _ { i _ { 2 } } ^ { \\otimes r _ { 2 } } \\otimes \\bar { x } _ { i _ { 3 } } ^ { \\otimes r _ { 2 } } \\otimes \\ldots \\otimes \\bar { x } _ { i _ { K } } ^ { \\otimes r _ { K } } } \\\\ & { \\qquad = Q _ { j } ^ { ( \\tilde { r } ) } ( \\bar { X } ) . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 290, + 438, + 761, + 551 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "Thus equation 20 holds, and so again we have that $\\begin{array} { r } { \\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac { 1 } { n } X 1 _ { n } \\mathbf { 1 } _ { n } ^ { T } ) \\in \\mathcal { F } _ { \\mathrm { f e a t } } ( D ) . } \\end{array}$ ", + "bbox": [ + 232, + 556, + 810, + 575 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "Finally we prove Lemma 4 ", + "bbox": [ + 173, + 626, + 352, + 641 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "Lemma 4. If all functions in $\\mathcal { Q } _ { D }$ can be written as ", + "bbox": [ + 173, + 645, + 513, + 661 + ], + "page_idx": 14 + }, + { + "type": "equation", + "img_path": "images/493a0cc1da7307942808fb8357e09c06774e8240571893dcab10a98a54fa1810.jpg", + "text": "$$\n\\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac { 1 } { n } X 1 _ { n } 1 _ { n } ^ { T } ) = \\sum _ { k = 1 } ^ { K } \\hat { A } _ { k } f _ { k } ( X ) ,\n$$", + "text_format": "latex", + "bbox": [ + 357, + 667, + 638, + 712 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "where $f _ { k } \\in \\mathcal { F } _ { \\mathrm { f e a t } }$ , $A _ { k } : W _ { \\mathrm { f e a t } } \\to W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } }$ and ${ \\hat { A } } _ { k } : W _ { \\mathrm { f e a t } } ^ { n } \\to ( W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } } ) ^ { n }$ is defined by elementwise application of $A _ { k }$ , then $\\mathcal { F } _ { \\mathrm { f e a t } }$ is $D$ -spanning. ", + "bbox": [ + 173, + 719, + 823, + 752 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "Proof. If the conditions in Lemma 4 hold, then since $\\mathcal { Q } _ { D }$ is $D$ -spanning, every translation invariant and permutation equivariant polynomials $p$ of degree $D$ can be written as ", + "bbox": [ + 173, + 770, + 823, + 800 + ], + "page_idx": 14 + }, + { + "type": "equation", + "img_path": "images/7ad44bc5e953bd5e76d62a99ae401620be4a17c0feaaf999df5c1d81b653db59.jpg", + "text": "$$\n\\begin{array} { l } { \\displaystyle p ( X ) = \\sum _ { \\vec { r } ^ { \\prime } | | \\vec { r } | | _ { 1 } \\le D } \\hat { \\Lambda } _ { \\vec { r } } \\bigg ( \\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac 1 n X \\mathbf { 1 } _ { n } \\mathbf { 1 } _ { n } ^ { T } ) \\bigg ) = \\sum _ { \\vec { r } ^ { \\prime } | | \\vec { r } ^ { \\prime } | | _ { 1 } \\le D } \\hat { \\Lambda } _ { \\vec { r } } \\bigg ( \\sum _ { k = 1 } ^ { K _ { \\vec { r } } } \\iota \\circ \\hat { A } _ { k , \\vec { r } } f _ { k , \\vec { r } } ( X ) \\bigg ) } \\\\ { = \\sum _ { \\vec { r } ^ { \\prime } | | \\vec { r } | | _ { 1 } \\le D } \\sum _ { k = 1 } ^ { K _ { \\vec { r } } } \\hat { \\Lambda } _ { k , \\vec { r } } ( f _ { k , \\vec { r } } ( X ) ) } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 204, + 808, + 795, + 902 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "where we denote $\\Lambda _ { k , \\vec { r } } = \\Lambda _ { \\vec { r } } \\circ \\iota \\circ A _ { k , \\vec { r } } .$ Thus we proved $\\mathcal { F } _ { \\mathrm { f e a t } }$ is $D$ -spanning. ", + "bbox": [ + 173, + 909, + 681, + 925 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "D PROOFS FOR SECTION 5 ", + "text_level": 1, + "bbox": [ + 176, + 101, + 411, + 118 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "We prove Lemma 5 ", + "bbox": [ + 174, + 133, + 305, + 148 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Lemma 5. Let $l ^ { ( 1 ) } = ( \\ell _ { 1 } ^ { ( 1 ) } , \\dots , \\ell _ { K _ { 1 } } ^ { ( 1 ) } )$ $l ^ { ( 2 ) } = ( \\ell _ { 1 } ^ { ( 2 ) } , \\dots , \\ell _ { K _ { 2 } } ^ { ( 2 ) } )$ . A function $\\boldsymbol { \\Lambda } = \\left( \\Lambda _ { 1 } , \\ldots , \\Lambda _ { K _ { 2 } } \\right)$ a linear equivariant mapping between $W _ { l ^ { ( 1 ) } }$ and $W _ { l ^ { ( 2 ) } }$ , if and only if there exists a $K _ { 1 } \\times K _ { 2 }$ matrix $M$ with $M _ { i j } = 0$ whenever $\\ell _ { i } ^ { ( 1 ) } \\neq \\ell _ { j } ^ { ( 2 ) }$ , such that ", + "bbox": [ + 173, + 152, + 825, + 204 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/c86fa12e12e78bc2dff340de63d341d041103c9ad97279aa1d5349fbec1ba3c2.jpg", + "text": "$$\n\\Lambda _ { j } ( V ) = \\sum _ { i = 1 } ^ { K _ { 1 } } M _ { i j } V _ { i }\n$$", + "text_format": "latex", + "bbox": [ + 429, + 213, + 568, + 256 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "where $V = ( V _ { i } ) _ { i = 1 } ^ { K _ { 1 } }$ and $V _ { i } \\in W _ { \\ell _ { i } ^ { ( 1 ) } }$ for all $i = 1 , \\ldots , K _ { 1 }$ ", + "bbox": [ + 173, + 262, + 552, + 284 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Proof. As mentioned in the main text, this lemma is based on Schur’s lemma. This lemma is typically stated for complex representations, but holds for odd dimensional real representation as well. We recount the lemma and its proof here for completeness (see also (Fulton & Harris, 2013)). ", + "bbox": [ + 173, + 301, + 825, + 344 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Lemma D.1 (Schur’s Lemma for SO(3)). Let $\\Lambda : W _ { \\ell _ { 1 } } \\to W _ { \\ell _ { 2 } }$ be a linear equivariant map. If $\\ell _ { 1 } \\neq \\ell _ { 2 }$ then $\\Lambda = 0$ . Otherwise $\\Lambda$ is a scalar multiply of the identity. ", + "bbox": [ + 169, + 353, + 825, + 383 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Proof. Let $\\Lambda : W _ { \\ell _ { 1 } } \\to W _ { \\ell _ { 2 } }$ be a linear equivariant map. The image and kernel of $\\Lambda$ are invariant subspaces of $W _ { \\ell _ { 1 } }$ and $W _ { \\ell _ { 2 } }$ , respectively. It follows that if $\\Lambda \\neq 0$ then $\\Lambda$ is a linear isomorphism so necessarily $\\ell _ { 1 } = \\ell _ { 2 }$ . Now assume $\\ell _ { 1 } = \\ell _ { 2 }$ . Since the dimension of $W _ { \\ell _ { 1 } }$ is odd, $\\Lambda$ has a real eigenvalue $\\lambda$ . The linear function $\\Lambda - \\lambda I$ is equivariant and has a non-trivial kernel, so $\\Lambda - \\lambda I =$ 0. □ ", + "bbox": [ + 173, + 398, + 826, + 470 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "We now return to the proof of Lemma 5. Note that each $\\Lambda _ { j } : W _ { l ^ { ( 1 ) } } \\to W _ { \\ell _ { \\vec { \\cdot } } ^ { ( 2 ) } }$ is linear and SO(3) equivariant. Next denote the restrictions of each $\\Lambda _ { j }$ to $W _ { \\ell _ { i } ^ { ( 1 ) } } , i = 1 , \\dots , K _ { 2 }$ by $\\Lambda _ { i j }$ , and note that ", + "bbox": [ + 174, + 487, + 823, + 526 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/1ba83e118e1d6d86913cdc6af9f0324cb8056d6b0463bb01413adffe4b0e3059.jpg", + "text": "$$\n\\Lambda _ { j } ( V _ { 1 } , \\dots , V _ { K _ { 1 } } ) = \\sum _ { i = 1 } ^ { K _ { 1 } } \\Lambda _ { i j } ( V _ { i } ) .\n$$", + "text_format": "latex", + "bbox": [ + 390, + 535, + 607, + 579 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "By considering vectors in $W _ { l ^ { ( 1 ) } }$ of the form $( 0 , \\ldots , 0 , V _ { i } , 0 \\ldots , 0 )$ we see that each $\\Lambda _ { i j } : W _ { \\ell _ { i } ^ { ( 1 ) } } \\to$ $W _ { \\ell _ { j } ^ { ( 2 ) } }$ is linear and SO(3)-equivariant. Thus by Schur’s lemma, if $\\ell _ { i } ^ { ( 1 ) } = \\ell _ { j } ^ { ( 2 ) }$ then $\\Lambda _ { i j } ( V _ { i } ) = M _ { i j } V _ { i }$ for some real $M _ { i j }$ , and otherwise $M _ { i j } = 0$ . Plugging this into equation 21 we obtain equation 12. ", + "bbox": [ + 173, + 584, + 825, + 643 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "We prove Theorem 2 which shows that the TFN network described in the main text is universal: ", + "bbox": [ + 171, + 680, + 802, + 696 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Theorem 2. For all $n \\in \\mathbb { N } , l _ { T } \\in \\mathbb { N } _ { + } ^ { * }$ , ", + "bbox": [ + 176, + 700, + 421, + 715 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "1. For $D \\in { \\mathbb { N } } _ { + }$ , every $G$ -equivariant polynomial $p : \\mathbb { R } ^ { 3 \\times n } W _ { T } ^ { n }$ of degree $D$ is in $\\mathcal { F } _ { C ( D ) , D } ^ { \\mathrm { T F N } }$ 2. Every continuous $G$ -equivariant function can be approximated uniformly on compact sets by functions in ∪D∈N+ FTFNC(D),D ", + "bbox": [ + 209, + 727, + 825, + 789 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Proof. As mentioned in the main text, we only need to show that the function space $\\mathcal { F } _ { \\mathrm { f e a t } } ( D )$ is $D$ -spanning. Recall that $\\mathcal { F } _ { \\mathrm { f e a t } } ( D )$ is obtained by $2 D$ consecutive convolutions with $D$ -filters. In general, we denote the space of functions defined by applying $J$ consecutive convolutions by $\\mathcal { G } _ { J , D }$ . ", + "bbox": [ + 174, + 805, + 826, + 849 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "If $\\mathcal { V }$ is a space of functions from $\\mathbb { R } ^ { 3 \\times n } Y ^ { n }$ , we denote by $\\langle \\mathcal { V } , \\mathcal { T } _ { T } \\rangle$ the space of all functions $p : \\mathbb { R } ^ { 3 \\times n } \\overset { \\cdot } { } \\mathcal { T } _ { T } ^ { n }$ of the form ", + "bbox": [ + 173, + 854, + 823, + 885 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/5b7d61c5e1f6bacf58ed590936872904db01b04c9207febcb1fb370a2b2e1cce.jpg", + "text": "$$\np ( X ) = \\sum _ { k = 1 } ^ { K } \\hat { A } _ { k } f _ { k } ( X ) ,\n$$", + "text_format": "latex", + "bbox": [ + 421, + 885, + 576, + 928 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "where $A _ { k } : Y \\mathcal { T } _ { T }$ are linear functions, $\\hat { A } _ { k } : Y ^ { n } \\mathcal { T } _ { T } ^ { n }$ are induced by elementwise application, and $f _ { k } \\in \\mathcal { V }$ . This notation is useful because: (i) by Lemma 4 it is sufficient to show that $Q ^ { ( \\vec { r } ) } ( \\bar { X } )$ is in $\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { T } \\rangle$ for all $\\vec { r } \\in \\Sigma _ { T }$ and all $T \\leq D$ , and because (ii) it enables comparison of the expressive power of function spaces $\\mathcal { \\scriptsize { N } } _ { 1 } , \\mathcal { \\scriptsize { N } } _ { 2 }$ whose elements map to different spaces $Y _ { 1 } ^ { n } , Y _ { 2 } ^ { n }$ , since the elements in $\\langle y _ { i } , \\mathcal { T } _ { T } \\rangle , i = \\bar { 1 } , 2$ both map to the same space. In particular, note that if for every $f \\in \\mathcal { V } _ { 2 }$ there is a $g \\in \\mathcal { V } 1$ and a linear map $A : Y _ { 1 } Y _ { 2 }$ such that $f ( X ) \\ = \\ \\hat { A } \\circ g ( X )$ , then $\\langle { \\mathcal { V } } _ { 2 } , { \\mathcal { T } } _ { T } \\rangle \\subseteq \\langle { \\mathcal { V } } _ { 1 } , { \\mathcal { T } } _ { T } \\rangle$ . ", + "bbox": [ + 173, + 101, + 825, + 208 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "We now use this abstract discussion to prove some useful results: the first is that for the purpose of this lemma, we can ‘forget about’ the multiplication by a unitary matrix in equation 14, used for decomposition into irreducible representations: To see this, denote by $\\tilde { \\mathcal { G } } _ { J , D }$ the function space obtained by taking $J$ consecutive convolutions with $D$ -filters without multiplying by a unitary matrix in equation 14. Since Kronecker products of unitary matrices are unitary matrices, we obtain that the elements in $\\mathcal { G } _ { J , D }$ and $\\tilde { \\mathcal { G } } _ { J , D }$ differ only by multiplication by a unitary matrix, and thus $\\langle \\tilde { \\mathcal { G } } _ { J , D } , \\mathcal { T } _ { T } \\rangle \\subseteq$ $\\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { T } \\rangle$ and $\\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { T } \\rangle \\subseteq \\langle \\tilde { \\mathcal { G } } _ { J , D } , \\mathcal { T } _ { T } \\rangle$ , so both sets are equal. ", + "bbox": [ + 173, + 212, + 825, + 318 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "Next, we prove that adding convolutional layers (enlarging $J$ ) or taking higher order filters (enlarging $D$ ) can only increase the expressive power of a network. ", + "bbox": [ + 174, + 323, + 821, + 352 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "Lemma D.2. For all $J , D , T \\in \\mathbb { N } _ { + }$ , $l . \\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { T } \\rangle \\subseteq \\langle \\mathcal { G } _ { J + 1 , D } , \\mathcal { T } _ { T } \\rangle .$ 2 $\\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { T } \\rangle \\subseteq \\langle \\mathcal { G } _ { J , D + 1 } , \\mathcal { T } _ { T } \\rangle .$ . ", + "bbox": [ + 173, + 357, + 413, + 373 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "", + "bbox": [ + 210, + 386, + 421, + 428 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "Proof. The first claim follows from the fact that every function $f$ in $\\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { T } \\rangle$ can be identified with a function in $\\langle \\mathcal { G } _ { J + 1 , D } , \\mathcal { T } _ { T } \\rangle$ by taking the $J + 1$ convolutional layer in equation 14 with $\\theta _ { 0 } =$ $1 , F = 0$ . ", + "bbox": [ + 176, + 439, + 825, + 483 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "The second claim follows from the fact that $D$ -filters can be identified with $D + 1$ -filters whose $D + 1$ -th entry is $0$ . □ ", + "bbox": [ + 173, + 488, + 825, + 518 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "The last preliminary lemma we will need is ", + "bbox": [ + 174, + 532, + 460, + 547 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "Lemma D.3. For every $J , D \\in \\mathbb { N } _ { + }$ , and every $t , s \\in \\mathbb { N } _ { + }$ , if $\\dot { \\boldsymbol { p } } \\in \\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { t } \\rangle$ , then the function $q$ defined by ", + "bbox": [ + 173, + 554, + 823, + 582 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/157b4eb8d42596b21df90c7c20da2c4e448ee8aa03cc6efccf2f970ca8ef510b.jpg", + "text": "$$\nq _ { a } ( X ) = \\sum _ { b = 1 } ^ { n } ( { \\bar { x } } _ { a } - { \\bar { x } } _ { b } ) ^ { \\otimes s } \\otimes p _ { b } ( X )\n$$", + "text_format": "latex", + "bbox": [ + 379, + 580, + 617, + 622 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "is in $\\langle \\mathcal { G } _ { J + 1 , D } , \\mathcal { T } _ { t + s } \\rangle$ ", + "bbox": [ + 173, + 622, + 312, + 638 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "Proof. This lemma is based on the fact that the space of $s$ homogeneous polynomial on $\\mathbb { R } ^ { 3 }$ is spanned by polynomials of the form $\\| x \\| ^ { s - \\ell } Y _ { m } ^ { \\ell } ( \\bar { x } )$ for $\\ell = s , s - 2 , s - 4 \\ldots$ (Dai & $\\mathrm { X u }$ , 2013). For each such $\\ell$ , and $s \\leq D$ , these polynomials can be realized by filters $F ^ { ( \\ell ) }$ by setting $R ^ { ( \\ell ) } ( \\| x \\| ) = \\| x \\| ^ { s }$ so that ", + "bbox": [ + 173, + 652, + 825, + 713 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/fe191eb6b097a60008f6c16047d29cdf855c55919dd9b7baf237fd7baa214758.jpg", + "text": "$$\nF _ { m } ^ { ( \\ell ) } ( x ) = \\| x \\| ^ { s } Y _ { m } ^ { \\ell } ( \\hat { x } ) = \\| x \\| ^ { s - \\ell } Y _ { m } ^ { \\ell } ( x ) .\n$$", + "text_format": "latex", + "bbox": [ + 362, + 715, + 633, + 736 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "For every $D \\ \\in \\ \\mathbb { N }$ and $s \\leq D$ , we can construct a $D$ -filter $F ^ { s , D } = ( F ^ { ( 0 ) } , \\ldots , F ^ { ( D ) } )$ where $F ^ { ( s ) } , F ^ { ( s - 2 ) } , \\ldots .$ are as defined above and the other filters are zero. Since both the entries of $F ^ { s , D } ( x )$ , and the entries of $x ^ { \\otimes s }$ , span the space of $s$ -homogeneous polynomials on $\\mathbb { R } ^ { 3 }$ , it follows that there exists a linear mapping $B _ { s } : W _ { l _ { D } } T _ { s }$ so that ", + "bbox": [ + 173, + 739, + 825, + 797 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/8fc3c35c727a8ae16fa51601ee4c0c1e54ea890358760d99bf70a12f5446c458.jpg", + "text": "$$\nx ^ { \\otimes s } = B _ { s } ( F ^ { s , D } ( { \\boldsymbol { x } } ) ) , \\forall { \\boldsymbol { x } } \\in \\mathbb { R } ^ { 3 } .\n$$", + "text_format": "latex", + "bbox": [ + 393, + 801, + 604, + 820 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "Thus, since $p$ can be written as a sum of compositions of linear mappings with functions in $\\mathcal { G } _ { J , D }$ as in equation 22, and similarly $x ^ { \\otimes s }$ is obtained as a linear image of functions in $\\mathcal { G } _ { 1 , D }$ as in equation 23, we deduce that ", + "bbox": [ + 176, + 829, + 826, + 869 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/fa10a55443e5f7a520ffd122229e37d3368fba368ccfb6597ac7da43ab02cec6.jpg", + "text": "$$\n\\sum _ { b = 1 } ^ { n } ( x _ { a } - x _ { b } ) \\otimes p _ { b } ( X ) = \\sum _ { b = 1 } ^ { n } ( \\bar { x } _ { a } - \\bar { x } _ { b } ) \\otimes p _ { b } ( X )\n$$", + "text_format": "latex", + "bbox": [ + 331, + 867, + 663, + 909 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "is in $\\langle \\mathcal { G } _ { J + 1 , D } , \\mathcal { T } _ { t + s } \\rangle$ ", + "bbox": [ + 173, + 909, + 308, + 925 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "As a final preliminary, we note that $D$ -filters can perform an averaging operation by setting $R ^ { ( 0 ) } = 1$ and $\\theta _ { 0 } , R ^ { ( \\mathrm { 1 } ) } , \\ldots , R ^ { ( \\dot { D } ) } = 0$ in equation 13 and equation 14 . We call this $D$ -filter an averaging filter. ", + "bbox": [ + 171, + 102, + 825, + 133 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "We are now ready to prove our claim: we need to show that for every $D , T \\in \\mathbb { N } _ { + }$ where $T \\leq D$ , for every $\\vec { r } \\in \\Sigma _ { T }$ , the function $Q ^ { ( \\vec { r } ) }$ is in $\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { T } \\rangle$ . Note that due to the inclusion relations in Lemma D.2 it is sufficient to prove this for the case $T = D$ . We prove this by induction on $D$ . For $D = 0$ , vectors $\\vec { r } \\in \\Sigma _ { 0 }$ contains only zeros and so ", + "bbox": [ + 173, + 138, + 825, + 198 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/6d9daefbbffc124f7a4c1212f098402ea43a3d5fd908e464d9176fcace8e1de8.jpg", + "text": "$$\nQ ^ { ( \\vec { r } ) } ( \\bar { X } ) = 1 _ { n } = \\pi _ { V } \\circ \\mathrm { e x t } ( X ) \\in \\langle \\mathcal { G } _ { 0 , 0 } , \\mathcal { T } _ { 0 } \\rangle .\n$$", + "text_format": "latex", + "bbox": [ + 351, + 202, + 647, + 222 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "We now assume the claim is true for all $D ^ { \\prime }$ with $D > D ^ { \\prime } \\geq 0$ and prove the claim is true for $D$ . We need to show that for every $\\vec { r } \\in \\Sigma _ { D }$ the function $Q ^ { ( \\vec { r } ) }$ is in $\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { D } \\rangle$ . We prove this yet again by induction, this time on the value of $r _ { 1 }$ : assume that $\\vec { r } \\in \\Sigma _ { D }$ and $r _ { 1 } = 0 .$ . Denote by $\\tilde { r }$ the vector in $\\Sigma _ { D - 1 }$ defined by ", + "bbox": [ + 173, + 232, + 825, + 290 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/46613d7f0da0ef0fce92ad630c0254ebf96380aefc66498e73438c4f46845f8a.jpg", + "text": "$$\n\\tilde { r } = ( r _ { 2 } - 1 , r _ { 3 } , \\ldots , r _ { K } ) .\n$$", + "text_format": "latex", + "bbox": [ + 413, + 289, + 583, + 305 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "By the induction assumption on $D$ , we know that $Q ^ { ( \\tilde { r } ) } ( \\bar { X } ) \\in \\mathcal { G } _ { 2 ( D - 1 ) , D - 1 , D - 1 }$ and so ", + "bbox": [ + 174, + 308, + 743, + 325 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/c752720d6430d8f7ff000a662ea0032fc8b642b3343d74d0bf331ac24b8629ac.jpg", + "text": "$$\n\\begin{array} { l } { \\displaystyle q _ { a } ( X ) = \\sum _ { b = 1 } ^ { n } ( \\bar { x } _ { a } - \\bar { x } _ { b } ) \\otimes Q _ { b } ^ { ( \\bar { r } ) } ( \\bar { X } ) = \\sum _ { b = 1 } ^ { n } ( \\bar { x } _ { a } - \\bar { x } _ { b } ) \\otimes \\bar { x } _ { b } ^ { \\otimes r _ { 2 } - 1 } \\otimes \\sum _ { i _ { 3 } , \\dots , i _ { K } = 1 } ^ { n } \\bar { x } _ { i _ { 3 } } ^ { \\otimes r _ { 3 } } \\otimes \\dots \\otimes \\bar { x } _ { i _ { K } } ^ { \\otimes r _ { K } } } \\\\ { \\displaystyle \\quad = \\left( \\bar { x } _ { a } \\otimes \\sum _ { b = 1 } ^ { n } Q _ { b } ^ { ( \\bar { r } ) } ( \\bar { X } ) \\right) - Q ^ { ( \\bar { r } ) } ( \\bar { X } ) } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 183, + 329, + 812, + 416 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "is in $\\langle \\mathcal { G } _ { 2 D - 1 , D - 1 } , \\mathcal { T } _ { D } \\rangle$ by Lemma D.3, which is contained in $\\langle \\mathcal { G } _ { 2 D - 1 , D } , \\mathcal { T } _ { D } \\rangle$ by Lemma D.2. Since $\\textstyle { \\bar { x } } _ { a }$ has zero mean, while $Q _ { a } ^ { ( \\vec { r } ) } ( \\bar { X } )$ does not depend on $a$ since $r _ { 1 } ~ = ~ 0$ , applying an averaging filter to $q _ { a }$ gives us a constant value $- Q _ { a } ^ { ( \\vec { r } ) } ( \\bar { X } )$ in each coordinate $a \\in [ n ]$ , and so $Q ^ { ( \\vec { r } ) } ( \\bar { X } )$ is in $\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { D } \\rangle$ . ", + "bbox": [ + 173, + 419, + 825, + 486 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Now assume the claim is true for all $\\vec { r } \\in \\Sigma _ { D }$ which sum to $D$ , and whose first coordinate is smaller than some $r _ { 1 } ^ { \\prime } \\geq 1$ , we now prove the claim is true when the first coordinate of $\\vec { r }$ is equal to $r _ { 1 } ^ { \\prime }$ . The vector $\\tilde { \\boldsymbol { r } } = \\left( r _ { 2 } , \\ldots , r _ { K } \\right)$ obtained from $\\vec { r }$ by removing the first coordinate, sums to $D ^ { \\prime } = D - r _ { 1 } ^ { \\prime } <$ $D$ , and so by the induction hypothesis on $D$ we know that $Q ^ { ( \\tilde { r } ) } \\in \\langle { \\mathcal { G } _ { 2 D ^ { \\prime } , D ^ { \\prime } } } , { \\mathcal { T } _ { D ^ { \\prime } } } \\rangle$ . By Lemma D.3 we obtain a function $q _ { a } \\in \\langle \\mathcal { G } _ { 2 D ^ { \\prime } + 1 , D ^ { \\prime } } , \\mathcal { T } _ { D } \\rangle \\subseteq \\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { D } \\rangle$ defined by ", + "bbox": [ + 174, + 489, + 825, + 564 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/34ee578e2afa8951963148d2fdb7aac5d4c4b3f96827ae3eda4207445f2efbad.jpg", + "text": "$$\n\\begin{array} { r l } & { q _ { a } ( X ) = \\displaystyle \\sum _ { b = 1 } ^ { n } ( \\bar { x } _ { a } - \\bar { x } _ { b } ) ^ { \\otimes r _ { 1 } } \\otimes Q _ { b } ^ { ( \\bar { r } ) } ( \\bar { X } ) } \\\\ & { \\qquad = \\displaystyle \\sum _ { b = 1 } ^ { n } ( \\bar { x } _ { a } - \\bar { x } _ { b } ) ^ { \\otimes r _ { 1 } } \\otimes \\bar { x } _ { b } ^ { \\otimes r _ { 2 } } \\otimes \\displaystyle \\sum _ { i _ { 3 } , \\ldots , i _ { K } = 1 } ^ { n } \\bar { x } _ { i _ { 3 } } ^ { \\otimes r _ { 3 } } \\otimes \\ldots \\otimes \\bar { x } _ { i _ { K } } ^ { \\otimes r _ { K } } } \\\\ & { \\qquad = Q _ { a } ^ { ( \\bar { r } ) } ( \\bar { X } ) + \\mathrm { a d d i t i o n a l ~ t e r m s } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 274, + 569, + 722, + 674 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "where the additional terms are linear combinations of functions of the form $P _ { D } Q _ { a } ^ { ( r ^ { \\prime } ) } ( \\bar { X } )$ where $r ^ { \\prime } \\in \\Sigma _ { D }$ and their first coordinate $r _ { 1 }$ is smaller than $r _ { 1 } ^ { \\prime }$ , and $P _ { D } : \\mathcal { T } _ { D } \\mathcal { T } _ { D }$ is a permutation. By the induction hypothesis on $r _ { 1 }$ , each such $Q ^ { ( r ^ { \\prime } ) }$ is in $\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { D } \\rangle$ . It follows that $P _ { D } Q _ { a } ^ { ( r ^ { \\prime } ) } ( \\bar { X } ) , a =$ $1 , \\ldots , n$ , and thus $Q ^ { ( \\vec { r } ) } ( \\bar { X } )$ , are in $\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { D } \\rangle$ as well. This concludes the proof of Theorem 2. ", + "bbox": [ + 173, + 680, + 825, + 746 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "E ALTERNATIVE TFN ARCHITECTURE ", + "text_level": 1, + "bbox": [ + 173, + 785, + 508, + 803 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "In this appendix we show that replacing the standard TFN convolutional layer with the layer defined in equation 15: ", + "bbox": [ + 171, + 815, + 825, + 845 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/eef53c7de4865ff017c01000e5c67857f897f3177ae29f9e8392e839781cbb2c.jpg", + "text": "$$\n\\tilde { V } _ { a } ( X , V ) = U ( l _ { f } , l _ { i } ) \\left( \\theta _ { 1 } \\sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \\otimes V _ { b } + \\theta _ { 2 } \\sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \\otimes V _ { a } \\right) ,\n$$", + "text_format": "latex", + "bbox": [ + 236, + 849, + 758, + 892 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "we can obtain $D$ -spanning networks using $2 D$ consecutive convolutions with 1-filters (that is, filters in $W _ { l _ { 1 } }$ , where $l _ { 1 } \\stackrel { \\cdot } { = } [ 0 , 1 ] ^ { \\frac { \\cdot } { T } }$ ). Our discussion here is somewhat informal, meant to provide the general ", + "bbox": [ + 171, + 895, + 823, + 924 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "ideas without delving into the details as we have done for the standard TFN architecture in the proof of Theorem 2. In the end of our discussion we will explain what is necessary to make this argument completely rigorous. ", + "bbox": [ + 174, + 103, + 823, + 146 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "We will only need two fixed filters for our argument here: The first is the 1-filter $F _ { I d } = ( F ^ { ( 0 ) } , F ^ { ( 1 ) } )$ defined by setting $R ^ { ( 0 ) } ( \\| x \\| ) = 0$ and $R ^ { ( 1 ) } ( \\| x \\| ) = \\| x \\|$ to obtain ", + "bbox": [ + 169, + 154, + 823, + 185 + ], + "page_idx": 18 + }, + { + "type": "equation", + "img_path": "images/9a1921dd046cebbb5ce79bcb65fe1fb560017efd5db92d99b0b00a6c1fa4eac1.jpg", + "text": "$$\nF _ { I d } ( x ) = \\| x \\| Y ^ { 1 } ( \\hat { x } ) = \\| x \\| \\hat { x } = x .\n$$", + "text_format": "latex", + "bbox": [ + 382, + 190, + 614, + 209 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "The second is the filter $F _ { 1 }$ defined by setting $R ^ { ( 0 ) } ( \\| x \\| ) = 1$ and $R ^ { ( 1 ) } ( \\| x \\| ) = 0$ , so that ", + "bbox": [ + 173, + 215, + 753, + 233 + ], + "page_idx": 18 + }, + { + "type": "equation", + "img_path": "images/5875576ebd591118816095a2c2dc95015639c6eac18d036fa8898751f9709c77.jpg", + "text": "$$\nF _ { 1 } ( x ) = 1 .\n$$", + "text_format": "latex", + "bbox": [ + 459, + 238, + 537, + 256 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "We prove our claim by showing that a pair of convolutions with 1-filters can construct any convolutional layer defined in equation 9 for the $D$ -spanning architecture using tensor representations. The claim then follows from the fact that $D$ convolutions of the latter architecture suffice for achieving $D$ -spanning, as shown in Lemma 3. ", + "bbox": [ + 173, + 268, + 825, + 325 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "Convolutions for tensor representations, defined in equation 9, are composed of two terms: ", + "bbox": [ + 169, + 332, + 766, + 348 + ], + "page_idx": 18 + }, + { + "type": "equation", + "img_path": "images/888294d450b7cdaa02d29f68368082ecd50c514cf58ba38d5813b40204ef78e0.jpg", + "text": "$$\n\\tilde { V } _ { a } ^ { \\mathrm { t e n s o r } , 1 } ( \\bar { X } , V ) = \\bar { x } _ { a } \\otimes V _ { a } \\mathrm { a n d } \\tilde { V } _ { a } ^ { \\mathrm { t e n s o r } , 2 } ( \\bar { X } , V ) = \\sum _ { b = 1 } ^ { n } \\bar { x } _ { b } \\otimes V _ { b } .\n$$", + "text_format": "latex", + "bbox": [ + 284, + 353, + 714, + 395 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "To obtain the first term $\\tilde { V } _ { a } ^ { \\mathrm { t e n s o r , 1 } }$ , we set $\\theta _ { 1 } = 0 , \\theta _ { 2 } = 1 / n , F = F _ { I d }$ in equation 15 we obtain (the decomposition into irreducibles of) $\\tilde { V } _ { a } ^ { \\mathrm { t e n s o r , 1 } } ( \\bar { X } , V ) = \\bar { x } _ { a } \\otimes V _ { a }$ . Thus this term can in fact be expressed by a single convolution. We can leave this outcome unchanged by a second convolution, defined by setting $\\bar { \\theta } _ { 1 } = 0 , \\theta _ { 2 } = 1 / n , F = F _ { 1 }$ . ", + "bbox": [ + 174, + 409, + 826, + 469 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "To obtain the second term $\\tilde { V } _ { a } ^ { \\mathrm { t e n s o r } , 2 }$ , we apply a first convolution with $\\theta _ { 1 } = - 1 , F = F _ { I d } , \\theta _ { 2 } = 0$ , to obtain ", + "bbox": [ + 171, + 477, + 825, + 507 + ], + "page_idx": 18 + }, + { + "type": "equation", + "img_path": "images/1c0c4b0d07b2326bc145517fc85e20b6caa46009e6b670e27b6aa2c2db475bc1.jpg", + "text": "$$\n\\sum _ { b = 1 } ^ { n } ( x _ { b } - x _ { a } ) \\otimes V _ { b } = \\sum _ { b = 1 } ^ { n } ( { \\bar { x } } _ { b } - { \\bar { x } } _ { a } ) \\otimes V _ { b } = { \\tilde { V } } _ { a } ^ { { \\mathrm { t e n s o r } } , 2 } ( V , { \\bar { X } } ) - { \\bar { x } } _ { a } \\otimes \\sum _ { b = 1 } ^ { n } V _ { b }\n$$", + "text_format": "latex", + "bbox": [ + 245, + 510, + 750, + 551 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "By applying an additional averaging filter, defined by setting $\\textstyle \\theta _ { 1 } = { \\frac { 1 } { n } } , F = F _ { 1 } , \\theta _ { 2 } = 0$ , we obtain $\\tilde { V } _ { a } ^ { \\mathrm { t e n s o r } , 2 } ( V , \\bar { X } )$ . This concludes our ‘informal proof’. ", + "bbox": [ + 173, + 559, + 823, + 592 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "Our discussion here has been somewhat inaccurate, since in practice $F _ { I d } ( x ) = ( 0 , x ) \\in W _ { 0 } \\oplus W _ { 1 }$ and $F _ { 1 } ( x ) = ( 1 , 0 ) \\in W _ { 0 } \\oplus W _ { 1 }$ . Moreover, in our proof we have glossed over the multiplication by the unitary matrix used to obtain decomposition into irreducible representations. However the ideas discussed here can be used to show that $2 D$ convolutions with 1-filters can satisfy the sufficient condition for $D$ -spanning defined in Lemma 4. See our treatment of Theorem 2 for more details. ", + "bbox": [ + 173, + 597, + 825, + 667 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "F COMPARISON WITH ORIGINAL TFN PAPER ", + "text_level": 1, + "bbox": [ + 174, + 688, + 560, + 704 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "In this Appendix we discuss three superficial differences between the presentation of the TFN architecture in Thomas et al. (2018) and our presentation here: ", + "bbox": [ + 174, + 719, + 823, + 747 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "1. We define convolutional layers between features residing in direct sums of irreducible representations, while (Thomas et al., 2018) focuses on features which inhabit a single irreducible representation. This difference is non-essential, as direct sums of irreducible representations can be represented as multiple channels where each feature inhabits a single irreducible representation. 2. The term $\\theta _ { 0 } V _ { a }$ in equation 14 appears in (Fuchs et al., 2020), but does not appear explicitly in (Thomas et al., 2018). However it can be obtained by concatenation of the input of a self-interaction layer to the output, and then applying a self-interaction layer. 3. We take the scalar functions $R ^ { ( \\ell ) }$ to be polynomials, while (Thomas et al., 2018) take them to be fully connected networks composed with radial basis functions. Using polynomial scalar bases is convenient for our presentation here since it enables exact expression of equivariant polynomials. Replacing polynomial bases with fully connected networks, we obtain approximation of equivariant polynomials instead of exact expression. It can be shown that if $p$ is a $G$ -equivariant polynomial which can be expressed by some network $\\mathcal { F } _ { C , D }$ defined with filters coming from a polynomial scalar basis, then $p$ can be approximated on a compact set $K$ , up to an arbitrary $\\epsilon$ error, by a similar network with scalar functions coming from a sufficiently large fully connected network. ", + "bbox": [ + 212, + 758, + 825, + 924 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "", + "bbox": [ + 232, + 103, + 825, + 188 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "G TENSOR UNIVERSALITY ", + "text_level": 1, + "bbox": [ + 174, + 207, + 413, + 223 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "In this section we show how to construct the complete set $\\mathcal { F } _ { \\mathrm { p o o l } }$ of linear $\\mathrm { S O ( 3 ) }$ invariant functionals from $W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } } = \\bigoplus _ { T = 0 } ^ { D } \\mathcal { T } _ { T }$ to $\\mathbb { R }$ . Since each such functional $\\Lambda$ is of the form ", + "bbox": [ + 173, + 238, + 823, + 272 + ], + "page_idx": 19 + }, + { + "type": "equation", + "img_path": "images/b57cd45c52bd31dbfd6a60a04df3da149300563b442ea80051d826ec75367796.jpg", + "text": "$$\n\\Lambda ( w _ { 0 } , . . . , w _ { D } ) = \\sum _ { T = 0 } ^ { D } \\Lambda _ { T } ( w _ { T } ) ,\n$$", + "text_format": "latex", + "bbox": [ + 390, + 276, + 606, + 321 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "where each $\\Lambda _ { T }$ is $\\mathrm { S O ( 3 ) }$ -invariant, it is sufficient to characterize all linear SO(3)-invariant functionals $\\Lambda : \\mathcal { T } _ { D } \\to \\mathbb { R }$ . ", + "bbox": [ + 173, + 327, + 823, + 354 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "It will be convenient to denote ", + "bbox": [ + 174, + 362, + 375, + 376 + ], + "page_idx": 19 + }, + { + "type": "equation", + "img_path": "images/45327efe75350802e18c9364be7f442c296b8749e8ed5baaa3d7bb1befff1543.jpg", + "text": "$$\nW = \\mathbb { R } ^ { 3 } \\operatorname { a n d } W ^ { \\otimes D } \\cong \\mathbb { R } ^ { 3 ^ { D } } = \\mathcal { T } _ { D } .\n$$", + "text_format": "latex", + "bbox": [ + 382, + 378, + 611, + 398 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "We achieve our characterization using the bijective correspondence between linear functional $\\Lambda :$ $W ^ { \\otimes D } \\to \\mathbb { R }$ and multi-linear functions $\\tilde { \\Lambda } : W ^ { D } \\to \\mathbb { R }$ : each such $\\Lambda$ corresponds to a unique $\\hat { \\Lambda }$ , such that ", + "bbox": [ + 174, + 405, + 825, + 446 + ], + "page_idx": 19 + }, + { + "type": "equation", + "img_path": "images/df8f8bbd5775a47d4532415c4acfb54c22bdcbfff704dc096711765598b481ea.jpg", + "text": "$$\n\\tilde { \\Lambda } ( e _ { i _ { 1 } } , \\ldots , e _ { i _ { D } } ) = \\Lambda ( e _ { i _ { 1 } } \\otimes \\ldots \\otimes e _ { i _ { D } } ) , \\forall ( i _ { 1 } , \\ldots , i _ { D } ) \\in [ 3 ] ^ { D } ,\n$$", + "text_format": "latex", + "bbox": [ + 295, + 445, + 700, + 465 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "where $e _ { 1 } , e _ { 2 } , e _ { 3 }$ denote the standard basis elements of $\\mathbb { R } ^ { 3 }$ . We define a spanning set of equivariant linear functionals on $W ^ { \\otimes D }$ via a corresponding characterization for multi-linear functionals on $W ^ { D }$ . Specifically, set ", + "bbox": [ + 174, + 467, + 826, + 507 + ], + "page_idx": 19 + }, + { + "type": "equation", + "img_path": "images/e2231805e6f80db582c8bc67d23c8cd5d73149e2fa25c05289d7491ad933fe3e.jpg", + "text": "$$\nK _ { D } = \\{ k \\in \\mathbb { N } _ { + } | D - 3 k { \\mathrm { i s ~ e v e n ~ a n d ~ n o n } } { \\mathrm { - n e g a t i v e . ~ } } \\}\n$$", + "text_format": "latex", + "bbox": [ + 321, + 507, + 673, + 525 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "For $k \\in K _ { D }$ we define a multi-linear functional: ", + "bbox": [ + 173, + 526, + 491, + 541 + ], + "page_idx": 19 + }, + { + "type": "equation", + "img_path": "images/b96e41e5ec77b65c2f7c30bdd31e0d9db744478990c4a7bf5915fcc7fd0a2048.jpg", + "text": "$$\n\\begin{array} { r } { \\tilde { \\Lambda } _ { k } ( w _ { 1 } , \\dots , w _ { D } ) = \\operatorname* { d e t } ( w _ { 1 } , w _ { 2 } , w _ { 3 } ) \\times \\dots \\times \\operatorname* { d e t } ( w _ { 3 k - 2 } , w _ { 3 k - 1 } , w _ { 3 k } ) \\times \\langle w _ { 3 k + 1 } , w _ { 3 k + 2 } \\rangle \\times \\dots } \\\\ { \\times \\langle w _ { D - 1 } , w _ { D } \\rangle , \\qquad ( 2 \\tilde { \\lambda } _ { 1 } - \\tilde { \\lambda } _ { 1 } ) \\times \\dots , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 184, + 546, + 812, + 583 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "and for $( k , \\sigma ) \\in K _ { D } \\times S _ { D }$ we define ", + "bbox": [ + 174, + 587, + 421, + 603 + ], + "page_idx": 19 + }, + { + "type": "equation", + "img_path": "images/ed66adc05cff02565a6b59ffff05877253169c4dcf9eadec0440a5192dfa0043.jpg", + "text": "$$\n\\tilde { \\Lambda } _ { k , \\sigma } ( w _ { 1 } , \\ldots , w _ { D } ) = \\tilde { \\Lambda } _ { k } ( w _ { \\sigma ( 1 ) } , \\ldots , w _ { \\sigma ( D ) } )\n$$", + "text_format": "latex", + "bbox": [ + 351, + 608, + 647, + 628 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "Proposition 1. The space of linear invariant functions from $\\mathcal { T } _ { D }$ to $\\mathbb { R }$ is spanned by the set of linear invariant functionals $\\lambda _ { D } = \\{ \\Lambda _ { k , \\sigma } | \\left( k , \\sigma \\right) \\in K _ { D } \\times S _ { D } \\}$ induced by the multi-linear functional $\\tilde { \\Lambda } _ { k , \\sigma }$ described in equation 25 and equation $2 6$ ", + "bbox": [ + 173, + 632, + 823, + 678 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "We note that (i) equation 24 provides a (cumbersome) way to compute all linear invariant functionals $\\Lambda _ { k , \\sigma }$ explicitly by evaluating the corresponding $\\tilde { \\Lambda } _ { k , \\sigma }$ on the $3 ^ { D }$ elements of the standard basis and (ii) the set $\\lambda _ { D }$ is spanning, but is not linearly independent. For example, since $\\langle w _ { 1 } , w _ { 2 } \\rangle = \\langle w _ { 2 } , w _ { 1 } \\rangle$ , the space of $\\mathrm { S O ( 3 ) }$ invariant functionals on ${ \\mathcal { T } } _ { 2 } = { \\overline { { W } } } ^ { \\otimes 2 }$ is one dimensional while $| \\lambda _ { 2 } | = 2$ . ", + "bbox": [ + 173, + 688, + 825, + 747 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "Proof of Proposition $^ { l }$ . We first show that the bijective correspondence between linear functional $\\Lambda : W ^ { \\otimes D } \\to \\mathbb { R }$ and multi-linear functions $\\tilde { \\Lambda } : \\dot { W } ^ { D } \\to \\mathbb { R }$ , extends to a bijective correspondence between SO(3)-invariant linear/multi-linear functionals. The action of SO(3) on $W ^ { D }$ is defined by ", + "bbox": [ + 173, + 761, + 825, + 806 + ], + "page_idx": 19 + }, + { + "type": "equation", + "img_path": "images/19eb1123b02d75c7439a3d1622ba1c49afde3b10a10160e4d16e44a73ee0dcef.jpg", + "text": "$$\n\\tilde { \\rho } ( R ) ( w _ { 1 } , \\ldots , w _ { D } ) = ( R w _ { 1 } , \\ldots , R w _ { D } ) .\n$$", + "text_format": "latex", + "bbox": [ + 362, + 810, + 633, + 828 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "The action $\\rho ( R ) = R ^ { \\otimes D }$ of SO(3) on $W ^ { \\otimes D }$ is such that the map ", + "bbox": [ + 173, + 832, + 604, + 849 + ], + "page_idx": 19 + }, + { + "type": "equation", + "img_path": "images/8cfd515d2d6196c80a511c7922c80f1e2366a1b77bbb16f659ac4cef32c23b68.jpg", + "text": "$$\n( w _ { 1 } , \\dots , w _ { D } ) \\mapsto w _ { 1 } \\otimes w _ { 2 } \\dots w _ { D }\n$$", + "text_format": "latex", + "bbox": [ + 385, + 854, + 611, + 872 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "is SO(3)- equivariant. It follows that if $\\tilde { \\Lambda }$ and $\\Lambda$ satisfy equation 24, then for all $R \\in \\mathrm { S O ( 3 ) }$ , the same equation holds for the pair $\\tilde { \\Lambda } \\circ \\tilde { \\rho } ( R )$ and $\\Lambda \\circ \\rho ( R )$ . Thus SO(3)-invariance of $\\tilde { \\Lambda }$ is equivalent to $\\mathrm { S O ( 3 ) }$ -invariance of $\\Lambda$ . ", + "bbox": [ + 174, + 878, + 825, + 924 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "Multi-linear functionals on $W ^ { D }$ invariant to $\\tilde { \\rho }$ are a subset of the set of polynomials on $W ^ { D }$ invariant to $\\tilde { \\rho }$ . It is known (see (Kraft & Procesi, 2000), page 114), that all such polynomials are algebraically generated by functions of the form ", + "bbox": [ + 173, + 102, + 823, + 146 + ], + "page_idx": 20 + }, + { + "type": "equation", + "img_path": "images/5c1bc250cffdd612aa0182d31b2a2ccdf1ef9897cc9acb334763c60639da03d5.jpg", + "text": "$$\n\\begin{array} { r } { \\operatorname* { d e t } ( w _ { i _ { 1 } } , w _ { i _ { 2 } } , w _ { i _ { 3 } } ) \\mathrm { ~ a n d ~ } \\langle w _ { j _ { 1 } } , w _ { j _ { 2 } } \\rangle , \\mathrm { ~ w h e r e ~ } i _ { 1 } , i _ { 2 } , i _ { 3 } , j _ { 1 } , j _ { 2 } \\in [ D ] . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 285, + 152, + 705, + 170 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "Equivalently, $\\mathrm { S O ( 3 ) }$ -invariant polynomials are spanned by linear combinations of polynomials of the form ", + "bbox": [ + 173, + 176, + 825, + 204 + ], + "page_idx": 20 + }, + { + "type": "equation", + "img_path": "images/aeb93f1dc0eb0f9e7c84a28c079f6e15a61d4012660eb399737422057ff3481f.jpg", + "text": "$$\n\\operatorname * { d e t } \\bigl ( w _ { i _ { 1 } } , w _ { i _ { 2 } } , w _ { i _ { 3 } } \\bigr ) \\operatorname * { d e t } \\bigl ( w _ { i _ { 4 } } , w _ { i _ { 5 } } , w _ { i _ { 6 } } \\bigr ) \\ldots \\langle w _ { j _ { 1 } } , w _ { j _ { 2 } } \\rangle \\langle w _ { j _ { 3 } } , w _ { j _ { 4 } } \\rangle \\ldots .\n$$", + "text_format": "latex", + "bbox": [ + 281, + 204, + 717, + 220 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "When considering the subset of multi-linear invariant polynomials, we see that they must be spanned by polynomials as in equation 27, where each $w _ { 1 } , \\ldots , w _ { D }$ appears exactly once in each polynomial in the spanning set. These precisely correspond to the functions in $\\lambda _ { D }$ . ", + "bbox": [ + 174, + 223, + 825, + 266 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "H EXPERIMENTS ", + "text_level": 1, + "bbox": [ + 174, + 315, + 331, + 332 + ], + "page_idx": 20 + }, + { + "type": "table", + "img_path": "images/30712253f5955d3ca6f795324f935ac71bda0d63fa1692f647560e32ee1bf29d.jpg", + "table_caption": [ + "Table 1: Results obtained on the QM9 dataset Ramakrishnan et al. (2014) for different design choices in the TFN architecture. Results are reported for the $\\epsilon _ { h o m o }$ target variable, and are multiplied by $1 0 ^ { 3 }$ . " + ], + "table_footnote": [], + "table_body": "
Model variant Mean l1error
Order 0irreps 111.2 ± 0.3
Order 0-1 ireps 82.1 ± 0.4
Order 0-2 irreps 62.1 ± 1.5
Order 0-3 irreps 53.2 ± 1.9
Order 0-4 irreps 51.2 ± 0.4
Order O-3 irreps without non-linearity 53.1 ± 0.8
Order O-3 irreps + self-interaction only in the final layer 55.2 ± 0.3
", + "bbox": [ + 250, + 401, + 748, + 516 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "This section provides an experimental evaluation of different design choices of the TFN architecture, inspired by our theoretical analysis. We study the following questions: ", + "bbox": [ + 173, + 532, + 821, + 561 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "1. The importance of non-linear activation. Our proof shows that using non-linear activation functions is not necessary for proving universality. Here, we empirically test the possibility of removing these layers. \n2. The importance of high-dimensional irreducible representations. Our theoretical analysis shows that in order to represent/approximate high degree polynomials, high-order representations should be used. Here, we check whether using high-order representations has practical benefits. \n3. The effect of self-interaction layers. Our proof suggests that it is enough to use self interaction linear layers at the end of the model. We empirically compare this approach with the more common approach of using self-interaction layers after each convolutional layer. ", + "bbox": [ + 210, + 574, + 825, + 737 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "Dataset. We use the QM9 (Ramakrishnan et al., 2014) dataset for our experiments. The dataset contains 134K molecules, with node 3D positions, 5 categorical node features and 4 categorical edge features. The task is a molecule property prediction regression task. ", + "bbox": [ + 176, + 752, + 821, + 795 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "Framework. We used pytorch (Paszke et al., 2017)as the deep learning framework and the Deep Graph Library (DGL) (Wang et al., 2019a) as the graph learning framework. All experiments ran on NVIDIA GV100 GPUs. ", + "bbox": [ + 174, + 809, + 825, + 852 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "Experimental setup. We use the the TFN implementation from Fuchs et al. (2020). We trained each model variant for 50 epochs on the $\\epsilon _ { h o m o }$ target variable using an $\\ell _ { 1 }$ loss function and the ADAM optimizer with learning rate $1 0 ^ { - 3 }$ and report results on the test set on the final epoch averaged over two runs. We used the default parameters and data splits from Fuchs et al. (2020). ", + "bbox": [ + 174, + 867, + 823, + 924 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "Architecture. The architecture consists of 4 TFN convolutional layers, each followed by a linear self-interaction layer. We used 16 copies of each irreducible representation used. We used normbased non-linearity as in the original TFN paper (Thomas et al., 2018). These convolutional layers are followed by a max-pooling layer and two fully connected layers with $1 6 d$ features in the hidden layer, where $d$ is the maximal degree of irreducible representations used. ", + "bbox": [ + 173, + 103, + 825, + 174 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "Results. Table 1 and Figure 1 present the results. The main conclusions are: (1) The experiments show that, at least for this task, using non-linear activations does not improve performance. This result fits our theoretical analysis which shows that these layers are not needed for universality. (2) Figure 1 presents a plot of error vs representation degrees used. The plot clearly shows that using high-dimensional representations (up to order 3) improves performance, which also fits our analysis. Using representation orders higher than 3 is significantly more time consuming, and was found to have little effect on the results (as in in Fuchs et al. (2020)), though we believe this to be application-dependent. (3) Using self interaction layers only at the end of the model is shown to have marginal negative effect on the results. ", + "bbox": [ + 174, + 188, + 483, + 439 + ], + "page_idx": 21 + }, + { + "type": "image", + "img_path": "images/0a3168963674044c22256599e104a6aa2cefc267903dcecc69ad907cfab638f0.jpg", + "image_caption": [ + "Figure 1: $\\ell _ { 1 }$ error versus maximal irreducible representation used. It is clear that the error is reduced as higher order representation are used. Semi-transparent color represents standard deviation. " + ], + "image_footnote": [], + "bbox": [ + 503, + 195, + 785, + 359 + ], + "page_idx": 21 + } +] \ No newline at end of file diff --git a/parse/train/6NFBvWlRXaG/6NFBvWlRXaG_middle.json b/parse/train/6NFBvWlRXaG/6NFBvWlRXaG_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..96279145b721333fb33549d27416b1347b37d915 --- /dev/null +++ b/parse/train/6NFBvWlRXaG/6NFBvWlRXaG_middle.json @@ -0,0 +1,87565 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 78, + 502, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 75, + 506, + 100 + ], + "spans": [ + { + "bbox": [ + 105, + 75, + 506, + 100 + ], + "score": 1.0, + "content": "ON THE UNIVERSALITY OF ROTATION EQUIVARIANT", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 97, + 304, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 97, + 304, + 118 + ], + "score": 1.0, + "content": "POINT CLOUD NETWORKS", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 112, + 135, + 222, + 168 + ], + "lines": [ + { + "bbox": [ + 111, + 134, + 166, + 148 + ], + "spans": [ + { + "bbox": [ + 111, + 134, + 166, + 148 + ], + "score": 1.0, + "content": "Nadav Dym", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 111, + 145, + 181, + 159 + ], + "spans": [ + { + "bbox": [ + 111, + 145, + 181, + 159 + ], + "score": 1.0, + "content": "Duke University", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 112, + 158, + 222, + 169 + ], + "spans": [ + { + "bbox": [ + 112, + 158, + 222, + 169 + ], + "score": 1.0, + "content": "nadavdym@gmail.com", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 313, + 135, + 417, + 168 + ], + "lines": [ + { + "bbox": [ + 312, + 135, + 379, + 148 + ], + "spans": [ + { + "bbox": [ + 312, + 135, + 379, + 148 + ], + "score": 1.0, + "content": "Haggai Maron", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 313, + 146, + 390, + 157 + ], + "spans": [ + { + "bbox": [ + 313, + 146, + 390, + 157 + ], + "score": 1.0, + "content": "NVIDIA Research", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 313, + 157, + 418, + 168 + ], + "spans": [ + { + "bbox": [ + 313, + 157, + 418, + 168 + ], + "score": 1.0, + "content": "hmaron@nvidia.com", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + }, + { + "type": "title", + "bbox": [ + 278, + 198, + 333, + 209 + ], + "lines": [ + { + "bbox": [ + 276, + 196, + 336, + 212 + ], + "spans": [ + { + "bbox": [ + 276, + 196, + 336, + 212 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 143, + 222, + 468, + 331 + ], + "lines": [ + { + "bbox": [ + 141, + 222, + 469, + 234 + ], + "spans": [ + { + "bbox": [ + 141, + 222, + 469, + 234 + ], + "score": 1.0, + "content": "Learning functions on point clouds has applications in many fields, including com-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 140, + 233, + 469, + 245 + ], + "spans": [ + { + "bbox": [ + 140, + 233, + 469, + 245 + ], + "score": 1.0, + "content": "puter vision, computer graphics, physics, and chemistry. Recently, there has been", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 140, + 244, + 470, + 255 + ], + "spans": [ + { + "bbox": [ + 140, + 244, + 470, + 255 + ], + "score": 1.0, + "content": "a growing interest in neural architectures that are invariant or equivariant to all", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 254, + 470, + 268 + ], + "spans": [ + { + "bbox": [ + 141, + 254, + 470, + 268 + ], + "score": 1.0, + "content": "three shape-preserving transformations of point clouds: translation, rotation, and", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 266, + 470, + 278 + ], + "spans": [ + { + "bbox": [ + 141, + 266, + 470, + 278 + ], + "score": 1.0, + "content": "permutation. In this paper, we present a first study of the approximation power", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 142, + 277, + 469, + 288 + ], + "spans": [ + { + "bbox": [ + 142, + 277, + 469, + 288 + ], + "score": 1.0, + "content": "of these architectures. We first derive two sufficient conditions for an equivariant", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 287, + 470, + 301 + ], + "spans": [ + { + "bbox": [ + 141, + 287, + 470, + 301 + ], + "score": 1.0, + "content": "architecture to have the universal approximation property, based on a novel char-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 142, + 299, + 469, + 309 + ], + "spans": [ + { + "bbox": [ + 142, + 299, + 469, + 309 + ], + "score": 1.0, + "content": "acterization of the space of equivariant polynomials. We then use these conditions", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 309, + 469, + 321 + ], + "spans": [ + { + "bbox": [ + 141, + 309, + 469, + 321 + ], + "score": 1.0, + "content": "to show that two recently suggested models (Thomas et al., 2018; Fuchs et al.,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 320, + 450, + 333 + ], + "spans": [ + { + "bbox": [ + 141, + 320, + 450, + 333 + ], + "score": 1.0, + "content": "2020) are universal, and for devising two other novel universal architectures.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 13.5 + }, + { + "type": "title", + "bbox": [ + 108, + 352, + 208, + 364 + ], + "lines": [ + { + "bbox": [ + 104, + 350, + 211, + 368 + ], + "spans": [ + { + "bbox": [ + 104, + 350, + 211, + 368 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 377, + 505, + 453 + ], + "lines": [ + { + "bbox": [ + 105, + 376, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 505, + 389 + ], + "score": 1.0, + "content": "Designing neural networks that respect data symmetry is a powerful approach for obtain-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 387, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 505, + 399 + ], + "score": 1.0, + "content": "ing efficient deep models. Prominent examples being convolutional networks which re-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 398, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 505, + 411 + ], + "score": 1.0, + "content": "spect the translational invariance of images, graph neural networks which respect the per-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 409, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 409, + 505, + 421 + ], + "score": 1.0, + "content": "mutation invariance of graphs (Gilmer et al., 2017; Maron et al., 2019b), networks such", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 419, + 507, + 433 + ], + "spans": [ + { + "bbox": [ + 104, + 419, + 507, + 433 + ], + "score": 1.0, + "content": "as (Zaheer et al., 2017; Qi et al., 2017a) which respect the permutation invariance of", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 431, + 506, + 443 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 506, + 443 + ], + "score": 1.0, + "content": "sets, and networks which respect 3D rotational symmetries (Cohen et al., 2018; Weiler", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 442, + 505, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 505, + 455 + ], + "score": 1.0, + "content": "et al., 2018; Esteves et al., 2018; Worrall & Brostow, 2018; Kondor et al., 2018a).", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 107, + 454, + 336, + 541 + ], + "lines": [ + { + "bbox": [ + 106, + 453, + 337, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 453, + 337, + 465 + ], + "score": 1.0, + "content": "While the expressive power of equivariant models is re-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 464, + 337, + 476 + ], + "spans": [ + { + "bbox": [ + 106, + 464, + 337, + 476 + ], + "score": 1.0, + "content": "duced by design to include only equivariant functions, a", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 475, + 337, + 487 + ], + "spans": [ + { + "bbox": [ + 106, + 475, + 337, + 487 + ], + "score": 1.0, + "content": "desirable property of equivariant networks is universal-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 486, + 337, + 498 + ], + "spans": [ + { + "bbox": [ + 106, + 486, + 337, + 498 + ], + "score": 1.0, + "content": "ity: the ability to approximate any continuous equivariant", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 497, + 337, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 337, + 509 + ], + "score": 1.0, + "content": "function. This is not always the case: while convolutional", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 507, + 337, + 520 + ], + "spans": [ + { + "bbox": [ + 105, + 507, + 337, + 520 + ], + "score": 1.0, + "content": "networks and networks for sets are universal (Yarotsky,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 519, + 337, + 531 + ], + "spans": [ + { + "bbox": [ + 106, + 519, + 337, + 531 + ], + "score": 1.0, + "content": "2018; Segol & Lipman, 2019), popular graph neural net-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 530, + 313, + 542 + ], + "spans": [ + { + "bbox": [ + 106, + 530, + 313, + 542 + ], + "score": 1.0, + "content": "works are not (Xu et al., 2019; Morris et al., 2018).", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 30.5 + }, + { + "type": "image", + "bbox": [ + 357, + 468, + 478, + 553 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 357, + 468, + 478, + 553 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 357, + 468, + 478, + 553 + ], + "spans": [ + { + "bbox": [ + 357, + 468, + 478, + 553 + ], + "score": 0.953, + "type": "image", + "image_path": "deeacb27baa383de014f051156e964f0944d6459137a61e1a3840aa1470e5f7e.jpg" + } + ] + } + ], + "index": 34.5, + "virtual_lines": [ + { + "bbox": [ + 357, + 468, + 478, + 510.5 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 357, + 510.5, + 478, + 553.0 + ], + "spans": [], + "index": 36 + } + ] + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 107, + 547, + 336, + 558 + ], + "lines": [ + { + "bbox": [ + 105, + 545, + 337, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 337, + 560 + ], + "score": 1.0, + "content": "In this paper, we consider the universality of networks", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 107, + 558, + 505, + 635 + ], + "lines": [ + { + "bbox": [ + 106, + 557, + 505, + 571 + ], + "spans": [ + { + "bbox": [ + 106, + 557, + 505, + 571 + ], + "score": 1.0, + "content": "that respect the symmetries of 3D point clouds: translations, rotations, and permutations. Design-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 569, + 505, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 569, + 505, + 581 + ], + "score": 1.0, + "content": "ing such networks is a popular paradigm in recent years (Thomas et al., 2018; Fuchs et al., 2020;", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 579, + 505, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 579, + 505, + 592 + ], + "score": 1.0, + "content": "Poulenard et al., 2019; Zhao et al., 2019). While there have been many works on the universality of", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 590, + 505, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 373, + 604 + ], + "score": 1.0, + "content": "permutation invariant networks (Zaheer et al., 2017; Maron et al.,", + "type": "text" + }, + { + "bbox": [ + 374, + 591, + 400, + 601 + ], + "score": 0.56, + "content": "2 0 1 9 \\mathrm { c }", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 590, + 505, + 604 + ], + "score": 1.0, + "content": "; Keriven & Peyre, 2019), ´", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 602, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 106, + 602, + 505, + 614 + ], + "score": 1.0, + "content": "and a recent work discussing the universality of rotation equivariant networks (Bogatskiy et al.,", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 612, + 506, + 626 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 506, + 626 + ], + "score": 1.0, + "content": "2020), this is a first paper which discusses the universality of networks which combine rotations,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 625, + 228, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 228, + 635 + ], + "score": 1.0, + "content": "permutations and translations.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 107, + 640, + 505, + 707 + ], + "lines": [ + { + "bbox": [ + 107, + 641, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 107, + 641, + 505, + 652 + ], + "score": 1.0, + "content": "We start the paper with a general, architecture-agnostic, discussion, and derive two sufficient con-", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 651, + 505, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 505, + 664 + ], + "score": 1.0, + "content": "ditions for universality. These conditions are a result of a novel characterization of equivariant", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 662, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 506, + 675 + ], + "score": 1.0, + "content": "polynomials for the symmetry group of interest. We use these conditions in order to prove univer-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 673, + 505, + 685 + ], + "spans": [ + { + "bbox": [ + 106, + 673, + 505, + 685 + ], + "score": 1.0, + "content": "sality of the prominent Tensor Field Networks (TFN) architecture (Thomas et al., 2018; Fuchs et al.,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 683, + 506, + 698 + ], + "spans": [ + { + "bbox": [ + 105, + 683, + 506, + 698 + ], + "score": 1.0, + "content": "2020). The following is a weakened and simplified statement of Theorem 2 stated later on in the", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 104, + 697, + 134, + 708 + ], + "spans": [ + { + "bbox": [ + 104, + 697, + 134, + 708 + ], + "score": 1.0, + "content": "paper:", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 47.5 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "Theorem (Simplification of Theorem 2). Any continuous equivariant function on point clouds can", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 721, + 413, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 413, + 733 + ], + "score": 1.0, + "content": "be approximated uniformly on compact sets by a composition of TFN layers.", + "type": "text" + } + ], + "index": 52 + } + ], + "index": 51.5 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 303, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 308, + 761 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 308, + 761 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 78, + 502, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 75, + 506, + 100 + ], + "spans": [ + { + "bbox": [ + 105, + 75, + 506, + 100 + ], + "score": 1.0, + "content": "ON THE UNIVERSALITY OF ROTATION EQUIVARIANT", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 97, + 304, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 97, + 304, + 118 + ], + "score": 1.0, + "content": "POINT CLOUD NETWORKS", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "list", + "bbox": [ + 112, + 135, + 222, + 168 + ], + "lines": [ + { + "bbox": [ + 111, + 134, + 166, + 148 + ], + "spans": [ + { + "bbox": [ + 111, + 134, + 166, + 148 + ], + "score": 1.0, + "content": "Nadav Dym", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 111, + 145, + 181, + 159 + ], + "spans": [ + { + "bbox": [ + 111, + 145, + 181, + 159 + ], + "score": 1.0, + "content": "Duke University", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 112, + 158, + 222, + 169 + ], + "spans": [ + { + "bbox": [ + 112, + 158, + 222, + 169 + ], + "score": 1.0, + "content": "nadavdym@gmail.com", + "type": "text" + } + ], + "index": 6, + "is_list_start_line": true + } + ], + "index": 4, + "bbox_fs": [ + 111, + 134, + 222, + 169 + ] + }, + { + "type": "text", + "bbox": [ + 313, + 135, + 417, + 168 + ], + "lines": [ + { + "bbox": [ + 312, + 135, + 379, + 148 + ], + "spans": [ + { + "bbox": [ + 312, + 135, + 379, + 148 + ], + "score": 1.0, + "content": "Haggai Maron", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 313, + 146, + 390, + 157 + ], + "spans": [ + { + "bbox": [ + 313, + 146, + 390, + 157 + ], + "score": 1.0, + "content": "NVIDIA Research", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 313, + 157, + 418, + 168 + ], + "spans": [ + { + "bbox": [ + 313, + 157, + 418, + 168 + ], + "score": 1.0, + "content": "hmaron@nvidia.com", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5, + "bbox_fs": [ + 312, + 135, + 418, + 168 + ] + }, + { + "type": "title", + "bbox": [ + 278, + 198, + 333, + 209 + ], + "lines": [ + { + "bbox": [ + 276, + 196, + 336, + 212 + ], + "spans": [ + { + "bbox": [ + 276, + 196, + 336, + 212 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 143, + 222, + 468, + 331 + ], + "lines": [ + { + "bbox": [ + 141, + 222, + 469, + 234 + ], + "spans": [ + { + "bbox": [ + 141, + 222, + 469, + 234 + ], + "score": 1.0, + "content": "Learning functions on point clouds has applications in many fields, including com-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 140, + 233, + 469, + 245 + ], + "spans": [ + { + "bbox": [ + 140, + 233, + 469, + 245 + ], + "score": 1.0, + "content": "puter vision, computer graphics, physics, and chemistry. Recently, there has been", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 140, + 244, + 470, + 255 + ], + "spans": [ + { + "bbox": [ + 140, + 244, + 470, + 255 + ], + "score": 1.0, + "content": "a growing interest in neural architectures that are invariant or equivariant to all", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 254, + 470, + 268 + ], + "spans": [ + { + "bbox": [ + 141, + 254, + 470, + 268 + ], + "score": 1.0, + "content": "three shape-preserving transformations of point clouds: translation, rotation, and", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 266, + 470, + 278 + ], + "spans": [ + { + "bbox": [ + 141, + 266, + 470, + 278 + ], + "score": 1.0, + "content": "permutation. In this paper, we present a first study of the approximation power", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 142, + 277, + 469, + 288 + ], + "spans": [ + { + "bbox": [ + 142, + 277, + 469, + 288 + ], + "score": 1.0, + "content": "of these architectures. We first derive two sufficient conditions for an equivariant", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 287, + 470, + 301 + ], + "spans": [ + { + "bbox": [ + 141, + 287, + 470, + 301 + ], + "score": 1.0, + "content": "architecture to have the universal approximation property, based on a novel char-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 142, + 299, + 469, + 309 + ], + "spans": [ + { + "bbox": [ + 142, + 299, + 469, + 309 + ], + "score": 1.0, + "content": "acterization of the space of equivariant polynomials. We then use these conditions", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 309, + 469, + 321 + ], + "spans": [ + { + "bbox": [ + 141, + 309, + 469, + 321 + ], + "score": 1.0, + "content": "to show that two recently suggested models (Thomas et al., 2018; Fuchs et al.,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 320, + 450, + 333 + ], + "spans": [ + { + "bbox": [ + 141, + 320, + 450, + 333 + ], + "score": 1.0, + "content": "2020) are universal, and for devising two other novel universal architectures.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 13.5, + "bbox_fs": [ + 140, + 222, + 470, + 333 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 352, + 208, + 364 + ], + "lines": [ + { + "bbox": [ + 104, + 350, + 211, + 368 + ], + "spans": [ + { + "bbox": [ + 104, + 350, + 211, + 368 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 377, + 505, + 453 + ], + "lines": [ + { + "bbox": [ + 105, + 376, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 505, + 389 + ], + "score": 1.0, + "content": "Designing neural networks that respect data symmetry is a powerful approach for obtain-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 387, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 505, + 399 + ], + "score": 1.0, + "content": "ing efficient deep models. Prominent examples being convolutional networks which re-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 398, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 505, + 411 + ], + "score": 1.0, + "content": "spect the translational invariance of images, graph neural networks which respect the per-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 409, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 409, + 505, + 421 + ], + "score": 1.0, + "content": "mutation invariance of graphs (Gilmer et al., 2017; Maron et al., 2019b), networks such", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 419, + 507, + 433 + ], + "spans": [ + { + "bbox": [ + 104, + 419, + 507, + 433 + ], + "score": 1.0, + "content": "as (Zaheer et al., 2017; Qi et al., 2017a) which respect the permutation invariance of", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 431, + 506, + 443 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 506, + 443 + ], + "score": 1.0, + "content": "sets, and networks which respect 3D rotational symmetries (Cohen et al., 2018; Weiler", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 442, + 505, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 505, + 455 + ], + "score": 1.0, + "content": "et al., 2018; Esteves et al., 2018; Worrall & Brostow, 2018; Kondor et al., 2018a).", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 23, + "bbox_fs": [ + 104, + 376, + 507, + 455 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 454, + 336, + 541 + ], + "lines": [ + { + "bbox": [ + 106, + 453, + 337, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 453, + 337, + 465 + ], + "score": 1.0, + "content": "While the expressive power of equivariant models is re-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 464, + 337, + 476 + ], + "spans": [ + { + "bbox": [ + 106, + 464, + 337, + 476 + ], + "score": 1.0, + "content": "duced by design to include only equivariant functions, a", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 475, + 337, + 487 + ], + "spans": [ + { + "bbox": [ + 106, + 475, + 337, + 487 + ], + "score": 1.0, + "content": "desirable property of equivariant networks is universal-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 486, + 337, + 498 + ], + "spans": [ + { + "bbox": [ + 106, + 486, + 337, + 498 + ], + "score": 1.0, + "content": "ity: the ability to approximate any continuous equivariant", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 497, + 337, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 337, + 509 + ], + "score": 1.0, + "content": "function. This is not always the case: while convolutional", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 507, + 337, + 520 + ], + "spans": [ + { + "bbox": [ + 105, + 507, + 337, + 520 + ], + "score": 1.0, + "content": "networks and networks for sets are universal (Yarotsky,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 519, + 337, + 531 + ], + "spans": [ + { + "bbox": [ + 106, + 519, + 337, + 531 + ], + "score": 1.0, + "content": "2018; Segol & Lipman, 2019), popular graph neural net-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 530, + 313, + 542 + ], + "spans": [ + { + "bbox": [ + 106, + 530, + 313, + 542 + ], + "score": 1.0, + "content": "works are not (Xu et al., 2019; Morris et al., 2018).", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 453, + 337, + 542 + ] + }, + { + "type": "image", + "bbox": [ + 357, + 468, + 478, + 553 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 357, + 468, + 478, + 553 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 357, + 468, + 478, + 553 + ], + "spans": [ + { + "bbox": [ + 357, + 468, + 478, + 553 + ], + "score": 0.953, + "type": "image", + "image_path": "deeacb27baa383de014f051156e964f0944d6459137a61e1a3840aa1470e5f7e.jpg" + } + ] + } + ], + "index": 34.5, + "virtual_lines": [ + { + "bbox": [ + 357, + 468, + 478, + 510.5 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 357, + 510.5, + 478, + 553.0 + ], + "spans": [], + "index": 36 + } + ] + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 107, + 547, + 336, + 558 + ], + "lines": [ + { + "bbox": [ + 105, + 545, + 337, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 337, + 560 + ], + "score": 1.0, + "content": "In this paper, we consider the universality of networks", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 557, + 505, + 571 + ], + "spans": [ + { + "bbox": [ + 106, + 557, + 505, + 571 + ], + "score": 1.0, + "content": "that respect the symmetries of 3D point clouds: translations, rotations, and permutations. Design-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 569, + 505, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 569, + 505, + 581 + ], + "score": 1.0, + "content": "ing such networks is a popular paradigm in recent years (Thomas et al., 2018; Fuchs et al., 2020;", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 579, + 505, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 579, + 505, + 592 + ], + "score": 1.0, + "content": "Poulenard et al., 2019; Zhao et al., 2019). While there have been many works on the universality of", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 590, + 505, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 373, + 604 + ], + "score": 1.0, + "content": "permutation invariant networks (Zaheer et al., 2017; Maron et al.,", + "type": "text" + }, + { + "bbox": [ + 374, + 591, + 400, + 601 + ], + "score": 0.56, + "content": "2 0 1 9 \\mathrm { c }", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 590, + 505, + 604 + ], + "score": 1.0, + "content": "; Keriven & Peyre, 2019), ´", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 602, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 106, + 602, + 505, + 614 + ], + "score": 1.0, + "content": "and a recent work discussing the universality of rotation equivariant networks (Bogatskiy et al.,", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 612, + 506, + 626 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 506, + 626 + ], + "score": 1.0, + "content": "2020), this is a first paper which discusses the universality of networks which combine rotations,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 625, + 228, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 228, + 635 + ], + "score": 1.0, + "content": "permutations and translations.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 37, + "bbox_fs": [ + 105, + 545, + 337, + 560 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 558, + 505, + 635 + ], + "lines": [], + "index": 41, + "bbox_fs": [ + 105, + 557, + 506, + 635 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 640, + 505, + 707 + ], + "lines": [ + { + "bbox": [ + 107, + 641, + 505, + 652 + ], + "spans": [ + { + "bbox": [ + 107, + 641, + 505, + 652 + ], + "score": 1.0, + "content": "We start the paper with a general, architecture-agnostic, discussion, and derive two sufficient con-", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 651, + 505, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 505, + 664 + ], + "score": 1.0, + "content": "ditions for universality. These conditions are a result of a novel characterization of equivariant", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 662, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 506, + 675 + ], + "score": 1.0, + "content": "polynomials for the symmetry group of interest. We use these conditions in order to prove univer-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 673, + 505, + 685 + ], + "spans": [ + { + "bbox": [ + 106, + 673, + 505, + 685 + ], + "score": 1.0, + "content": "sality of the prominent Tensor Field Networks (TFN) architecture (Thomas et al., 2018; Fuchs et al.,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 683, + 506, + 698 + ], + "spans": [ + { + "bbox": [ + 105, + 683, + 506, + 698 + ], + "score": 1.0, + "content": "2020). The following is a weakened and simplified statement of Theorem 2 stated later on in the", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 104, + 697, + 134, + 708 + ], + "spans": [ + { + "bbox": [ + 104, + 697, + 134, + 708 + ], + "score": 1.0, + "content": "paper:", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 47.5, + "bbox_fs": [ + 104, + 641, + 506, + 708 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "Theorem (Simplification of Theorem 2). Any continuous equivariant function on point clouds can", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 721, + 413, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 413, + 733 + ], + "score": 1.0, + "content": "be approximated uniformly on compact sets by a composition of TFN layers.", + "type": "text" + } + ], + "index": 52 + } + ], + "index": 51.5, + "bbox_fs": [ + 105, + 708, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 504, + 138 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "We use our general discussion to prove the universality of two additional equivariant models: the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "first is a simple modification of the TFN architecture which allows for universality using only low", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 103, + 506, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 506, + 118 + ], + "score": 1.0, + "content": "dimensional filters. The second is a minimal architecture which is based on tensor product represen-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 115, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 115, + 506, + 128 + ], + "score": 1.0, + "content": "tations, rather than the more commonly used irreducible representations of SO(3). We discuss the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 308, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 308, + 139 + ], + "score": 1.0, + "content": "advantages and disadvantages of both approaches.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 107, + 143, + 504, + 187 + ], + "lines": [ + { + "bbox": [ + 105, + 143, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 143, + 505, + 156 + ], + "score": 1.0, + "content": "To summarize, the contributions of this paper are: (1) A general approach for proving the universal-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 154, + 505, + 166 + ], + "spans": [ + { + "bbox": [ + 106, + 154, + 505, + 166 + ], + "score": 1.0, + "content": "ity of rotation equivariant models for point clouds; (2) A proof that two recent equivariant models", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 164, + 506, + 178 + ], + "spans": [ + { + "bbox": [ + 105, + 164, + 506, + 178 + ], + "score": 1.0, + "content": "(Thomas et al., 2018; Fuchs et al., 2020) are universal; (3) Two additional simple and novel universal", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 177, + 163, + 187 + ], + "spans": [ + { + "bbox": [ + 106, + 177, + 163, + 187 + ], + "score": 1.0, + "content": "architectures.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5 + }, + { + "type": "title", + "bbox": [ + 108, + 205, + 213, + 218 + ], + "lines": [ + { + "bbox": [ + 105, + 205, + 215, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 205, + 215, + 221 + ], + "score": 1.0, + "content": "2 PREVIOUS WORK", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 106, + 232, + 505, + 364 + ], + "lines": [ + { + "bbox": [ + 105, + 232, + 505, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 232, + 505, + 245 + ], + "score": 1.0, + "content": "Deep learning on point clouds. (Qi et al., 2017a; Zaheer et al., 2017) were the first to apply neural", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 104, + 243, + 506, + 257 + ], + "spans": [ + { + "bbox": [ + 104, + 243, + 506, + 257 + ], + "score": 1.0, + "content": "networks directly to the raw point cloud data, by using pointwise functions and pooling operations.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 254, + 505, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 505, + 267 + ], + "score": 1.0, + "content": "Many subsequent works used local neighborhood information (Qi et al., 2017b; Wang et al., 2019b;", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 264, + 505, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 505, + 277 + ], + "score": 1.0, + "content": "Atzmon et al., 2018). We refer the reader to a recent survey for more details (Guo et al., 2020).", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 275, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 505, + 289 + ], + "score": 1.0, + "content": "In contrast with the aforementioned works which focused solely on permutation invariance, more", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 287, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 287, + 505, + 299 + ], + "score": 1.0, + "content": "related to this paper are works that additionally incorporated invariance to rigid motions. (Thomas", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 298, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 505, + 311 + ], + "score": 1.0, + "content": "et al., 2018) proposed Tensor Field Networks (TFN) and showed their efficacy on physics and chem-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 309, + 505, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 309, + 505, + 321 + ], + "score": 1.0, + "content": "istry tasks.(Kondor et al., 2018b) also suggested an equivariant model for continuous rotations. (Li", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 320, + 505, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 505, + 333 + ], + "score": 1.0, + "content": "et al., 2019) suggested models that are equivariant to discrete subgroups of SO(3). (Poulenard et al.,", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 331, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 106, + 331, + 505, + 343 + ], + "score": 1.0, + "content": "2019) suggested an invariant model based on spherical harmonics. (Fuchs et al., 2020) followed TFN", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 342, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 505, + 354 + ], + "score": 1.0, + "content": "and added an attention mechanism. Recently, (Zhao et al., 2019) proposed a quaternion equivariant", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 353, + 410, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 410, + 365 + ], + "score": 1.0, + "content": "point capsule network that also achieves rotation and translation invariance.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 107, + 370, + 505, + 502 + ], + "lines": [ + { + "bbox": [ + 106, + 370, + 505, + 383 + ], + "spans": [ + { + "bbox": [ + 106, + 370, + 505, + 383 + ], + "score": 1.0, + "content": "Universal approximation for invariant networks. Understanding the approximation power of in-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 380, + 505, + 394 + ], + "spans": [ + { + "bbox": [ + 105, + 380, + 505, + 394 + ], + "score": 1.0, + "content": "variant models is a popular research goal. Most of the current results assume that the symmetry", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 392, + 506, + 405 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 349, + 405 + ], + "score": 1.0, + "content": "group is a permutation group. (Zaheer et al., 2017; Qi et al.,", + "type": "text" + }, + { + "bbox": [ + 349, + 392, + 375, + 402 + ], + "score": 0.44, + "content": "2 0 1 7 \\mathrm { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 392, + 506, + 405 + ], + "score": 1.0, + "content": "; Segol & Lipman, 2019; Maron", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 403, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 387, + 415 + ], + "score": 1.0, + "content": "et al., 2020; Serviansky et al., 2020) proved universality for several", + "type": "text" + }, + { + "bbox": [ + 387, + 403, + 400, + 414 + ], + "score": 0.89, + "content": "S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 403, + 505, + 415 + ], + "score": 1.0, + "content": "-invariant and equivariant", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 412, + 505, + 427 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 505, + 427 + ], + "score": 1.0, + "content": "models. (Maron et al., 2019b;a; Keriven & Peyre, 2019; Maehara & NT, 2019) studied the approx- ´", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 424, + 504, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 504, + 437 + ], + "score": 1.0, + "content": "imation power of high-order graph neural networks. (Maron et al., 2019c; Ravanbakhsh, 2020) tar-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 435, + 505, + 449 + ], + "spans": [ + { + "bbox": [ + 104, + 435, + 505, + 449 + ], + "score": 1.0, + "content": "geted universality of networks that use high-order representations for permutation groups(Yarotsky,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 446, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 505, + 459 + ], + "score": 1.0, + "content": "2018) provided several theoretical constructions of universal equivariant neural network models", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 457, + 505, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 286, + 470 + ], + "score": 1.0, + "content": "based on polynomial invariants, including an", + "type": "text" + }, + { + "bbox": [ + 286, + 457, + 315, + 469 + ], + "score": 0.92, + "content": "S E ( 2 )", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 457, + 505, + 470 + ], + "score": 1.0, + "content": "equivariant model. In a recent work (Bogatskiy", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 468, + 506, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 468, + 506, + 482 + ], + "score": 1.0, + "content": "et al., 2020) presented a universal approximation theorem for networks that are equivariant to sev-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 479, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 208, + 493 + ], + "score": 1.0, + "content": "eral Lie groups including", + "type": "text" + }, + { + "bbox": [ + 208, + 479, + 236, + 491 + ], + "score": 0.77, + "content": "S O ( 3 )", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 479, + 505, + 493 + ], + "score": 1.0, + "content": ". The main difference from our paper is that we prove a universality", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 491, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 505, + 503 + ], + "score": 1.0, + "content": "theorem for a more complex group that besides rotations also includes translations and permutations.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 27.5 + }, + { + "type": "title", + "bbox": [ + 108, + 521, + 357, + 532 + ], + "lines": [ + { + "bbox": [ + 104, + 519, + 359, + 535 + ], + "spans": [ + { + "bbox": [ + 104, + 519, + 359, + 535 + ], + "score": 1.0, + "content": "3 A FRAMEWORK FOR PROVING UNIVERSALITY", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 107, + 546, + 504, + 569 + ], + "lines": [ + { + "bbox": [ + 105, + 545, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 505, + 559 + ], + "score": 1.0, + "content": "In this section, we describe a framework for proving the universality of equivariant networks. We", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 558, + 289, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 289, + 569 + ], + "score": 1.0, + "content": "begin with some mathematical preliminaries:", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5 + }, + { + "type": "title", + "bbox": [ + 108, + 584, + 232, + 596 + ], + "lines": [ + { + "bbox": [ + 106, + 584, + 233, + 597 + ], + "spans": [ + { + "bbox": [ + 106, + 584, + 233, + 597 + ], + "score": 1.0, + "content": "3.1 MATHEMATICAL SETUP", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 107, + 605, + 505, + 661 + ], + "lines": [ + { + "bbox": [ + 105, + 604, + 506, + 619 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 194, + 619 + ], + "score": 1.0, + "content": "An action of a group", + "type": "text" + }, + { + "bbox": [ + 194, + 606, + 203, + 616 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 604, + 295, + 619 + ], + "score": 1.0, + "content": "on a real vector space", + "type": "text" + }, + { + "bbox": [ + 295, + 606, + 307, + 616 + ], + "score": 0.74, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 604, + 403, + 619 + ], + "score": 1.0, + "content": "is a collection of maps", + "type": "text" + }, + { + "bbox": [ + 403, + 606, + 471, + 618 + ], + "score": 0.91, + "content": "\\rho ( g ) : W \\to W", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 604, + 506, + 619 + ], + "score": 1.0, + "content": "defined", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 616, + 504, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 137, + 630 + ], + "score": 1.0, + "content": "for any", + "type": "text" + }, + { + "bbox": [ + 138, + 617, + 165, + 629 + ], + "score": 0.9, + "content": "g \\in G", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 616, + 207, + 630 + ], + "score": 1.0, + "content": ", such that", + "type": "text" + }, + { + "bbox": [ + 208, + 617, + 307, + 629 + ], + "score": 0.91, + "content": "\\rho ( g _ { 1 } ) \\circ \\rho ( g _ { 2 } ) = \\rho ( g _ { 1 } g _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 616, + 335, + 630 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 336, + 617, + 380, + 629 + ], + "score": 0.92, + "content": "g _ { 1 } , g _ { 2 } \\in G", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 616, + 495, + 630 + ], + "score": 1.0, + "content": ", and the identity element of", + "type": "text" + }, + { + "bbox": [ + 495, + 617, + 504, + 627 + ], + "score": 0.81, + "content": "G", + "type": "inline_equation" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 625, + 506, + 643 + ], + "spans": [ + { + "bbox": [ + 104, + 625, + 259, + 643 + ], + "score": 1.0, + "content": "is mapped to the identity mapping on", + "type": "text" + }, + { + "bbox": [ + 259, + 629, + 271, + 638 + ], + "score": 0.72, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 625, + 307, + 643 + ], + "score": 1.0, + "content": ". We say", + "type": "text" + }, + { + "bbox": [ + 308, + 630, + 315, + 639 + ], + "score": 0.82, + "content": "\\rho", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 625, + 403, + 643 + ], + "score": 1.0, + "content": "is a representation of", + "type": "text" + }, + { + "bbox": [ + 403, + 628, + 413, + 638 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 625, + 423, + 643 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 423, + 628, + 442, + 640 + ], + "score": 0.92, + "content": "\\rho ( g )", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 625, + 506, + 643 + ], + "score": 1.0, + "content": "is a linear map", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 638, + 507, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 145, + 653 + ], + "score": 1.0, + "content": "for every", + "type": "text" + }, + { + "bbox": [ + 145, + 639, + 172, + 651 + ], + "score": 0.9, + "content": "g \\in G", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 638, + 453, + 653 + ], + "score": 1.0, + "content": ". As is customary, when it does not cause confusion we often say that", + "type": "text" + }, + { + "bbox": [ + 453, + 639, + 465, + 649 + ], + "score": 0.8, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 638, + 507, + 653 + ], + "score": 1.0, + "content": "itself is a", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 649, + 192, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 176, + 662 + ], + "score": 1.0, + "content": "representation of", + "type": "text" + }, + { + "bbox": [ + 177, + 650, + 186, + 660 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 649, + 192, + 662 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 105, + 666, + 504, + 689 + ], + "lines": [ + { + "bbox": [ + 105, + 666, + 504, + 679 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 491, + 679 + ], + "score": 1.0, + "content": "In this paper, we are interested in functions on point clouds. Point clouds are sets of vectors in", + "type": "text" + }, + { + "bbox": [ + 491, + 666, + 504, + 677 + ], + "score": 0.87, + "content": "\\mathbb { R } ^ { 3 }", + "type": "inline_equation" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 678, + 194, + 691 + ], + "spans": [ + { + "bbox": [ + 106, + 678, + 194, + 691 + ], + "score": 1.0, + "content": "arranged as matrices:", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5 + }, + { + "type": "interline_equation", + "bbox": [ + 249, + 689, + 362, + 704 + ], + "lines": [ + { + "bbox": [ + 249, + 689, + 362, + 704 + ], + "spans": [ + { + "bbox": [ + 249, + 689, + 362, + 704 + ], + "score": 0.91, + "content": "X = ( x _ { 1 } , \\ldots , x _ { n } ) \\in \\mathbb { R } ^ { 3 \\times n } .", + "type": "interline_equation", + "image_path": "2b2080272f92a716470478cabee758a436313e049083470d0833964e2e318b44.jpg" + } + ] + } + ], + "index": 45, + "virtual_lines": [ + { + "bbox": [ + 249, + 689, + 362, + 704 + ], + "spans": [], + "index": 45 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Many machine learning tasks on point clouds, such as classification, aim to learn a function which", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "score": 1.0, + "content": "is invariant to rigid motions and relabeling of the points. Put differently, such functions are required", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 46.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 105, + 25, + 293, + 39 + ], + "spans": [ + { + "bbox": [ + 105, + 25, + 293, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 504, + 138 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "We use our general discussion to prove the universality of two additional equivariant models: the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "first is a simple modification of the TFN architecture which allows for universality using only low", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 103, + 506, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 506, + 118 + ], + "score": 1.0, + "content": "dimensional filters. The second is a minimal architecture which is based on tensor product represen-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 115, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 115, + 506, + 128 + ], + "score": 1.0, + "content": "tations, rather than the more commonly used irreducible representations of SO(3). We discuss the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 308, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 308, + 139 + ], + "score": 1.0, + "content": "advantages and disadvantages of both approaches.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 82, + 506, + 139 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 143, + 504, + 187 + ], + "lines": [ + { + "bbox": [ + 105, + 143, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 143, + 505, + 156 + ], + "score": 1.0, + "content": "To summarize, the contributions of this paper are: (1) A general approach for proving the universal-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 154, + 505, + 166 + ], + "spans": [ + { + "bbox": [ + 106, + 154, + 505, + 166 + ], + "score": 1.0, + "content": "ity of rotation equivariant models for point clouds; (2) A proof that two recent equivariant models", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 164, + 506, + 178 + ], + "spans": [ + { + "bbox": [ + 105, + 164, + 506, + 178 + ], + "score": 1.0, + "content": "(Thomas et al., 2018; Fuchs et al., 2020) are universal; (3) Two additional simple and novel universal", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 177, + 163, + 187 + ], + "spans": [ + { + "bbox": [ + 106, + 177, + 163, + 187 + ], + "score": 1.0, + "content": "architectures.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5, + "bbox_fs": [ + 105, + 143, + 506, + 187 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 205, + 213, + 218 + ], + "lines": [ + { + "bbox": [ + 105, + 205, + 215, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 205, + 215, + 221 + ], + "score": 1.0, + "content": "2 PREVIOUS WORK", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 106, + 232, + 505, + 364 + ], + "lines": [ + { + "bbox": [ + 105, + 232, + 505, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 232, + 505, + 245 + ], + "score": 1.0, + "content": "Deep learning on point clouds. (Qi et al., 2017a; Zaheer et al., 2017) were the first to apply neural", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 104, + 243, + 506, + 257 + ], + "spans": [ + { + "bbox": [ + 104, + 243, + 506, + 257 + ], + "score": 1.0, + "content": "networks directly to the raw point cloud data, by using pointwise functions and pooling operations.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 254, + 505, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 505, + 267 + ], + "score": 1.0, + "content": "Many subsequent works used local neighborhood information (Qi et al., 2017b; Wang et al., 2019b;", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 264, + 505, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 505, + 277 + ], + "score": 1.0, + "content": "Atzmon et al., 2018). We refer the reader to a recent survey for more details (Guo et al., 2020).", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 275, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 505, + 289 + ], + "score": 1.0, + "content": "In contrast with the aforementioned works which focused solely on permutation invariance, more", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 287, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 287, + 505, + 299 + ], + "score": 1.0, + "content": "related to this paper are works that additionally incorporated invariance to rigid motions. (Thomas", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 298, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 505, + 311 + ], + "score": 1.0, + "content": "et al., 2018) proposed Tensor Field Networks (TFN) and showed their efficacy on physics and chem-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 309, + 505, + 321 + ], + "spans": [ + { + "bbox": [ + 105, + 309, + 505, + 321 + ], + "score": 1.0, + "content": "istry tasks.(Kondor et al., 2018b) also suggested an equivariant model for continuous rotations. (Li", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 320, + 505, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 505, + 333 + ], + "score": 1.0, + "content": "et al., 2019) suggested models that are equivariant to discrete subgroups of SO(3). (Poulenard et al.,", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 331, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 106, + 331, + 505, + 343 + ], + "score": 1.0, + "content": "2019) suggested an invariant model based on spherical harmonics. (Fuchs et al., 2020) followed TFN", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 342, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 505, + 354 + ], + "score": 1.0, + "content": "and added an attention mechanism. Recently, (Zhao et al., 2019) proposed a quaternion equivariant", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 353, + 410, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 410, + 365 + ], + "score": 1.0, + "content": "point capsule network that also achieves rotation and translation invariance.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 15.5, + "bbox_fs": [ + 104, + 232, + 506, + 365 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 370, + 505, + 502 + ], + "lines": [ + { + "bbox": [ + 106, + 370, + 505, + 383 + ], + "spans": [ + { + "bbox": [ + 106, + 370, + 505, + 383 + ], + "score": 1.0, + "content": "Universal approximation for invariant networks. Understanding the approximation power of in-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 380, + 505, + 394 + ], + "spans": [ + { + "bbox": [ + 105, + 380, + 505, + 394 + ], + "score": 1.0, + "content": "variant models is a popular research goal. Most of the current results assume that the symmetry", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 392, + 506, + 405 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 349, + 405 + ], + "score": 1.0, + "content": "group is a permutation group. (Zaheer et al., 2017; Qi et al.,", + "type": "text" + }, + { + "bbox": [ + 349, + 392, + 375, + 402 + ], + "score": 0.44, + "content": "2 0 1 7 \\mathrm { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 392, + 506, + 405 + ], + "score": 1.0, + "content": "; Segol & Lipman, 2019; Maron", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 403, + 505, + 415 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 387, + 415 + ], + "score": 1.0, + "content": "et al., 2020; Serviansky et al., 2020) proved universality for several", + "type": "text" + }, + { + "bbox": [ + 387, + 403, + 400, + 414 + ], + "score": 0.89, + "content": "S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 403, + 505, + 415 + ], + "score": 1.0, + "content": "-invariant and equivariant", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 412, + 505, + 427 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 505, + 427 + ], + "score": 1.0, + "content": "models. (Maron et al., 2019b;a; Keriven & Peyre, 2019; Maehara & NT, 2019) studied the approx- ´", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 424, + 504, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 504, + 437 + ], + "score": 1.0, + "content": "imation power of high-order graph neural networks. (Maron et al., 2019c; Ravanbakhsh, 2020) tar-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 435, + 505, + 449 + ], + "spans": [ + { + "bbox": [ + 104, + 435, + 505, + 449 + ], + "score": 1.0, + "content": "geted universality of networks that use high-order representations for permutation groups(Yarotsky,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 446, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 505, + 459 + ], + "score": 1.0, + "content": "2018) provided several theoretical constructions of universal equivariant neural network models", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 457, + 505, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 286, + 470 + ], + "score": 1.0, + "content": "based on polynomial invariants, including an", + "type": "text" + }, + { + "bbox": [ + 286, + 457, + 315, + 469 + ], + "score": 0.92, + "content": "S E ( 2 )", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 457, + 505, + 470 + ], + "score": 1.0, + "content": "equivariant model. In a recent work (Bogatskiy", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 468, + 506, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 468, + 506, + 482 + ], + "score": 1.0, + "content": "et al., 2020) presented a universal approximation theorem for networks that are equivariant to sev-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 479, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 208, + 493 + ], + "score": 1.0, + "content": "eral Lie groups including", + "type": "text" + }, + { + "bbox": [ + 208, + 479, + 236, + 491 + ], + "score": 0.77, + "content": "S O ( 3 )", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 479, + 505, + 493 + ], + "score": 1.0, + "content": ". The main difference from our paper is that we prove a universality", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 491, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 505, + 503 + ], + "score": 1.0, + "content": "theorem for a more complex group that besides rotations also includes translations and permutations.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 27.5, + "bbox_fs": [ + 104, + 370, + 506, + 503 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 521, + 357, + 532 + ], + "lines": [ + { + "bbox": [ + 104, + 519, + 359, + 535 + ], + "spans": [ + { + "bbox": [ + 104, + 519, + 359, + 535 + ], + "score": 1.0, + "content": "3 A FRAMEWORK FOR PROVING UNIVERSALITY", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 107, + 546, + 504, + 569 + ], + "lines": [ + { + "bbox": [ + 105, + 545, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 505, + 559 + ], + "score": 1.0, + "content": "In this section, we describe a framework for proving the universality of equivariant networks. We", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 558, + 289, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 289, + 569 + ], + "score": 1.0, + "content": "begin with some mathematical preliminaries:", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5, + "bbox_fs": [ + 105, + 545, + 505, + 569 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 584, + 232, + 596 + ], + "lines": [ + { + "bbox": [ + 106, + 584, + 233, + 597 + ], + "spans": [ + { + "bbox": [ + 106, + 584, + 233, + 597 + ], + "score": 1.0, + "content": "3.1 MATHEMATICAL SETUP", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 107, + 605, + 505, + 661 + ], + "lines": [ + { + "bbox": [ + 105, + 604, + 506, + 619 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 194, + 619 + ], + "score": 1.0, + "content": "An action of a group", + "type": "text" + }, + { + "bbox": [ + 194, + 606, + 203, + 616 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 604, + 295, + 619 + ], + "score": 1.0, + "content": "on a real vector space", + "type": "text" + }, + { + "bbox": [ + 295, + 606, + 307, + 616 + ], + "score": 0.74, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 604, + 403, + 619 + ], + "score": 1.0, + "content": "is a collection of maps", + "type": "text" + }, + { + "bbox": [ + 403, + 606, + 471, + 618 + ], + "score": 0.91, + "content": "\\rho ( g ) : W \\to W", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 604, + 506, + 619 + ], + "score": 1.0, + "content": "defined", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 616, + 504, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 137, + 630 + ], + "score": 1.0, + "content": "for any", + "type": "text" + }, + { + "bbox": [ + 138, + 617, + 165, + 629 + ], + "score": 0.9, + "content": "g \\in G", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 616, + 207, + 630 + ], + "score": 1.0, + "content": ", such that", + "type": "text" + }, + { + "bbox": [ + 208, + 617, + 307, + 629 + ], + "score": 0.91, + "content": "\\rho ( g _ { 1 } ) \\circ \\rho ( g _ { 2 } ) = \\rho ( g _ { 1 } g _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 616, + 335, + 630 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 336, + 617, + 380, + 629 + ], + "score": 0.92, + "content": "g _ { 1 } , g _ { 2 } \\in G", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 616, + 495, + 630 + ], + "score": 1.0, + "content": ", and the identity element of", + "type": "text" + }, + { + "bbox": [ + 495, + 617, + 504, + 627 + ], + "score": 0.81, + "content": "G", + "type": "inline_equation" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 625, + 506, + 643 + ], + "spans": [ + { + "bbox": [ + 104, + 625, + 259, + 643 + ], + "score": 1.0, + "content": "is mapped to the identity mapping on", + "type": "text" + }, + { + "bbox": [ + 259, + 629, + 271, + 638 + ], + "score": 0.72, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 625, + 307, + 643 + ], + "score": 1.0, + "content": ". We say", + "type": "text" + }, + { + "bbox": [ + 308, + 630, + 315, + 639 + ], + "score": 0.82, + "content": "\\rho", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 625, + 403, + 643 + ], + "score": 1.0, + "content": "is a representation of", + "type": "text" + }, + { + "bbox": [ + 403, + 628, + 413, + 638 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 625, + 423, + 643 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 423, + 628, + 442, + 640 + ], + "score": 0.92, + "content": "\\rho ( g )", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 625, + 506, + 643 + ], + "score": 1.0, + "content": "is a linear map", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 638, + 507, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 145, + 653 + ], + "score": 1.0, + "content": "for every", + "type": "text" + }, + { + "bbox": [ + 145, + 639, + 172, + 651 + ], + "score": 0.9, + "content": "g \\in G", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 638, + 453, + 653 + ], + "score": 1.0, + "content": ". As is customary, when it does not cause confusion we often say that", + "type": "text" + }, + { + "bbox": [ + 453, + 639, + 465, + 649 + ], + "score": 0.8, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 638, + 507, + 653 + ], + "score": 1.0, + "content": "itself is a", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 649, + 192, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 176, + 662 + ], + "score": 1.0, + "content": "representation of", + "type": "text" + }, + { + "bbox": [ + 177, + 650, + 186, + 660 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 649, + 192, + 662 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40, + "bbox_fs": [ + 104, + 604, + 507, + 662 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 666, + 504, + 689 + ], + "lines": [ + { + "bbox": [ + 105, + 666, + 504, + 679 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 491, + 679 + ], + "score": 1.0, + "content": "In this paper, we are interested in functions on point clouds. Point clouds are sets of vectors in", + "type": "text" + }, + { + "bbox": [ + 491, + 666, + 504, + 677 + ], + "score": 0.87, + "content": "\\mathbb { R } ^ { 3 }", + "type": "inline_equation" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 678, + 194, + 691 + ], + "spans": [ + { + "bbox": [ + 106, + 678, + 194, + 691 + ], + "score": 1.0, + "content": "arranged as matrices:", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5, + "bbox_fs": [ + 105, + 666, + 504, + 691 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 249, + 689, + 362, + 704 + ], + "lines": [ + { + "bbox": [ + 249, + 689, + 362, + 704 + ], + "spans": [ + { + "bbox": [ + 249, + 689, + 362, + 704 + ], + "score": 0.91, + "content": "X = ( x _ { 1 } , \\ldots , x _ { n } ) \\in \\mathbb { R } ^ { 3 \\times n } .", + "type": "interline_equation", + "image_path": "2b2080272f92a716470478cabee758a436313e049083470d0833964e2e318b44.jpg" + } + ] + } + ], + "index": 45, + "virtual_lines": [ + { + "bbox": [ + 249, + 689, + 362, + 704 + ], + "spans": [], + "index": 45 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Many machine learning tasks on point clouds, such as classification, aim to learn a function which", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 505, + 732 + ], + "score": 1.0, + "content": "is invariant to rigid motions and relabeling of the points. Put differently, such functions are required", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 46.5, + "bbox_fs": [ + 105, + 709, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 378, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 379, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 228, + 96 + ], + "score": 1.0, + "content": "to be invariant to the action of", + "type": "text" + }, + { + "bbox": [ + 229, + 82, + 324, + 95 + ], + "score": 0.93, + "content": "G = \\mathbb { R } ^ { 3 } \\rtimes \\mathrm { S O ( 3 ) } \\times S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 80, + 338, + 96 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 338, + 82, + 362, + 93 + ], + "score": 0.89, + "content": "\\mathbb { R } ^ { 3 \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 80, + 379, + 96 + ], + "score": 1.0, + "content": "via", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "interline_equation", + "bbox": [ + 231, + 96, + 378, + 110 + ], + "lines": [ + { + "bbox": [ + 231, + 96, + 378, + 110 + ], + "spans": [ + { + "bbox": [ + 231, + 96, + 378, + 110 + ], + "score": 0.86, + "content": "\\rho _ { G } ( t , R , P ) ( X ) = R ( X - t 1 _ { n } ^ { T } ) P ^ { T } ,", + "type": "interline_equation", + "image_path": "80e403e3b0b33933be7a0061a3d3d29a6dfa55bc07f442f7a89f4af8dbd9fd6f.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 231, + 96, + 378, + 110 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 111, + 432, + 123 + ], + "lines": [ + { + "bbox": [ + 106, + 109, + 432, + 124 + ], + "spans": [ + { + "bbox": [ + 106, + 109, + 133, + 124 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 110, + 162, + 121 + ], + "score": 0.91, + "content": "t \\in \\mathbb { R } ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 109, + 248, + 124 + ], + "score": 1.0, + "content": "defines a translation,", + "type": "text" + }, + { + "bbox": [ + 248, + 111, + 257, + 121 + ], + "score": 0.84, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 109, + 324, + 124 + ], + "score": 1.0, + "content": "is a rotation and", + "type": "text" + }, + { + "bbox": [ + 325, + 111, + 334, + 121 + ], + "score": 0.84, + "content": "P", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 109, + 432, + 124 + ], + "score": 1.0, + "content": "is a permutation matrix.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 107, + 127, + 505, + 160 + ], + "lines": [ + { + "bbox": [ + 105, + 128, + 504, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 128, + 378, + 140 + ], + "score": 1.0, + "content": "Equivariant functions are generalizations of invariant functions: If", + "type": "text" + }, + { + "bbox": [ + 379, + 128, + 388, + 138 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 128, + 421, + 140 + ], + "score": 1.0, + "content": "acts on", + "type": "text" + }, + { + "bbox": [ + 421, + 128, + 437, + 139 + ], + "score": 0.9, + "content": "W _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 128, + 504, + 140 + ], + "score": 1.0, + "content": "via some action", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 138, + 505, + 152 + ], + "spans": [ + { + "bbox": [ + 106, + 138, + 130, + 151 + ], + "score": 0.91, + "content": "\\bar { \\rho _ { 1 } ( g ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 130, + 138, + 167, + 152 + ], + "score": 1.0, + "content": ", and on", + "type": "text" + }, + { + "bbox": [ + 167, + 139, + 182, + 150 + ], + "score": 0.9, + "content": "W _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 138, + 303, + 152 + ], + "score": 1.0, + "content": "via some other group action", + "type": "text" + }, + { + "bbox": [ + 303, + 138, + 326, + 151 + ], + "score": 0.92, + "content": "\\rho _ { 2 } ( g )", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 138, + 427, + 152 + ], + "score": 1.0, + "content": ", we say that a function", + "type": "text" + }, + { + "bbox": [ + 428, + 139, + 493, + 150 + ], + "score": 0.92, + "content": "f : W _ { 1 } \\to W _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 138, + 505, + 152 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 150, + 165, + 161 + ], + "spans": [ + { + "bbox": [ + 105, + 150, + 165, + 161 + ], + "score": 1.0, + "content": "equivariant if", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + }, + { + "type": "interline_equation", + "bbox": [ + 209, + 160, + 354, + 173 + ], + "lines": [], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 209, + 160, + 354, + 173 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 173, + 503, + 185 + ], + "lines": [ + { + "bbox": [ + 106, + 171, + 501, + 186 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 330, + 186 + ], + "score": 1.0, + "content": "Invariant functions correspond to the special case where", + "type": "text" + }, + { + "bbox": [ + 331, + 173, + 354, + 185 + ], + "score": 0.92, + "content": "\\rho _ { 2 } ( g )", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 171, + 475, + 186 + ], + "score": 1.0, + "content": "is the identity mapping for all", + "type": "text" + }, + { + "bbox": [ + 475, + 173, + 501, + 185 + ], + "score": 0.9, + "content": "g \\in G", + "type": "inline_equation" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 106, + 190, + 505, + 256 + ], + "lines": [ + { + "bbox": [ + 106, + 190, + 505, + 202 + ], + "spans": [ + { + "bbox": [ + 106, + 190, + 505, + 202 + ], + "score": 1.0, + "content": "In some machine learning tasks on point clouds, the functions learned are not invariant but rather", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 201, + 506, + 214 + ], + "spans": [ + { + "bbox": [ + 105, + 201, + 506, + 214 + ], + "score": 1.0, + "content": "equivariant. For example, segmentation tasks assign a discrete label to each point. They are invariant", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 211, + 506, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 211, + 506, + 225 + ], + "score": 1.0, + "content": "to translations and rotations but equivariant to permutations – in the sense that permuting the input", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 224, + 505, + 235 + ], + "spans": [ + { + "bbox": [ + 106, + 224, + 505, + 235 + ], + "score": 1.0, + "content": "causes a corresponding permutation of the output. Another example is predicting a normal for each", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 233, + 506, + 246 + ], + "spans": [ + { + "bbox": [ + 104, + 233, + 506, + 246 + ], + "score": 1.0, + "content": "point of a point cloud. This task is invariant to translations but equivariant to both rotations and", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 246, + 163, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 163, + 257 + ], + "score": 1.0, + "content": "permutations.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 106, + 261, + 505, + 361 + ], + "lines": [ + { + "bbox": [ + 105, + 261, + 506, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 456, + 275 + ], + "score": 1.0, + "content": "In this paper, we are interested in learning equivariant functions from point clouds into", + "type": "text" + }, + { + "bbox": [ + 457, + 262, + 474, + 274 + ], + "score": 0.9, + "content": "W _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 261, + 506, + 275 + ], + "score": 1.0, + "content": ", where", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 107, + 273, + 505, + 285 + ], + "spans": [ + { + "bbox": [ + 107, + 273, + 123, + 284 + ], + "score": 0.89, + "content": "W _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 273, + 505, + 285 + ], + "score": 1.0, + "content": "is some representation of SO(3). The equivariance of these functions is with respect to the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 283, + 506, + 298 + ], + "spans": [ + { + "bbox": [ + 105, + 283, + 133, + 298 + ], + "score": 1.0, + "content": "action", + "type": "text" + }, + { + "bbox": [ + 133, + 285, + 146, + 295 + ], + "score": 0.85, + "content": "\\rho _ { G }", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 283, + 368, + 298 + ], + "score": 1.0, + "content": "on point clouds defined in equation 1, and the action of", + "type": "text" + }, + { + "bbox": [ + 368, + 284, + 377, + 294 + ], + "score": 0.81, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 283, + 390, + 298 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 390, + 284, + 408, + 296 + ], + "score": 0.9, + "content": "W _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 283, + 506, + 298 + ], + "score": 1.0, + "content": "defined by applying the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 294, + 505, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 505, + 307 + ], + "score": 1.0, + "content": "rotation action from the left and permutation action from the right as in 1, but ‘ignoring’ the trans-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 306, + 505, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 208, + 318 + ], + "score": 1.0, + "content": "lation component. Thus,", + "type": "text" + }, + { + "bbox": [ + 209, + 306, + 218, + 316 + ], + "score": 0.81, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 306, + 505, + 318 + ], + "score": 1.0, + "content": "-equivariant functions will be translation invariant. This formulation of", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 316, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 316, + 365, + 329 + ], + "score": 1.0, + "content": "equivariance includes the normals prediction example by taking", + "type": "text" + }, + { + "bbox": [ + 365, + 316, + 407, + 327 + ], + "score": 0.92, + "content": "W _ { T } = \\mathbb { R } ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 316, + 505, + 329 + ], + "score": 1.0, + "content": ", as well as the segmen-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 327, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 106, + 327, + 193, + 340 + ], + "score": 1.0, + "content": "tation case by setting", + "type": "text" + }, + { + "bbox": [ + 194, + 327, + 232, + 339 + ], + "score": 0.92, + "content": "W _ { T } = \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 327, + 505, + 340 + ], + "score": 1.0, + "content": "with the trivial identity representation. We focus on the harder case", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 338, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 176, + 352 + ], + "score": 1.0, + "content": "of functions into", + "type": "text" + }, + { + "bbox": [ + 176, + 339, + 194, + 350 + ], + "score": 0.94, + "content": "W _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 338, + 505, + 352 + ], + "score": 1.0, + "content": "which are equivariant to permutations, since it easily implies the easier case", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 349, + 275, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 350, + 255, + 362 + ], + "score": 1.0, + "content": "of permutation invariant functions to", + "type": "text" + }, + { + "bbox": [ + 255, + 349, + 272, + 360 + ], + "score": 0.9, + "content": "W _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 350, + 275, + 362 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 365, + 503, + 391 + ], + "lines": [ + { + "bbox": [ + 104, + 363, + 506, + 380 + ], + "spans": [ + { + "bbox": [ + 104, + 363, + 231, + 380 + ], + "score": 1.0, + "content": "Notation. We use the notation", + "type": "text" + }, + { + "bbox": [ + 231, + 366, + 293, + 378 + ], + "score": 0.93, + "content": " { \\mathbb { N } } _ { + } = { \\mathbb { N } } \\cup \\lbrace 0 \\rbrace", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 363, + 311, + 380 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 311, + 366, + 378, + 380 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\mathbb { N } _ { + } ^ { * } = \\bigcup _ { r \\in \\mathbb { N } } \\mathbb { N } _ { + } ^ { r } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 363, + 412, + 380 + ], + "score": 1.0, + "content": ". We set", + "type": "text" + }, + { + "bbox": [ + 413, + 366, + 487, + 378 + ], + "score": 0.93, + "content": "[ D ] = \\{ 1 , \\dots , D \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 363, + 506, + 380 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 378, + 189, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 185, + 391 + ], + "score": 0.91, + "content": "[ D ] _ { 0 } = \\{ 0 , \\dots , D \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 378, + 189, + 392 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 107, + 395, + 397, + 407 + ], + "lines": [ + { + "bbox": [ + 105, + 394, + 398, + 409 + ], + "spans": [ + { + "bbox": [ + 105, + 394, + 398, + 409 + ], + "score": 1.0, + "content": "Proofs. Proofs appear in the appendices, arranged according to sections.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "title", + "bbox": [ + 109, + 419, + 272, + 430 + ], + "lines": [ + { + "bbox": [ + 106, + 419, + 274, + 432 + ], + "spans": [ + { + "bbox": [ + 106, + 419, + 274, + 432 + ], + "score": 1.0, + "content": "3.2 CONDITIONS FOR UNIVERSALITY", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 106, + 439, + 505, + 517 + ], + "lines": [ + { + "bbox": [ + 105, + 439, + 506, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 506, + 453 + ], + "score": 1.0, + "content": "The semi-lifted approach In general, highly expressive equivariant neural networks can be", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 451, + 505, + 463 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 505, + 463 + ], + "score": 1.0, + "content": "achieved by using a ‘lifted approach’, where intermediate features in the network belong to high", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 462, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 106, + 462, + 470, + 474 + ], + "score": 1.0, + "content": "dimensional representations of the group. In the context of point clouds where typically", + "type": "text" + }, + { + "bbox": [ + 470, + 462, + 501, + 473 + ], + "score": 0.89, + "content": "n \\gg 3", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 462, + 505, + 474 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 473, + 505, + 485 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 505, + 485 + ], + "score": 1.0, + "content": "many papers, e.g., (Thomas et al., 2018; Kondor, 2018; Bogatskiy et al., 2020) use a ‘semi-lifted’", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 483, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 505, + 497 + ], + "score": 1.0, + "content": "approach, where hidden layers hold only higher dimensional representations of SO(3), but not high", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 494, + 505, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 494, + 505, + 508 + ], + "score": 1.0, + "content": "order permutation representations. In this subsection, we propose a strategy for achieving universal-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 506, + 241, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 241, + 518 + ], + "score": 1.0, + "content": "ity with the semi-lifted approach.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 106, + 523, + 505, + 578 + ], + "lines": [ + { + "bbox": [ + 105, + 521, + 506, + 535 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 506, + 535 + ], + "score": 1.0, + "content": "We begin by an axiomatic formulation of the semi-lifted approach (see illustration", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 532, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 104, + 532, + 505, + 546 + ], + "score": 1.0, + "content": "in inset): we assume that our neural networks are composed of two main compo-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 543, + 506, + 558 + ], + "spans": [ + { + "bbox": [ + 104, + 543, + 300, + 558 + ], + "score": 1.0, + "content": "nents: the first component is a family", + "type": "text" + }, + { + "bbox": [ + 300, + 545, + 322, + 556 + ], + "score": 0.88, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 543, + 447, + 558 + ], + "score": 1.0, + "content": "of parametric continuous", + "type": "text" + }, + { + "bbox": [ + 447, + 545, + 456, + 555 + ], + "score": 0.78, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 543, + 506, + 558 + ], + "score": 1.0, + "content": "-equivariant", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 104, + 553, + 507, + 569 + ], + "spans": [ + { + "bbox": [ + 104, + 553, + 150, + 569 + ], + "score": 1.0, + "content": "functions", + "type": "text" + }, + { + "bbox": [ + 151, + 556, + 172, + 567 + ], + "score": 0.83, + "content": "f _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 553, + 345, + 569 + ], + "score": 1.0, + "content": "which map the original point cloud", + "type": "text" + }, + { + "bbox": [ + 346, + 555, + 370, + 566 + ], + "score": 0.89, + "content": "\\mathbb { R } ^ { 3 \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 553, + 507, + 569 + ], + "score": 1.0, + "content": "to a semi-lifted point cloud", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 561, + 507, + 584 + ], + "spans": [ + { + "bbox": [ + 106, + 567, + 205, + 579 + ], + "score": 0.92, + "content": "\\begin{array} { c c l } { W _ { \\mathrm { f e a t } } ^ { n } } & { = } & { \\bigoplus _ { i = 1 } ^ { n } W _ { \\mathrm { f e a t } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 561, + 243, + 584 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 243, + 567, + 268, + 578 + ], + "score": 0.9, + "content": "W _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 561, + 507, + 584 + ], + "score": 1.0, + "content": "is a lifted (i.e., high-order) representation of SO(3).", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 107, + 579, + 297, + 700 + ], + "lines": [ + { + "bbox": [ + 106, + 577, + 297, + 590 + ], + "spans": [ + { + "bbox": [ + 106, + 577, + 297, + 590 + ], + "score": 1.0, + "content": "The second component is a family of paramet-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 586, + 298, + 602 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 269, + 602 + ], + "score": 1.0, + "content": "ric linear SO(3)-equivariant functions", + "type": "text" + }, + { + "bbox": [ + 269, + 588, + 293, + 601 + ], + "score": 0.91, + "content": "\\mathcal { F } _ { \\mathrm { p o o l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 586, + 298, + 602 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 599, + 297, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 297, + 612 + ], + "score": 1.0, + "content": "which map from the high order representa-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 610, + 297, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 126, + 623 + ], + "score": 1.0, + "content": "tion", + "type": "text" + }, + { + "bbox": [ + 127, + 610, + 151, + 621 + ], + "score": 0.89, + "content": "W _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 610, + 297, + 623 + ], + "score": 1.0, + "content": "down to the target representation", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 107, + 621, + 297, + 633 + ], + "spans": [ + { + "bbox": [ + 107, + 622, + 123, + 632 + ], + "score": 0.89, + "content": "W _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 621, + 169, + 633 + ], + "score": 1.0, + "content": ". Each such", + "type": "text" + }, + { + "bbox": [ + 170, + 621, + 197, + 633 + ], + "score": 0.26, + "content": "\\mathrm { S O ( 3 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 621, + 282, + 633 + ], + "score": 1.0, + "content": "–equivariant function", + "type": "text" + }, + { + "bbox": [ + 282, + 622, + 290, + 631 + ], + "score": 0.53, + "content": "\\Lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 621, + 297, + 633 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 107, + 632, + 295, + 645 + ], + "spans": [ + { + "bbox": [ + 107, + 632, + 162, + 644 + ], + "score": 0.91, + "content": "W _ { \\mathrm { f e a t } } \\to W _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 632, + 246, + 645 + ], + "score": 1.0, + "content": "can be extended to a", + "type": "text" + }, + { + "bbox": [ + 246, + 632, + 295, + 644 + ], + "score": 0.91, + "content": "\\mathrm { S O } ( 3 ) \\times S _ { n }", + "type": "inline_equation" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 644, + 298, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 190, + 660 + ], + "score": 1.0, + "content": "equivariant function", + "type": "text" + }, + { + "bbox": [ + 191, + 644, + 267, + 658 + ], + "score": 0.92, + "content": "\\hat { \\Lambda } : W _ { \\mathrm { f e a t } } ^ { n } \\to W _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 644, + 298, + 660 + ], + "score": 1.0, + "content": "by ap-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 655, + 297, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 135, + 668 + ], + "score": 1.0, + "content": "plying", + "type": "text" + }, + { + "bbox": [ + 135, + 657, + 143, + 666 + ], + "score": 0.77, + "content": "\\Lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 655, + 297, + 668 + ], + "score": 1.0, + "content": "elementwise. For every positive inte-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 667, + 298, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 122, + 680 + ], + "score": 1.0, + "content": "ger", + "type": "text" + }, + { + "bbox": [ + 122, + 668, + 131, + 677 + ], + "score": 0.81, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 131, + 667, + 298, + 680 + ], + "score": 1.0, + "content": ", these two families of functions induce a", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 676, + 298, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 188, + 691 + ], + "score": 1.0, + "content": "family of functions", + "type": "text" + }, + { + "bbox": [ + 189, + 678, + 204, + 689 + ], + "score": 0.9, + "content": "\\mathcal { F } _ { C }", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 676, + 298, + 691 + ], + "score": 1.0, + "content": "obtained by summing", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 107, + 689, + 286, + 701 + ], + "spans": [ + { + "bbox": [ + 107, + 690, + 115, + 699 + ], + "score": 0.83, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 116, + 689, + 286, + 701 + ], + "score": 1.0, + "content": "different compositions of these functions:", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 44 + }, + { + "type": "image", + "bbox": [ + 319, + 581, + 469, + 683 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 319, + 581, + 469, + 683 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 319, + 581, + 469, + 683 + ], + "spans": [ + { + "bbox": [ + 319, + 581, + 469, + 683 + ], + "score": 0.259, + "type": "image", + "image_path": "b54a82d14eb02b10dbee355a8abcb92b8088b25aadb6409a1bedc597d43bf20d.jpg" + } + ] + } + ], + "index": 48.0, + "virtual_lines": [ + { + "bbox": [ + 319, + 581, + 469, + 632.0 + ], + "spans": [], + "index": 45 + }, + { + "bbox": [ + 319, + 632.0, + 469, + 683.0 + ], + "spans": [], + "index": 51 + } + ] + } + ], + "index": 48.0 + }, + { + "type": "interline_equation", + "bbox": [ + 150, + 701, + 460, + 735 + ], + "lines": [ + { + "bbox": [ + 150, + 701, + 460, + 735 + ], + "spans": [ + { + "bbox": [ + 150, + 701, + 460, + 735 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } , \\mathcal { F } _ { \\mathrm { p o o l } } ) = \\{ f | f ( X ) = \\sum _ { c = 1 } ^ { C } \\hat { \\Lambda } _ { c } ( g _ { c } ( X ) ) , \\ : ( \\Lambda _ { c } , g _ { c } ) \\in \\mathcal { F } _ { \\mathrm { p o o l } } \\times \\mathcal { F } _ { \\mathrm { f e a t } } \\} .", + "type": "interline_equation", + "image_path": "b06c45d6a0b3b6d8ed3f7c06f8d25672199ece3ea9ef5d00ecf9d267757b82bb.jpg" + } + ] + } + ], + "index": 53, + "virtual_lines": [ + { + "bbox": [ + 150, + 701, + 460, + 712.3333333333334 + ], + "spans": [], + "index": 52 + }, + { + "bbox": [ + 150, + 712.3333333333334, + 460, + 723.6666666666667 + ], + "spans": [], + "index": 53 + }, + { + "bbox": [ + 150, + 723.6666666666667, + 460, + 735.0000000000001 + ], + "spans": [], + "index": 54 + } + ] + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 378, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 379, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 228, + 96 + ], + "score": 1.0, + "content": "to be invariant to the action of", + "type": "text" + }, + { + "bbox": [ + 229, + 82, + 324, + 95 + ], + "score": 0.93, + "content": "G = \\mathbb { R } ^ { 3 } \\rtimes \\mathrm { S O ( 3 ) } \\times S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 80, + 338, + 96 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 338, + 82, + 362, + 93 + ], + "score": 0.89, + "content": "\\mathbb { R } ^ { 3 \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 80, + 379, + 96 + ], + "score": 1.0, + "content": "via", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 105, + 80, + 379, + 96 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 231, + 96, + 378, + 110 + ], + "lines": [ + { + "bbox": [ + 231, + 96, + 378, + 110 + ], + "spans": [ + { + "bbox": [ + 231, + 96, + 378, + 110 + ], + "score": 0.86, + "content": "\\rho _ { G } ( t , R , P ) ( X ) = R ( X - t 1 _ { n } ^ { T } ) P ^ { T } ,", + "type": "interline_equation", + "image_path": "80e403e3b0b33933be7a0061a3d3d29a6dfa55bc07f442f7a89f4af8dbd9fd6f.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 231, + 96, + 378, + 110 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 111, + 432, + 123 + ], + "lines": [ + { + "bbox": [ + 106, + 109, + 432, + 124 + ], + "spans": [ + { + "bbox": [ + 106, + 109, + 133, + 124 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 110, + 162, + 121 + ], + "score": 0.91, + "content": "t \\in \\mathbb { R } ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 109, + 248, + 124 + ], + "score": 1.0, + "content": "defines a translation,", + "type": "text" + }, + { + "bbox": [ + 248, + 111, + 257, + 121 + ], + "score": 0.84, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 109, + 324, + 124 + ], + "score": 1.0, + "content": "is a rotation and", + "type": "text" + }, + { + "bbox": [ + 325, + 111, + 334, + 121 + ], + "score": 0.84, + "content": "P", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 109, + 432, + 124 + ], + "score": 1.0, + "content": "is a permutation matrix.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2, + "bbox_fs": [ + 106, + 109, + 432, + 124 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 127, + 505, + 160 + ], + "lines": [ + { + "bbox": [ + 105, + 128, + 504, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 128, + 378, + 140 + ], + "score": 1.0, + "content": "Equivariant functions are generalizations of invariant functions: If", + "type": "text" + }, + { + "bbox": [ + 379, + 128, + 388, + 138 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 128, + 421, + 140 + ], + "score": 1.0, + "content": "acts on", + "type": "text" + }, + { + "bbox": [ + 421, + 128, + 437, + 139 + ], + "score": 0.9, + "content": "W _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 128, + 504, + 140 + ], + "score": 1.0, + "content": "via some action", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 138, + 505, + 152 + ], + "spans": [ + { + "bbox": [ + 106, + 138, + 130, + 151 + ], + "score": 0.91, + "content": "\\bar { \\rho _ { 1 } ( g ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 130, + 138, + 167, + 152 + ], + "score": 1.0, + "content": ", and on", + "type": "text" + }, + { + "bbox": [ + 167, + 139, + 182, + 150 + ], + "score": 0.9, + "content": "W _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 138, + 303, + 152 + ], + "score": 1.0, + "content": "via some other group action", + "type": "text" + }, + { + "bbox": [ + 303, + 138, + 326, + 151 + ], + "score": 0.92, + "content": "\\rho _ { 2 } ( g )", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 138, + 427, + 152 + ], + "score": 1.0, + "content": ", we say that a function", + "type": "text" + }, + { + "bbox": [ + 428, + 139, + 493, + 150 + ], + "score": 0.92, + "content": "f : W _ { 1 } \\to W _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 138, + 505, + 152 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 150, + 165, + 161 + ], + "spans": [ + { + "bbox": [ + 105, + 150, + 165, + 161 + ], + "score": 1.0, + "content": "equivariant if", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4, + "bbox_fs": [ + 105, + 128, + 505, + 161 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 209, + 160, + 354, + 173 + ], + "lines": [], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 209, + 160, + 354, + 173 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 173, + 503, + 185 + ], + "lines": [ + { + "bbox": [ + 106, + 171, + 501, + 186 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 330, + 186 + ], + "score": 1.0, + "content": "Invariant functions correspond to the special case where", + "type": "text" + }, + { + "bbox": [ + 331, + 173, + 354, + 185 + ], + "score": 0.92, + "content": "\\rho _ { 2 } ( g )", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 171, + 475, + 186 + ], + "score": 1.0, + "content": "is the identity mapping for all", + "type": "text" + }, + { + "bbox": [ + 475, + 173, + 501, + 185 + ], + "score": 0.9, + "content": "g \\in G", + "type": "inline_equation" + } + ], + "index": 7 + } + ], + "index": 7, + "bbox_fs": [ + 106, + 171, + 501, + 186 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 190, + 505, + 256 + ], + "lines": [ + { + "bbox": [ + 106, + 190, + 505, + 202 + ], + "spans": [ + { + "bbox": [ + 106, + 190, + 505, + 202 + ], + "score": 1.0, + "content": "In some machine learning tasks on point clouds, the functions learned are not invariant but rather", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 201, + 506, + 214 + ], + "spans": [ + { + "bbox": [ + 105, + 201, + 506, + 214 + ], + "score": 1.0, + "content": "equivariant. For example, segmentation tasks assign a discrete label to each point. They are invariant", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 211, + 506, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 211, + 506, + 225 + ], + "score": 1.0, + "content": "to translations and rotations but equivariant to permutations – in the sense that permuting the input", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 224, + 505, + 235 + ], + "spans": [ + { + "bbox": [ + 106, + 224, + 505, + 235 + ], + "score": 1.0, + "content": "causes a corresponding permutation of the output. Another example is predicting a normal for each", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 233, + 506, + 246 + ], + "spans": [ + { + "bbox": [ + 104, + 233, + 506, + 246 + ], + "score": 1.0, + "content": "point of a point cloud. This task is invariant to translations but equivariant to both rotations and", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 246, + 163, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 163, + 257 + ], + "score": 1.0, + "content": "permutations.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 10.5, + "bbox_fs": [ + 104, + 190, + 506, + 257 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 261, + 505, + 361 + ], + "lines": [ + { + "bbox": [ + 105, + 261, + 506, + 275 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 456, + 275 + ], + "score": 1.0, + "content": "In this paper, we are interested in learning equivariant functions from point clouds into", + "type": "text" + }, + { + "bbox": [ + 457, + 262, + 474, + 274 + ], + "score": 0.9, + "content": "W _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 261, + 506, + 275 + ], + "score": 1.0, + "content": ", where", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 107, + 273, + 505, + 285 + ], + "spans": [ + { + "bbox": [ + 107, + 273, + 123, + 284 + ], + "score": 0.89, + "content": "W _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 273, + 505, + 285 + ], + "score": 1.0, + "content": "is some representation of SO(3). The equivariance of these functions is with respect to the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 283, + 506, + 298 + ], + "spans": [ + { + "bbox": [ + 105, + 283, + 133, + 298 + ], + "score": 1.0, + "content": "action", + "type": "text" + }, + { + "bbox": [ + 133, + 285, + 146, + 295 + ], + "score": 0.85, + "content": "\\rho _ { G }", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 283, + 368, + 298 + ], + "score": 1.0, + "content": "on point clouds defined in equation 1, and the action of", + "type": "text" + }, + { + "bbox": [ + 368, + 284, + 377, + 294 + ], + "score": 0.81, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 283, + 390, + 298 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 390, + 284, + 408, + 296 + ], + "score": 0.9, + "content": "W _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 283, + 506, + 298 + ], + "score": 1.0, + "content": "defined by applying the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 294, + 505, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 505, + 307 + ], + "score": 1.0, + "content": "rotation action from the left and permutation action from the right as in 1, but ‘ignoring’ the trans-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 306, + 505, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 208, + 318 + ], + "score": 1.0, + "content": "lation component. Thus,", + "type": "text" + }, + { + "bbox": [ + 209, + 306, + 218, + 316 + ], + "score": 0.81, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 306, + 505, + 318 + ], + "score": 1.0, + "content": "-equivariant functions will be translation invariant. This formulation of", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 316, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 316, + 365, + 329 + ], + "score": 1.0, + "content": "equivariance includes the normals prediction example by taking", + "type": "text" + }, + { + "bbox": [ + 365, + 316, + 407, + 327 + ], + "score": 0.92, + "content": "W _ { T } = \\mathbb { R } ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 316, + 505, + 329 + ], + "score": 1.0, + "content": ", as well as the segmen-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 327, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 106, + 327, + 193, + 340 + ], + "score": 1.0, + "content": "tation case by setting", + "type": "text" + }, + { + "bbox": [ + 194, + 327, + 232, + 339 + ], + "score": 0.92, + "content": "W _ { T } = \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 327, + 505, + 340 + ], + "score": 1.0, + "content": "with the trivial identity representation. We focus on the harder case", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 338, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 176, + 352 + ], + "score": 1.0, + "content": "of functions into", + "type": "text" + }, + { + "bbox": [ + 176, + 339, + 194, + 350 + ], + "score": 0.94, + "content": "W _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 338, + 505, + 352 + ], + "score": 1.0, + "content": "which are equivariant to permutations, since it easily implies the easier case", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 349, + 275, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 350, + 255, + 362 + ], + "score": 1.0, + "content": "of permutation invariant functions to", + "type": "text" + }, + { + "bbox": [ + 255, + 349, + 272, + 360 + ], + "score": 0.9, + "content": "W _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 350, + 275, + 362 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 261, + 506, + 362 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 365, + 503, + 391 + ], + "lines": [ + { + "bbox": [ + 104, + 363, + 506, + 380 + ], + "spans": [ + { + "bbox": [ + 104, + 363, + 231, + 380 + ], + "score": 1.0, + "content": "Notation. We use the notation", + "type": "text" + }, + { + "bbox": [ + 231, + 366, + 293, + 378 + ], + "score": 0.93, + "content": " { \\mathbb { N } } _ { + } = { \\mathbb { N } } \\cup \\lbrace 0 \\rbrace", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 363, + 311, + 380 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 311, + 366, + 378, + 380 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\mathbb { N } _ { + } ^ { * } = \\bigcup _ { r \\in \\mathbb { N } } \\mathbb { N } _ { + } ^ { r } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 363, + 412, + 380 + ], + "score": 1.0, + "content": ". We set", + "type": "text" + }, + { + "bbox": [ + 413, + 366, + 487, + 378 + ], + "score": 0.93, + "content": "[ D ] = \\{ 1 , \\dots , D \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 363, + 506, + 380 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 378, + 189, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 185, + 391 + ], + "score": 0.91, + "content": "[ D ] _ { 0 } = \\{ 0 , \\dots , D \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 378, + 189, + 392 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 104, + 363, + 506, + 392 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 395, + 397, + 407 + ], + "lines": [ + { + "bbox": [ + 105, + 394, + 398, + 409 + ], + "spans": [ + { + "bbox": [ + 105, + 394, + 398, + 409 + ], + "score": 1.0, + "content": "Proofs. Proofs appear in the appendices, arranged according to sections.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25, + "bbox_fs": [ + 105, + 394, + 398, + 409 + ] + }, + { + "type": "title", + "bbox": [ + 109, + 419, + 272, + 430 + ], + "lines": [ + { + "bbox": [ + 106, + 419, + 274, + 432 + ], + "spans": [ + { + "bbox": [ + 106, + 419, + 274, + 432 + ], + "score": 1.0, + "content": "3.2 CONDITIONS FOR UNIVERSALITY", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 106, + 439, + 505, + 517 + ], + "lines": [ + { + "bbox": [ + 105, + 439, + 506, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 506, + 453 + ], + "score": 1.0, + "content": "The semi-lifted approach In general, highly expressive equivariant neural networks can be", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 451, + 505, + 463 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 505, + 463 + ], + "score": 1.0, + "content": "achieved by using a ‘lifted approach’, where intermediate features in the network belong to high", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 462, + 505, + 474 + ], + "spans": [ + { + "bbox": [ + 106, + 462, + 470, + 474 + ], + "score": 1.0, + "content": "dimensional representations of the group. In the context of point clouds where typically", + "type": "text" + }, + { + "bbox": [ + 470, + 462, + 501, + 473 + ], + "score": 0.89, + "content": "n \\gg 3", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 462, + 505, + 474 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 473, + 505, + 485 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 505, + 485 + ], + "score": 1.0, + "content": "many papers, e.g., (Thomas et al., 2018; Kondor, 2018; Bogatskiy et al., 2020) use a ‘semi-lifted’", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 483, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 505, + 497 + ], + "score": 1.0, + "content": "approach, where hidden layers hold only higher dimensional representations of SO(3), but not high", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 494, + 505, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 494, + 505, + 508 + ], + "score": 1.0, + "content": "order permutation representations. In this subsection, we propose a strategy for achieving universal-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 506, + 241, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 241, + 518 + ], + "score": 1.0, + "content": "ity with the semi-lifted approach.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 439, + 506, + 518 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 523, + 505, + 578 + ], + "lines": [ + { + "bbox": [ + 105, + 521, + 506, + 535 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 506, + 535 + ], + "score": 1.0, + "content": "We begin by an axiomatic formulation of the semi-lifted approach (see illustration", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 532, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 104, + 532, + 505, + 546 + ], + "score": 1.0, + "content": "in inset): we assume that our neural networks are composed of two main compo-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 543, + 506, + 558 + ], + "spans": [ + { + "bbox": [ + 104, + 543, + 300, + 558 + ], + "score": 1.0, + "content": "nents: the first component is a family", + "type": "text" + }, + { + "bbox": [ + 300, + 545, + 322, + 556 + ], + "score": 0.88, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 543, + 447, + 558 + ], + "score": 1.0, + "content": "of parametric continuous", + "type": "text" + }, + { + "bbox": [ + 447, + 545, + 456, + 555 + ], + "score": 0.78, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 543, + 506, + 558 + ], + "score": 1.0, + "content": "-equivariant", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 104, + 553, + 507, + 569 + ], + "spans": [ + { + "bbox": [ + 104, + 553, + 150, + 569 + ], + "score": 1.0, + "content": "functions", + "type": "text" + }, + { + "bbox": [ + 151, + 556, + 172, + 567 + ], + "score": 0.83, + "content": "f _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 553, + 345, + 569 + ], + "score": 1.0, + "content": "which map the original point cloud", + "type": "text" + }, + { + "bbox": [ + 346, + 555, + 370, + 566 + ], + "score": 0.89, + "content": "\\mathbb { R } ^ { 3 \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 553, + 507, + 569 + ], + "score": 1.0, + "content": "to a semi-lifted point cloud", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 561, + 507, + 584 + ], + "spans": [ + { + "bbox": [ + 106, + 567, + 205, + 579 + ], + "score": 0.92, + "content": "\\begin{array} { c c l } { W _ { \\mathrm { f e a t } } ^ { n } } & { = } & { \\bigoplus _ { i = 1 } ^ { n } W _ { \\mathrm { f e a t } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 561, + 243, + 584 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 243, + 567, + 268, + 578 + ], + "score": 0.9, + "content": "W _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 561, + 507, + 584 + ], + "score": 1.0, + "content": "is a lifted (i.e., high-order) representation of SO(3).", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36, + "bbox_fs": [ + 104, + 521, + 507, + 584 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 579, + 297, + 700 + ], + "lines": [ + { + "bbox": [ + 106, + 577, + 297, + 590 + ], + "spans": [ + { + "bbox": [ + 106, + 577, + 297, + 590 + ], + "score": 1.0, + "content": "The second component is a family of paramet-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 586, + 298, + 602 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 269, + 602 + ], + "score": 1.0, + "content": "ric linear SO(3)-equivariant functions", + "type": "text" + }, + { + "bbox": [ + 269, + 588, + 293, + 601 + ], + "score": 0.91, + "content": "\\mathcal { F } _ { \\mathrm { p o o l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 586, + 298, + 602 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 599, + 297, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 297, + 612 + ], + "score": 1.0, + "content": "which map from the high order representa-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 610, + 297, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 126, + 623 + ], + "score": 1.0, + "content": "tion", + "type": "text" + }, + { + "bbox": [ + 127, + 610, + 151, + 621 + ], + "score": 0.89, + "content": "W _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 610, + 297, + 623 + ], + "score": 1.0, + "content": "down to the target representation", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 107, + 621, + 297, + 633 + ], + "spans": [ + { + "bbox": [ + 107, + 622, + 123, + 632 + ], + "score": 0.89, + "content": "W _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 621, + 169, + 633 + ], + "score": 1.0, + "content": ". Each such", + "type": "text" + }, + { + "bbox": [ + 170, + 621, + 197, + 633 + ], + "score": 0.26, + "content": "\\mathrm { S O ( 3 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 621, + 282, + 633 + ], + "score": 1.0, + "content": "–equivariant function", + "type": "text" + }, + { + "bbox": [ + 282, + 622, + 290, + 631 + ], + "score": 0.53, + "content": "\\Lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 621, + 297, + 633 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 107, + 632, + 295, + 645 + ], + "spans": [ + { + "bbox": [ + 107, + 632, + 162, + 644 + ], + "score": 0.91, + "content": "W _ { \\mathrm { f e a t } } \\to W _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 632, + 246, + 645 + ], + "score": 1.0, + "content": "can be extended to a", + "type": "text" + }, + { + "bbox": [ + 246, + 632, + 295, + 644 + ], + "score": 0.91, + "content": "\\mathrm { S O } ( 3 ) \\times S _ { n }", + "type": "inline_equation" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 644, + 298, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 190, + 660 + ], + "score": 1.0, + "content": "equivariant function", + "type": "text" + }, + { + "bbox": [ + 191, + 644, + 267, + 658 + ], + "score": 0.92, + "content": "\\hat { \\Lambda } : W _ { \\mathrm { f e a t } } ^ { n } \\to W _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 644, + 298, + 660 + ], + "score": 1.0, + "content": "by ap-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 655, + 297, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 135, + 668 + ], + "score": 1.0, + "content": "plying", + "type": "text" + }, + { + "bbox": [ + 135, + 657, + 143, + 666 + ], + "score": 0.77, + "content": "\\Lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 655, + 297, + 668 + ], + "score": 1.0, + "content": "elementwise. For every positive inte-", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 667, + 298, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 122, + 680 + ], + "score": 1.0, + "content": "ger", + "type": "text" + }, + { + "bbox": [ + 122, + 668, + 131, + 677 + ], + "score": 0.81, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 131, + 667, + 298, + 680 + ], + "score": 1.0, + "content": ", these two families of functions induce a", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 676, + 298, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 188, + 691 + ], + "score": 1.0, + "content": "family of functions", + "type": "text" + }, + { + "bbox": [ + 189, + 678, + 204, + 689 + ], + "score": 0.9, + "content": "\\mathcal { F } _ { C }", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 676, + 298, + 691 + ], + "score": 1.0, + "content": "obtained by summing", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 107, + 689, + 286, + 701 + ], + "spans": [ + { + "bbox": [ + 107, + 690, + 115, + 699 + ], + "score": 0.83, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 116, + 689, + 286, + 701 + ], + "score": 1.0, + "content": "different compositions of these functions:", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 44, + "bbox_fs": [ + 105, + 577, + 298, + 701 + ] + }, + { + "type": "image", + "bbox": [ + 319, + 581, + 469, + 683 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 319, + 581, + 469, + 683 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 319, + 581, + 469, + 683 + ], + "spans": [ + { + "bbox": [ + 319, + 581, + 469, + 683 + ], + "score": 0.259, + "type": "image", + "image_path": "b54a82d14eb02b10dbee355a8abcb92b8088b25aadb6409a1bedc597d43bf20d.jpg" + } + ] + } + ], + "index": 48.0, + "virtual_lines": [ + { + "bbox": [ + 319, + 581, + 469, + 632.0 + ], + "spans": [], + "index": 45 + }, + { + "bbox": [ + 319, + 632.0, + 469, + 683.0 + ], + "spans": [], + "index": 51 + } + ] + } + ], + "index": 48.0 + }, + { + "type": "interline_equation", + "bbox": [ + 150, + 701, + 460, + 735 + ], + "lines": [ + { + "bbox": [ + 150, + 701, + 460, + 735 + ], + "spans": [ + { + "bbox": [ + 150, + 701, + 460, + 735 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } , \\mathcal { F } _ { \\mathrm { p o o l } } ) = \\{ f | f ( X ) = \\sum _ { c = 1 } ^ { C } \\hat { \\Lambda } _ { c } ( g _ { c } ( X ) ) , \\ : ( \\Lambda _ { c } , g _ { c } ) \\in \\mathcal { F } _ { \\mathrm { p o o l } } \\times \\mathcal { F } _ { \\mathrm { f e a t } } \\} .", + "type": "interline_equation", + "image_path": "b06c45d6a0b3b6d8ed3f7c06f8d25672199ece3ea9ef5d00ecf9d267757b82bb.jpg" + } + ] + } + ], + "index": 53, + "virtual_lines": [ + { + "bbox": [ + 150, + 701, + 460, + 712.3333333333334 + ], + "spans": [], + "index": 52 + }, + { + "bbox": [ + 150, + 712.3333333333334, + 460, + 723.6666666666667 + ], + "spans": [], + "index": 53 + }, + { + "bbox": [ + 150, + 723.6666666666667, + 460, + 735.0000000000001 + ], + "spans": [], + "index": 54 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 504, + 117 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "Conditions for universality We now describe two conditions that guarantee universality using the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 447, + 106 + ], + "score": 1.0, + "content": "semi-lifted approach. The first step is showing, as in (Yarotsky, 2018), that continuous", + "type": "text" + }, + { + "bbox": [ + 447, + 94, + 456, + 104 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "-equivariant", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 104, + 101, + 490, + 119 + ], + "spans": [ + { + "bbox": [ + 104, + 101, + 146, + 119 + ], + "score": 1.0, + "content": "functions", + "type": "text" + }, + { + "bbox": [ + 146, + 104, + 210, + 117 + ], + "score": 0.92, + "content": "\\mathcal { C } _ { G } \\mathrm { ( } \\mathbb { R } ^ { 3 \\times n } , W _ { T } ^ { n } \\mathrm { ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 101, + 309, + 119 + ], + "score": 1.0, + "content": "can be approximated by", + "type": "text" + }, + { + "bbox": [ + 310, + 105, + 319, + 114 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 319, + 101, + 419, + 119 + ], + "score": 1.0, + "content": "-equivariant polynomials", + "type": "text" + }, + { + "bbox": [ + 420, + 104, + 485, + 117 + ], + "score": 0.92, + "content": "\\mathcal { P } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , \\bar { W } _ { T } ^ { n } )", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 101, + 490, + 119 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 106, + 119, + 504, + 143 + ], + "lines": [ + { + "bbox": [ + 105, + 117, + 506, + 134 + ], + "spans": [ + { + "bbox": [ + 105, + 117, + 222, + 134 + ], + "score": 1.0, + "content": "Lemma 1. Any continuous", + "type": "text" + }, + { + "bbox": [ + 222, + 120, + 231, + 131 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 117, + 331, + 134 + ], + "score": 1.0, + "content": "-equivariant function in", + "type": "text" + }, + { + "bbox": [ + 331, + 119, + 395, + 132 + ], + "score": 0.88, + "content": "\\mathcal { C } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , W _ { T } ^ { n } )", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 117, + 506, + 134 + ], + "score": 1.0, + "content": "can be approximated uni-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 129, + 405, + 145 + ], + "spans": [ + { + "bbox": [ + 104, + 129, + 213, + 145 + ], + "score": 1.0, + "content": "formly on compact sets by", + "type": "text" + }, + { + "bbox": [ + 213, + 132, + 222, + 141 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 129, + 334, + 145 + ], + "score": 1.0, + "content": "-equivariant polynomials in", + "type": "text" + }, + { + "bbox": [ + 334, + 131, + 399, + 144 + ], + "score": 0.78, + "content": "\\mathcal { P } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , \\bar { W } _ { T } ^ { n } )", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 129, + 405, + 145 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 107, + 152, + 505, + 274 + ], + "lines": [ + { + "bbox": [ + 106, + 153, + 505, + 165 + ], + "spans": [ + { + "bbox": [ + 106, + 153, + 319, + 165 + ], + "score": 1.0, + "content": "Universality is now reduced to the approximation of", + "type": "text" + }, + { + "bbox": [ + 319, + 153, + 328, + 163 + ], + "score": 0.81, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 153, + 505, + 165 + ], + "score": 1.0, + "content": "-equivariant polynomials. Next, we provide", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 164, + 505, + 177 + ], + "spans": [ + { + "bbox": [ + 105, + 164, + 257, + 177 + ], + "score": 1.0, + "content": "two conditions which guarantee that", + "type": "text" + }, + { + "bbox": [ + 257, + 164, + 266, + 174 + ], + "score": 0.81, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 164, + 408, + 177 + ], + "score": 1.0, + "content": "-equivariant polynomials of degree", + "type": "text" + }, + { + "bbox": [ + 409, + 164, + 419, + 174 + ], + "score": 0.79, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 164, + 505, + 177 + ], + "score": 1.0, + "content": "can be expressed by", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 174, + 506, + 188 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 171, + 188 + ], + "score": 1.0, + "content": "function spaces", + "type": "text" + }, + { + "bbox": [ + 171, + 175, + 241, + 187 + ], + "score": 0.91, + "content": "\\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } , \\mathcal { F } _ { \\mathrm { p o o l } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 174, + 506, + 188 + ], + "score": 1.0, + "content": "as defined in equation 2. The idea behind these conditions is that", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 186, + 506, + 199 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 506, + 199 + ], + "score": 1.0, + "content": "explicit characterizations of polynomials equivariant to the joint action of translations, rotations and", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 196, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 196, + 505, + 210 + ], + "score": 1.0, + "content": "permutations is challenging. However, it is possible to explicitly characterize polynomials equiv-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 208, + 505, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 208, + 505, + 221 + ], + "score": 1.0, + "content": "ariant to translations and permutations (but not rotations). The key observation is then that this", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 103, + 215, + 507, + 235 + ], + "spans": [ + { + "bbox": [ + 103, + 215, + 347, + 235 + ], + "score": 1.0, + "content": "characterization can be rewritten as a sum of functions to", + "type": "text" + }, + { + "bbox": [ + 347, + 219, + 371, + 231 + ], + "score": 0.91, + "content": "W _ { \\mathrm { f e a t } } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 215, + 507, + 235 + ], + "score": 1.0, + "content": ", a high dimensional representa-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 230, + 505, + 241 + ], + "spans": [ + { + "bbox": [ + 106, + 230, + 505, + 241 + ], + "score": 1.0, + "content": "tions of SO(3) which is equivariant to translations, permutations and rotations, composed with a", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 240, + 506, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 240, + 506, + 253 + ], + "score": 1.0, + "content": "linear map which is permutation equivariant (but does not respect rotations). Accordingly, our first", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 104, + 250, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 104, + 250, + 176, + 264 + ], + "score": 1.0, + "content": "condition is that", + "type": "text" + }, + { + "bbox": [ + 176, + 252, + 198, + 263 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 250, + 382, + 264 + ], + "score": 1.0, + "content": "contains a spanning set of such functions to", + "type": "text" + }, + { + "bbox": [ + 382, + 252, + 406, + 264 + ], + "score": 0.9, + "content": "W _ { \\mathrm { f e a t } } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 250, + 408, + 264 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 412, + 252, + 505, + 263 + ], + "score": 1.0, + "content": "We call this condition", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 107, + 261, + 160, + 277 + ], + "spans": [ + { + "bbox": [ + 107, + 263, + 116, + 272 + ], + "score": 0.78, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 261, + 160, + 277 + ], + "score": 1.0, + "content": "-spanning:", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 277, + 505, + 312 + ], + "lines": [ + { + "bbox": [ + 102, + 271, + 508, + 295 + ], + "spans": [ + { + "bbox": [ + 102, + 271, + 163, + 295 + ], + "score": 1.0, + "content": "Definition 1 (", + "type": "text" + }, + { + "bbox": [ + 163, + 279, + 173, + 288 + ], + "score": 0.71, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 271, + 239, + 295 + ], + "score": 1.0, + "content": "-spanning). For", + "type": "text" + }, + { + "bbox": [ + 239, + 278, + 276, + 290 + ], + "score": 0.9, + "content": "D \\in { \\mathbb { N } } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 277, + 271, + 295, + 295 + ], + "score": 1.0, + "content": ", let", + "type": "text" + }, + { + "bbox": [ + 295, + 279, + 316, + 289 + ], + "score": 0.89, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 271, + 378, + 295 + ], + "score": 1.0, + "content": "be a subset of", + "type": "text" + }, + { + "bbox": [ + 379, + 277, + 449, + 290 + ], + "score": 0.89, + "content": "\\mathcal { C } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , W _ { \\mathrm { f e a t } } ^ { n } )", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 271, + 508, + 295 + ], + "score": 1.0, + "content": ". We say that", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 107, + 288, + 503, + 302 + ], + "spans": [ + { + "bbox": [ + 107, + 289, + 128, + 300 + ], + "score": 0.88, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 129, + 288, + 139, + 302 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 139, + 289, + 149, + 299 + ], + "score": 0.79, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 288, + 244, + 302 + ], + "score": 1.0, + "content": "-spanning, if there exist", + "type": "text" + }, + { + "bbox": [ + 244, + 290, + 322, + 301 + ], + "score": 0.89, + "content": "f _ { 1 } , \\dots , f _ { K } \\in \\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 288, + 437, + 302 + ], + "score": 1.0, + "content": ", such that every polynomial", + "type": "text" + }, + { + "bbox": [ + 437, + 289, + 503, + 300 + ], + "score": 0.9, + "content": "p : \\mathbb { R } ^ { 3 \\times n } \\mathbb { R } ^ { n }", + "type": "inline_equation" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 299, + 496, + 313 + ], + "spans": [ + { + "bbox": [ + 106, + 299, + 145, + 313 + ], + "score": 1.0, + "content": "of degree", + "type": "text" + }, + { + "bbox": [ + 146, + 300, + 155, + 310 + ], + "score": 0.74, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 299, + 496, + 313 + ], + "score": 1.0, + "content": "which is invariant to translations and equivariant to permutations, can be written as", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17 + }, + { + "type": "interline_equation", + "bbox": [ + 254, + 318, + 356, + 353 + ], + "lines": [ + { + "bbox": [ + 254, + 318, + 356, + 353 + ], + "spans": [ + { + "bbox": [ + 254, + 318, + 356, + 353 + ], + "score": 0.95, + "content": "p ( X ) = \\sum _ { k = 1 } ^ { K } \\hat { \\Lambda } _ { k } ( f _ { k } ( X ) ) ,", + "type": "interline_equation", + "image_path": "43e6309bd685ec48eb69a7abb94cee27f9f294cbd921a8df42aa6dc02eebdbe9.jpg" + } + ] + } + ], + "index": 19.5, + "virtual_lines": [ + { + "bbox": [ + 254, + 318, + 356, + 335.5 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 254, + 335.5, + 356, + 353.0 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 360, + 505, + 385 + ], + "lines": [ + { + "bbox": [ + 103, + 357, + 506, + 380 + ], + "spans": [ + { + "bbox": [ + 103, + 357, + 133, + 380 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 362, + 203, + 373 + ], + "score": 0.91, + "content": "\\Lambda _ { k } : W _ { \\mathrm { f e a t } } \\to \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 357, + 327, + 380 + ], + "score": 1.0, + "content": "are all linear functionals, and", + "type": "text" + }, + { + "bbox": [ + 328, + 360, + 402, + 374 + ], + "score": 0.94, + "content": "{ \\hat { \\boldsymbol { \\Lambda } } } _ { k } : { \\cal W } _ { \\mathrm { f e a t } } ^ { n } \\mathbb { R } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 357, + 506, + 380 + ], + "score": 1.0, + "content": "are the functions defined", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 372, + 248, + 386 + ], + "spans": [ + { + "bbox": [ + 106, + 372, + 231, + 386 + ], + "score": 1.0, + "content": "by elementwise applications of", + "type": "text" + }, + { + "bbox": [ + 232, + 373, + 244, + 384 + ], + "score": 0.88, + "content": "\\Lambda _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 372, + 248, + 386 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 107, + 394, + 504, + 417 + ], + "lines": [ + { + "bbox": [ + 106, + 394, + 506, + 407 + ], + "spans": [ + { + "bbox": [ + 106, + 394, + 266, + 407 + ], + "score": 1.0, + "content": "In Lemma 4 we explicitly construct a", + "type": "text" + }, + { + "bbox": [ + 266, + 395, + 276, + 405 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 394, + 506, + 407 + ], + "score": 1.0, + "content": "-spanning family of functions. This provides us with a", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 405, + 433, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 241, + 417 + ], + "score": 1.0, + "content": "concrete condition which implies", + "type": "text" + }, + { + "bbox": [ + 242, + 406, + 251, + 415 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 405, + 433, + 417 + ], + "score": 1.0, + "content": "-spanning for other function families as well.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 107, + 421, + 505, + 466 + ], + "lines": [ + { + "bbox": [ + 105, + 421, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 224, + 436 + ], + "score": 1.0, + "content": "The second condition is that", + "type": "text" + }, + { + "bbox": [ + 225, + 423, + 249, + 434 + ], + "score": 0.91, + "content": "\\mathcal { F } _ { \\mathrm { p o o l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 421, + 505, + 436 + ], + "score": 1.0, + "content": "contains all SO(3) linear equivariant layers. We call this con-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 433, + 505, + 446 + ], + "spans": [ + { + "bbox": [ + 106, + 433, + 400, + 446 + ], + "score": 1.0, + "content": "dition Linear universality. Intuitively, taking linear rotation equivariant", + "type": "text" + }, + { + "bbox": [ + 401, + 434, + 414, + 444 + ], + "score": 0.89, + "content": "\\Lambda _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 433, + 505, + 446 + ], + "score": 1.0, + "content": "in equation 3 ensures", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 444, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 105, + 444, + 216, + 457 + ], + "score": 1.0, + "content": "that the resulting function", + "type": "text" + }, + { + "bbox": [ + 216, + 446, + 223, + 456 + ], + "score": 0.79, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 444, + 400, + 457 + ], + "score": 1.0, + "content": "will be rotation equivariant and thus fully", + "type": "text" + }, + { + "bbox": [ + 400, + 445, + 409, + 454 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 444, + 505, + 457 + ], + "score": 1.0, + "content": "-equivariant, and linear", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 455, + 408, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 455, + 319, + 467 + ], + "score": 1.0, + "content": "universality guarantees the ability to express all such", + "type": "text" + }, + { + "bbox": [ + 319, + 456, + 328, + 465 + ], + "score": 0.79, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 455, + 408, + 467 + ], + "score": 1.0, + "content": "invariant functions.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 107, + 470, + 505, + 505 + ], + "lines": [ + { + "bbox": [ + 105, + 470, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 347, + 484 + ], + "score": 1.0, + "content": "Definition 2 (Linear universality). We say that a collection", + "type": "text" + }, + { + "bbox": [ + 347, + 471, + 371, + 483 + ], + "score": 0.9, + "content": "\\mathcal { F } _ { \\mathrm { p o o l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 470, + 505, + 484 + ], + "score": 1.0, + "content": "of equivariant linear functionals", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 482, + 506, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 223, + 495 + ], + "score": 1.0, + "content": "between two representations", + "type": "text" + }, + { + "bbox": [ + 223, + 482, + 248, + 493 + ], + "score": 0.9, + "content": "W _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 482, + 267, + 495 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 268, + 482, + 284, + 493 + ], + "score": 0.89, + "content": "W _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 482, + 506, + 495 + ], + "score": 1.0, + "content": "of SO(3) is linearly universal, if it contains all linear", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 492, + 358, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 358, + 507 + ], + "score": 1.0, + "content": "SO(3)-equivariant mappings between the two representations.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 514, + 504, + 537 + ], + "lines": [ + { + "bbox": [ + 105, + 513, + 506, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 506, + 528 + ], + "score": 1.0, + "content": "When these two conditions apply, a rather simple symmetrization arguments leads to the following", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 524, + 145, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 145, + 538 + ], + "score": 1.0, + "content": "theorem:", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5 + }, + { + "type": "text", + "bbox": [ + 106, + 540, + 505, + 575 + ], + "lines": [ + { + "bbox": [ + 105, + 539, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 169, + 555 + ], + "score": 1.0, + "content": "Theorem 1. If", + "type": "text" + }, + { + "bbox": [ + 169, + 542, + 191, + 552 + ], + "score": 0.9, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 539, + 201, + 555 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 201, + 542, + 210, + 551 + ], + "score": 0.78, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 539, + 269, + 555 + ], + "score": 1.0, + "content": "-spanning and", + "type": "text" + }, + { + "bbox": [ + 270, + 542, + 294, + 553 + ], + "score": 0.93, + "content": "\\mathcal { F } _ { \\mathrm { p o o l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 539, + 469, + 555 + ], + "score": 1.0, + "content": "is linearly universal, then there exists some", + "type": "text" + }, + { + "bbox": [ + 469, + 541, + 505, + 553 + ], + "score": 0.9, + "content": "C ( D ) \\in", + "type": "inline_equation" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 550, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 106, + 552, + 116, + 563 + ], + "score": 0.55, + "content": "\\mathbb { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 116, + 550, + 188, + 567 + ], + "score": 1.0, + "content": "such that for all", + "type": "text" + }, + { + "bbox": [ + 188, + 552, + 239, + 564 + ], + "score": 0.92, + "content": "C \\geq C ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 550, + 320, + 567 + ], + "score": 1.0, + "content": "the function space", + "type": "text" + }, + { + "bbox": [ + 321, + 552, + 390, + 565 + ], + "score": 0.91, + "content": "\\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } , \\mathcal { F } _ { \\mathrm { p o o l } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 550, + 445, + 567 + ], + "score": 1.0, + "content": "contains all", + "type": "text" + }, + { + "bbox": [ + 446, + 553, + 455, + 562 + ], + "score": 0.75, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 550, + 506, + 567 + ], + "score": 1.0, + "content": "-equivariant", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 563, + 220, + 575 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 196, + 575 + ], + "score": 1.0, + "content": "polynomials of degree", + "type": "text" + }, + { + "bbox": [ + 197, + 564, + 217, + 574 + ], + "score": 0.8, + "content": "\\le D", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 563, + 220, + 575 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 106, + 596, + 504, + 619 + ], + "lines": [ + { + "bbox": [ + 105, + 596, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 186, + 610 + ], + "score": 1.0, + "content": "Proof idea. By the", + "type": "text" + }, + { + "bbox": [ + 187, + 597, + 196, + 607 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 596, + 335, + 610 + ], + "score": 1.0, + "content": "-spanning assumption, there exist", + "type": "text" + }, + { + "bbox": [ + 336, + 597, + 417, + 609 + ], + "score": 0.92, + "content": "f _ { 1 } , \\dots , f _ { K } \\in { \\mathcal { F } } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 596, + 505, + 610 + ], + "score": 1.0, + "content": "such that any vector", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 608, + 464, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 608, + 464, + 621 + ], + "score": 1.0, + "content": "valued polynomial invariant to translations and equivariant to permutations is of the form", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + }, + { + "type": "interline_equation", + "bbox": [ + 254, + 626, + 356, + 661 + ], + "lines": [ + { + "bbox": [ + 254, + 626, + 356, + 661 + ], + "spans": [ + { + "bbox": [ + 254, + 626, + 356, + 661 + ], + "score": 0.95, + "content": "p ( X ) = \\sum _ { k = 1 } ^ { K } \\hat { \\Lambda } _ { k } ( f _ { k } ( X ) ) ,", + "type": "interline_equation", + "image_path": "65688a9ed2ba7a86da01fc534b78beb1d3c5369d1355607144fa4b5d092569e0.jpg" + } + ] + } + ], + "index": 39.5, + "virtual_lines": [ + { + "bbox": [ + 254, + 626, + 356, + 643.5 + ], + "spans": [], + "index": 39 + }, + { + "bbox": [ + 254, + 643.5, + 356, + 661.0 + ], + "spans": [], + "index": 40 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 666, + 505, + 690 + ], + "lines": [ + { + "bbox": [ + 105, + 666, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 279, + 680 + ], + "score": 1.0, + "content": "While by definition this holds for functions", + "type": "text" + }, + { + "bbox": [ + 279, + 670, + 285, + 679 + ], + "score": 0.8, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 666, + 349, + 680 + ], + "score": 1.0, + "content": "whose image is", + "type": "text" + }, + { + "bbox": [ + 349, + 668, + 363, + 677 + ], + "score": 0.87, + "content": "\\mathbb { R } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 666, + 505, + 680 + ], + "score": 1.0, + "content": ", this is easily extended to functions", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 678, + 169, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 117, + 691 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 117, + 678, + 134, + 690 + ], + "score": 0.91, + "content": "W _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 678, + 169, + 691 + ], + "score": 1.0, + "content": "as well.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41.5 + }, + { + "type": "text", + "bbox": [ + 105, + 694, + 504, + 718 + ], + "lines": [ + { + "bbox": [ + 105, + 695, + 505, + 708 + ], + "spans": [ + { + "bbox": [ + 105, + 695, + 221, + 708 + ], + "score": 1.0, + "content": "It remains to show that when", + "type": "text" + }, + { + "bbox": [ + 222, + 699, + 227, + 707 + ], + "score": 0.78, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 695, + 393, + 708 + ], + "score": 1.0, + "content": "is also SO(3)-equivariant, we can choose", + "type": "text" + }, + { + "bbox": [ + 393, + 696, + 406, + 706 + ], + "score": 0.9, + "content": "\\Lambda _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 695, + 505, + 708 + ], + "score": 1.0, + "content": "to be SO(3) equivariant.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 705, + 503, + 718 + ], + "spans": [ + { + "bbox": [ + 106, + 705, + 298, + 718 + ], + "score": 1.0, + "content": "This is accomplished by averaging over SO(3).", + "type": "text" + }, + { + "bbox": [ + 497, + 709, + 503, + 715 + ], + "score": 0.0, + "content": "", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 8 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 504, + 117 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "Conditions for universality We now describe two conditions that guarantee universality using the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 447, + 106 + ], + "score": 1.0, + "content": "semi-lifted approach. The first step is showing, as in (Yarotsky, 2018), that continuous", + "type": "text" + }, + { + "bbox": [ + 447, + 94, + 456, + 104 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "-equivariant", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 104, + 101, + 490, + 119 + ], + "spans": [ + { + "bbox": [ + 104, + 101, + 146, + 119 + ], + "score": 1.0, + "content": "functions", + "type": "text" + }, + { + "bbox": [ + 146, + 104, + 210, + 117 + ], + "score": 0.92, + "content": "\\mathcal { C } _ { G } \\mathrm { ( } \\mathbb { R } ^ { 3 \\times n } , W _ { T } ^ { n } \\mathrm { ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 101, + 309, + 119 + ], + "score": 1.0, + "content": "can be approximated by", + "type": "text" + }, + { + "bbox": [ + 310, + 105, + 319, + 114 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 319, + 101, + 419, + 119 + ], + "score": 1.0, + "content": "-equivariant polynomials", + "type": "text" + }, + { + "bbox": [ + 420, + 104, + 485, + 117 + ], + "score": 0.92, + "content": "\\mathcal { P } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , \\bar { W } _ { T } ^ { n } )", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 101, + 490, + 119 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 104, + 81, + 505, + 119 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 119, + 504, + 143 + ], + "lines": [ + { + "bbox": [ + 105, + 117, + 506, + 134 + ], + "spans": [ + { + "bbox": [ + 105, + 117, + 222, + 134 + ], + "score": 1.0, + "content": "Lemma 1. Any continuous", + "type": "text" + }, + { + "bbox": [ + 222, + 120, + 231, + 131 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 117, + 331, + 134 + ], + "score": 1.0, + "content": "-equivariant function in", + "type": "text" + }, + { + "bbox": [ + 331, + 119, + 395, + 132 + ], + "score": 0.88, + "content": "\\mathcal { C } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , W _ { T } ^ { n } )", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 117, + 506, + 134 + ], + "score": 1.0, + "content": "can be approximated uni-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 129, + 405, + 145 + ], + "spans": [ + { + "bbox": [ + 104, + 129, + 213, + 145 + ], + "score": 1.0, + "content": "formly on compact sets by", + "type": "text" + }, + { + "bbox": [ + 213, + 132, + 222, + 141 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 129, + 334, + 145 + ], + "score": 1.0, + "content": "-equivariant polynomials in", + "type": "text" + }, + { + "bbox": [ + 334, + 131, + 399, + 144 + ], + "score": 0.78, + "content": "\\mathcal { P } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , \\bar { W } _ { T } ^ { n } )", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 129, + 405, + 145 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 104, + 117, + 506, + 145 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 152, + 505, + 274 + ], + "lines": [ + { + "bbox": [ + 106, + 153, + 505, + 165 + ], + "spans": [ + { + "bbox": [ + 106, + 153, + 319, + 165 + ], + "score": 1.0, + "content": "Universality is now reduced to the approximation of", + "type": "text" + }, + { + "bbox": [ + 319, + 153, + 328, + 163 + ], + "score": 0.81, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 153, + 505, + 165 + ], + "score": 1.0, + "content": "-equivariant polynomials. Next, we provide", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 164, + 505, + 177 + ], + "spans": [ + { + "bbox": [ + 105, + 164, + 257, + 177 + ], + "score": 1.0, + "content": "two conditions which guarantee that", + "type": "text" + }, + { + "bbox": [ + 257, + 164, + 266, + 174 + ], + "score": 0.81, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 164, + 408, + 177 + ], + "score": 1.0, + "content": "-equivariant polynomials of degree", + "type": "text" + }, + { + "bbox": [ + 409, + 164, + 419, + 174 + ], + "score": 0.79, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 164, + 505, + 177 + ], + "score": 1.0, + "content": "can be expressed by", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 174, + 506, + 188 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 171, + 188 + ], + "score": 1.0, + "content": "function spaces", + "type": "text" + }, + { + "bbox": [ + 171, + 175, + 241, + 187 + ], + "score": 0.91, + "content": "\\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } , \\mathcal { F } _ { \\mathrm { p o o l } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 174, + 506, + 188 + ], + "score": 1.0, + "content": "as defined in equation 2. The idea behind these conditions is that", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 186, + 506, + 199 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 506, + 199 + ], + "score": 1.0, + "content": "explicit characterizations of polynomials equivariant to the joint action of translations, rotations and", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 196, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 196, + 505, + 210 + ], + "score": 1.0, + "content": "permutations is challenging. However, it is possible to explicitly characterize polynomials equiv-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 208, + 505, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 208, + 505, + 221 + ], + "score": 1.0, + "content": "ariant to translations and permutations (but not rotations). The key observation is then that this", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 103, + 215, + 507, + 235 + ], + "spans": [ + { + "bbox": [ + 103, + 215, + 347, + 235 + ], + "score": 1.0, + "content": "characterization can be rewritten as a sum of functions to", + "type": "text" + }, + { + "bbox": [ + 347, + 219, + 371, + 231 + ], + "score": 0.91, + "content": "W _ { \\mathrm { f e a t } } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 215, + 507, + 235 + ], + "score": 1.0, + "content": ", a high dimensional representa-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 230, + 505, + 241 + ], + "spans": [ + { + "bbox": [ + 106, + 230, + 505, + 241 + ], + "score": 1.0, + "content": "tions of SO(3) which is equivariant to translations, permutations and rotations, composed with a", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 240, + 506, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 240, + 506, + 253 + ], + "score": 1.0, + "content": "linear map which is permutation equivariant (but does not respect rotations). Accordingly, our first", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 104, + 250, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 104, + 250, + 176, + 264 + ], + "score": 1.0, + "content": "condition is that", + "type": "text" + }, + { + "bbox": [ + 176, + 252, + 198, + 263 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 250, + 382, + 264 + ], + "score": 1.0, + "content": "contains a spanning set of such functions to", + "type": "text" + }, + { + "bbox": [ + 382, + 252, + 406, + 264 + ], + "score": 0.9, + "content": "W _ { \\mathrm { f e a t } } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 250, + 408, + 264 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 412, + 252, + 505, + 263 + ], + "score": 1.0, + "content": "We call this condition", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 107, + 261, + 160, + 277 + ], + "spans": [ + { + "bbox": [ + 107, + 263, + 116, + 272 + ], + "score": 0.78, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 261, + 160, + 277 + ], + "score": 1.0, + "content": "-spanning:", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 10, + "bbox_fs": [ + 103, + 153, + 507, + 277 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 277, + 505, + 312 + ], + "lines": [ + { + "bbox": [ + 102, + 271, + 508, + 295 + ], + "spans": [ + { + "bbox": [ + 102, + 271, + 163, + 295 + ], + "score": 1.0, + "content": "Definition 1 (", + "type": "text" + }, + { + "bbox": [ + 163, + 279, + 173, + 288 + ], + "score": 0.71, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 271, + 239, + 295 + ], + "score": 1.0, + "content": "-spanning). For", + "type": "text" + }, + { + "bbox": [ + 239, + 278, + 276, + 290 + ], + "score": 0.9, + "content": "D \\in { \\mathbb { N } } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 277, + 271, + 295, + 295 + ], + "score": 1.0, + "content": ", let", + "type": "text" + }, + { + "bbox": [ + 295, + 279, + 316, + 289 + ], + "score": 0.89, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 271, + 378, + 295 + ], + "score": 1.0, + "content": "be a subset of", + "type": "text" + }, + { + "bbox": [ + 379, + 277, + 449, + 290 + ], + "score": 0.89, + "content": "\\mathcal { C } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , W _ { \\mathrm { f e a t } } ^ { n } )", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 271, + 508, + 295 + ], + "score": 1.0, + "content": ". We say that", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 107, + 288, + 503, + 302 + ], + "spans": [ + { + "bbox": [ + 107, + 289, + 128, + 300 + ], + "score": 0.88, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 129, + 288, + 139, + 302 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 139, + 289, + 149, + 299 + ], + "score": 0.79, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 288, + 244, + 302 + ], + "score": 1.0, + "content": "-spanning, if there exist", + "type": "text" + }, + { + "bbox": [ + 244, + 290, + 322, + 301 + ], + "score": 0.89, + "content": "f _ { 1 } , \\dots , f _ { K } \\in \\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 288, + 437, + 302 + ], + "score": 1.0, + "content": ", such that every polynomial", + "type": "text" + }, + { + "bbox": [ + 437, + 289, + 503, + 300 + ], + "score": 0.9, + "content": "p : \\mathbb { R } ^ { 3 \\times n } \\mathbb { R } ^ { n }", + "type": "inline_equation" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 299, + 496, + 313 + ], + "spans": [ + { + "bbox": [ + 106, + 299, + 145, + 313 + ], + "score": 1.0, + "content": "of degree", + "type": "text" + }, + { + "bbox": [ + 146, + 300, + 155, + 310 + ], + "score": 0.74, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 299, + 496, + 313 + ], + "score": 1.0, + "content": "which is invariant to translations and equivariant to permutations, can be written as", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17, + "bbox_fs": [ + 102, + 271, + 508, + 313 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 254, + 318, + 356, + 353 + ], + "lines": [ + { + "bbox": [ + 254, + 318, + 356, + 353 + ], + "spans": [ + { + "bbox": [ + 254, + 318, + 356, + 353 + ], + "score": 0.95, + "content": "p ( X ) = \\sum _ { k = 1 } ^ { K } \\hat { \\Lambda } _ { k } ( f _ { k } ( X ) ) ,", + "type": "interline_equation", + "image_path": "43e6309bd685ec48eb69a7abb94cee27f9f294cbd921a8df42aa6dc02eebdbe9.jpg" + } + ] + } + ], + "index": 19.5, + "virtual_lines": [ + { + "bbox": [ + 254, + 318, + 356, + 335.5 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 254, + 335.5, + 356, + 353.0 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 360, + 505, + 385 + ], + "lines": [ + { + "bbox": [ + 103, + 357, + 506, + 380 + ], + "spans": [ + { + "bbox": [ + 103, + 357, + 133, + 380 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 362, + 203, + 373 + ], + "score": 0.91, + "content": "\\Lambda _ { k } : W _ { \\mathrm { f e a t } } \\to \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 357, + 327, + 380 + ], + "score": 1.0, + "content": "are all linear functionals, and", + "type": "text" + }, + { + "bbox": [ + 328, + 360, + 402, + 374 + ], + "score": 0.94, + "content": "{ \\hat { \\boldsymbol { \\Lambda } } } _ { k } : { \\cal W } _ { \\mathrm { f e a t } } ^ { n } \\mathbb { R } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 357, + 506, + 380 + ], + "score": 1.0, + "content": "are the functions defined", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 372, + 248, + 386 + ], + "spans": [ + { + "bbox": [ + 106, + 372, + 231, + 386 + ], + "score": 1.0, + "content": "by elementwise applications of", + "type": "text" + }, + { + "bbox": [ + 232, + 373, + 244, + 384 + ], + "score": 0.88, + "content": "\\Lambda _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 372, + 248, + 386 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5, + "bbox_fs": [ + 103, + 357, + 506, + 386 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 394, + 504, + 417 + ], + "lines": [ + { + "bbox": [ + 106, + 394, + 506, + 407 + ], + "spans": [ + { + "bbox": [ + 106, + 394, + 266, + 407 + ], + "score": 1.0, + "content": "In Lemma 4 we explicitly construct a", + "type": "text" + }, + { + "bbox": [ + 266, + 395, + 276, + 405 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 394, + 506, + 407 + ], + "score": 1.0, + "content": "-spanning family of functions. This provides us with a", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 405, + 433, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 241, + 417 + ], + "score": 1.0, + "content": "concrete condition which implies", + "type": "text" + }, + { + "bbox": [ + 242, + 406, + 251, + 415 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 405, + 433, + 417 + ], + "score": 1.0, + "content": "-spanning for other function families as well.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 105, + 394, + 506, + 417 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 421, + 505, + 466 + ], + "lines": [ + { + "bbox": [ + 105, + 421, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 224, + 436 + ], + "score": 1.0, + "content": "The second condition is that", + "type": "text" + }, + { + "bbox": [ + 225, + 423, + 249, + 434 + ], + "score": 0.91, + "content": "\\mathcal { F } _ { \\mathrm { p o o l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 421, + 505, + 436 + ], + "score": 1.0, + "content": "contains all SO(3) linear equivariant layers. We call this con-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 433, + 505, + 446 + ], + "spans": [ + { + "bbox": [ + 106, + 433, + 400, + 446 + ], + "score": 1.0, + "content": "dition Linear universality. Intuitively, taking linear rotation equivariant", + "type": "text" + }, + { + "bbox": [ + 401, + 434, + 414, + 444 + ], + "score": 0.89, + "content": "\\Lambda _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 433, + 505, + 446 + ], + "score": 1.0, + "content": "in equation 3 ensures", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 444, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 105, + 444, + 216, + 457 + ], + "score": 1.0, + "content": "that the resulting function", + "type": "text" + }, + { + "bbox": [ + 216, + 446, + 223, + 456 + ], + "score": 0.79, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 444, + 400, + 457 + ], + "score": 1.0, + "content": "will be rotation equivariant and thus fully", + "type": "text" + }, + { + "bbox": [ + 400, + 445, + 409, + 454 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 444, + 505, + 457 + ], + "score": 1.0, + "content": "-equivariant, and linear", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 455, + 408, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 455, + 319, + 467 + ], + "score": 1.0, + "content": "universality guarantees the ability to express all such", + "type": "text" + }, + { + "bbox": [ + 319, + 456, + 328, + 465 + ], + "score": 0.79, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 455, + 408, + 467 + ], + "score": 1.0, + "content": "invariant functions.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 421, + 505, + 467 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 470, + 505, + 505 + ], + "lines": [ + { + "bbox": [ + 105, + 470, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 347, + 484 + ], + "score": 1.0, + "content": "Definition 2 (Linear universality). We say that a collection", + "type": "text" + }, + { + "bbox": [ + 347, + 471, + 371, + 483 + ], + "score": 0.9, + "content": "\\mathcal { F } _ { \\mathrm { p o o l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 470, + 505, + 484 + ], + "score": 1.0, + "content": "of equivariant linear functionals", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 482, + 506, + 495 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 223, + 495 + ], + "score": 1.0, + "content": "between two representations", + "type": "text" + }, + { + "bbox": [ + 223, + 482, + 248, + 493 + ], + "score": 0.9, + "content": "W _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 482, + 267, + 495 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 268, + 482, + 284, + 493 + ], + "score": 0.89, + "content": "W _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 482, + 506, + 495 + ], + "score": 1.0, + "content": "of SO(3) is linearly universal, if it contains all linear", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 492, + 358, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 358, + 507 + ], + "score": 1.0, + "content": "SO(3)-equivariant mappings between the two representations.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 470, + 506, + 507 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 514, + 504, + 537 + ], + "lines": [ + { + "bbox": [ + 105, + 513, + 506, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 506, + 528 + ], + "score": 1.0, + "content": "When these two conditions apply, a rather simple symmetrization arguments leads to the following", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 524, + 145, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 145, + 538 + ], + "score": 1.0, + "content": "theorem:", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5, + "bbox_fs": [ + 105, + 513, + 506, + 538 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 540, + 505, + 575 + ], + "lines": [ + { + "bbox": [ + 105, + 539, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 169, + 555 + ], + "score": 1.0, + "content": "Theorem 1. If", + "type": "text" + }, + { + "bbox": [ + 169, + 542, + 191, + 552 + ], + "score": 0.9, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 539, + 201, + 555 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 201, + 542, + 210, + 551 + ], + "score": 0.78, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 539, + 269, + 555 + ], + "score": 1.0, + "content": "-spanning and", + "type": "text" + }, + { + "bbox": [ + 270, + 542, + 294, + 553 + ], + "score": 0.93, + "content": "\\mathcal { F } _ { \\mathrm { p o o l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 539, + 469, + 555 + ], + "score": 1.0, + "content": "is linearly universal, then there exists some", + "type": "text" + }, + { + "bbox": [ + 469, + 541, + 505, + 553 + ], + "score": 0.9, + "content": "C ( D ) \\in", + "type": "inline_equation" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 550, + 506, + 567 + ], + "spans": [ + { + "bbox": [ + 106, + 552, + 116, + 563 + ], + "score": 0.55, + "content": "\\mathbb { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 116, + 550, + 188, + 567 + ], + "score": 1.0, + "content": "such that for all", + "type": "text" + }, + { + "bbox": [ + 188, + 552, + 239, + 564 + ], + "score": 0.92, + "content": "C \\geq C ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 550, + 320, + 567 + ], + "score": 1.0, + "content": "the function space", + "type": "text" + }, + { + "bbox": [ + 321, + 552, + 390, + 565 + ], + "score": 0.91, + "content": "\\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } , \\mathcal { F } _ { \\mathrm { p o o l } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 550, + 445, + 567 + ], + "score": 1.0, + "content": "contains all", + "type": "text" + }, + { + "bbox": [ + 446, + 553, + 455, + 562 + ], + "score": 0.75, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 550, + 506, + 567 + ], + "score": 1.0, + "content": "-equivariant", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 563, + 220, + 575 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 196, + 575 + ], + "score": 1.0, + "content": "polynomials of degree", + "type": "text" + }, + { + "bbox": [ + 197, + 564, + 217, + 574 + ], + "score": 0.8, + "content": "\\le D", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 563, + 220, + 575 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35, + "bbox_fs": [ + 105, + 539, + 506, + 575 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 596, + 504, + 619 + ], + "lines": [ + { + "bbox": [ + 105, + 596, + 505, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 186, + 610 + ], + "score": 1.0, + "content": "Proof idea. By the", + "type": "text" + }, + { + "bbox": [ + 187, + 597, + 196, + 607 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 596, + 335, + 610 + ], + "score": 1.0, + "content": "-spanning assumption, there exist", + "type": "text" + }, + { + "bbox": [ + 336, + 597, + 417, + 609 + ], + "score": 0.92, + "content": "f _ { 1 } , \\dots , f _ { K } \\in { \\mathcal { F } } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 596, + 505, + 610 + ], + "score": 1.0, + "content": "such that any vector", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 608, + 464, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 608, + 464, + 621 + ], + "score": 1.0, + "content": "valued polynomial invariant to translations and equivariant to permutations is of the form", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 596, + 505, + 621 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 254, + 626, + 356, + 661 + ], + "lines": [ + { + "bbox": [ + 254, + 626, + 356, + 661 + ], + "spans": [ + { + "bbox": [ + 254, + 626, + 356, + 661 + ], + "score": 0.95, + "content": "p ( X ) = \\sum _ { k = 1 } ^ { K } \\hat { \\Lambda } _ { k } ( f _ { k } ( X ) ) ,", + "type": "interline_equation", + "image_path": "65688a9ed2ba7a86da01fc534b78beb1d3c5369d1355607144fa4b5d092569e0.jpg" + } + ] + } + ], + "index": 39.5, + "virtual_lines": [ + { + "bbox": [ + 254, + 626, + 356, + 643.5 + ], + "spans": [], + "index": 39 + }, + { + "bbox": [ + 254, + 643.5, + 356, + 661.0 + ], + "spans": [], + "index": 40 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 666, + 505, + 690 + ], + "lines": [ + { + "bbox": [ + 105, + 666, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 279, + 680 + ], + "score": 1.0, + "content": "While by definition this holds for functions", + "type": "text" + }, + { + "bbox": [ + 279, + 670, + 285, + 679 + ], + "score": 0.8, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 666, + 349, + 680 + ], + "score": 1.0, + "content": "whose image is", + "type": "text" + }, + { + "bbox": [ + 349, + 668, + 363, + 677 + ], + "score": 0.87, + "content": "\\mathbb { R } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 666, + 505, + 680 + ], + "score": 1.0, + "content": ", this is easily extended to functions", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 678, + 169, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 678, + 117, + 691 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 117, + 678, + 134, + 690 + ], + "score": 0.91, + "content": "W _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 678, + 169, + 691 + ], + "score": 1.0, + "content": "as well.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41.5, + "bbox_fs": [ + 105, + 666, + 505, + 691 + ] + }, + { + "type": "list", + "bbox": [ + 105, + 694, + 504, + 718 + ], + "lines": [ + { + "bbox": [ + 105, + 695, + 505, + 708 + ], + "spans": [ + { + "bbox": [ + 105, + 695, + 221, + 708 + ], + "score": 1.0, + "content": "It remains to show that when", + "type": "text" + }, + { + "bbox": [ + 222, + 699, + 227, + 707 + ], + "score": 0.78, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 695, + 393, + 708 + ], + "score": 1.0, + "content": "is also SO(3)-equivariant, we can choose", + "type": "text" + }, + { + "bbox": [ + 393, + 696, + 406, + 706 + ], + "score": 0.9, + "content": "\\Lambda _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 695, + 505, + 708 + ], + "score": 1.0, + "content": "to be SO(3) equivariant.", + "type": "text" + } + ], + "index": 43, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 705, + 503, + 718 + ], + "spans": [ + { + "bbox": [ + 106, + 705, + 298, + 718 + ], + "score": 1.0, + "content": "This is accomplished by averaging over SO(3).", + "type": "text" + }, + { + "bbox": [ + 497, + 709, + 503, + 715 + ], + "score": 0.0, + "content": "", + "type": "text" + } + ], + "index": 44, + "is_list_start_line": true, + "is_list_end_line": true + } + ], + "index": 43.5, + "bbox_fs": [ + 105, + 695, + 505, + 718 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 336, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 337, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 337, + 95 + ], + "score": 1.0, + "content": "As a result of Theorem 1 and Lemma 1 we obtain our", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 284, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 284, + 106 + ], + "score": 1.0, + "content": "universality result (see inset for illustration)", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 106, + 107, + 336, + 152 + ], + "lines": [ + { + "bbox": [ + 106, + 106, + 337, + 120 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 201, + 120 + ], + "score": 1.0, + "content": "Corollary 1. For all", + "type": "text" + }, + { + "bbox": [ + 201, + 108, + 258, + 119 + ], + "score": 0.91, + "content": "{ \\cal C } , { \\cal D } \\in \\mathbb { N } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 106, + 280, + 120 + ], + "score": 1.0, + "content": ", let", + "type": "text" + }, + { + "bbox": [ + 280, + 108, + 304, + 120 + ], + "score": 0.91, + "content": "\\mathcal { F } _ { C , D }", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 106, + 337, + 120 + ], + "score": 1.0, + "content": "denote", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 119, + 337, + 132 + ], + "spans": [ + { + "bbox": [ + 105, + 119, + 337, + 132 + ], + "score": 1.0, + "content": "function spaces generated by a pair of functions spaces", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 130, + 337, + 142 + ], + "spans": [ + { + "bbox": [ + 106, + 130, + 148, + 142 + ], + "score": 1.0, + "content": "which are", + "type": "text" + }, + { + "bbox": [ + 149, + 130, + 158, + 140 + ], + "score": 0.75, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 130, + 337, + 142 + ], + "score": 1.0, + "content": "-spanning and linearly universal as in equa-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 140, + 337, + 153 + ], + "spans": [ + { + "bbox": [ + 105, + 140, + 229, + 153 + ], + "score": 1.0, + "content": "tion 2. Then any continuous", + "type": "text" + }, + { + "bbox": [ + 229, + 141, + 238, + 151 + ], + "score": 0.79, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 140, + 337, + 153 + ], + "score": 1.0, + "content": "-equivariant function in", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3.5 + }, + { + "type": "image", + "bbox": [ + 352, + 68, + 482, + 132 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 352, + 68, + 482, + 132 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 352, + 68, + 482, + 132 + ], + "spans": [ + { + "bbox": [ + 352, + 68, + 482, + 132 + ], + "score": 0.786, + "type": "image", + "image_path": "e9ffa6a3d2fc7c2c788f5f90cefc79b1dd49ce1bf98aa7f80e26cd6b3b4cedca.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 352, + 68, + 482, + 100.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 352, + 100.0, + 482, + 132.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 103, + 152, + 475, + 164 + ], + "lines": [ + { + "bbox": [ + 107, + 148, + 477, + 166 + ], + "spans": [ + { + "bbox": [ + 107, + 151, + 171, + 164 + ], + "score": 0.92, + "content": "\\mathcal { C } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , W _ { T } ^ { n } )", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 148, + 477, + 166 + ], + "score": 1.0, + "content": "can be approximated uniformly on compact sets by equivariant functions in", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "interline_equation", + "bbox": [ + 264, + 168, + 346, + 195 + ], + "lines": [ + { + "bbox": [ + 264, + 168, + 346, + 195 + ], + "spans": [ + { + "bbox": [ + 264, + 168, + 346, + 195 + ], + "score": 0.94, + "content": "{ \\mathcal { F } } = \\bigcup _ { D \\in \\mathbb { N } } { \\mathcal { F } } _ { C ( D ) , D } .", + "type": "interline_equation", + "image_path": "1c264ee43605c8770f7fcfe0fb711e379d5b74553fd50d1b112fdb2da713aff9.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 264, + 168, + 346, + 195 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "title", + "bbox": [ + 108, + 206, + 299, + 217 + ], + "lines": [ + { + "bbox": [ + 105, + 206, + 301, + 219 + ], + "spans": [ + { + "bbox": [ + 105, + 206, + 301, + 219 + ], + "score": 1.0, + "content": "3.3 UNIVERSALITY CONDITIONS IN ACTION", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 226, + 505, + 316 + ], + "lines": [ + { + "bbox": [ + 104, + 226, + 506, + 240 + ], + "spans": [ + { + "bbox": [ + 104, + 226, + 366, + 240 + ], + "score": 1.0, + "content": "In the remainder of the paper, we prove the universality of several", + "type": "text" + }, + { + "bbox": [ + 367, + 227, + 376, + 237 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 226, + 506, + 240 + ], + "score": 1.0, + "content": "-equivariant architectures, based", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 239, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 239, + 505, + 249 + ], + "score": 1.0, + "content": "on the framework we discussed in the previous subsection. We discuss two different strategies for", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 249, + 506, + 262 + ], + "spans": [ + { + "bbox": [ + 105, + 249, + 506, + 262 + ], + "score": 1.0, + "content": "achieving universality, which differ mainly in the type of lifted representations of SO(3) they use:", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 260, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 260, + 505, + 272 + ], + "score": 1.0, + "content": "(i) The first strategy uses (direct sums of) tensor-product representations; (ii) The second uses (direct", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 270, + 506, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 506, + 284 + ], + "score": 1.0, + "content": "sums of) irreducible representations. The main advantage of the first strategy from the perspective of", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 281, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 259, + 295 + ], + "score": 1.0, + "content": "our methodology is that achieving the", + "type": "text" + }, + { + "bbox": [ + 259, + 282, + 269, + 291 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 281, + 505, + 295 + ], + "score": 1.0, + "content": "-spanning property is more straightforward. The advantage", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 293, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 293, + 505, + 306 + ], + "score": 1.0, + "content": "of irreducible representations is that they almost automatically guarantees the linear universality", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 304, + 145, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 145, + 317 + ], + "score": 1.0, + "content": "property.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 14.5 + }, + { + "type": "text", + "bbox": [ + 106, + 320, + 505, + 387 + ], + "lines": [ + { + "bbox": [ + 105, + 320, + 505, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 505, + 333 + ], + "score": 1.0, + "content": "In Section 4 we discuss universality through tensor product representations, and give an example", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 332, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 332, + 505, + 344 + ], + "score": 1.0, + "content": "of a minimal tensor representation network architecture that would satisfy universality. In section 5", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 342, + 506, + 355 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 506, + 355 + ], + "score": 1.0, + "content": "we discuss universality through irreducible representations, which is currently the more common", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 353, + 505, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 505, + 366 + ], + "score": 1.0, + "content": "strategy. We show that the TFN architecture (Thomas et al., 2018; Fuchs et al., 2020) which follows", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 363, + 506, + 377 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 506, + 377 + ], + "score": 1.0, + "content": "this strategy is universal, and describe a simple tweak that achieves universality using only low order", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 375, + 427, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 427, + 388 + ], + "score": 1.0, + "content": "filters, though the representations throughout the network are high dimensional.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21.5 + }, + { + "type": "title", + "bbox": [ + 107, + 403, + 378, + 415 + ], + "lines": [ + { + "bbox": [ + 105, + 402, + 379, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 379, + 417 + ], + "score": 1.0, + "content": "4 UNIVERSALITY WITH TENSOR REPRESENTATIONS", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 106, + 427, + 504, + 471 + ], + "lines": [ + { + "bbox": [ + 105, + 426, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 505, + 441 + ], + "score": 1.0, + "content": "In this section, we prove universality for models that are based on tensor product repre-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 437, + 506, + 451 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 454, + 451 + ], + "score": 1.0, + "content": "sentations, as defined below. The main advantage of this approach is that", + "type": "text" + }, + { + "bbox": [ + 455, + 438, + 464, + 448 + ], + "score": 0.73, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 437, + 506, + 451 + ], + "score": 1.0, + "content": "-spanning", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 449, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 449, + 505, + 462 + ], + "score": 1.0, + "content": "is achieved rather easily. The main drawbacks are that its data representation is some-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 461, + 505, + 472 + ], + "spans": [ + { + "bbox": [ + 106, + 461, + 505, + 472 + ], + "score": 1.0, + "content": "what redundant and that characterizing the linear equivariant layers is more laborious.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 106, + 488, + 337, + 545 + ], + "lines": [ + { + "bbox": [ + 106, + 487, + 336, + 501 + ], + "spans": [ + { + "bbox": [ + 106, + 487, + 336, + 501 + ], + "score": 1.0, + "content": "Tensor representations We begin by defining tensor rep-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 499, + 338, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 499, + 177, + 514 + ], + "score": 1.0, + "content": "resentations. For", + "type": "text" + }, + { + "bbox": [ + 177, + 501, + 211, + 513 + ], + "score": 0.91, + "content": "k \\in \\mathbb { N } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 499, + 242, + 514 + ], + "score": 1.0, + "content": "denote", + "type": "text" + }, + { + "bbox": [ + 243, + 499, + 284, + 513 + ], + "score": 0.92, + "content": "\\mathcal { T } _ { k } = \\mathbb { R } ^ { 3 ^ { k } }", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 499, + 338, + 514 + ], + "score": 1.0, + "content": ". SO(3) acts", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 511, + 336, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 120, + 525 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 120, + 513, + 132, + 523 + ], + "score": 0.87, + "content": "\\mathcal { T } _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 511, + 336, + 525 + ], + "score": 1.0, + "content": "by the tensor product representation, i.e., by ap-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 523, + 336, + 535 + ], + "spans": [ + { + "bbox": [ + 106, + 524, + 257, + 535 + ], + "score": 1.0, + "content": "plying the matrix Kronecker product", + "type": "text" + }, + { + "bbox": [ + 257, + 524, + 264, + 533 + ], + "score": 0.79, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 265, + 524, + 294, + 535 + ], + "score": 1.0, + "content": "times:", + "type": "text" + }, + { + "bbox": [ + 294, + 523, + 336, + 535 + ], + "score": 0.91, + "content": "\\rho _ { k } ( R ) : =", + "type": "inline_equation" + } + ], + "index": 33 + }, + { + "bbox": [ + 107, + 533, + 337, + 547 + ], + "spans": [ + { + "bbox": [ + 107, + 534, + 126, + 545 + ], + "score": 0.89, + "content": "\\bar { R } ^ { \\otimes k }", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 533, + 337, + 547 + ], + "score": 1.0, + "content": ". The inset illustrates the vector spaces and action for", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 106, + 545, + 469, + 557 + ], + "lines": [ + { + "bbox": [ + 106, + 543, + 470, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 150, + 556 + ], + "score": 0.91, + "content": "k = 1 , 2 , 3", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 543, + 254, + 559 + ], + "score": 1.0, + "content": ". With this action, for any", + "type": "text" + }, + { + "bbox": [ + 254, + 545, + 317, + 557 + ], + "score": 0.92, + "content": "\\bar { i } _ { 1 } , \\ldots , i _ { k } \\in [ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 543, + 376, + 559 + ], + "score": 1.0, + "content": ", the map from", + "type": "text" + }, + { + "bbox": [ + 377, + 545, + 401, + 555 + ], + "score": 0.91, + "content": "\\mathbb { R } ^ { 3 \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 543, + 412, + 559 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 412, + 545, + 424, + 556 + ], + "score": 0.88, + "content": "\\mathcal { T } _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 543, + 470, + 559 + ], + "score": 1.0, + "content": "defined by", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "image", + "bbox": [ + 349, + 486, + 485, + 543 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 349, + 486, + 485, + 543 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 349, + 486, + 485, + 543 + ], + "spans": [ + { + "bbox": [ + 349, + 486, + 485, + 543 + ], + "score": 0.944, + "type": "image", + "image_path": "50d9d12099cc8cbcdfa045b7f9ad7b985e39b62e026d157b80b6dee7accd6a00.jpg" + } + ] + } + ], + "index": 36.0, + "virtual_lines": [ + { + "bbox": [ + 349, + 486, + 485, + 514.5 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 349, + 514.5, + 485, + 543.0 + ], + "spans": [], + "index": 37 + } + ] + } + ], + "index": 36.0 + }, + { + "type": "interline_equation", + "bbox": [ + 232, + 561, + 378, + 574 + ], + "lines": [ + { + "bbox": [ + 232, + 561, + 378, + 574 + ], + "spans": [ + { + "bbox": [ + 232, + 561, + 378, + 574 + ], + "score": 0.91, + "content": "( x _ { 1 } , \\ldots , x _ { n } ) \\mapsto x _ { i _ { 1 } } \\otimes x _ { i _ { 2 } } \\ldots \\otimes x _ { i _ { k } }", + "type": "interline_equation", + "image_path": "3a715a86537245fcd6b8c4ec6286409899ac3e9af73d7db3cea35849934105da.jpg" + } + ] + } + ], + "index": 38, + "virtual_lines": [ + { + "bbox": [ + 232, + 561, + 378, + 574 + ], + "spans": [], + "index": 38 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 577, + 192, + 590 + ], + "lines": [ + { + "bbox": [ + 105, + 576, + 194, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 194, + 591 + ], + "score": 1.0, + "content": "is SO(3) equivariant.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 106, + 595, + 506, + 618 + ], + "lines": [ + { + "bbox": [ + 106, + 594, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 116, + 607 + ], + "score": 1.0, + "content": "A", + "type": "text" + }, + { + "bbox": [ + 117, + 595, + 126, + 605 + ], + "score": 0.74, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 126, + 594, + 505, + 607 + ], + "score": 1.0, + "content": "-spanning family We now show that tensor representations can be used to define a finite set of", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 107, + 605, + 392, + 619 + ], + "spans": [ + { + "bbox": [ + 107, + 606, + 116, + 616 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 605, + 300, + 619 + ], + "score": 1.0, + "content": "-spanning functions. The lifted representation", + "type": "text" + }, + { + "bbox": [ + 300, + 606, + 324, + 617 + ], + "score": 0.9, + "content": "W _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 605, + 392, + 619 + ], + "score": 1.0, + "content": "will be given by", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5 + }, + { + "type": "interline_equation", + "bbox": [ + 270, + 621, + 341, + 655 + ], + "lines": [ + { + "bbox": [ + 270, + 621, + 341, + 655 + ], + "spans": [ + { + "bbox": [ + 270, + 621, + 341, + 655 + ], + "score": 0.94, + "content": "W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } } = \\bigoplus _ { T = 0 } ^ { D } \\mathcal { T } _ { T } .", + "type": "interline_equation", + "image_path": "61830aefac19c1277d057845830e6bfae00c53974c5a2f9063ff38a8ceec272e.jpg" + } + ] + } + ], + "index": 42.5, + "virtual_lines": [ + { + "bbox": [ + 270, + 621, + 341, + 638.0 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 270, + 638.0, + 341, + 655.0 + ], + "spans": [], + "index": 43 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 659, + 505, + 698 + ], + "lines": [ + { + "bbox": [ + 105, + 658, + 505, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 124, + 673 + ], + "score": 1.0, + "content": "The", + "type": "text" + }, + { + "bbox": [ + 124, + 660, + 134, + 669 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 658, + 304, + 673 + ], + "score": 1.0, + "content": "-spanning functions are indexed by vectors", + "type": "text" + }, + { + "bbox": [ + 304, + 659, + 374, + 672 + ], + "score": 0.94, + "content": "\\vec { r } = \\left( r _ { 1 } , \\ldots , r _ { K } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 658, + 424, + 673 + ], + "score": 1.0, + "content": ", where each", + "type": "text" + }, + { + "bbox": [ + 424, + 662, + 434, + 671 + ], + "score": 0.83, + "content": "r _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 658, + 505, + 673 + ], + "score": 1.0, + "content": "is a non-negative", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 100, + 668, + 507, + 692 + ], + "spans": [ + { + "bbox": [ + 100, + 668, + 179, + 692 + ], + "score": 1.0, + "content": "integer. Denoting", + "type": "text" + }, + { + "bbox": [ + 180, + 673, + 221, + 685 + ], + "score": 0.93, + "content": "T = \\| r \\| _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 668, + 280, + 692 + ], + "score": 1.0, + "content": ", the functions", + "type": "text" + }, + { + "bbox": [ + 280, + 672, + 361, + 685 + ], + "score": 0.93, + "content": "Q ^ { ( \\vec { r } ) } : \\mathbb { R } ^ { 3 \\times n } \\mathcal { T } _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 668, + 368, + 692 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 368, + 671, + 442, + 687 + ], + "score": 0.92, + "content": "Q ^ { ( \\vec { r } ) } = ( Q _ { j } ^ { ( \\vec { r } ) } ) _ { j = 1 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 669, + 507, + 688 + ], + "score": 1.0, + "content": "))nj=1 are defined for", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 683, + 173, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 683, + 128, + 700 + ], + "score": 1.0, + "content": "fixed", + "type": "text" + }, + { + "bbox": [ + 129, + 685, + 158, + 698 + ], + "score": 0.92, + "content": "j \\in [ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 683, + 173, + 700 + ], + "score": 1.0, + "content": "by", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45 + }, + { + "type": "interline_equation", + "bbox": [ + 187, + 701, + 424, + 735 + ], + "lines": [ + { + "bbox": [ + 187, + 701, + 424, + 735 + ], + "spans": [ + { + "bbox": [ + 187, + 701, + 424, + 735 + ], + "score": 0.93, + "content": "Q _ { j } ^ { ( \\vec { r } ) } ( X ) = \\sum _ { i _ { 2 } , \\ldots , i _ { K } = 1 } ^ { n } x _ { j } ^ { \\otimes r _ { 1 } } \\otimes x _ { i _ { 2 } } ^ { \\otimes r _ { 2 } } \\otimes x _ { i _ { 3 } } ^ { \\otimes r _ { 3 } } \\otimes \\ldots \\otimes x _ { i _ { K } } ^ { \\otimes r _ { K } } .", + "type": "interline_equation", + "image_path": "e6ca06060048c51bdbe91ba75c3f61ec795f6819e3895e4f638adf7baba11d0e.jpg" + } + ] + } + ], + "index": 47.5, + "virtual_lines": [ + { + "bbox": [ + 187, + 701, + 424, + 718.0 + ], + "spans": [], + "index": 47 + }, + { + "bbox": [ + 187, + 718.0, + 424, + 735.0 + ], + "spans": [], + "index": 48 + } + ] + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 336, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 337, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 337, + 95 + ], + "score": 1.0, + "content": "As a result of Theorem 1 and Lemma 1 we obtain our", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 284, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 284, + 106 + ], + "score": 1.0, + "content": "universality result (see inset for illustration)", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 106, + 81, + 337, + 106 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 107, + 336, + 152 + ], + "lines": [ + { + "bbox": [ + 106, + 106, + 337, + 120 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 201, + 120 + ], + "score": 1.0, + "content": "Corollary 1. For all", + "type": "text" + }, + { + "bbox": [ + 201, + 108, + 258, + 119 + ], + "score": 0.91, + "content": "{ \\cal C } , { \\cal D } \\in \\mathbb { N } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 106, + 280, + 120 + ], + "score": 1.0, + "content": ", let", + "type": "text" + }, + { + "bbox": [ + 280, + 108, + 304, + 120 + ], + "score": 0.91, + "content": "\\mathcal { F } _ { C , D }", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 106, + 337, + 120 + ], + "score": 1.0, + "content": "denote", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 119, + 337, + 132 + ], + "spans": [ + { + "bbox": [ + 105, + 119, + 337, + 132 + ], + "score": 1.0, + "content": "function spaces generated by a pair of functions spaces", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 130, + 337, + 142 + ], + "spans": [ + { + "bbox": [ + 106, + 130, + 148, + 142 + ], + "score": 1.0, + "content": "which are", + "type": "text" + }, + { + "bbox": [ + 149, + 130, + 158, + 140 + ], + "score": 0.75, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 130, + 337, + 142 + ], + "score": 1.0, + "content": "-spanning and linearly universal as in equa-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 140, + 337, + 153 + ], + "spans": [ + { + "bbox": [ + 105, + 140, + 229, + 153 + ], + "score": 1.0, + "content": "tion 2. Then any continuous", + "type": "text" + }, + { + "bbox": [ + 229, + 141, + 238, + 151 + ], + "score": 0.79, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 140, + 337, + 153 + ], + "score": 1.0, + "content": "-equivariant function in", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 107, + 148, + 477, + 166 + ], + "spans": [ + { + "bbox": [ + 107, + 151, + 171, + 164 + ], + "score": 0.92, + "content": "\\mathcal { C } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , W _ { T } ^ { n } )", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 148, + 477, + 166 + ], + "score": 1.0, + "content": "can be approximated uniformly on compact sets by equivariant functions in", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 3.5, + "bbox_fs": [ + 105, + 106, + 337, + 153 + ] + }, + { + "type": "image", + "bbox": [ + 352, + 68, + 482, + 132 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 352, + 68, + 482, + 132 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 352, + 68, + 482, + 132 + ], + "spans": [ + { + "bbox": [ + 352, + 68, + 482, + 132 + ], + "score": 0.786, + "type": "image", + "image_path": "e9ffa6a3d2fc7c2c788f5f90cefc79b1dd49ce1bf98aa7f80e26cd6b3b4cedca.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 352, + 68, + 482, + 100.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 352, + 100.0, + 482, + 132.0 + ], + "spans": [], + "index": 7 + } + ] + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 103, + 152, + 475, + 164 + ], + "lines": [], + "index": 8, + "bbox_fs": [ + 107, + 148, + 477, + 166 + ], + "lines_deleted": true + }, + { + "type": "interline_equation", + "bbox": [ + 264, + 168, + 346, + 195 + ], + "lines": [ + { + "bbox": [ + 264, + 168, + 346, + 195 + ], + "spans": [ + { + "bbox": [ + 264, + 168, + 346, + 195 + ], + "score": 0.94, + "content": "{ \\mathcal { F } } = \\bigcup _ { D \\in \\mathbb { N } } { \\mathcal { F } } _ { C ( D ) , D } .", + "type": "interline_equation", + "image_path": "1c264ee43605c8770f7fcfe0fb711e379d5b74553fd50d1b112fdb2da713aff9.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 264, + 168, + 346, + 195 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "title", + "bbox": [ + 108, + 206, + 299, + 217 + ], + "lines": [ + { + "bbox": [ + 105, + 206, + 301, + 219 + ], + "spans": [ + { + "bbox": [ + 105, + 206, + 301, + 219 + ], + "score": 1.0, + "content": "3.3 UNIVERSALITY CONDITIONS IN ACTION", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 226, + 505, + 316 + ], + "lines": [ + { + "bbox": [ + 104, + 226, + 506, + 240 + ], + "spans": [ + { + "bbox": [ + 104, + 226, + 366, + 240 + ], + "score": 1.0, + "content": "In the remainder of the paper, we prove the universality of several", + "type": "text" + }, + { + "bbox": [ + 367, + 227, + 376, + 237 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 226, + 506, + 240 + ], + "score": 1.0, + "content": "-equivariant architectures, based", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 239, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 239, + 505, + 249 + ], + "score": 1.0, + "content": "on the framework we discussed in the previous subsection. We discuss two different strategies for", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 249, + 506, + 262 + ], + "spans": [ + { + "bbox": [ + 105, + 249, + 506, + 262 + ], + "score": 1.0, + "content": "achieving universality, which differ mainly in the type of lifted representations of SO(3) they use:", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 260, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 260, + 505, + 272 + ], + "score": 1.0, + "content": "(i) The first strategy uses (direct sums of) tensor-product representations; (ii) The second uses (direct", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 270, + 506, + 284 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 506, + 284 + ], + "score": 1.0, + "content": "sums of) irreducible representations. The main advantage of the first strategy from the perspective of", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 281, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 259, + 295 + ], + "score": 1.0, + "content": "our methodology is that achieving the", + "type": "text" + }, + { + "bbox": [ + 259, + 282, + 269, + 291 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 281, + 505, + 295 + ], + "score": 1.0, + "content": "-spanning property is more straightforward. The advantage", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 293, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 105, + 293, + 505, + 306 + ], + "score": 1.0, + "content": "of irreducible representations is that they almost automatically guarantees the linear universality", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 304, + 145, + 317 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 145, + 317 + ], + "score": 1.0, + "content": "property.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 14.5, + "bbox_fs": [ + 104, + 226, + 506, + 317 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 320, + 505, + 387 + ], + "lines": [ + { + "bbox": [ + 105, + 320, + 505, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 505, + 333 + ], + "score": 1.0, + "content": "In Section 4 we discuss universality through tensor product representations, and give an example", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 332, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 332, + 505, + 344 + ], + "score": 1.0, + "content": "of a minimal tensor representation network architecture that would satisfy universality. In section 5", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 342, + 506, + 355 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 506, + 355 + ], + "score": 1.0, + "content": "we discuss universality through irreducible representations, which is currently the more common", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 353, + 505, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 505, + 366 + ], + "score": 1.0, + "content": "strategy. We show that the TFN architecture (Thomas et al., 2018; Fuchs et al., 2020) which follows", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 363, + 506, + 377 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 506, + 377 + ], + "score": 1.0, + "content": "this strategy is universal, and describe a simple tweak that achieves universality using only low order", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 375, + 427, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 427, + 388 + ], + "score": 1.0, + "content": "filters, though the representations throughout the network are high dimensional.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 320, + 506, + 388 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 403, + 378, + 415 + ], + "lines": [ + { + "bbox": [ + 105, + 402, + 379, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 379, + 417 + ], + "score": 1.0, + "content": "4 UNIVERSALITY WITH TENSOR REPRESENTATIONS", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 106, + 427, + 504, + 471 + ], + "lines": [ + { + "bbox": [ + 105, + 426, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 105, + 426, + 505, + 441 + ], + "score": 1.0, + "content": "In this section, we prove universality for models that are based on tensor product repre-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 437, + 506, + 451 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 454, + 451 + ], + "score": 1.0, + "content": "sentations, as defined below. The main advantage of this approach is that", + "type": "text" + }, + { + "bbox": [ + 455, + 438, + 464, + 448 + ], + "score": 0.73, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 437, + 506, + 451 + ], + "score": 1.0, + "content": "-spanning", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 449, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 449, + 505, + 462 + ], + "score": 1.0, + "content": "is achieved rather easily. The main drawbacks are that its data representation is some-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 461, + 505, + 472 + ], + "spans": [ + { + "bbox": [ + 106, + 461, + 505, + 472 + ], + "score": 1.0, + "content": "what redundant and that characterizing the linear equivariant layers is more laborious.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5, + "bbox_fs": [ + 105, + 426, + 506, + 472 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 488, + 337, + 545 + ], + "lines": [ + { + "bbox": [ + 106, + 487, + 336, + 501 + ], + "spans": [ + { + "bbox": [ + 106, + 487, + 336, + 501 + ], + "score": 1.0, + "content": "Tensor representations We begin by defining tensor rep-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 499, + 338, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 499, + 177, + 514 + ], + "score": 1.0, + "content": "resentations. For", + "type": "text" + }, + { + "bbox": [ + 177, + 501, + 211, + 513 + ], + "score": 0.91, + "content": "k \\in \\mathbb { N } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 499, + 242, + 514 + ], + "score": 1.0, + "content": "denote", + "type": "text" + }, + { + "bbox": [ + 243, + 499, + 284, + 513 + ], + "score": 0.92, + "content": "\\mathcal { T } _ { k } = \\mathbb { R } ^ { 3 ^ { k } }", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 499, + 338, + 514 + ], + "score": 1.0, + "content": ". SO(3) acts", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 511, + 336, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 120, + 525 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 120, + 513, + 132, + 523 + ], + "score": 0.87, + "content": "\\mathcal { T } _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 511, + 336, + 525 + ], + "score": 1.0, + "content": "by the tensor product representation, i.e., by ap-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 523, + 336, + 535 + ], + "spans": [ + { + "bbox": [ + 106, + 524, + 257, + 535 + ], + "score": 1.0, + "content": "plying the matrix Kronecker product", + "type": "text" + }, + { + "bbox": [ + 257, + 524, + 264, + 533 + ], + "score": 0.79, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 265, + 524, + 294, + 535 + ], + "score": 1.0, + "content": "times:", + "type": "text" + }, + { + "bbox": [ + 294, + 523, + 336, + 535 + ], + "score": 0.91, + "content": "\\rho _ { k } ( R ) : =", + "type": "inline_equation" + } + ], + "index": 33 + }, + { + "bbox": [ + 107, + 533, + 337, + 547 + ], + "spans": [ + { + "bbox": [ + 107, + 534, + 126, + 545 + ], + "score": 0.89, + "content": "\\bar { R } ^ { \\otimes k }", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 533, + 337, + 547 + ], + "score": 1.0, + "content": ". The inset illustrates the vector spaces and action for", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 543, + 470, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 150, + 556 + ], + "score": 0.91, + "content": "k = 1 , 2 , 3", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 543, + 254, + 559 + ], + "score": 1.0, + "content": ". With this action, for any", + "type": "text" + }, + { + "bbox": [ + 254, + 545, + 317, + 557 + ], + "score": 0.92, + "content": "\\bar { i } _ { 1 } , \\ldots , i _ { k } \\in [ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 543, + 376, + 559 + ], + "score": 1.0, + "content": ", the map from", + "type": "text" + }, + { + "bbox": [ + 377, + 545, + 401, + 555 + ], + "score": 0.91, + "content": "\\mathbb { R } ^ { 3 \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 543, + 412, + 559 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 412, + 545, + 424, + 556 + ], + "score": 0.88, + "content": "\\mathcal { T } _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 543, + 470, + 559 + ], + "score": 1.0, + "content": "defined by", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 32, + "bbox_fs": [ + 105, + 487, + 338, + 547 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 545, + 469, + 557 + ], + "lines": [], + "index": 36, + "bbox_fs": [ + 106, + 543, + 470, + 559 + ], + "lines_deleted": true + }, + { + "type": "image", + "bbox": [ + 349, + 486, + 485, + 543 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 349, + 486, + 485, + 543 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 349, + 486, + 485, + 543 + ], + "spans": [ + { + "bbox": [ + 349, + 486, + 485, + 543 + ], + "score": 0.944, + "type": "image", + "image_path": "50d9d12099cc8cbcdfa045b7f9ad7b985e39b62e026d157b80b6dee7accd6a00.jpg" + } + ] + } + ], + "index": 36.0, + "virtual_lines": [ + { + "bbox": [ + 349, + 486, + 485, + 514.5 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 349, + 514.5, + 485, + 543.0 + ], + "spans": [], + "index": 37 + } + ] + } + ], + "index": 36.0 + }, + { + "type": "interline_equation", + "bbox": [ + 232, + 561, + 378, + 574 + ], + "lines": [ + { + "bbox": [ + 232, + 561, + 378, + 574 + ], + "spans": [ + { + "bbox": [ + 232, + 561, + 378, + 574 + ], + "score": 0.91, + "content": "( x _ { 1 } , \\ldots , x _ { n } ) \\mapsto x _ { i _ { 1 } } \\otimes x _ { i _ { 2 } } \\ldots \\otimes x _ { i _ { k } }", + "type": "interline_equation", + "image_path": "3a715a86537245fcd6b8c4ec6286409899ac3e9af73d7db3cea35849934105da.jpg" + } + ] + } + ], + "index": 38, + "virtual_lines": [ + { + "bbox": [ + 232, + 561, + 378, + 574 + ], + "spans": [], + "index": 38 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 577, + 192, + 590 + ], + "lines": [ + { + "bbox": [ + 105, + 576, + 194, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 194, + 591 + ], + "score": 1.0, + "content": "is SO(3) equivariant.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39, + "bbox_fs": [ + 105, + 576, + 194, + 591 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 595, + 506, + 618 + ], + "lines": [ + { + "bbox": [ + 106, + 594, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 116, + 607 + ], + "score": 1.0, + "content": "A", + "type": "text" + }, + { + "bbox": [ + 117, + 595, + 126, + 605 + ], + "score": 0.74, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 126, + 594, + 505, + 607 + ], + "score": 1.0, + "content": "-spanning family We now show that tensor representations can be used to define a finite set of", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 107, + 605, + 392, + 619 + ], + "spans": [ + { + "bbox": [ + 107, + 606, + 116, + 616 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 605, + 300, + 619 + ], + "score": 1.0, + "content": "-spanning functions. The lifted representation", + "type": "text" + }, + { + "bbox": [ + 300, + 606, + 324, + 617 + ], + "score": 0.9, + "content": "W _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 605, + 392, + 619 + ], + "score": 1.0, + "content": "will be given by", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5, + "bbox_fs": [ + 106, + 594, + 505, + 619 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 270, + 621, + 341, + 655 + ], + "lines": [ + { + "bbox": [ + 270, + 621, + 341, + 655 + ], + "spans": [ + { + "bbox": [ + 270, + 621, + 341, + 655 + ], + "score": 0.94, + "content": "W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } } = \\bigoplus _ { T = 0 } ^ { D } \\mathcal { T } _ { T } .", + "type": "interline_equation", + "image_path": "61830aefac19c1277d057845830e6bfae00c53974c5a2f9063ff38a8ceec272e.jpg" + } + ] + } + ], + "index": 42.5, + "virtual_lines": [ + { + "bbox": [ + 270, + 621, + 341, + 638.0 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 270, + 638.0, + 341, + 655.0 + ], + "spans": [], + "index": 43 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 659, + 505, + 698 + ], + "lines": [ + { + "bbox": [ + 105, + 658, + 505, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 124, + 673 + ], + "score": 1.0, + "content": "The", + "type": "text" + }, + { + "bbox": [ + 124, + 660, + 134, + 669 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 658, + 304, + 673 + ], + "score": 1.0, + "content": "-spanning functions are indexed by vectors", + "type": "text" + }, + { + "bbox": [ + 304, + 659, + 374, + 672 + ], + "score": 0.94, + "content": "\\vec { r } = \\left( r _ { 1 } , \\ldots , r _ { K } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 658, + 424, + 673 + ], + "score": 1.0, + "content": ", where each", + "type": "text" + }, + { + "bbox": [ + 424, + 662, + 434, + 671 + ], + "score": 0.83, + "content": "r _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 658, + 505, + 673 + ], + "score": 1.0, + "content": "is a non-negative", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 100, + 668, + 507, + 692 + ], + "spans": [ + { + "bbox": [ + 100, + 668, + 179, + 692 + ], + "score": 1.0, + "content": "integer. Denoting", + "type": "text" + }, + { + "bbox": [ + 180, + 673, + 221, + 685 + ], + "score": 0.93, + "content": "T = \\| r \\| _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 668, + 280, + 692 + ], + "score": 1.0, + "content": ", the functions", + "type": "text" + }, + { + "bbox": [ + 280, + 672, + 361, + 685 + ], + "score": 0.93, + "content": "Q ^ { ( \\vec { r } ) } : \\mathbb { R } ^ { 3 \\times n } \\mathcal { T } _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 668, + 368, + 692 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 368, + 671, + 442, + 687 + ], + "score": 0.92, + "content": "Q ^ { ( \\vec { r } ) } = ( Q _ { j } ^ { ( \\vec { r } ) } ) _ { j = 1 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 669, + 507, + 688 + ], + "score": 1.0, + "content": "))nj=1 are defined for", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 683, + 173, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 683, + 128, + 700 + ], + "score": 1.0, + "content": "fixed", + "type": "text" + }, + { + "bbox": [ + 129, + 685, + 158, + 698 + ], + "score": 0.92, + "content": "j \\in [ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 683, + 173, + 700 + ], + "score": 1.0, + "content": "by", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45, + "bbox_fs": [ + 100, + 658, + 507, + 700 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 187, + 701, + 424, + 735 + ], + "lines": [ + { + "bbox": [ + 187, + 701, + 424, + 735 + ], + "spans": [ + { + "bbox": [ + 187, + 701, + 424, + 735 + ], + "score": 0.93, + "content": "Q _ { j } ^ { ( \\vec { r } ) } ( X ) = \\sum _ { i _ { 2 } , \\ldots , i _ { K } = 1 } ^ { n } x _ { j } ^ { \\otimes r _ { 1 } } \\otimes x _ { i _ { 2 } } ^ { \\otimes r _ { 2 } } \\otimes x _ { i _ { 3 } } ^ { \\otimes r _ { 3 } } \\otimes \\ldots \\otimes x _ { i _ { K } } ^ { \\otimes r _ { K } } .", + "type": "interline_equation", + "image_path": "e6ca06060048c51bdbe91ba75c3f61ec795f6819e3895e4f638adf7baba11d0e.jpg" + } + ] + } + ], + "index": 47.5, + "virtual_lines": [ + { + "bbox": [ + 187, + 701, + 424, + 718.0 + ], + "spans": [], + "index": 47 + }, + { + "bbox": [ + 187, + 718.0, + 424, + 735.0 + ], + "spans": [], + "index": 48 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 505, + 142 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 163, + 96 + ], + "score": 1.0, + "content": "The functions", + "type": "text" + }, + { + "bbox": [ + 163, + 80, + 183, + 96 + ], + "score": 0.92, + "content": "Q _ { j } ^ { ( \\vec { r } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 80, + 506, + 96 + ], + "score": 1.0, + "content": "are SO(3) equivariant as they are a sum of equivariant functions from equation 5.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 95, + 506, + 110 + ], + "spans": [ + { + "bbox": [ + 105, + 95, + 129, + 110 + ], + "score": 1.0, + "content": "Thus", + "type": "text" + }, + { + "bbox": [ + 129, + 96, + 148, + 109 + ], + "score": 0.91, + "content": "Q ^ { ( \\vec { r } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 95, + 158, + 110 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 159, + 97, + 208, + 109 + ], + "score": 0.91, + "content": "\\mathrm { S O } ( 3 ) \\times S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 95, + 506, + 110 + ], + "score": 1.0, + "content": "equivariant. The motivation behind the definition of these functions is that", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 107, + 506, + 121 + ], + "spans": [ + { + "bbox": [ + 105, + 107, + 506, + 121 + ], + "score": 1.0, + "content": "known characterizations of permutation equivariant polynomials Segol & Lipman (2019) tell us that", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 118, + 506, + 132 + ], + "spans": [ + { + "bbox": [ + 105, + 118, + 506, + 132 + ], + "score": 1.0, + "content": "the entries of these tensor valued functions span all permutation equivariant polynomials (see the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 130, + 253, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 130, + 253, + 141 + ], + "score": 1.0, + "content": "proof of Lemma 2 for more details).", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 107, + 146, + 503, + 170 + ], + "lines": [ + { + "bbox": [ + 105, + 145, + 505, + 161 + ], + "spans": [ + { + "bbox": [ + 105, + 145, + 364, + 161 + ], + "score": 1.0, + "content": "To account for translation invariance, we compose the functions", + "type": "text" + }, + { + "bbox": [ + 365, + 146, + 384, + 159 + ], + "score": 0.92, + "content": "Q ^ { ( \\vec { r } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 145, + 505, + 161 + ], + "score": 1.0, + "content": "with the centralization opera-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 159, + 247, + 171 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 247, + 171 + ], + "score": 1.0, + "content": "tion and define the set of functions", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "interline_equation", + "bbox": [ + 211, + 173, + 399, + 198 + ], + "lines": [ + { + "bbox": [ + 211, + 173, + 399, + 198 + ], + "spans": [ + { + "bbox": [ + 211, + 173, + 399, + 198 + ], + "score": 0.94, + "content": "\\mathcal { Q } _ { D } = \\{ \\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac { 1 } { n } X 1 _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) | \\| \\vec { r } \\| _ { 1 } \\leq D \\} ,", + "type": "interline_equation", + "image_path": "577822362a69d996a5b8f585aa8be41f18d57409dd20b244abea1af4e17b8356.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 211, + 173, + 399, + 198 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 204, + 504, + 228 + ], + "lines": [ + { + "bbox": [ + 105, + 201, + 507, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 201, + 134, + 221 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 208, + 139, + 215 + ], + "score": 0.73, + "content": "\\iota", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 201, + 311, + 221 + ], + "score": 1.0, + "content": "is the natural embedding that takes each", + "type": "text" + }, + { + "bbox": [ + 311, + 205, + 324, + 216 + ], + "score": 0.89, + "content": "\\mathcal { T } _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 201, + 346, + 221 + ], + "score": 1.0, + "content": "into", + "type": "text" + }, + { + "bbox": [ + 346, + 203, + 428, + 218 + ], + "score": 0.93, + "content": "W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } } = \\oplus _ { T = 0 } ^ { D } \\mathcal { T } _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 201, + 507, + 221 + ], + "score": 1.0, + "content": ". In the following", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 215, + 288, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 235, + 229 + ], + "score": 1.0, + "content": "lemma, we prove that this set is", + "type": "text" + }, + { + "bbox": [ + 235, + 217, + 244, + 226 + ], + "score": 0.85, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 215, + 288, + 229 + ], + "score": 1.0, + "content": "-spanning.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 106, + 230, + 342, + 243 + ], + "lines": [ + { + "bbox": [ + 105, + 228, + 343, + 246 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 194, + 246 + ], + "score": 1.0, + "content": "Lemma 2. For every", + "type": "text" + }, + { + "bbox": [ + 195, + 231, + 230, + 243 + ], + "score": 0.91, + "content": "D \\in { \\mathbb { N } } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 228, + 262, + 246 + ], + "score": 1.0, + "content": ", the set", + "type": "text" + }, + { + "bbox": [ + 263, + 231, + 278, + 243 + ], + "score": 0.88, + "content": "\\mathcal { Q } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 228, + 289, + 246 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 289, + 232, + 299, + 241 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 228, + 343, + 246 + ], + "score": 1.0, + "content": "-spanning.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 105, + 259, + 504, + 285 + ], + "lines": [ + { + "bbox": [ + 104, + 258, + 504, + 273 + ], + "spans": [ + { + "bbox": [ + 104, + 258, + 433, + 273 + ], + "score": 1.0, + "content": "Proof idea. It is known (Segol & Lipman, 2019) (Theorem 2) that polynomials", + "type": "text" + }, + { + "bbox": [ + 433, + 259, + 504, + 271 + ], + "score": 0.9, + "content": "p : \\mathbb { R } ^ { 3 \\times n } \\mathbb { R } ^ { n }", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 270, + 479, + 287 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 148, + 287 + ], + "score": 1.0, + "content": "which are", + "type": "text" + }, + { + "bbox": [ + 148, + 273, + 160, + 284 + ], + "score": 0.89, + "content": "S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 270, + 373, + 287 + ], + "score": 1.0, + "content": "-equivariant, are spanned by polynomials of the form", + "type": "text" + }, + { + "bbox": [ + 374, + 271, + 431, + 286 + ], + "score": 0.95, + "content": "p _ { \\vec { \\alpha } } = ( p _ { \\vec { \\alpha } } ^ { j } ) _ { j = 1 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 270, + 479, + 287 + ], + "score": 1.0, + "content": ", defined as", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + }, + { + "type": "interline_equation", + "bbox": [ + 232, + 291, + 377, + 325 + ], + "lines": [ + { + "bbox": [ + 232, + 291, + 377, + 325 + ], + "spans": [ + { + "bbox": [ + 232, + 291, + 377, + 325 + ], + "score": 0.95, + "content": "p _ { \\vec { \\alpha } } ^ { j } ( X ) = \\sum _ { i _ { 2 } , \\ldots , i _ { K } = 1 } ^ { n } x _ { j } ^ { \\alpha _ { 1 } } x _ { i _ { 2 } } ^ { \\alpha _ { 2 } } \\cdot \\cdot \\cdot x _ { i _ { k } } ^ { \\alpha _ { k } }", + "type": "interline_equation", + "image_path": "99ab3f48ed1763969604fac776352a2c075cf44d5e166c7a767f93ed4702223c.jpg" + } + ] + } + ], + "index": 13.5, + "virtual_lines": [ + { + "bbox": [ + 232, + 291, + 377, + 308.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 232, + 308.0, + 377, + 325.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 331, + 505, + 357 + ], + "lines": [ + { + "bbox": [ + 106, + 331, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 331, + 133, + 344 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 331, + 209, + 344 + ], + "score": 0.92, + "content": "\\vec { \\alpha } = ( \\alpha _ { 1 } , \\ldots , \\alpha _ { K } )", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 331, + 248, + 344 + ], + "score": 1.0, + "content": "and each", + "type": "text" + }, + { + "bbox": [ + 248, + 331, + 286, + 344 + ], + "score": 0.93, + "content": "\\alpha _ { k } \\in \\mathbb { N } _ { + } ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 331, + 505, + 344 + ], + "score": 1.0, + "content": "is a multi-index. We first show that these polynomials", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 342, + 505, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 195, + 360 + ], + "score": 1.0, + "content": "can be extracted from", + "type": "text" + }, + { + "bbox": [ + 196, + 344, + 215, + 357 + ], + "score": 0.92, + "content": "Q ^ { ( \\vec { r } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 342, + 340, + 360 + ], + "score": 1.0, + "content": "and then use them to represent", + "type": "text" + }, + { + "bbox": [ + 340, + 348, + 347, + 357 + ], + "score": 0.79, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 342, + 351, + 360 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 495, + 346, + 505, + 356 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 106, + 369, + 506, + 414 + ], + "lines": [ + { + "bbox": [ + 105, + 369, + 505, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 362, + 382 + ], + "score": 1.0, + "content": "A minimal universal architecture Once we have shown that", + "type": "text" + }, + { + "bbox": [ + 362, + 370, + 378, + 381 + ], + "score": 0.9, + "content": "\\mathcal { Q } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 369, + 390, + 382 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 390, + 370, + 400, + 380 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 369, + 505, + 382 + ], + "score": 1.0, + "content": "-spanning, we can design", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 107, + 380, + 506, + 393 + ], + "spans": [ + { + "bbox": [ + 107, + 381, + 116, + 390 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 380, + 468, + 393 + ], + "score": 1.0, + "content": "-spanning architectures, by devising architectures that are able to span all elements of", + "type": "text" + }, + { + "bbox": [ + 469, + 381, + 484, + 392 + ], + "score": 0.89, + "content": "\\mathcal { Q } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 380, + 506, + 393 + ], + "score": 1.0, + "content": ". As", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 392, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 106, + 392, + 505, + 404 + ], + "score": 1.0, + "content": "we will now show, the compositional nature of neural networks allows us to do this in a very clean", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 404, + 141, + 414 + ], + "spans": [ + { + "bbox": [ + 106, + 404, + 141, + 414 + ], + "score": 1.0, + "content": "manner.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 106, + 418, + 502, + 446 + ], + "lines": [ + { + "bbox": [ + 105, + 416, + 506, + 435 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 235, + 435 + ], + "score": 1.0, + "content": "We define a parametric function", + "type": "text" + }, + { + "bbox": [ + 235, + 419, + 296, + 432 + ], + "score": 0.92, + "content": "f ( X , V | \\theta _ { 1 } , \\theta _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 416, + 347, + 435 + ], + "score": 1.0, + "content": "which maps", + "type": "text" + }, + { + "bbox": [ + 347, + 419, + 393, + 432 + ], + "score": 0.92, + "content": "\\mathbb { R } ^ { 3 \\times n } \\oplus \\mathcal { T } _ { k } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 416, + 405, + 435 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 405, + 419, + 459, + 433 + ], + "score": 0.95, + "content": "\\mathbb { R } ^ { 3 \\times n } \\oplus \\mathcal { T } _ { k + 1 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 416, + 506, + 435 + ], + "score": 1.0, + "content": "as follows:", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 432, + 348, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 135, + 447 + ], + "score": 1.0, + "content": "For all", + "type": "text" + }, + { + "bbox": [ + 135, + 433, + 164, + 446 + ], + "score": 0.92, + "content": "j \\in [ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 432, + 204, + 447 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + }, + { + "bbox": [ + 204, + 432, + 316, + 446 + ], + "score": 0.92, + "content": "f _ { j } ( X , V ) = ( x _ { j } , \\tilde { V } _ { j } ( X , V ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 432, + 348, + 447 + ], + "score": 1.0, + "content": ", where", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + }, + { + "type": "interline_equation", + "bbox": [ + 201, + 457, + 410, + 484 + ], + "lines": [ + { + "bbox": [ + 201, + 457, + 410, + 484 + ], + "spans": [ + { + "bbox": [ + 201, + 457, + 410, + 484 + ], + "score": 0.9, + "content": "\\tilde { V } _ { j } ( X , V | \\theta _ { 1 } , \\theta _ { 2 } ) = \\theta _ { 1 } \\left( x _ { j } \\otimes V _ { j } \\right) + \\theta _ { 2 } \\sum _ { i } \\left( x _ { i } \\otimes V _ { i } \\right)", + "type": "interline_equation", + "image_path": "d38e84166719d7c03679c4c996ad32c73150b74c54182a5d36ccac62eb944303.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 201, + 457, + 410, + 484 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 492, + 505, + 537 + ], + "lines": [ + { + "bbox": [ + 105, + 492, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 238, + 506 + ], + "score": 1.0, + "content": "We denote the set of functions", + "type": "text" + }, + { + "bbox": [ + 239, + 493, + 348, + 505 + ], + "score": 0.93, + "content": "( X , V ) \\mapsto f ( X , V | \\theta _ { 1 } , \\theta _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 492, + 505, + 506 + ], + "score": 1.0, + "content": "obtained by choosing the parameters", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 107, + 504, + 506, + 517 + ], + "spans": [ + { + "bbox": [ + 107, + 504, + 150, + 515 + ], + "score": 0.91, + "content": "\\theta _ { 1 } , \\theta _ { 2 } \\in \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 504, + 167, + 517 + ], + "score": 1.0, + "content": ", by", + "type": "text" + }, + { + "bbox": [ + 167, + 505, + 191, + 515 + ], + "score": 0.88, + "content": "\\mathcal { F } _ { m i n }", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 504, + 506, + 517 + ], + "score": 1.0, + "content": ". While in the hidden layers of our network the data is represented using both", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 515, + 505, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 156, + 527 + ], + "score": 1.0, + "content": "coordinates", + "type": "text" + }, + { + "bbox": [ + 156, + 516, + 186, + 527 + ], + "score": 0.91, + "content": "( X , V )", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 515, + 364, + 527 + ], + "score": 1.0, + "content": ", the input to the network only contains an", + "type": "text" + }, + { + "bbox": [ + 365, + 516, + 375, + 525 + ], + "score": 0.83, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 515, + 505, + 527 + ], + "score": 1.0, + "content": "coordinate and the output only", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 526, + 352, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 526, + 149, + 538 + ], + "score": 1.0, + "content": "contains a", + "type": "text" + }, + { + "bbox": [ + 149, + 527, + 158, + 536 + ], + "score": 0.82, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 526, + 352, + 538 + ], + "score": 1.0, + "content": "coordinate. To this end, we define the functions", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5 + }, + { + "type": "interline_equation", + "bbox": [ + 224, + 543, + 386, + 557 + ], + "lines": [ + { + "bbox": [ + 224, + 543, + 386, + 557 + ], + "spans": [ + { + "bbox": [ + 224, + 543, + 386, + 557 + ], + "score": 0.91, + "content": "\\operatorname { e x t } ( X ) = ( X , 1 _ { n } ) \\operatorname { a n d } \\pi _ { V } ( X , V ) = V .", + "type": "interline_equation", + "image_path": "a5eede119b22fdd477db38131758e79a5298c33c1d635ab71ae68ec67c6b3b40.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 224, + 543, + 386, + 557 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 562, + 506, + 586 + ], + "lines": [ + { + "bbox": [ + 105, + 561, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 171, + 576 + ], + "score": 1.0, + "content": "We can achieve", + "type": "text" + }, + { + "bbox": [ + 172, + 563, + 181, + 572 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 561, + 347, + 576 + ], + "score": 1.0, + "content": "-spanning by composition of functions in", + "type": "text" + }, + { + "bbox": [ + 347, + 563, + 371, + 574 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { m i n }", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 561, + 505, + 576 + ], + "score": 1.0, + "content": "with these functions and central-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 572, + 133, + 588 + ], + "spans": [ + { + "bbox": [ + 104, + 572, + 133, + 588 + ], + "score": 1.0, + "content": "izing:", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 105, + 588, + 299, + 600 + ], + "lines": [ + { + "bbox": [ + 106, + 587, + 299, + 602 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 221, + 602 + ], + "score": 1.0, + "content": "Lemma 3. The function set", + "type": "text" + }, + { + "bbox": [ + 221, + 589, + 236, + 600 + ], + "score": 0.89, + "content": "\\mathcal { Q } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 587, + 299, + 602 + ], + "score": 1.0, + "content": "is contained in", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "interline_equation", + "bbox": [ + 139, + 605, + 471, + 629 + ], + "lines": [ + { + "bbox": [ + 139, + 605, + 471, + 629 + ], + "spans": [ + { + "bbox": [ + 139, + 605, + 471, + 629 + ], + "score": 0.9, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } } = \\big \\{ \\iota \\circ \\pi _ { V } \\circ f ^ { 1 } \\circ f ^ { 2 } \\circ \\ldots \\circ f ^ { T } \\circ \\mathrm { e x t } ( X - \\frac { 1 } { n } X \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) | f ^ { j } \\in \\mathcal { F } _ { m i n } , T \\leq D \\big \\} .", + "type": "interline_equation", + "image_path": "93ede582cc29e75094f8d3a9e1daafa85c8f9c1049c9e766450977511ecf598a.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 139, + 605, + 471, + 629 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 633, + 212, + 646 + ], + "lines": [ + { + "bbox": [ + 105, + 632, + 214, + 648 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 128, + 648 + ], + "score": 1.0, + "content": "Thus", + "type": "text" + }, + { + "bbox": [ + 129, + 634, + 150, + 645 + ], + "score": 0.88, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 632, + 160, + 648 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 161, + 635, + 170, + 644 + ], + "score": 0.78, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 632, + 214, + 648 + ], + "score": 1.0, + "content": "-spanning.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 106, + 662, + 372, + 675 + ], + "lines": [ + { + "bbox": [ + 105, + 662, + 373, + 676 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 360, + 676 + ], + "score": 1.0, + "content": "Proof idea. The proof is technical and follows by induction on", + "type": "text" + }, + { + "bbox": [ + 360, + 664, + 369, + 673 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 662, + 373, + 676 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 686, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 686, + 505, + 700 + ], + "score": 1.0, + "content": "To complete the construction of a universal network, we now need to characterize all linear equivari-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 102, + 695, + 509, + 716 + ], + "spans": [ + { + "bbox": [ + 102, + 695, + 183, + 716 + ], + "score": 1.0, + "content": "ant functions from", + "type": "text" + }, + { + "bbox": [ + 183, + 698, + 207, + 711 + ], + "score": 0.92, + "content": "W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 695, + 318, + 716 + ], + "score": 1.0, + "content": "to the target representation", + "type": "text" + }, + { + "bbox": [ + 318, + 699, + 335, + 710 + ], + "score": 0.89, + "content": "W _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 695, + 392, + 716 + ], + "score": 1.0, + "content": ". In Appendix", + "type": "text" + }, + { + "bbox": [ + 392, + 699, + 402, + 709 + ], + "score": 0.38, + "content": "\\mathbf { G }", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 695, + 509, + 716 + ], + "score": 1.0, + "content": "we show how this can be", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 243, + 722 + ], + "score": 1.0, + "content": "done for the trivial representation", + "type": "text" + }, + { + "bbox": [ + 244, + 710, + 281, + 721 + ], + "score": 0.92, + "content": "W _ { T } = \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 709, + 505, + 722 + ], + "score": 1.0, + "content": ". This characterization gives us a set of linear functions", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 720, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 130, + 733 + ], + "score": 0.91, + "content": "\\mathcal { F } _ { \\mathrm { p o o l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 131, + 720, + 221, + 734 + ], + "score": 1.0, + "content": ", which combined with", + "type": "text" + }, + { + "bbox": [ + 222, + 721, + 243, + 732 + ], + "score": 0.9, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 720, + 506, + 734 + ], + "score": 1.0, + "content": "defined in equation 11 (corresponds to SO(3) invariant functions)", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36.5 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 495, + 663, + 504, + 673 + ], + "lines": [ + { + "bbox": [ + 496, + 665, + 504, + 673 + ], + "spans": [ + { + "bbox": [ + 496, + 665, + 504, + 673 + ], + "score": 0.997, + "content": "□", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 310, + 761 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 310, + 761 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 505, + 142 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 163, + 96 + ], + "score": 1.0, + "content": "The functions", + "type": "text" + }, + { + "bbox": [ + 163, + 80, + 183, + 96 + ], + "score": 0.92, + "content": "Q _ { j } ^ { ( \\vec { r } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 80, + 506, + 96 + ], + "score": 1.0, + "content": "are SO(3) equivariant as they are a sum of equivariant functions from equation 5.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 95, + 506, + 110 + ], + "spans": [ + { + "bbox": [ + 105, + 95, + 129, + 110 + ], + "score": 1.0, + "content": "Thus", + "type": "text" + }, + { + "bbox": [ + 129, + 96, + 148, + 109 + ], + "score": 0.91, + "content": "Q ^ { ( \\vec { r } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 95, + 158, + 110 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 159, + 97, + 208, + 109 + ], + "score": 0.91, + "content": "\\mathrm { S O } ( 3 ) \\times S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 95, + 506, + 110 + ], + "score": 1.0, + "content": "equivariant. The motivation behind the definition of these functions is that", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 107, + 506, + 121 + ], + "spans": [ + { + "bbox": [ + 105, + 107, + 506, + 121 + ], + "score": 1.0, + "content": "known characterizations of permutation equivariant polynomials Segol & Lipman (2019) tell us that", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 118, + 506, + 132 + ], + "spans": [ + { + "bbox": [ + 105, + 118, + 506, + 132 + ], + "score": 1.0, + "content": "the entries of these tensor valued functions span all permutation equivariant polynomials (see the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 130, + 253, + 141 + ], + "spans": [ + { + "bbox": [ + 105, + 130, + 253, + 141 + ], + "score": 1.0, + "content": "proof of Lemma 2 for more details).", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 80, + 506, + 141 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 146, + 503, + 170 + ], + "lines": [ + { + "bbox": [ + 105, + 145, + 505, + 161 + ], + "spans": [ + { + "bbox": [ + 105, + 145, + 364, + 161 + ], + "score": 1.0, + "content": "To account for translation invariance, we compose the functions", + "type": "text" + }, + { + "bbox": [ + 365, + 146, + 384, + 159 + ], + "score": 0.92, + "content": "Q ^ { ( \\vec { r } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 145, + 505, + 161 + ], + "score": 1.0, + "content": "with the centralization opera-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 159, + 247, + 171 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 247, + 171 + ], + "score": 1.0, + "content": "tion and define the set of functions", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5, + "bbox_fs": [ + 105, + 145, + 505, + 171 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 211, + 173, + 399, + 198 + ], + "lines": [ + { + "bbox": [ + 211, + 173, + 399, + 198 + ], + "spans": [ + { + "bbox": [ + 211, + 173, + 399, + 198 + ], + "score": 0.94, + "content": "\\mathcal { Q } _ { D } = \\{ \\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac { 1 } { n } X 1 _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) | \\| \\vec { r } \\| _ { 1 } \\leq D \\} ,", + "type": "interline_equation", + "image_path": "577822362a69d996a5b8f585aa8be41f18d57409dd20b244abea1af4e17b8356.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 211, + 173, + 399, + 198 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 204, + 504, + 228 + ], + "lines": [ + { + "bbox": [ + 105, + 201, + 507, + 221 + ], + "spans": [ + { + "bbox": [ + 105, + 201, + 134, + 221 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 208, + 139, + 215 + ], + "score": 0.73, + "content": "\\iota", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 201, + 311, + 221 + ], + "score": 1.0, + "content": "is the natural embedding that takes each", + "type": "text" + }, + { + "bbox": [ + 311, + 205, + 324, + 216 + ], + "score": 0.89, + "content": "\\mathcal { T } _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 201, + 346, + 221 + ], + "score": 1.0, + "content": "into", + "type": "text" + }, + { + "bbox": [ + 346, + 203, + 428, + 218 + ], + "score": 0.93, + "content": "W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } } = \\oplus _ { T = 0 } ^ { D } \\mathcal { T } _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 201, + 507, + 221 + ], + "score": 1.0, + "content": ". In the following", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 215, + 288, + 229 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 235, + 229 + ], + "score": 1.0, + "content": "lemma, we prove that this set is", + "type": "text" + }, + { + "bbox": [ + 235, + 217, + 244, + 226 + ], + "score": 0.85, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 215, + 288, + 229 + ], + "score": 1.0, + "content": "-spanning.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 201, + 507, + 229 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 230, + 342, + 243 + ], + "lines": [ + { + "bbox": [ + 105, + 228, + 343, + 246 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 194, + 246 + ], + "score": 1.0, + "content": "Lemma 2. For every", + "type": "text" + }, + { + "bbox": [ + 195, + 231, + 230, + 243 + ], + "score": 0.91, + "content": "D \\in { \\mathbb { N } } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 228, + 262, + 246 + ], + "score": 1.0, + "content": ", the set", + "type": "text" + }, + { + "bbox": [ + 263, + 231, + 278, + 243 + ], + "score": 0.88, + "content": "\\mathcal { Q } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 228, + 289, + 246 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 289, + 232, + 299, + 241 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 228, + 343, + 246 + ], + "score": 1.0, + "content": "-spanning.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 228, + 343, + 246 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 259, + 504, + 285 + ], + "lines": [ + { + "bbox": [ + 104, + 258, + 504, + 273 + ], + "spans": [ + { + "bbox": [ + 104, + 258, + 433, + 273 + ], + "score": 1.0, + "content": "Proof idea. It is known (Segol & Lipman, 2019) (Theorem 2) that polynomials", + "type": "text" + }, + { + "bbox": [ + 433, + 259, + 504, + 271 + ], + "score": 0.9, + "content": "p : \\mathbb { R } ^ { 3 \\times n } \\mathbb { R } ^ { n }", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 270, + 479, + 287 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 148, + 287 + ], + "score": 1.0, + "content": "which are", + "type": "text" + }, + { + "bbox": [ + 148, + 273, + 160, + 284 + ], + "score": 0.89, + "content": "S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 270, + 373, + 287 + ], + "score": 1.0, + "content": "-equivariant, are spanned by polynomials of the form", + "type": "text" + }, + { + "bbox": [ + 374, + 271, + 431, + 286 + ], + "score": 0.95, + "content": "p _ { \\vec { \\alpha } } = ( p _ { \\vec { \\alpha } } ^ { j } ) _ { j = 1 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 270, + 479, + 287 + ], + "score": 1.0, + "content": ", defined as", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5, + "bbox_fs": [ + 104, + 258, + 504, + 287 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 232, + 291, + 377, + 325 + ], + "lines": [ + { + "bbox": [ + 232, + 291, + 377, + 325 + ], + "spans": [ + { + "bbox": [ + 232, + 291, + 377, + 325 + ], + "score": 0.95, + "content": "p _ { \\vec { \\alpha } } ^ { j } ( X ) = \\sum _ { i _ { 2 } , \\ldots , i _ { K } = 1 } ^ { n } x _ { j } ^ { \\alpha _ { 1 } } x _ { i _ { 2 } } ^ { \\alpha _ { 2 } } \\cdot \\cdot \\cdot x _ { i _ { k } } ^ { \\alpha _ { k } }", + "type": "interline_equation", + "image_path": "99ab3f48ed1763969604fac776352a2c075cf44d5e166c7a767f93ed4702223c.jpg" + } + ] + } + ], + "index": 13.5, + "virtual_lines": [ + { + "bbox": [ + 232, + 291, + 377, + 308.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 232, + 308.0, + 377, + 325.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 331, + 505, + 357 + ], + "lines": [ + { + "bbox": [ + 106, + 331, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 331, + 133, + 344 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 331, + 209, + 344 + ], + "score": 0.92, + "content": "\\vec { \\alpha } = ( \\alpha _ { 1 } , \\ldots , \\alpha _ { K } )", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 331, + 248, + 344 + ], + "score": 1.0, + "content": "and each", + "type": "text" + }, + { + "bbox": [ + 248, + 331, + 286, + 344 + ], + "score": 0.93, + "content": "\\alpha _ { k } \\in \\mathbb { N } _ { + } ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 331, + 505, + 344 + ], + "score": 1.0, + "content": "is a multi-index. We first show that these polynomials", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 342, + 505, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 195, + 360 + ], + "score": 1.0, + "content": "can be extracted from", + "type": "text" + }, + { + "bbox": [ + 196, + 344, + 215, + 357 + ], + "score": 0.92, + "content": "Q ^ { ( \\vec { r } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 342, + 340, + 360 + ], + "score": 1.0, + "content": "and then use them to represent", + "type": "text" + }, + { + "bbox": [ + 340, + 348, + 347, + 357 + ], + "score": 0.79, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 342, + 351, + 360 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 495, + 346, + 505, + 356 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 331, + 505, + 360 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 369, + 506, + 414 + ], + "lines": [ + { + "bbox": [ + 105, + 369, + 505, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 362, + 382 + ], + "score": 1.0, + "content": "A minimal universal architecture Once we have shown that", + "type": "text" + }, + { + "bbox": [ + 362, + 370, + 378, + 381 + ], + "score": 0.9, + "content": "\\mathcal { Q } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 369, + 390, + 382 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 390, + 370, + 400, + 380 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 369, + 505, + 382 + ], + "score": 1.0, + "content": "-spanning, we can design", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 107, + 380, + 506, + 393 + ], + "spans": [ + { + "bbox": [ + 107, + 381, + 116, + 390 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 380, + 468, + 393 + ], + "score": 1.0, + "content": "-spanning architectures, by devising architectures that are able to span all elements of", + "type": "text" + }, + { + "bbox": [ + 469, + 381, + 484, + 392 + ], + "score": 0.89, + "content": "\\mathcal { Q } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 380, + 506, + 393 + ], + "score": 1.0, + "content": ". As", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 392, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 106, + 392, + 505, + 404 + ], + "score": 1.0, + "content": "we will now show, the compositional nature of neural networks allows us to do this in a very clean", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 404, + 141, + 414 + ], + "spans": [ + { + "bbox": [ + 106, + 404, + 141, + 414 + ], + "score": 1.0, + "content": "manner.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 369, + 506, + 414 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 418, + 502, + 446 + ], + "lines": [ + { + "bbox": [ + 105, + 416, + 506, + 435 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 235, + 435 + ], + "score": 1.0, + "content": "We define a parametric function", + "type": "text" + }, + { + "bbox": [ + 235, + 419, + 296, + 432 + ], + "score": 0.92, + "content": "f ( X , V | \\theta _ { 1 } , \\theta _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 416, + 347, + 435 + ], + "score": 1.0, + "content": "which maps", + "type": "text" + }, + { + "bbox": [ + 347, + 419, + 393, + 432 + ], + "score": 0.92, + "content": "\\mathbb { R } ^ { 3 \\times n } \\oplus \\mathcal { T } _ { k } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 416, + 405, + 435 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 405, + 419, + 459, + 433 + ], + "score": 0.95, + "content": "\\mathbb { R } ^ { 3 \\times n } \\oplus \\mathcal { T } _ { k + 1 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 416, + 506, + 435 + ], + "score": 1.0, + "content": "as follows:", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 432, + 348, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 135, + 447 + ], + "score": 1.0, + "content": "For all", + "type": "text" + }, + { + "bbox": [ + 135, + 433, + 164, + 446 + ], + "score": 0.92, + "content": "j \\in [ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 432, + 204, + 447 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + }, + { + "bbox": [ + 204, + 432, + 316, + 446 + ], + "score": 0.92, + "content": "f _ { j } ( X , V ) = ( x _ { j } , \\tilde { V } _ { j } ( X , V ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 432, + 348, + 447 + ], + "score": 1.0, + "content": ", where", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 416, + 506, + 447 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 201, + 457, + 410, + 484 + ], + "lines": [ + { + "bbox": [ + 201, + 457, + 410, + 484 + ], + "spans": [ + { + "bbox": [ + 201, + 457, + 410, + 484 + ], + "score": 0.9, + "content": "\\tilde { V } _ { j } ( X , V | \\theta _ { 1 } , \\theta _ { 2 } ) = \\theta _ { 1 } \\left( x _ { j } \\otimes V _ { j } \\right) + \\theta _ { 2 } \\sum _ { i } \\left( x _ { i } \\otimes V _ { i } \\right)", + "type": "interline_equation", + "image_path": "d38e84166719d7c03679c4c996ad32c73150b74c54182a5d36ccac62eb944303.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 201, + 457, + 410, + 484 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 492, + 505, + 537 + ], + "lines": [ + { + "bbox": [ + 105, + 492, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 238, + 506 + ], + "score": 1.0, + "content": "We denote the set of functions", + "type": "text" + }, + { + "bbox": [ + 239, + 493, + 348, + 505 + ], + "score": 0.93, + "content": "( X , V ) \\mapsto f ( X , V | \\theta _ { 1 } , \\theta _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 492, + 505, + 506 + ], + "score": 1.0, + "content": "obtained by choosing the parameters", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 107, + 504, + 506, + 517 + ], + "spans": [ + { + "bbox": [ + 107, + 504, + 150, + 515 + ], + "score": 0.91, + "content": "\\theta _ { 1 } , \\theta _ { 2 } \\in \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 504, + 167, + 517 + ], + "score": 1.0, + "content": ", by", + "type": "text" + }, + { + "bbox": [ + 167, + 505, + 191, + 515 + ], + "score": 0.88, + "content": "\\mathcal { F } _ { m i n }", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 504, + 506, + 517 + ], + "score": 1.0, + "content": ". While in the hidden layers of our network the data is represented using both", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 515, + 505, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 156, + 527 + ], + "score": 1.0, + "content": "coordinates", + "type": "text" + }, + { + "bbox": [ + 156, + 516, + 186, + 527 + ], + "score": 0.91, + "content": "( X , V )", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 515, + 364, + 527 + ], + "score": 1.0, + "content": ", the input to the network only contains an", + "type": "text" + }, + { + "bbox": [ + 365, + 516, + 375, + 525 + ], + "score": 0.83, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 515, + 505, + 527 + ], + "score": 1.0, + "content": "coordinate and the output only", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 526, + 352, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 526, + 149, + 538 + ], + "score": 1.0, + "content": "contains a", + "type": "text" + }, + { + "bbox": [ + 149, + 527, + 158, + 536 + ], + "score": 0.82, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 526, + 352, + 538 + ], + "score": 1.0, + "content": "coordinate. To this end, we define the functions", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 492, + 506, + 538 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 224, + 543, + 386, + 557 + ], + "lines": [ + { + "bbox": [ + 224, + 543, + 386, + 557 + ], + "spans": [ + { + "bbox": [ + 224, + 543, + 386, + 557 + ], + "score": 0.91, + "content": "\\operatorname { e x t } ( X ) = ( X , 1 _ { n } ) \\operatorname { a n d } \\pi _ { V } ( X , V ) = V .", + "type": "interline_equation", + "image_path": "a5eede119b22fdd477db38131758e79a5298c33c1d635ab71ae68ec67c6b3b40.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 224, + 543, + 386, + 557 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 562, + 506, + 586 + ], + "lines": [ + { + "bbox": [ + 105, + 561, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 171, + 576 + ], + "score": 1.0, + "content": "We can achieve", + "type": "text" + }, + { + "bbox": [ + 172, + 563, + 181, + 572 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 561, + 347, + 576 + ], + "score": 1.0, + "content": "-spanning by composition of functions in", + "type": "text" + }, + { + "bbox": [ + 347, + 563, + 371, + 574 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { m i n }", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 561, + 505, + 576 + ], + "score": 1.0, + "content": "with these functions and central-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 572, + 133, + 588 + ], + "spans": [ + { + "bbox": [ + 104, + 572, + 133, + 588 + ], + "score": 1.0, + "content": "izing:", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5, + "bbox_fs": [ + 104, + 561, + 505, + 588 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 588, + 299, + 600 + ], + "lines": [ + { + "bbox": [ + 106, + 587, + 299, + 602 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 221, + 602 + ], + "score": 1.0, + "content": "Lemma 3. The function set", + "type": "text" + }, + { + "bbox": [ + 221, + 589, + 236, + 600 + ], + "score": 0.89, + "content": "\\mathcal { Q } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 587, + 299, + 602 + ], + "score": 1.0, + "content": "is contained in", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31, + "bbox_fs": [ + 106, + 587, + 299, + 602 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 139, + 605, + 471, + 629 + ], + "lines": [ + { + "bbox": [ + 139, + 605, + 471, + 629 + ], + "spans": [ + { + "bbox": [ + 139, + 605, + 471, + 629 + ], + "score": 0.9, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } } = \\big \\{ \\iota \\circ \\pi _ { V } \\circ f ^ { 1 } \\circ f ^ { 2 } \\circ \\ldots \\circ f ^ { T } \\circ \\mathrm { e x t } ( X - \\frac { 1 } { n } X \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) | f ^ { j } \\in \\mathcal { F } _ { m i n } , T \\leq D \\big \\} .", + "type": "interline_equation", + "image_path": "93ede582cc29e75094f8d3a9e1daafa85c8f9c1049c9e766450977511ecf598a.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 139, + 605, + 471, + 629 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 633, + 212, + 646 + ], + "lines": [ + { + "bbox": [ + 105, + 632, + 214, + 648 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 128, + 648 + ], + "score": 1.0, + "content": "Thus", + "type": "text" + }, + { + "bbox": [ + 129, + 634, + 150, + 645 + ], + "score": 0.88, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 632, + 160, + 648 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 161, + 635, + 170, + 644 + ], + "score": 0.78, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 632, + 214, + 648 + ], + "score": 1.0, + "content": "-spanning.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33, + "bbox_fs": [ + 105, + 632, + 214, + 648 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 662, + 372, + 675 + ], + "lines": [ + { + "bbox": [ + 105, + 662, + 373, + 676 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 360, + 676 + ], + "score": 1.0, + "content": "Proof idea. The proof is technical and follows by induction on", + "type": "text" + }, + { + "bbox": [ + 360, + 664, + 369, + 673 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 662, + 373, + 676 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 662, + 373, + 676 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 686, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 686, + 505, + 700 + ], + "score": 1.0, + "content": "To complete the construction of a universal network, we now need to characterize all linear equivari-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 102, + 695, + 509, + 716 + ], + "spans": [ + { + "bbox": [ + 102, + 695, + 183, + 716 + ], + "score": 1.0, + "content": "ant functions from", + "type": "text" + }, + { + "bbox": [ + 183, + 698, + 207, + 711 + ], + "score": 0.92, + "content": "W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 695, + 318, + 716 + ], + "score": 1.0, + "content": "to the target representation", + "type": "text" + }, + { + "bbox": [ + 318, + 699, + 335, + 710 + ], + "score": 0.89, + "content": "W _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 695, + 392, + 716 + ], + "score": 1.0, + "content": ". In Appendix", + "type": "text" + }, + { + "bbox": [ + 392, + 699, + 402, + 709 + ], + "score": 0.38, + "content": "\\mathbf { G }", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 695, + 509, + 716 + ], + "score": 1.0, + "content": "we show how this can be", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 243, + 722 + ], + "score": 1.0, + "content": "done for the trivial representation", + "type": "text" + }, + { + "bbox": [ + 244, + 710, + 281, + 721 + ], + "score": 0.92, + "content": "W _ { T } = \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 709, + 505, + 722 + ], + "score": 1.0, + "content": ". This characterization gives us a set of linear functions", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 720, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 130, + 733 + ], + "score": 0.91, + "content": "\\mathcal { F } _ { \\mathrm { p o o l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 131, + 720, + 221, + 734 + ], + "score": 1.0, + "content": ", which combined with", + "type": "text" + }, + { + "bbox": [ + 222, + 721, + 243, + 732 + ], + "score": 0.9, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 720, + 506, + 734 + ], + "score": 1.0, + "content": "defined in equation 11 (corresponds to SO(3) invariant functions)", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36.5, + "bbox_fs": [ + 102, + 686, + 509, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 104, + 82, + 504, + 106 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 96 + ], + "score": 1.0, + "content": "gives us a universal architecture as in Theorem 1. However, the disadvantage of this approach is that", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 405, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 272, + 106 + ], + "score": 1.0, + "content": "implementation of the linear functions in", + "type": "text" + }, + { + "bbox": [ + 272, + 94, + 296, + 106 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { \\mathrm { p o o l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 93, + 405, + 106 + ], + "score": 1.0, + "content": "is somewhat cumbersome.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 107, + 110, + 504, + 144 + ], + "lines": [ + { + "bbox": [ + 105, + 110, + 505, + 123 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 505, + 123 + ], + "score": 1.0, + "content": "In the next section we discuss irreducible representations, which give us a systematic way to address", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 122, + 505, + 134 + ], + "spans": [ + { + "bbox": [ + 106, + 122, + 255, + 134 + ], + "score": 1.0, + "content": "linear equivariant mappings into any", + "type": "text" + }, + { + "bbox": [ + 255, + 122, + 272, + 132 + ], + "score": 0.89, + "content": "W _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 122, + 311, + 134 + ], + "score": 1.0, + "content": ". Proving", + "type": "text" + }, + { + "bbox": [ + 312, + 122, + 321, + 131 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 122, + 505, + 134 + ], + "score": 1.0, + "content": "-spanning for these networks is accomplished", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 133, + 444, + 145 + ], + "spans": [ + { + "bbox": [ + 106, + 133, + 136, + 145 + ], + "score": 1.0, + "content": "via the", + "type": "text" + }, + { + "bbox": [ + 136, + 133, + 146, + 142 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 133, + 444, + 145 + ], + "score": 1.0, + "content": "-spanning property of tensor representations, through the following lemma", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 106, + 146, + 314, + 159 + ], + "lines": [ + { + "bbox": [ + 105, + 146, + 314, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 146, + 225, + 160 + ], + "score": 1.0, + "content": "Lemma 4. If all functions in", + "type": "text" + }, + { + "bbox": [ + 225, + 147, + 241, + 158 + ], + "score": 0.88, + "content": "\\mathcal { Q } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 146, + 314, + 160 + ], + "score": 1.0, + "content": "can be written as", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "interline_equation", + "bbox": [ + 218, + 162, + 391, + 197 + ], + "lines": [ + { + "bbox": [ + 218, + 162, + 391, + 197 + ], + "spans": [ + { + "bbox": [ + 218, + 162, + 391, + 197 + ], + "score": 0.94, + "content": "\\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac { 1 } { n } X 1 _ { n } 1 _ { n } ^ { T } ) = \\sum _ { k = 1 } ^ { K } \\hat { A } _ { k } f _ { k } ( X ) ,", + "type": "interline_equation", + "image_path": "e30e8f7ccf0c360604f9bbc4ab4bfc3b7f392383f90a98dc50473b858cbedd5b.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 218, + 162, + 391, + 179.5 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 218, + 179.5, + 391, + 197.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 201, + 505, + 226 + ], + "lines": [ + { + "bbox": [ + 103, + 198, + 508, + 222 + ], + "spans": [ + { + "bbox": [ + 103, + 198, + 134, + 222 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 204, + 180, + 215 + ], + "score": 0.67, + "content": "f _ { k } \\in \\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 198, + 185, + 222 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 186, + 202, + 274, + 215 + ], + "score": 0.87, + "content": "A _ { k } : W _ { \\mathrm { f e a t } } \\to W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 198, + 294, + 222 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 295, + 202, + 397, + 216 + ], + "score": 0.92, + "content": "{ \\hat { A } } _ { k } : W _ { \\mathrm { f e a t } } ^ { n } \\to ( W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } } ) ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 198, + 508, + 222 + ], + "score": 1.0, + "content": "is defined by elementwise", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 213, + 286, + 228 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 164, + 228 + ], + "score": 1.0, + "content": "application of", + "type": "text" + }, + { + "bbox": [ + 165, + 215, + 178, + 226 + ], + "score": 0.87, + "content": "A _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 213, + 201, + 228 + ], + "score": 1.0, + "content": ", then", + "type": "text" + }, + { + "bbox": [ + 202, + 215, + 223, + 226 + ], + "score": 0.81, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 213, + 234, + 228 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 234, + 215, + 244, + 224 + ], + "score": 0.67, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 213, + 286, + 228 + ], + "score": 1.0, + "content": "-spanning.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 107, + 234, + 372, + 246 + ], + "lines": [ + { + "bbox": [ + 106, + 234, + 372, + 248 + ], + "spans": [ + { + "bbox": [ + 106, + 234, + 200, + 248 + ], + "score": 1.0, + "content": "We note that as before,", + "type": "text" + }, + { + "bbox": [ + 200, + 235, + 214, + 246 + ], + "score": 0.89, + "content": "A _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 234, + 372, + 248 + ], + "score": 1.0, + "content": "are not necessarily SO(3)- equivariant.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 258, + 358, + 270 + ], + "lines": [ + { + "bbox": [ + 105, + 257, + 359, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 257, + 359, + 272 + ], + "score": 1.0, + "content": "Proof idea. The lemma follows directly from the assumptions.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "title", + "bbox": [ + 108, + 293, + 407, + 306 + ], + "lines": [ + { + "bbox": [ + 104, + 291, + 409, + 310 + ], + "spans": [ + { + "bbox": [ + 104, + 291, + 409, + 310 + ], + "score": 1.0, + "content": "5 UNIVERSALITY WITH IRREDUCIBLE REPRESENTATIONS", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 317, + 505, + 384 + ], + "lines": [ + { + "bbox": [ + 106, + 318, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 318, + 505, + 329 + ], + "score": 1.0, + "content": "In this section, we discuss how to achieve universality when using irreducible representations of", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 328, + 505, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 505, + 341 + ], + "score": 1.0, + "content": "SO(3). We will begin by defining irreducible representations, and explaining how linear univer-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 339, + 506, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 282, + 352 + ], + "score": 1.0, + "content": "sality is easily achieved by them, while the", + "type": "text" + }, + { + "bbox": [ + 282, + 340, + 292, + 349 + ], + "score": 0.78, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 339, + 506, + 352 + ], + "score": 1.0, + "content": "-spanning properties of tensor representations can be", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 351, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 505, + 362 + ], + "score": 1.0, + "content": "preserved. This discussion can be seen as an interpretation of the choices made in the construction", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 362, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 505, + 373 + ], + "score": 1.0, + "content": "of TFN and similar networks in the literature. We then show that these architectures are indeed", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 372, + 148, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 148, + 384 + ], + "score": 1.0, + "content": "universal.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 15.5 + }, + { + "type": "title", + "bbox": [ + 107, + 397, + 316, + 409 + ], + "lines": [ + { + "bbox": [ + 105, + 397, + 316, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 287, + 410 + ], + "score": 1.0, + "content": "5.1 IRREDUCIBLE REPRESENTATIONS OF", + "type": "text" + }, + { + "bbox": [ + 288, + 398, + 316, + 410 + ], + "score": 0.84, + "content": "S O ( 3 )", + "type": "inline_equation" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 417, + 505, + 487 + ], + "lines": [ + { + "bbox": [ + 105, + 417, + 506, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 302, + 431 + ], + "score": 1.0, + "content": "In general, any finite-dimensional representation", + "type": "text" + }, + { + "bbox": [ + 302, + 418, + 314, + 428 + ], + "score": 0.71, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 417, + 395, + 431 + ], + "score": 1.0, + "content": "of a compact group", + "type": "text" + }, + { + "bbox": [ + 395, + 419, + 405, + 428 + ], + "score": 0.81, + "content": "H", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 417, + 506, + 431 + ], + "score": 1.0, + "content": "can be decomposed into", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 428, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 262, + 442 + ], + "score": 1.0, + "content": "irreducible representations: a subspace", + "type": "text" + }, + { + "bbox": [ + 262, + 429, + 301, + 440 + ], + "score": 0.92, + "content": "W _ { 0 } \\subset W", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 428, + 311, + 442 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 311, + 429, + 321, + 439 + ], + "score": 0.83, + "content": "H", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 428, + 369, + 442 + ], + "score": 1.0, + "content": "-invariant if", + "type": "text" + }, + { + "bbox": [ + 369, + 429, + 409, + 440 + ], + "score": 0.92, + "content": "h w \\in W _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 428, + 435, + 442 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 436, + 429, + 501, + 440 + ], + "score": 0.91, + "content": "h \\in H , w \\in W _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 428, + 505, + 442 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 439, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 174, + 452 + ], + "score": 1.0, + "content": "A representation", + "type": "text" + }, + { + "bbox": [ + 174, + 440, + 186, + 450 + ], + "score": 0.7, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 439, + 505, + 452 + ], + "score": 1.0, + "content": "is irreducible if it has no non-trivial invariant subspaces. In the case of SO(3), all", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 451, + 506, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 327, + 465 + ], + "score": 1.0, + "content": "irreducible real representations are defined by matrices", + "type": "text" + }, + { + "bbox": [ + 327, + 451, + 362, + 464 + ], + "score": 0.94, + "content": "D ^ { ( \\ell ) } ( R )", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 451, + 506, + 465 + ], + "score": 1.0, + "content": ", called the real Wigner D-matrices,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 460, + 507, + 477 + ], + "spans": [ + { + "bbox": [ + 104, + 460, + 146, + 477 + ], + "score": 1.0, + "content": "acting on", + "type": "text" + }, + { + "bbox": [ + 147, + 462, + 204, + 474 + ], + "score": 0.92, + "content": "W _ { \\ell } : = \\overset { \\bullet } { \\mathbb { R } } ^ { 2 \\ell + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 460, + 454, + 477 + ], + "score": 1.0, + "content": "by matrix multiplication. In particular, the representation for", + "type": "text" + }, + { + "bbox": [ + 454, + 463, + 489, + 474 + ], + "score": 0.92, + "content": "\\ell = 0 , 1", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 460, + 507, + 477 + ], + "score": 1.0, + "content": "are", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 473, + 239, + 487 + ], + "spans": [ + { + "bbox": [ + 106, + 474, + 161, + 487 + ], + "score": 0.93, + "content": "D ^ { ( 0 ) } ( R ) = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 473, + 178, + 487 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 179, + 474, + 236, + 487 + ], + "score": 0.93, + "content": "D ^ { ( 1 ) } ( R ) = R", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 473, + 239, + 487 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 106, + 492, + 505, + 539 + ], + "lines": [ + { + "bbox": [ + 105, + 492, + 505, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 505, + 504 + ], + "score": 1.0, + "content": "Linear maps between irreducible representations As mentioned above, one of the main advan-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 503, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 503, + 506, + 515 + ], + "score": 1.0, + "content": "tages of using irreducible representations is that there is a very simple characterization of all linear", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 514, + 504, + 526 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 489, + 526 + ], + "score": 1.0, + "content": "equivariant maps between two direct sums of irreducible representations. We use the notation", + "type": "text" + }, + { + "bbox": [ + 490, + 514, + 504, + 525 + ], + "score": 0.85, + "content": "W _ { l }", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 102, + 522, + 501, + 545 + ], + "spans": [ + { + "bbox": [ + 102, + 522, + 316, + 545 + ], + "score": 1.0, + "content": "for direct sums of irreducible representations, where", + "type": "text" + }, + { + "bbox": [ + 316, + 525, + 411, + 540 + ], + "score": 0.91, + "content": "\\boldsymbol { l } = ( \\ell _ { 1 } , \\dots , \\ell _ { K } ) \\in \\mathbb { N } _ { + } ^ { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 522, + 429, + 545 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 429, + 524, + 501, + 540 + ], + "score": 0.93, + "content": "W _ { l } = \\bigoplus _ { k = 1 } ^ { K } W _ { \\ell _ { k } }", + "type": "inline_equation" + } + ], + "index": 29 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 106, + 549, + 505, + 590 + ], + "lines": [ + { + "bbox": [ + 102, + 546, + 509, + 569 + ], + "spans": [ + { + "bbox": [ + 102, + 546, + 169, + 569 + ], + "score": 1.0, + "content": "Lemma 5. Let", + "type": "text" + }, + { + "bbox": [ + 170, + 550, + 258, + 565 + ], + "score": 0.91, + "content": "l ^ { ( 1 ) } = ( \\ell _ { 1 } ^ { ( 1 ) } , \\dots , \\ell _ { K _ { 1 } } ^ { ( 1 ) } )", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 546, + 277, + 569 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 277, + 550, + 366, + 564 + ], + "score": 0.91, + "content": "l ^ { ( 2 ) } = ( \\ell _ { 1 } ^ { ( 2 ) } , \\dots , \\ell _ { K _ { 2 } } ^ { ( 2 ) } )", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 546, + 414, + 569 + ], + "score": 1.0, + "content": ". A function", + "type": "text" + }, + { + "bbox": [ + 414, + 552, + 494, + 564 + ], + "score": 0.9, + "content": "\\boldsymbol { \\Lambda } = \\left( \\Lambda _ { 1 } , \\ldots , \\Lambda _ { K _ { 2 } } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 546, + 509, + 569 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 563, + 506, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 261, + 577 + ], + "score": 1.0, + "content": "a linear equivariant mapping between", + "type": "text" + }, + { + "bbox": [ + 261, + 564, + 283, + 575 + ], + "score": 0.89, + "content": "W _ { l ^ { ( 1 ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 283, + 563, + 303, + 577 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 303, + 564, + 325, + 575 + ], + "score": 0.89, + "content": "W _ { l ^ { ( 2 ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 563, + 436, + 577 + ], + "score": 1.0, + "content": ", if and only if there exists a", + "type": "text" + }, + { + "bbox": [ + 437, + 564, + 475, + 574 + ], + "score": 0.89, + "content": "K _ { 1 } \\times K _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 563, + 506, + 577 + ], + "score": 1.0, + "content": "matrix", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 107, + 570, + 308, + 594 + ], + "spans": [ + { + "bbox": [ + 107, + 577, + 119, + 587 + ], + "score": 0.72, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 570, + 139, + 594 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 140, + 577, + 176, + 590 + ], + "score": 0.91, + "content": "M _ { i j } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 570, + 217, + 594 + ], + "score": 1.0, + "content": "whenever", + "type": "text" + }, + { + "bbox": [ + 218, + 574, + 261, + 591 + ], + "score": 0.94, + "content": "\\bar { \\ell } _ { i } ^ { ( 1 ) } \\neq \\ell _ { j } ^ { ( 2 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 570, + 308, + 594 + ], + "score": 1.0, + "content": ", such that", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "interline_equation", + "bbox": [ + 263, + 594, + 348, + 629 + ], + "lines": [ + { + "bbox": [ + 263, + 594, + 348, + 629 + ], + "spans": [ + { + "bbox": [ + 263, + 594, + 348, + 629 + ], + "score": 0.94, + "content": "\\Lambda _ { j } ( V ) = \\sum _ { i = 1 } ^ { K _ { 1 } } M _ { i j } V _ { i }", + "type": "interline_equation", + "image_path": "c582f18bbf70a46c752752da5f8838c3a628cadb9e49ebbc6e5a8d85c122efdf.jpg" + } + ] + } + ], + "index": 33.5, + "virtual_lines": [ + { + "bbox": [ + 263, + 594, + 348, + 611.5 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 263, + 611.5, + 348, + 629.0 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 633, + 338, + 650 + ], + "lines": [ + { + "bbox": [ + 104, + 631, + 335, + 651 + ], + "spans": [ + { + "bbox": [ + 104, + 631, + 133, + 650 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 633, + 185, + 648 + ], + "score": 0.93, + "content": "V = ( V _ { i } ) _ { i = 1 } ^ { K _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 631, + 204, + 650 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 205, + 635, + 248, + 651 + ], + "score": 0.92, + "content": "V _ { i } \\in W _ { \\ell _ { i } ^ { ( 1 ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 631, + 278, + 650 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 278, + 636, + 335, + 647 + ], + "score": 0.92, + "content": "i = 1 , \\ldots , K _ { 1 }", + "type": "inline_equation" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 106, + 665, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "Proof idea. This lemma is a simple generalization of Schur’s lemma, a classical tool in representa-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "score": 1.0, + "content": "tion theory, which asserts that a non-zero linear map between irreducible representations is a scaled", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "multiply of the identity mapping. Lemma 5 was stated in the complex setting in Kondor (2018).", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "While Schur’s lemma, and thus Lemma 5, does not always hold for representations over the reals,", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 710, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 506, + 722 + ], + "score": 1.0, + "content": "we observe here that it holds for real irreducible representations of SO(3) since their dimension is", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 720, + 506, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 156, + 733 + ], + "score": 1.0, + "content": "always odd.", + "type": "text" + }, + { + "bbox": [ + 494, + 721, + 506, + 732 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 38.5 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 259, + 505, + 269 + ], + "lines": [ + { + "bbox": [ + 496, + 260, + 505, + 270 + ], + "spans": [ + { + "bbox": [ + 496, + 260, + 505, + 270 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 104, + 82, + 504, + 106 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 96 + ], + "score": 1.0, + "content": "gives us a universal architecture as in Theorem 1. However, the disadvantage of this approach is that", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 405, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 272, + 106 + ], + "score": 1.0, + "content": "implementation of the linear functions in", + "type": "text" + }, + { + "bbox": [ + 272, + 94, + 296, + 106 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { \\mathrm { p o o l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 93, + 405, + 106 + ], + "score": 1.0, + "content": "is somewhat cumbersome.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 105, + 82, + 505, + 106 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 110, + 504, + 144 + ], + "lines": [ + { + "bbox": [ + 105, + 110, + 505, + 123 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 505, + 123 + ], + "score": 1.0, + "content": "In the next section we discuss irreducible representations, which give us a systematic way to address", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 122, + 505, + 134 + ], + "spans": [ + { + "bbox": [ + 106, + 122, + 255, + 134 + ], + "score": 1.0, + "content": "linear equivariant mappings into any", + "type": "text" + }, + { + "bbox": [ + 255, + 122, + 272, + 132 + ], + "score": 0.89, + "content": "W _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 122, + 311, + 134 + ], + "score": 1.0, + "content": ". Proving", + "type": "text" + }, + { + "bbox": [ + 312, + 122, + 321, + 131 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 122, + 505, + 134 + ], + "score": 1.0, + "content": "-spanning for these networks is accomplished", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 133, + 444, + 145 + ], + "spans": [ + { + "bbox": [ + 106, + 133, + 136, + 145 + ], + "score": 1.0, + "content": "via the", + "type": "text" + }, + { + "bbox": [ + 136, + 133, + 146, + 142 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 133, + 444, + 145 + ], + "score": 1.0, + "content": "-spanning property of tensor representations, through the following lemma", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3, + "bbox_fs": [ + 105, + 110, + 505, + 145 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 146, + 314, + 159 + ], + "lines": [ + { + "bbox": [ + 105, + 146, + 314, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 146, + 225, + 160 + ], + "score": 1.0, + "content": "Lemma 4. If all functions in", + "type": "text" + }, + { + "bbox": [ + 225, + 147, + 241, + 158 + ], + "score": 0.88, + "content": "\\mathcal { Q } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 146, + 314, + 160 + ], + "score": 1.0, + "content": "can be written as", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 146, + 314, + 160 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 218, + 162, + 391, + 197 + ], + "lines": [ + { + "bbox": [ + 218, + 162, + 391, + 197 + ], + "spans": [ + { + "bbox": [ + 218, + 162, + 391, + 197 + ], + "score": 0.94, + "content": "\\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac { 1 } { n } X 1 _ { n } 1 _ { n } ^ { T } ) = \\sum _ { k = 1 } ^ { K } \\hat { A } _ { k } f _ { k } ( X ) ,", + "type": "interline_equation", + "image_path": "e30e8f7ccf0c360604f9bbc4ab4bfc3b7f392383f90a98dc50473b858cbedd5b.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 218, + 162, + 391, + 179.5 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 218, + 179.5, + 391, + 197.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 201, + 505, + 226 + ], + "lines": [ + { + "bbox": [ + 103, + 198, + 508, + 222 + ], + "spans": [ + { + "bbox": [ + 103, + 198, + 134, + 222 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 204, + 180, + 215 + ], + "score": 0.67, + "content": "f _ { k } \\in \\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 198, + 185, + 222 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 186, + 202, + 274, + 215 + ], + "score": 0.87, + "content": "A _ { k } : W _ { \\mathrm { f e a t } } \\to W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 198, + 294, + 222 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 295, + 202, + 397, + 216 + ], + "score": 0.92, + "content": "{ \\hat { A } } _ { k } : W _ { \\mathrm { f e a t } } ^ { n } \\to ( W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } } ) ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 198, + 508, + 222 + ], + "score": 1.0, + "content": "is defined by elementwise", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 213, + 286, + 228 + ], + "spans": [ + { + "bbox": [ + 106, + 213, + 164, + 228 + ], + "score": 1.0, + "content": "application of", + "type": "text" + }, + { + "bbox": [ + 165, + 215, + 178, + 226 + ], + "score": 0.87, + "content": "A _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 213, + 201, + 228 + ], + "score": 1.0, + "content": ", then", + "type": "text" + }, + { + "bbox": [ + 202, + 215, + 223, + 226 + ], + "score": 0.81, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 213, + 234, + 228 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 234, + 215, + 244, + 224 + ], + "score": 0.67, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 213, + 286, + 228 + ], + "score": 1.0, + "content": "-spanning.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 103, + 198, + 508, + 228 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 234, + 372, + 246 + ], + "lines": [ + { + "bbox": [ + 106, + 234, + 372, + 248 + ], + "spans": [ + { + "bbox": [ + 106, + 234, + 200, + 248 + ], + "score": 1.0, + "content": "We note that as before,", + "type": "text" + }, + { + "bbox": [ + 200, + 235, + 214, + 246 + ], + "score": 0.89, + "content": "A _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 234, + 372, + 248 + ], + "score": 1.0, + "content": "are not necessarily SO(3)- equivariant.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10, + "bbox_fs": [ + 106, + 234, + 372, + 248 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 258, + 358, + 270 + ], + "lines": [ + { + "bbox": [ + 105, + 257, + 359, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 257, + 359, + 272 + ], + "score": 1.0, + "content": "Proof idea. The lemma follows directly from the assumptions.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11, + "bbox_fs": [ + 105, + 257, + 359, + 272 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 293, + 407, + 306 + ], + "lines": [ + { + "bbox": [ + 104, + 291, + 409, + 310 + ], + "spans": [ + { + "bbox": [ + 104, + 291, + 409, + 310 + ], + "score": 1.0, + "content": "5 UNIVERSALITY WITH IRREDUCIBLE REPRESENTATIONS", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 317, + 505, + 384 + ], + "lines": [ + { + "bbox": [ + 106, + 318, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 318, + 505, + 329 + ], + "score": 1.0, + "content": "In this section, we discuss how to achieve universality when using irreducible representations of", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 328, + 505, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 505, + 341 + ], + "score": 1.0, + "content": "SO(3). We will begin by defining irreducible representations, and explaining how linear univer-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 339, + 506, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 282, + 352 + ], + "score": 1.0, + "content": "sality is easily achieved by them, while the", + "type": "text" + }, + { + "bbox": [ + 282, + 340, + 292, + 349 + ], + "score": 0.78, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 339, + 506, + 352 + ], + "score": 1.0, + "content": "-spanning properties of tensor representations can be", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 351, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 505, + 362 + ], + "score": 1.0, + "content": "preserved. This discussion can be seen as an interpretation of the choices made in the construction", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 362, + 505, + 373 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 505, + 373 + ], + "score": 1.0, + "content": "of TFN and similar networks in the literature. We then show that these architectures are indeed", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 372, + 148, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 148, + 384 + ], + "score": 1.0, + "content": "universal.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 318, + 506, + 384 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 397, + 316, + 409 + ], + "lines": [ + { + "bbox": [ + 105, + 397, + 316, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 287, + 410 + ], + "score": 1.0, + "content": "5.1 IRREDUCIBLE REPRESENTATIONS OF", + "type": "text" + }, + { + "bbox": [ + 288, + 398, + 316, + 410 + ], + "score": 0.84, + "content": "S O ( 3 )", + "type": "inline_equation" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 417, + 505, + 487 + ], + "lines": [ + { + "bbox": [ + 105, + 417, + 506, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 417, + 302, + 431 + ], + "score": 1.0, + "content": "In general, any finite-dimensional representation", + "type": "text" + }, + { + "bbox": [ + 302, + 418, + 314, + 428 + ], + "score": 0.71, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 417, + 395, + 431 + ], + "score": 1.0, + "content": "of a compact group", + "type": "text" + }, + { + "bbox": [ + 395, + 419, + 405, + 428 + ], + "score": 0.81, + "content": "H", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 417, + 506, + 431 + ], + "score": 1.0, + "content": "can be decomposed into", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 428, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 262, + 442 + ], + "score": 1.0, + "content": "irreducible representations: a subspace", + "type": "text" + }, + { + "bbox": [ + 262, + 429, + 301, + 440 + ], + "score": 0.92, + "content": "W _ { 0 } \\subset W", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 428, + 311, + 442 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 311, + 429, + 321, + 439 + ], + "score": 0.83, + "content": "H", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 428, + 369, + 442 + ], + "score": 1.0, + "content": "-invariant if", + "type": "text" + }, + { + "bbox": [ + 369, + 429, + 409, + 440 + ], + "score": 0.92, + "content": "h w \\in W _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 428, + 435, + 442 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 436, + 429, + 501, + 440 + ], + "score": 0.91, + "content": "h \\in H , w \\in W _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 428, + 505, + 442 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 439, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 174, + 452 + ], + "score": 1.0, + "content": "A representation", + "type": "text" + }, + { + "bbox": [ + 174, + 440, + 186, + 450 + ], + "score": 0.7, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 439, + 505, + 452 + ], + "score": 1.0, + "content": "is irreducible if it has no non-trivial invariant subspaces. In the case of SO(3), all", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 451, + 506, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 327, + 465 + ], + "score": 1.0, + "content": "irreducible real representations are defined by matrices", + "type": "text" + }, + { + "bbox": [ + 327, + 451, + 362, + 464 + ], + "score": 0.94, + "content": "D ^ { ( \\ell ) } ( R )", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 451, + 506, + 465 + ], + "score": 1.0, + "content": ", called the real Wigner D-matrices,", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 460, + 507, + 477 + ], + "spans": [ + { + "bbox": [ + 104, + 460, + 146, + 477 + ], + "score": 1.0, + "content": "acting on", + "type": "text" + }, + { + "bbox": [ + 147, + 462, + 204, + 474 + ], + "score": 0.92, + "content": "W _ { \\ell } : = \\overset { \\bullet } { \\mathbb { R } } ^ { 2 \\ell + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 460, + 454, + 477 + ], + "score": 1.0, + "content": "by matrix multiplication. In particular, the representation for", + "type": "text" + }, + { + "bbox": [ + 454, + 463, + 489, + 474 + ], + "score": 0.92, + "content": "\\ell = 0 , 1", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 460, + 507, + 477 + ], + "score": 1.0, + "content": "are", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 473, + 239, + 487 + ], + "spans": [ + { + "bbox": [ + 106, + 474, + 161, + 487 + ], + "score": 0.93, + "content": "D ^ { ( 0 ) } ( R ) = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 473, + 178, + 487 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 179, + 474, + 236, + 487 + ], + "score": 0.93, + "content": "D ^ { ( 1 ) } ( R ) = R", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 473, + 239, + 487 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 22.5, + "bbox_fs": [ + 104, + 417, + 507, + 487 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 492, + 505, + 539 + ], + "lines": [ + { + "bbox": [ + 105, + 492, + 505, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 505, + 504 + ], + "score": 1.0, + "content": "Linear maps between irreducible representations As mentioned above, one of the main advan-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 503, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 503, + 506, + 515 + ], + "score": 1.0, + "content": "tages of using irreducible representations is that there is a very simple characterization of all linear", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 514, + 504, + 526 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 489, + 526 + ], + "score": 1.0, + "content": "equivariant maps between two direct sums of irreducible representations. We use the notation", + "type": "text" + }, + { + "bbox": [ + 490, + 514, + 504, + 525 + ], + "score": 0.85, + "content": "W _ { l }", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 102, + 522, + 501, + 545 + ], + "spans": [ + { + "bbox": [ + 102, + 522, + 316, + 545 + ], + "score": 1.0, + "content": "for direct sums of irreducible representations, where", + "type": "text" + }, + { + "bbox": [ + 316, + 525, + 411, + 540 + ], + "score": 0.91, + "content": "\\boldsymbol { l } = ( \\ell _ { 1 } , \\dots , \\ell _ { K } ) \\in \\mathbb { N } _ { + } ^ { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 522, + 429, + 545 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 429, + 524, + 501, + 540 + ], + "score": 0.93, + "content": "W _ { l } = \\bigoplus _ { k = 1 } ^ { K } W _ { \\ell _ { k } }", + "type": "inline_equation" + } + ], + "index": 29 + } + ], + "index": 27.5, + "bbox_fs": [ + 102, + 492, + 506, + 545 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 549, + 505, + 590 + ], + "lines": [ + { + "bbox": [ + 102, + 546, + 509, + 569 + ], + "spans": [ + { + "bbox": [ + 102, + 546, + 169, + 569 + ], + "score": 1.0, + "content": "Lemma 5. Let", + "type": "text" + }, + { + "bbox": [ + 170, + 550, + 258, + 565 + ], + "score": 0.91, + "content": "l ^ { ( 1 ) } = ( \\ell _ { 1 } ^ { ( 1 ) } , \\dots , \\ell _ { K _ { 1 } } ^ { ( 1 ) } )", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 546, + 277, + 569 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 277, + 550, + 366, + 564 + ], + "score": 0.91, + "content": "l ^ { ( 2 ) } = ( \\ell _ { 1 } ^ { ( 2 ) } , \\dots , \\ell _ { K _ { 2 } } ^ { ( 2 ) } )", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 546, + 414, + 569 + ], + "score": 1.0, + "content": ". A function", + "type": "text" + }, + { + "bbox": [ + 414, + 552, + 494, + 564 + ], + "score": 0.9, + "content": "\\boldsymbol { \\Lambda } = \\left( \\Lambda _ { 1 } , \\ldots , \\Lambda _ { K _ { 2 } } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 546, + 509, + 569 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 563, + 506, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 261, + 577 + ], + "score": 1.0, + "content": "a linear equivariant mapping between", + "type": "text" + }, + { + "bbox": [ + 261, + 564, + 283, + 575 + ], + "score": 0.89, + "content": "W _ { l ^ { ( 1 ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 283, + 563, + 303, + 577 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 303, + 564, + 325, + 575 + ], + "score": 0.89, + "content": "W _ { l ^ { ( 2 ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 563, + 436, + 577 + ], + "score": 1.0, + "content": ", if and only if there exists a", + "type": "text" + }, + { + "bbox": [ + 437, + 564, + 475, + 574 + ], + "score": 0.89, + "content": "K _ { 1 } \\times K _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 563, + 506, + 577 + ], + "score": 1.0, + "content": "matrix", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 107, + 570, + 308, + 594 + ], + "spans": [ + { + "bbox": [ + 107, + 577, + 119, + 587 + ], + "score": 0.72, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 570, + 139, + 594 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 140, + 577, + 176, + 590 + ], + "score": 0.91, + "content": "M _ { i j } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 570, + 217, + 594 + ], + "score": 1.0, + "content": "whenever", + "type": "text" + }, + { + "bbox": [ + 218, + 574, + 261, + 591 + ], + "score": 0.94, + "content": "\\bar { \\ell } _ { i } ^ { ( 1 ) } \\neq \\ell _ { j } ^ { ( 2 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 570, + 308, + 594 + ], + "score": 1.0, + "content": ", such that", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 102, + 546, + 509, + 594 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 263, + 594, + 348, + 629 + ], + "lines": [ + { + "bbox": [ + 263, + 594, + 348, + 629 + ], + "spans": [ + { + "bbox": [ + 263, + 594, + 348, + 629 + ], + "score": 0.94, + "content": "\\Lambda _ { j } ( V ) = \\sum _ { i = 1 } ^ { K _ { 1 } } M _ { i j } V _ { i }", + "type": "interline_equation", + "image_path": "c582f18bbf70a46c752752da5f8838c3a628cadb9e49ebbc6e5a8d85c122efdf.jpg" + } + ] + } + ], + "index": 33.5, + "virtual_lines": [ + { + "bbox": [ + 263, + 594, + 348, + 611.5 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 263, + 611.5, + 348, + 629.0 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 633, + 338, + 650 + ], + "lines": [ + { + "bbox": [ + 104, + 631, + 335, + 651 + ], + "spans": [ + { + "bbox": [ + 104, + 631, + 133, + 650 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 633, + 185, + 648 + ], + "score": 0.93, + "content": "V = ( V _ { i } ) _ { i = 1 } ^ { K _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 631, + 204, + 650 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 205, + 635, + 248, + 651 + ], + "score": 0.92, + "content": "V _ { i } \\in W _ { \\ell _ { i } ^ { ( 1 ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 631, + 278, + 650 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 278, + 636, + 335, + 647 + ], + "score": 0.92, + "content": "i = 1 , \\ldots , K _ { 1 }", + "type": "inline_equation" + } + ], + "index": 35 + } + ], + "index": 35, + "bbox_fs": [ + 104, + 631, + 335, + 651 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 665, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 505, + 678 + ], + "score": 1.0, + "content": "Proof idea. This lemma is a simple generalization of Schur’s lemma, a classical tool in representa-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 506, + 689 + ], + "score": 1.0, + "content": "tion theory, which asserts that a non-zero linear map between irreducible representations is a scaled", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "multiply of the identity mapping. Lemma 5 was stated in the complex setting in Kondor (2018).", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "While Schur’s lemma, and thus Lemma 5, does not always hold for representations over the reals,", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 710, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 506, + 722 + ], + "score": 1.0, + "content": "we observe here that it holds for real irreducible representations of SO(3) since their dimension is", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 720, + 506, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 156, + 733 + ], + "score": 1.0, + "content": "always odd.", + "type": "text" + }, + { + "bbox": [ + 494, + 721, + 506, + 732 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 38.5, + "bbox_fs": [ + 105, + 665, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 182 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 506, + 95 + ], + "score": 1.0, + "content": "Clebsch-Gordan decomposition of tensor products As any finite-dimensional representation of", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "SO(3) can be decomposed into a direct sum of irreducible representations, this is true for tensor", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "score": 1.0, + "content": "representations as well. In particular, the Clebsch-Gordan coefficients provide an explicit formula", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 115, + 383, + 128 + ], + "score": 1.0, + "content": "for decomposing the tensor product of two irreducible representations", + "type": "text" + }, + { + "bbox": [ + 383, + 116, + 401, + 127 + ], + "score": 0.91, + "content": "W _ { \\ell _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 115, + 418, + 128 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 419, + 115, + 436, + 127 + ], + "score": 0.92, + "content": "W _ { \\ell _ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 436, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "into a direct sum", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 505, + 140 + ], + "score": 1.0, + "content": "of irreducible representations. This decomposition can be easily extended to decompose the tensor", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 505, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 140, + 150 + ], + "score": 1.0, + "content": "product", + "type": "text" + }, + { + "bbox": [ + 140, + 137, + 187, + 149 + ], + "score": 0.93, + "content": "W _ { l _ { 1 } } \\otimes W _ { l _ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 137, + 412, + 150 + ], + "score": 1.0, + "content": "into a direct sum of irreducible representations, where", + "type": "text" + }, + { + "bbox": [ + 412, + 137, + 434, + 149 + ], + "score": 0.91, + "content": "l _ { 1 } , l _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 137, + 505, + 150 + ], + "score": 1.0, + "content": "are now vectors.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 146, + 503, + 163 + ], + "spans": [ + { + "bbox": [ + 105, + 146, + 370, + 163 + ], + "score": 1.0, + "content": "In matrix notation, this means there is a unitary linear equivariant", + "type": "text" + }, + { + "bbox": [ + 370, + 149, + 407, + 160 + ], + "score": 0.89, + "content": "U ( l _ { 1 } , l _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 146, + 457, + 163 + ], + "score": 1.0, + "content": "mapping of", + "type": "text" + }, + { + "bbox": [ + 457, + 149, + 503, + 160 + ], + "score": 0.91, + "content": "W _ { l _ { 1 } } \\otimes W _ { l _ { 2 } }", + "type": "inline_equation" + } + ], + "index": 6 + }, + { + "bbox": [ + 104, + 157, + 505, + 173 + ], + "spans": [ + { + "bbox": [ + 104, + 157, + 127, + 173 + ], + "score": 1.0, + "content": "onto", + "type": "text" + }, + { + "bbox": [ + 127, + 159, + 141, + 170 + ], + "score": 0.88, + "content": "W _ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 157, + 258, + 173 + ], + "score": 1.0, + "content": ", where the explicit values of", + "type": "text" + }, + { + "bbox": [ + 258, + 159, + 308, + 171 + ], + "score": 0.93, + "content": "l = l ( l _ { 1 } , l _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 157, + 370, + 173 + ], + "score": 1.0, + "content": "and the matrix", + "type": "text" + }, + { + "bbox": [ + 371, + 160, + 407, + 171 + ], + "score": 0.91, + "content": "U ( l _ { 1 } , l _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 157, + 505, + 173 + ], + "score": 1.0, + "content": "can be inferred directly", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 170, + 275, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 170, + 189, + 182 + ], + "score": 1.0, + "content": "from the case where", + "type": "text" + }, + { + "bbox": [ + 189, + 171, + 199, + 181 + ], + "score": 0.88, + "content": "\\ell _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 170, + 217, + 182 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 217, + 171, + 227, + 181 + ], + "score": 0.86, + "content": "\\ell _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 170, + 275, + 182 + ], + "score": 1.0, + "content": "are scalars.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 107, + 187, + 505, + 231 + ], + "lines": [ + { + "bbox": [ + 106, + 187, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 106, + 187, + 505, + 199 + ], + "score": 1.0, + "content": "By repeatedly taking tensor products and applying Clebsch-Gordan decompositions to the result,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 197, + 505, + 212 + ], + "spans": [ + { + "bbox": [ + 105, + 197, + 290, + 212 + ], + "score": 1.0, + "content": "TFN and similar architectures can achieve the", + "type": "text" + }, + { + "bbox": [ + 290, + 199, + 300, + 208 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 197, + 505, + 212 + ], + "score": 1.0, + "content": "-spanning property in a manner analogous to tensor", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 209, + 505, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 505, + 222 + ], + "score": 1.0, + "content": "representations, and also enjoy linear universality since they maintain irreducible representations", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 221, + 204, + 232 + ], + "spans": [ + { + "bbox": [ + 106, + 221, + 204, + 232 + ], + "score": 1.0, + "content": "throughout the network.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5 + }, + { + "type": "title", + "bbox": [ + 108, + 244, + 245, + 255 + ], + "lines": [ + { + "bbox": [ + 105, + 243, + 246, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 243, + 246, + 257 + ], + "score": 1.0, + "content": "5.2 TENSOR FIELD NETWORKS", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 264, + 505, + 321 + ], + "lines": [ + { + "bbox": [ + 106, + 265, + 504, + 277 + ], + "spans": [ + { + "bbox": [ + 106, + 265, + 504, + 277 + ], + "score": 1.0, + "content": "We now describe the basic layers of the TFN architecture (Thomas et al., 2018), which are based on", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 276, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 505, + 288 + ], + "score": 1.0, + "content": "irreducible representations, and suggest an architecture based on these layers which can approximate", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 107, + 284, + 507, + 302 + ], + "spans": [ + { + "bbox": [ + 107, + 288, + 115, + 297 + ], + "score": 0.73, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 116, + 284, + 284, + 302 + ], + "score": 1.0, + "content": "-equivariant maps into any representation", + "type": "text" + }, + { + "bbox": [ + 284, + 287, + 344, + 300 + ], + "score": 0.92, + "content": "W _ { l _ { T } } ^ { n } , l _ { T } \\in \\mathbb { N } _ { + } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 284, + 507, + 302 + ], + "score": 1.0, + "content": ". There are some superficial differences", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 296, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 104, + 296, + 505, + 312 + ], + "score": 1.0, + "content": "Tbetween our description of TFN and the description in the original paper, for more details see Ap-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 309, + 147, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 309, + 147, + 320 + ], + "score": 1.0, + "content": "pendix F.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 107, + 325, + 505, + 359 + ], + "lines": [ + { + "bbox": [ + 106, + 326, + 505, + 338 + ], + "spans": [ + { + "bbox": [ + 106, + 326, + 505, + 338 + ], + "score": 1.0, + "content": "We note that the universality of TFN also implies the universality of Fuchs et al. (2020), which", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 337, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 105, + 337, + 505, + 349 + ], + "score": 1.0, + "content": "is a generalization of TFN that enables adding an attention mechanism. Assuming the attention", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 348, + 493, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 348, + 493, + 360 + ], + "score": 1.0, + "content": "mechanism is not restricted to local neighborhoods, this method is at least as expressive as TFN.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 108, + 364, + 504, + 388 + ], + "lines": [ + { + "bbox": [ + 107, + 365, + 504, + 377 + ], + "spans": [ + { + "bbox": [ + 107, + 365, + 504, + 377 + ], + "score": 1.0, + "content": "TFNs are composed of three types of layers: (i) Convolution (ii) Self-interaction and (iii) Non-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 375, + 504, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 504, + 389 + ], + "score": 1.0, + "content": "linearities. In our architecture, we only use the first two layers types, which we will now describe:1.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 106, + 392, + 505, + 439 + ], + "lines": [ + { + "bbox": [ + 105, + 392, + 506, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 506, + 406 + ], + "score": 1.0, + "content": "Convolution. Convolutional layers involve taking tensor products of a filter and a feature vector", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 404, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 105, + 404, + 505, + 416 + ], + "score": 1.0, + "content": "to create a new feature vector, and then decomposing into irreducible representations. Unlike in", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 104, + 412, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 104, + 412, + 381, + 428 + ], + "score": 1.0, + "content": "standard CNN, a filter here depends on the input, and is a function", + "type": "text" + }, + { + "bbox": [ + 381, + 414, + 448, + 426 + ], + "score": 0.91, + "content": "F : \\bar { \\mathbb { R } ^ { 3 } } W _ { l _ { D } }", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 412, + 480, + 428 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 480, + 415, + 505, + 426 + ], + "score": 0.88, + "content": "\\boldsymbol { l } _ { D } =", + "type": "inline_equation" + } + ], + "index": 26 + }, + { + "bbox": [ + 107, + 425, + 500, + 442 + ], + "spans": [ + { + "bbox": [ + 107, + 426, + 164, + 439 + ], + "score": 0.92, + "content": "[ 0 , 1 , \\ldots , D ] ^ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 425, + 187, + 442 + ], + "score": 1.0, + "content": ". The", + "type": "text" + }, + { + "bbox": [ + 187, + 427, + 192, + 437 + ], + "score": 0.77, + "content": "\\ell", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 425, + 298, + 442 + ], + "score": 1.0, + "content": "-th component of the filter", + "type": "text" + }, + { + "bbox": [ + 299, + 425, + 430, + 440 + ], + "score": 0.93, + "content": "F ( \\boldsymbol { x } ) = \\left[ F ^ { ( 0 ) } ( x ) , \\ldots , F ^ { ( D ) } ( x ) \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 425, + 500, + 442 + ], + "score": 1.0, + "content": "will be given by", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5 + }, + { + "type": "interline_equation", + "bbox": [ + 214, + 443, + 396, + 458 + ], + "lines": [ + { + "bbox": [ + 214, + 443, + 396, + 458 + ], + "spans": [ + { + "bbox": [ + 214, + 443, + 396, + 458 + ], + "score": 0.91, + "content": "F _ { m } ^ { \\ell } ( x ) = R ^ { ( \\ell ) } ( \\| x \\| ) Y _ { m } ^ { \\ell } ( \\hat { x } ) , m = - \\ell , \\ldots , \\ell", + "type": "interline_equation", + "image_path": "ab43c4063a462588d15cfce3a32fb29492e677c37687cb4a82ed38c439545931.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 214, + 443, + 396, + 458 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 462, + 505, + 497 + ], + "lines": [ + { + "bbox": [ + 105, + 460, + 506, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 135, + 477 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 463, + 186, + 475 + ], + "score": 0.92, + "content": "\\hat { x } = x / \\| x \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 460, + 198, + 477 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 199, + 463, + 229, + 474 + ], + "score": 0.91, + "content": "x \\neq 0", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 460, + 249, + 477 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 249, + 463, + 280, + 473 + ], + "score": 0.91, + "content": "\\hat { x } ~ = ~ 0", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 460, + 327, + 477 + ], + "score": 1.0, + "content": "otherwise,", + "type": "text" + }, + { + "bbox": [ + 327, + 462, + 342, + 475 + ], + "score": 0.91, + "content": "Y _ { m } ^ { \\ell }", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 460, + 467, + 477 + ], + "score": 1.0, + "content": "are spherical harmonics, and", + "type": "text" + }, + { + "bbox": [ + 467, + 462, + 486, + 473 + ], + "score": 0.9, + "content": "R ^ { ( \\ell ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 460, + 506, + 477 + ], + "score": 1.0, + "content": "any", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 474, + 505, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 193, + 487 + ], + "score": 1.0, + "content": "polynomial of degree", + "type": "text" + }, + { + "bbox": [ + 193, + 475, + 213, + 485 + ], + "score": 0.87, + "content": "\\leq D", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 474, + 505, + 487 + ], + "score": 1.0, + "content": ". In Appendix F we show that these polynomial functions can be replaced", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 485, + 462, + 497 + ], + "spans": [ + { + "bbox": [ + 106, + 485, + 462, + 497 + ], + "score": 1.0, + "content": "by fully connected networks, since the latter can approximate all polynomials uniformly.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 501, + 504, + 528 + ], + "lines": [ + { + "bbox": [ + 105, + 500, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 253, + 515 + ], + "score": 1.0, + "content": "The convolution of an input feature", + "type": "text" + }, + { + "bbox": [ + 254, + 502, + 293, + 515 + ], + "score": 0.91, + "content": "V \\in W _ { l _ { i } } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 500, + 342, + 515 + ], + "score": 1.0, + "content": "and a filter", + "type": "text" + }, + { + "bbox": [ + 342, + 502, + 351, + 512 + ], + "score": 0.84, + "content": "F", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 500, + 505, + 515 + ], + "score": 1.0, + "content": "as defined above, will give an output", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 513, + 382, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 137, + 531 + ], + "score": 1.0, + "content": "feature", + "type": "text" + }, + { + "bbox": [ + 137, + 514, + 221, + 529 + ], + "score": 0.93, + "content": "\\tilde { V } = ( \\tilde { V } _ { a } ) _ { a = 1 } ^ { n } \\in W _ { l _ { 0 } } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 513, + 252, + 531 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 252, + 516, + 305, + 528 + ], + "score": 0.92, + "content": "\\boldsymbol { l _ { o } } = \\boldsymbol { l } ( \\boldsymbol { l _ { f } } , \\boldsymbol { l _ { i } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 513, + 382, + 531 + ], + "score": 1.0, + "content": ", which is given by", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5 + }, + { + "type": "interline_equation", + "bbox": [ + 190, + 532, + 421, + 566 + ], + "lines": [ + { + "bbox": [ + 190, + 532, + 421, + 566 + ], + "spans": [ + { + "bbox": [ + 190, + 532, + 421, + 566 + ], + "score": 0.94, + "content": "\\tilde { V } _ { a } ( X , V ) = U ( l _ { f } , l _ { i } ) \\left( \\theta _ { 0 } V _ { a } + \\sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \\otimes V _ { b } \\right) .", + "type": "interline_equation", + "image_path": "965407aa486d2589a2966a08fbbe90a36b2b8b549c1c3a671e196f7ef48eb5f8.jpg" + } + ] + } + ], + "index": 34.5, + "virtual_lines": [ + { + "bbox": [ + 190, + 532, + 421, + 549.0 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 190, + 549.0, + 421, + 566.0 + ], + "spans": [], + "index": 35 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 569, + 505, + 618 + ], + "lines": [ + { + "bbox": [ + 105, + 569, + 504, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 569, + 451, + 583 + ], + "score": 1.0, + "content": "More formally we will think of convolutional layer as functions of the form", + "type": "text" + }, + { + "bbox": [ + 452, + 569, + 504, + 582 + ], + "score": 0.89, + "content": "f ( X , V ) \\quad = \\quad", + "type": "inline_equation" + } + ], + "index": 36 + }, + { + "bbox": [ + 107, + 581, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 107, + 581, + 164, + 595 + ], + "score": 0.89, + "content": "( X , { \\tilde { V } } ( X , V ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 582, + 356, + 594 + ], + "score": 1.0, + "content": ". These functions are defined by a choice of", + "type": "text" + }, + { + "bbox": [ + 356, + 583, + 366, + 592 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 582, + 505, + 594 + ], + "score": 1.0, + "content": ", a choice of a scalar polynomial", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 107, + 592, + 506, + 608 + ], + "spans": [ + { + "bbox": [ + 107, + 595, + 184, + 606 + ], + "score": 0.87, + "content": "R ^ { ( \\ell ) } , \\ell = 0 , \\dots , D", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 592, + 310, + 608 + ], + "score": 1.0, + "content": ", and a choice of the parameter", + "type": "text" + }, + { + "bbox": [ + 311, + 595, + 342, + 606 + ], + "score": 0.92, + "content": "\\theta _ { 0 } \\in \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 592, + 506, + 608 + ], + "score": 1.0, + "content": "in equation 14. We denote the set of all", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 604, + 207, + 619 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 167, + 619 + ], + "score": 1.0, + "content": "such functions", + "type": "text" + }, + { + "bbox": [ + 167, + 607, + 174, + 617 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 604, + 187, + 619 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 188, + 606, + 203, + 617 + ], + "score": 0.89, + "content": "\\mathcal { F } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 604, + 207, + 619 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 106, + 623, + 505, + 658 + ], + "lines": [ + { + "bbox": [ + 104, + 622, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 104, + 622, + 394, + 639 + ], + "score": 1.0, + "content": "Self Interaction layers. Self interaction layers are linear functions from", + "type": "text" + }, + { + "bbox": [ + 394, + 623, + 460, + 636 + ], + "score": 0.9, + "content": "\\hat { \\Lambda } : W _ { l } ^ { n } \\to W _ { l , r } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 460, + 622, + 506, + 639 + ], + "score": 1.0, + "content": ", which are", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 635, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 105, + 635, + 386, + 648 + ], + "score": 1.0, + "content": "obtained from elementwise application of equivariant linear functions", + "type": "text" + }, + { + "bbox": [ + 387, + 636, + 449, + 647 + ], + "score": 0.91, + "content": "\\Lambda : W _ { l } \\stackrel { \\cdot } { } W _ { l _ { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 635, + 505, + 648 + ], + "score": 1.0, + "content": ". These linear", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 645, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 106, + 645, + 299, + 659 + ], + "score": 1.0, + "content": "functions can be specified by a choice of matrix", + "type": "text" + }, + { + "bbox": [ + 299, + 646, + 312, + 656 + ], + "score": 0.78, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 645, + 505, + 659 + ], + "score": 1.0, + "content": "with the sparsity pattern described in Lemma 5.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 107, + 663, + 504, + 686 + ], + "lines": [ + { + "bbox": [ + 106, + 663, + 505, + 676 + ], + "spans": [ + { + "bbox": [ + 106, + 663, + 505, + 676 + ], + "score": 1.0, + "content": "Activation functions. TFN, as well as other papers, proposed several activation functions. We find", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 675, + 456, + 686 + ], + "spans": [ + { + "bbox": [ + 106, + 675, + 456, + 686 + ], + "score": 1.0, + "content": "that these layers are not necessary for universality and thus we do not define them here.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5 + }, + { + "type": "text", + "bbox": [ + 107, + 690, + 504, + 714 + ], + "lines": [ + { + "bbox": [ + 105, + 690, + 505, + 703 + ], + "spans": [ + { + "bbox": [ + 105, + 690, + 505, + 703 + ], + "score": 1.0, + "content": "Network architecture. For our universality proof, we suggest a simple architecture which depends", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 702, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 105, + 702, + 253, + 714 + ], + "score": 1.0, + "content": "on two positive integer parameters", + "type": "text" + }, + { + "bbox": [ + 253, + 703, + 281, + 714 + ], + "score": 0.91, + "content": "( C , D )", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 702, + 332, + 714 + ], + "score": 1.0, + "content": ": For given", + "type": "text" + }, + { + "bbox": [ + 333, + 703, + 342, + 712 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 702, + 410, + 714 + ], + "score": 1.0, + "content": ", we will define", + "type": "text" + }, + { + "bbox": [ + 411, + 702, + 448, + 714 + ], + "score": 0.93, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 702, + 506, + 714 + ], + "score": 1.0, + "content": "as the set of", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 117, + 721, + 446, + 732 + ], + "lines": [ + { + "bbox": [ + 119, + 720, + 446, + 733 + ], + "spans": [ + { + "bbox": [ + 119, + 720, + 446, + 733 + ], + "score": 1.0, + "content": "1Since convolution layers in TFN are not linear, the non-linearities are formally redundant", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 108, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "spans": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "score": 1.0, + "content": "8", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 182 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 506, + 95 + ], + "score": 1.0, + "content": "Clebsch-Gordan decomposition of tensor products As any finite-dimensional representation of", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "SO(3) can be decomposed into a direct sum of irreducible representations, this is true for tensor", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "score": 1.0, + "content": "representations as well. In particular, the Clebsch-Gordan coefficients provide an explicit formula", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 115, + 383, + 128 + ], + "score": 1.0, + "content": "for decomposing the tensor product of two irreducible representations", + "type": "text" + }, + { + "bbox": [ + 383, + 116, + 401, + 127 + ], + "score": 0.91, + "content": "W _ { \\ell _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 115, + 418, + 128 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 419, + 115, + 436, + 127 + ], + "score": 0.92, + "content": "W _ { \\ell _ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 436, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "into a direct sum", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 505, + 140 + ], + "score": 1.0, + "content": "of irreducible representations. This decomposition can be easily extended to decompose the tensor", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 505, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 140, + 150 + ], + "score": 1.0, + "content": "product", + "type": "text" + }, + { + "bbox": [ + 140, + 137, + 187, + 149 + ], + "score": 0.93, + "content": "W _ { l _ { 1 } } \\otimes W _ { l _ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 137, + 412, + 150 + ], + "score": 1.0, + "content": "into a direct sum of irreducible representations, where", + "type": "text" + }, + { + "bbox": [ + 412, + 137, + 434, + 149 + ], + "score": 0.91, + "content": "l _ { 1 } , l _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 137, + 505, + 150 + ], + "score": 1.0, + "content": "are now vectors.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 146, + 503, + 163 + ], + "spans": [ + { + "bbox": [ + 105, + 146, + 370, + 163 + ], + "score": 1.0, + "content": "In matrix notation, this means there is a unitary linear equivariant", + "type": "text" + }, + { + "bbox": [ + 370, + 149, + 407, + 160 + ], + "score": 0.89, + "content": "U ( l _ { 1 } , l _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 146, + 457, + 163 + ], + "score": 1.0, + "content": "mapping of", + "type": "text" + }, + { + "bbox": [ + 457, + 149, + 503, + 160 + ], + "score": 0.91, + "content": "W _ { l _ { 1 } } \\otimes W _ { l _ { 2 } }", + "type": "inline_equation" + } + ], + "index": 6 + }, + { + "bbox": [ + 104, + 157, + 505, + 173 + ], + "spans": [ + { + "bbox": [ + 104, + 157, + 127, + 173 + ], + "score": 1.0, + "content": "onto", + "type": "text" + }, + { + "bbox": [ + 127, + 159, + 141, + 170 + ], + "score": 0.88, + "content": "W _ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 157, + 258, + 173 + ], + "score": 1.0, + "content": ", where the explicit values of", + "type": "text" + }, + { + "bbox": [ + 258, + 159, + 308, + 171 + ], + "score": 0.93, + "content": "l = l ( l _ { 1 } , l _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 157, + 370, + 173 + ], + "score": 1.0, + "content": "and the matrix", + "type": "text" + }, + { + "bbox": [ + 371, + 160, + 407, + 171 + ], + "score": 0.91, + "content": "U ( l _ { 1 } , l _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 408, + 157, + 505, + 173 + ], + "score": 1.0, + "content": "can be inferred directly", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 170, + 275, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 170, + 189, + 182 + ], + "score": 1.0, + "content": "from the case where", + "type": "text" + }, + { + "bbox": [ + 189, + 171, + 199, + 181 + ], + "score": 0.88, + "content": "\\ell _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 170, + 217, + 182 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 217, + 171, + 227, + 181 + ], + "score": 0.86, + "content": "\\ell _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 170, + 275, + 182 + ], + "score": 1.0, + "content": "are scalars.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 4, + "bbox_fs": [ + 104, + 82, + 506, + 182 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 187, + 505, + 231 + ], + "lines": [ + { + "bbox": [ + 106, + 187, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 106, + 187, + 505, + 199 + ], + "score": 1.0, + "content": "By repeatedly taking tensor products and applying Clebsch-Gordan decompositions to the result,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 197, + 505, + 212 + ], + "spans": [ + { + "bbox": [ + 105, + 197, + 290, + 212 + ], + "score": 1.0, + "content": "TFN and similar architectures can achieve the", + "type": "text" + }, + { + "bbox": [ + 290, + 199, + 300, + 208 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 197, + 505, + 212 + ], + "score": 1.0, + "content": "-spanning property in a manner analogous to tensor", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 209, + 505, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 505, + 222 + ], + "score": 1.0, + "content": "representations, and also enjoy linear universality since they maintain irreducible representations", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 221, + 204, + 232 + ], + "spans": [ + { + "bbox": [ + 106, + 221, + 204, + 232 + ], + "score": 1.0, + "content": "throughout the network.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 187, + 505, + 232 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 244, + 245, + 255 + ], + "lines": [ + { + "bbox": [ + 105, + 243, + 246, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 243, + 246, + 257 + ], + "score": 1.0, + "content": "5.2 TENSOR FIELD NETWORKS", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 264, + 505, + 321 + ], + "lines": [ + { + "bbox": [ + 106, + 265, + 504, + 277 + ], + "spans": [ + { + "bbox": [ + 106, + 265, + 504, + 277 + ], + "score": 1.0, + "content": "We now describe the basic layers of the TFN architecture (Thomas et al., 2018), which are based on", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 276, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 505, + 288 + ], + "score": 1.0, + "content": "irreducible representations, and suggest an architecture based on these layers which can approximate", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 107, + 284, + 507, + 302 + ], + "spans": [ + { + "bbox": [ + 107, + 288, + 115, + 297 + ], + "score": 0.73, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 116, + 284, + 284, + 302 + ], + "score": 1.0, + "content": "-equivariant maps into any representation", + "type": "text" + }, + { + "bbox": [ + 284, + 287, + 344, + 300 + ], + "score": 0.92, + "content": "W _ { l _ { T } } ^ { n } , l _ { T } \\in \\mathbb { N } _ { + } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 284, + 507, + 302 + ], + "score": 1.0, + "content": ". There are some superficial differences", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 296, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 104, + 296, + 505, + 312 + ], + "score": 1.0, + "content": "Tbetween our description of TFN and the description in the original paper, for more details see Ap-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 309, + 147, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 309, + 147, + 320 + ], + "score": 1.0, + "content": "pendix F.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16, + "bbox_fs": [ + 104, + 265, + 507, + 320 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 325, + 505, + 359 + ], + "lines": [ + { + "bbox": [ + 106, + 326, + 505, + 338 + ], + "spans": [ + { + "bbox": [ + 106, + 326, + 505, + 338 + ], + "score": 1.0, + "content": "We note that the universality of TFN also implies the universality of Fuchs et al. (2020), which", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 337, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 105, + 337, + 505, + 349 + ], + "score": 1.0, + "content": "is a generalization of TFN that enables adding an attention mechanism. Assuming the attention", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 348, + 493, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 348, + 493, + 360 + ], + "score": 1.0, + "content": "mechanism is not restricted to local neighborhoods, this method is at least as expressive as TFN.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 326, + 505, + 360 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 364, + 504, + 388 + ], + "lines": [ + { + "bbox": [ + 107, + 365, + 504, + 377 + ], + "spans": [ + { + "bbox": [ + 107, + 365, + 504, + 377 + ], + "score": 1.0, + "content": "TFNs are composed of three types of layers: (i) Convolution (ii) Self-interaction and (iii) Non-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 375, + 504, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 504, + 389 + ], + "score": 1.0, + "content": "linearities. In our architecture, we only use the first two layers types, which we will now describe:1.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 365, + 504, + 389 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 392, + 505, + 439 + ], + "lines": [ + { + "bbox": [ + 105, + 392, + 506, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 506, + 406 + ], + "score": 1.0, + "content": "Convolution. Convolutional layers involve taking tensor products of a filter and a feature vector", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 404, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 105, + 404, + 505, + 416 + ], + "score": 1.0, + "content": "to create a new feature vector, and then decomposing into irreducible representations. Unlike in", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 104, + 412, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 104, + 412, + 381, + 428 + ], + "score": 1.0, + "content": "standard CNN, a filter here depends on the input, and is a function", + "type": "text" + }, + { + "bbox": [ + 381, + 414, + 448, + 426 + ], + "score": 0.91, + "content": "F : \\bar { \\mathbb { R } ^ { 3 } } W _ { l _ { D } }", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 412, + 480, + 428 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 480, + 415, + 505, + 426 + ], + "score": 0.88, + "content": "\\boldsymbol { l } _ { D } =", + "type": "inline_equation" + } + ], + "index": 26 + }, + { + "bbox": [ + 107, + 425, + 500, + 442 + ], + "spans": [ + { + "bbox": [ + 107, + 426, + 164, + 439 + ], + "score": 0.92, + "content": "[ 0 , 1 , \\ldots , D ] ^ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 425, + 187, + 442 + ], + "score": 1.0, + "content": ". The", + "type": "text" + }, + { + "bbox": [ + 187, + 427, + 192, + 437 + ], + "score": 0.77, + "content": "\\ell", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 425, + 298, + 442 + ], + "score": 1.0, + "content": "-th component of the filter", + "type": "text" + }, + { + "bbox": [ + 299, + 425, + 430, + 440 + ], + "score": 0.93, + "content": "F ( \\boldsymbol { x } ) = \\left[ F ^ { ( 0 ) } ( x ) , \\ldots , F ^ { ( D ) } ( x ) \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 425, + 500, + 442 + ], + "score": 1.0, + "content": "will be given by", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5, + "bbox_fs": [ + 104, + 392, + 506, + 442 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 214, + 443, + 396, + 458 + ], + "lines": [ + { + "bbox": [ + 214, + 443, + 396, + 458 + ], + "spans": [ + { + "bbox": [ + 214, + 443, + 396, + 458 + ], + "score": 0.91, + "content": "F _ { m } ^ { \\ell } ( x ) = R ^ { ( \\ell ) } ( \\| x \\| ) Y _ { m } ^ { \\ell } ( \\hat { x } ) , m = - \\ell , \\ldots , \\ell", + "type": "interline_equation", + "image_path": "ab43c4063a462588d15cfce3a32fb29492e677c37687cb4a82ed38c439545931.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 214, + 443, + 396, + 458 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 462, + 505, + 497 + ], + "lines": [ + { + "bbox": [ + 105, + 460, + 506, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 460, + 135, + 477 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 463, + 186, + 475 + ], + "score": 0.92, + "content": "\\hat { x } = x / \\| x \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 460, + 198, + 477 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 199, + 463, + 229, + 474 + ], + "score": 0.91, + "content": "x \\neq 0", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 460, + 249, + 477 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 249, + 463, + 280, + 473 + ], + "score": 0.91, + "content": "\\hat { x } ~ = ~ 0", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 460, + 327, + 477 + ], + "score": 1.0, + "content": "otherwise,", + "type": "text" + }, + { + "bbox": [ + 327, + 462, + 342, + 475 + ], + "score": 0.91, + "content": "Y _ { m } ^ { \\ell }", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 460, + 467, + 477 + ], + "score": 1.0, + "content": "are spherical harmonics, and", + "type": "text" + }, + { + "bbox": [ + 467, + 462, + 486, + 473 + ], + "score": 0.9, + "content": "R ^ { ( \\ell ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 460, + 506, + 477 + ], + "score": 1.0, + "content": "any", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 474, + 505, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 474, + 193, + 487 + ], + "score": 1.0, + "content": "polynomial of degree", + "type": "text" + }, + { + "bbox": [ + 193, + 475, + 213, + 485 + ], + "score": 0.87, + "content": "\\leq D", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 474, + 505, + 487 + ], + "score": 1.0, + "content": ". In Appendix F we show that these polynomial functions can be replaced", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 485, + 462, + 497 + ], + "spans": [ + { + "bbox": [ + 106, + 485, + 462, + 497 + ], + "score": 1.0, + "content": "by fully connected networks, since the latter can approximate all polynomials uniformly.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 460, + 506, + 497 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 501, + 504, + 528 + ], + "lines": [ + { + "bbox": [ + 105, + 500, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 253, + 515 + ], + "score": 1.0, + "content": "The convolution of an input feature", + "type": "text" + }, + { + "bbox": [ + 254, + 502, + 293, + 515 + ], + "score": 0.91, + "content": "V \\in W _ { l _ { i } } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 500, + 342, + 515 + ], + "score": 1.0, + "content": "and a filter", + "type": "text" + }, + { + "bbox": [ + 342, + 502, + 351, + 512 + ], + "score": 0.84, + "content": "F", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 500, + 505, + 515 + ], + "score": 1.0, + "content": "as defined above, will give an output", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 513, + 382, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 137, + 531 + ], + "score": 1.0, + "content": "feature", + "type": "text" + }, + { + "bbox": [ + 137, + 514, + 221, + 529 + ], + "score": 0.93, + "content": "\\tilde { V } = ( \\tilde { V } _ { a } ) _ { a = 1 } ^ { n } \\in W _ { l _ { 0 } } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 513, + 252, + 531 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 252, + 516, + 305, + 528 + ], + "score": 0.92, + "content": "\\boldsymbol { l _ { o } } = \\boldsymbol { l } ( \\boldsymbol { l _ { f } } , \\boldsymbol { l _ { i } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 513, + 382, + 531 + ], + "score": 1.0, + "content": ", which is given by", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5, + "bbox_fs": [ + 105, + 500, + 505, + 531 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 190, + 532, + 421, + 566 + ], + "lines": [ + { + "bbox": [ + 190, + 532, + 421, + 566 + ], + "spans": [ + { + "bbox": [ + 190, + 532, + 421, + 566 + ], + "score": 0.94, + "content": "\\tilde { V } _ { a } ( X , V ) = U ( l _ { f } , l _ { i } ) \\left( \\theta _ { 0 } V _ { a } + \\sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \\otimes V _ { b } \\right) .", + "type": "interline_equation", + "image_path": "965407aa486d2589a2966a08fbbe90a36b2b8b549c1c3a671e196f7ef48eb5f8.jpg" + } + ] + } + ], + "index": 34.5, + "virtual_lines": [ + { + "bbox": [ + 190, + 532, + 421, + 549.0 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 190, + 549.0, + 421, + 566.0 + ], + "spans": [], + "index": 35 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 569, + 505, + 618 + ], + "lines": [ + { + "bbox": [ + 105, + 569, + 504, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 569, + 451, + 583 + ], + "score": 1.0, + "content": "More formally we will think of convolutional layer as functions of the form", + "type": "text" + }, + { + "bbox": [ + 452, + 569, + 504, + 582 + ], + "score": 0.89, + "content": "f ( X , V ) \\quad = \\quad", + "type": "inline_equation" + } + ], + "index": 36 + }, + { + "bbox": [ + 107, + 581, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 107, + 581, + 164, + 595 + ], + "score": 0.89, + "content": "( X , { \\tilde { V } } ( X , V ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 582, + 356, + 594 + ], + "score": 1.0, + "content": ". These functions are defined by a choice of", + "type": "text" + }, + { + "bbox": [ + 356, + 583, + 366, + 592 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 582, + 505, + 594 + ], + "score": 1.0, + "content": ", a choice of a scalar polynomial", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 107, + 592, + 506, + 608 + ], + "spans": [ + { + "bbox": [ + 107, + 595, + 184, + 606 + ], + "score": 0.87, + "content": "R ^ { ( \\ell ) } , \\ell = 0 , \\dots , D", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 592, + 310, + 608 + ], + "score": 1.0, + "content": ", and a choice of the parameter", + "type": "text" + }, + { + "bbox": [ + 311, + 595, + 342, + 606 + ], + "score": 0.92, + "content": "\\theta _ { 0 } \\in \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 592, + 506, + 608 + ], + "score": 1.0, + "content": "in equation 14. We denote the set of all", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 604, + 207, + 619 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 167, + 619 + ], + "score": 1.0, + "content": "such functions", + "type": "text" + }, + { + "bbox": [ + 167, + 607, + 174, + 617 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 604, + 187, + 619 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 188, + 606, + 203, + 617 + ], + "score": 0.89, + "content": "\\mathcal { F } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 604, + 207, + 619 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 569, + 506, + 619 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 623, + 505, + 658 + ], + "lines": [ + { + "bbox": [ + 104, + 622, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 104, + 622, + 394, + 639 + ], + "score": 1.0, + "content": "Self Interaction layers. Self interaction layers are linear functions from", + "type": "text" + }, + { + "bbox": [ + 394, + 623, + 460, + 636 + ], + "score": 0.9, + "content": "\\hat { \\Lambda } : W _ { l } ^ { n } \\to W _ { l , r } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 460, + 622, + 506, + 639 + ], + "score": 1.0, + "content": ", which are", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 635, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 105, + 635, + 386, + 648 + ], + "score": 1.0, + "content": "obtained from elementwise application of equivariant linear functions", + "type": "text" + }, + { + "bbox": [ + 387, + 636, + 449, + 647 + ], + "score": 0.91, + "content": "\\Lambda : W _ { l } \\stackrel { \\cdot } { } W _ { l _ { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 635, + 505, + 648 + ], + "score": 1.0, + "content": ". These linear", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 645, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 106, + 645, + 299, + 659 + ], + "score": 1.0, + "content": "functions can be specified by a choice of matrix", + "type": "text" + }, + { + "bbox": [ + 299, + 646, + 312, + 656 + ], + "score": 0.78, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 645, + 505, + 659 + ], + "score": 1.0, + "content": "with the sparsity pattern described in Lemma 5.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41, + "bbox_fs": [ + 104, + 622, + 506, + 659 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 663, + 504, + 686 + ], + "lines": [ + { + "bbox": [ + 106, + 663, + 505, + 676 + ], + "spans": [ + { + "bbox": [ + 106, + 663, + 505, + 676 + ], + "score": 1.0, + "content": "Activation functions. TFN, as well as other papers, proposed several activation functions. We find", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 675, + 456, + 686 + ], + "spans": [ + { + "bbox": [ + 106, + 675, + 456, + 686 + ], + "score": 1.0, + "content": "that these layers are not necessary for universality and thus we do not define them here.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5, + "bbox_fs": [ + 106, + 663, + 505, + 686 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 690, + 504, + 714 + ], + "lines": [ + { + "bbox": [ + 105, + 690, + 505, + 703 + ], + "spans": [ + { + "bbox": [ + 105, + 690, + 505, + 703 + ], + "score": 1.0, + "content": "Network architecture. For our universality proof, we suggest a simple architecture which depends", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 702, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 105, + 702, + 253, + 714 + ], + "score": 1.0, + "content": "on two positive integer parameters", + "type": "text" + }, + { + "bbox": [ + 253, + 703, + 281, + 714 + ], + "score": 0.91, + "content": "( C , D )", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 702, + 332, + 714 + ], + "score": 1.0, + "content": ": For given", + "type": "text" + }, + { + "bbox": [ + 333, + 703, + 342, + 712 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 702, + 410, + 714 + ], + "score": 1.0, + "content": ", we will define", + "type": "text" + }, + { + "bbox": [ + 411, + 702, + 448, + 714 + ], + "score": 0.93, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 702, + 506, + 714 + ], + "score": 1.0, + "content": "as the set of", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5, + "bbox_fs": [ + 105, + 690, + 506, + 714 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 298, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 299, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 191, + 95 + ], + "score": 1.0, + "content": "function obtained by", + "type": "text" + }, + { + "bbox": [ + 192, + 83, + 206, + 92 + ], + "score": 0.86, + "content": "2 D", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 82, + 299, + 95 + ], + "score": 1.0, + "content": "recursive convolutions", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "interline_equation", + "bbox": [ + 186, + 97, + 424, + 112 + ], + "lines": [ + { + "bbox": [ + 186, + 97, + 424, + 112 + ], + "spans": [ + { + "bbox": [ + 186, + 97, + 424, + 112 + ], + "score": 0.89, + "content": "{ \\mathcal { F } } _ { \\mathrm { f e a t } } ( D ) = \\{ \\pi _ { V } \\circ f ^ { 2 D } \\circ \\ldots f ^ { 2 } \\circ f ^ { 1 } \\circ \\operatorname { e x t } ( X ) | f ^ { j } \\in { \\mathcal { F } } _ { D } \\} ,", + "type": "interline_equation", + "image_path": "adc6540c8611808e70ae0a52bb304fe48558b096f6639f463c82fbedde064dd0.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 186, + 97, + 424, + 112 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 114, + 505, + 177 + ], + "lines": [ + { + "bbox": [ + 104, + 111, + 507, + 131 + ], + "spans": [ + { + "bbox": [ + 104, + 111, + 507, + 131 + ], + "score": 1.0, + "content": "where ext and πV are defined as in equation 10. The output of a function in Ffeat(D) is in W nl(D),", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 127, + 506, + 142 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 146, + 142 + ], + "score": 1.0, + "content": "for some", + "type": "text" + }, + { + "bbox": [ + 146, + 128, + 167, + 140 + ], + "score": 0.91, + "content": "\\boldsymbol { l } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 127, + 248, + 142 + ], + "score": 1.0, + "content": "which depends on", + "type": "text" + }, + { + "bbox": [ + 248, + 129, + 258, + 138 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 127, + 331, + 142 + ], + "score": 1.0, + "content": ". We then define", + "type": "text" + }, + { + "bbox": [ + 331, + 128, + 372, + 140 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { \\mathrm { p o o l } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 127, + 506, + 142 + ], + "score": 1.0, + "content": "to be the self-interaction layers", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 103, + 136, + 508, + 157 + ], + "spans": [ + { + "bbox": [ + 103, + 136, + 154, + 157 + ], + "score": 1.0, + "content": "which map", + "type": "text" + }, + { + "bbox": [ + 155, + 141, + 181, + 154 + ], + "score": 0.93, + "content": "W _ { l ( D ) } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 136, + 194, + 157 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 195, + 140, + 213, + 153 + ], + "score": 0.92, + "content": "W _ { l _ { T } } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 136, + 282, + 157 + ], + "score": 1.0, + "content": ". This choice of", + "type": "text" + }, + { + "bbox": [ + 282, + 140, + 320, + 152 + ], + "score": 0.93, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 136, + 340, + 157 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 340, + 140, + 380, + 152 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { \\mathrm { p o o l } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 136, + 508, + 157 + ], + "score": 1.0, + "content": ", together with a choice of the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 101, + 147, + 505, + 173 + ], + "spans": [ + { + "bbox": [ + 101, + 147, + 187, + 173 + ], + "score": 1.0, + "content": "number of channels", + "type": "text" + }, + { + "bbox": [ + 188, + 155, + 196, + 164 + ], + "score": 0.81, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 147, + 351, + 173 + ], + "score": 1.0, + "content": "T, defines the final network architecture", + "type": "text" + }, + { + "bbox": [ + 351, + 153, + 493, + 167 + ], + "score": 0.91, + "content": "\\mathcal { F } _ { C , D } ^ { \\mathrm { T F N } } = \\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } ( D ) , \\mathcal { F } _ { \\mathrm { p o o l } } ( D ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 147, + 505, + 173 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 165, + 367, + 177 + ], + "spans": [ + { + "bbox": [ + 105, + 165, + 367, + 177 + ], + "score": 1.0, + "content": "in equation 2. In the appendix we prove the universality of TFN:", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 107, + 178, + 258, + 191 + ], + "lines": [ + { + "bbox": [ + 105, + 176, + 255, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 189, + 194 + ], + "score": 1.0, + "content": "Theorem 2. For all", + "type": "text" + }, + { + "bbox": [ + 189, + 178, + 255, + 192 + ], + "score": 0.93, + "content": "n \\in \\mathbb { N } , l _ { T } \\in \\mathbb { N } _ { + } ^ { * }", + "type": "inline_equation" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 127, + 199, + 505, + 247 + ], + "lines": [ + { + "bbox": [ + 124, + 192, + 501, + 222 + ], + "spans": [ + { + "bbox": [ + 124, + 192, + 158, + 222 + ], + "score": 1.0, + "content": "1. For", + "type": "text" + }, + { + "bbox": [ + 158, + 201, + 194, + 213 + ], + "score": 0.91, + "content": "D \\in { \\mathbb { N } } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 192, + 221, + 222 + ], + "score": 1.0, + "content": ", every", + "type": "text" + }, + { + "bbox": [ + 221, + 201, + 230, + 211 + ], + "score": 0.78, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 192, + 327, + 222 + ], + "score": 1.0, + "content": "-equivariant polynomial", + "type": "text" + }, + { + "bbox": [ + 327, + 200, + 396, + 213 + ], + "score": 0.93, + "content": "p : \\mathbb { R } ^ { 3 \\times n } W _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 192, + 435, + 222 + ], + "score": 1.0, + "content": "of degree", + "type": "text" + }, + { + "bbox": [ + 435, + 201, + 445, + 211 + ], + "score": 0.78, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 192, + 464, + 222 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 464, + 200, + 501, + 216 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { C ( D ) , D } ^ { \\mathrm { T F N } }", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 128, + 220, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 128, + 220, + 203, + 253 + ], + "score": 1.0, + "content": "2. Every continuoby functions in", + "type": "text" + }, + { + "bbox": [ + 215, + 222, + 224, + 231 + ], + "score": 0.79, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 220, + 505, + 253 + ], + "score": 1.0, + "content": "t function can be approximated uniformly on compact sets", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 204, + 231, + 269, + 247 + ], + "spans": [ + { + "bbox": [ + 204, + 231, + 269, + 247 + ], + "score": 0.91, + "content": "\\cup _ { D \\in \\mathbb { N } _ { + } } \\mathcal { F } _ { C ( D ) , D } ^ { \\mathrm { T F N } }", + "type": "inline_equation" + } + ], + "index": 10 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 109, + 254, + 505, + 288 + ], + "lines": [ + { + "bbox": [ + 106, + 254, + 505, + 267 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 312, + 267 + ], + "score": 1.0, + "content": "As discussed previously, the linear universality of", + "type": "text" + }, + { + "bbox": [ + 313, + 255, + 337, + 266 + ], + "score": 0.91, + "content": "\\mathcal { F } _ { \\mathrm { p o o l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 337, + 254, + 505, + 267 + ], + "score": 1.0, + "content": "is guaranteed. Thus proving Theorem 2", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 265, + 505, + 277 + ], + "spans": [ + { + "bbox": [ + 106, + 265, + 209, + 277 + ], + "score": 1.0, + "content": "amounts to showing that", + "type": "text" + }, + { + "bbox": [ + 210, + 265, + 248, + 277 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 265, + 260, + 277 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 260, + 266, + 270, + 275 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 265, + 505, + 277 + ], + "score": 1.0, + "content": "-spanning. This is done using the sufficient condition for", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 107, + 276, + 243, + 289 + ], + "spans": [ + { + "bbox": [ + 107, + 277, + 116, + 286 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 276, + 243, + 289 + ], + "score": 1.0, + "content": "-spanning defined in Lemma 4.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 107, + 299, + 505, + 345 + ], + "lines": [ + { + "bbox": [ + 106, + 300, + 504, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 504, + 312 + ], + "score": 1.0, + "content": "Proof idea. The proof is rather technical and involved. A useful observation (see Dai & Xu (2013))", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 311, + 505, + 323 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 295, + 323 + ], + "score": 1.0, + "content": "used in the proof is that the filters of orders", + "type": "text" + }, + { + "bbox": [ + 296, + 311, + 365, + 322 + ], + "score": 0.93, + "content": "\\ell = 0 , 1 , \\ldots , D", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 311, + 505, + 323 + ], + "score": 1.0, + "content": ", defined in equation 13, span all", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 104, + 321, + 504, + 335 + ], + "spans": [ + { + "bbox": [ + 104, + 321, + 234, + 335 + ], + "score": 1.0, + "content": "polynomial functions of degree", + "type": "text" + }, + { + "bbox": [ + 234, + 322, + 244, + 332 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 321, + 257, + 335 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 258, + 321, + 270, + 332 + ], + "score": 0.86, + "content": "\\mathbb { R } ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 321, + 487, + 335 + ], + "score": 1.0, + "content": ". This observation is used to show that all functions in", + "type": "text" + }, + { + "bbox": [ + 488, + 322, + 504, + 333 + ], + "score": 0.87, + "content": "\\mathcal { Q } _ { D }", + "type": "inline_equation" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 333, + 504, + 346 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 189, + 346 + ], + "score": 1.0, + "content": "can be expressed by", + "type": "text" + }, + { + "bbox": [ + 189, + 333, + 227, + 344 + ], + "score": 0.93, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 333, + 257, + 346 + ], + "score": 1.0, + "content": "and so", + "type": "text" + }, + { + "bbox": [ + 257, + 333, + 279, + 344 + ], + "score": 0.9, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 333, + 289, + 346 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 289, + 333, + 299, + 343 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 333, + 432, + 346 + ], + "score": 1.0, + "content": "-spanning, as stated in Lemma 2.", + "type": "text" + }, + { + "bbox": [ + 495, + 334, + 504, + 343 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 107, + 356, + 505, + 390 + ], + "lines": [ + { + "bbox": [ + 106, + 356, + 505, + 369 + ], + "spans": [ + { + "bbox": [ + 106, + 356, + 447, + 369 + ], + "score": 1.0, + "content": "Alternative architecture The complexity of the TFN network used to construct", + "type": "text" + }, + { + "bbox": [ + 447, + 357, + 456, + 367 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 356, + 505, + 369 + ], + "score": 1.0, + "content": "-equivariant", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 367, + 506, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 198, + 380 + ], + "score": 1.0, + "content": "polynomials of degree", + "type": "text" + }, + { + "bbox": [ + 198, + 368, + 208, + 378 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 367, + 506, + 380 + ], + "score": 1.0, + "content": ", can be reduced using a simple modifications of the convolutional layer in", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 378, + 505, + 391 + ], + "spans": [ + { + "bbox": [ + 105, + 378, + 254, + 391 + ], + "score": 1.0, + "content": "equation 14: We add two parameters", + "type": "text" + }, + { + "bbox": [ + 255, + 379, + 298, + 390 + ], + "score": 0.92, + "content": "\\theta _ { 1 } , \\theta _ { 2 } \\in \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 378, + 505, + 391 + ], + "score": 1.0, + "content": "to the convolutional layer, which is now defined as:", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19 + }, + { + "type": "interline_equation", + "bbox": [ + 146, + 393, + 465, + 427 + ], + "lines": [ + { + "bbox": [ + 146, + 393, + 465, + 427 + ], + "spans": [ + { + "bbox": [ + 146, + 393, + 465, + 427 + ], + "score": 0.93, + "content": "\\tilde { V } _ { a } ( X , V ) = U ( l _ { f } , l _ { i } ) \\left( \\theta _ { 1 } \\sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \\otimes V _ { b } + \\theta _ { 2 } \\sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \\otimes V _ { a } \\right) .", + "type": "interline_equation", + "image_path": "a62cee15c1d813f1fad5e4a74cd43f39c9d1db3184a1e47f48c5df6dc91c07d9.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 146, + 393, + 465, + 404.3333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 146, + 404.3333333333333, + 465, + 415.66666666666663 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 146, + 415.66666666666663, + 465, + 426.99999999999994 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 435, + 506, + 459 + ], + "lines": [ + { + "bbox": [ + 105, + 435, + 506, + 449 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 283, + 449 + ], + "score": 1.0, + "content": "With this simple change, we can show that", + "type": "text" + }, + { + "bbox": [ + 283, + 436, + 321, + 448 + ], + "score": 0.93, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 435, + 333, + 449 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 333, + 437, + 342, + 446 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 435, + 506, + 449 + ], + "score": 1.0, + "content": "-spanning even if we only take filters of", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 447, + 382, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 382, + 459 + ], + "score": 1.0, + "content": "order 0 and 1 throughout the network. This is shown in Appendix E.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5 + }, + { + "type": "title", + "bbox": [ + 107, + 474, + 196, + 487 + ], + "lines": [ + { + "bbox": [ + 105, + 471, + 198, + 490 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 198, + 490 + ], + "score": 1.0, + "content": "6 CONCLUSION", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 106, + 498, + 505, + 555 + ], + "lines": [ + { + "bbox": [ + 105, + 498, + 505, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 498, + 425, + 512 + ], + "score": 1.0, + "content": "In this paper, we have presented a new framework for proving the universality of", + "type": "text" + }, + { + "bbox": [ + 425, + 500, + 434, + 509 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 498, + 505, + 512 + ], + "score": 1.0, + "content": "-equivariant point", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 509, + 504, + 523 + ], + "spans": [ + { + "bbox": [ + 106, + 509, + 504, + 523 + ], + "score": 1.0, + "content": "cloud networks. We used this framework for proving the universality of the TFN model Thomas et al.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 521, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 106, + 521, + 505, + 533 + ], + "score": 1.0, + "content": "(2018); Fuchs et al. (2020), and for devising two additional novel simple universal architectures. In", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 531, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 388, + 545 + ], + "score": 1.0, + "content": "the future we hope to extend these simple constructions to operational", + "type": "text" + }, + { + "bbox": [ + 388, + 532, + 397, + 542 + ], + "score": 0.81, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 531, + 505, + 545 + ], + "score": 1.0, + "content": "-equivariant networks with", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 543, + 358, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 358, + 556 + ], + "score": 1.0, + "content": "universality guarantees and competitive practical performance.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 559, + 505, + 648 + ], + "lines": [ + { + "bbox": [ + 106, + 560, + 505, + 572 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 505, + 572 + ], + "score": 1.0, + "content": "Our universal architectures do not require activation functions, and use a single self-interaction layer.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 570, + 505, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 505, + 583 + ], + "score": 1.0, + "content": "In Appendix H we present an experiment indicating that the performance of TFN is not significantly", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 582, + 505, + 594 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 505, + 594 + ], + "score": 1.0, + "content": "altered by these simplifications. Our architectures also require high order representations, and our", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 593, + 505, + 605 + ], + "spans": [ + { + "bbox": [ + 105, + 593, + 505, + 605 + ], + "score": 1.0, + "content": "experiments show that using increasingly high order representations does indeed improve perfor-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 604, + 505, + 616 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 505, + 616 + ], + "score": 1.0, + "content": "mance. To date, practical TFN implementation included a relatively small amount of layers, and", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 614, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 505, + 627 + ], + "score": 1.0, + "content": "did not use very high order representations. We believe our theoretical results will inspire interest", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 625, + 506, + 638 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 506, + 638 + ], + "score": 1.0, + "content": "in stable implementation of larger architectures. On the other hand, an interesting open problem is", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 636, + 493, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 493, + 649 + ], + "score": 1.0, + "content": "understanding whether universality can be achieved using only low-dimensional representations.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 106, + 653, + 505, + 698 + ], + "lines": [ + { + "bbox": [ + 105, + 653, + 504, + 666 + ], + "spans": [ + { + "bbox": [ + 105, + 653, + 504, + 666 + ], + "score": 1.0, + "content": "Finally, we believe that the framework we developed here will be useful for proving the universality", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 664, + 505, + 677 + ], + "spans": [ + { + "bbox": [ + 105, + 664, + 141, + 677 + ], + "score": 1.0, + "content": "of other", + "type": "text" + }, + { + "bbox": [ + 141, + 665, + 150, + 675 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 664, + 505, + 677 + ], + "score": 1.0, + "content": "-equivariant models for point cloud networks, and other related equivariant models. We", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 675, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 675, + 505, + 689 + ], + "score": 1.0, + "content": "note that large parts of our discussion can be easily generalized to symmetry groups of the form", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 107, + 684, + 450, + 700 + ], + "spans": [ + { + "bbox": [ + 107, + 686, + 186, + 697 + ], + "score": 0.91, + "content": "G = \\mathbb { R } ^ { d } \\rtimes \\breve { H } \\times \\bar { S } _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 684, + 226, + 700 + ], + "score": 1.0, + "content": "acting on", + "type": "text" + }, + { + "bbox": [ + 227, + 686, + 250, + 696 + ], + "score": 0.9, + "content": "\\mathbb { R } ^ { d \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 684, + 281, + 700 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 281, + 687, + 291, + 696 + ], + "score": 0.79, + "content": "H", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 684, + 450, + 700 + ], + "score": 1.0, + "content": "can be any compact topological group.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 41.5 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Acknowledgments The authors would like to thank Fabian B. Fuchs for making code available", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 720, + 504, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 504, + 733 + ], + "score": 1.0, + "content": "and Taco Cohen for helpful discussion. N.D. is supported by THEORINET Simons award 814643.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44.5 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 292, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 298, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 299, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 191, + 95 + ], + "score": 1.0, + "content": "function obtained by", + "type": "text" + }, + { + "bbox": [ + 192, + 83, + 206, + 92 + ], + "score": 0.86, + "content": "2 D", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 82, + 299, + 95 + ], + "score": 1.0, + "content": "recursive convolutions", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 106, + 82, + 299, + 95 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 186, + 97, + 424, + 112 + ], + "lines": [ + { + "bbox": [ + 186, + 97, + 424, + 112 + ], + "spans": [ + { + "bbox": [ + 186, + 97, + 424, + 112 + ], + "score": 0.89, + "content": "{ \\mathcal { F } } _ { \\mathrm { f e a t } } ( D ) = \\{ \\pi _ { V } \\circ f ^ { 2 D } \\circ \\ldots f ^ { 2 } \\circ f ^ { 1 } \\circ \\operatorname { e x t } ( X ) | f ^ { j } \\in { \\mathcal { F } } _ { D } \\} ,", + "type": "interline_equation", + "image_path": "adc6540c8611808e70ae0a52bb304fe48558b096f6639f463c82fbedde064dd0.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 186, + 97, + 424, + 112 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 114, + 505, + 177 + ], + "lines": [ + { + "bbox": [ + 104, + 111, + 507, + 131 + ], + "spans": [ + { + "bbox": [ + 104, + 111, + 507, + 131 + ], + "score": 1.0, + "content": "where ext and πV are defined as in equation 10. The output of a function in Ffeat(D) is in W nl(D),", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 127, + 506, + 142 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 146, + 142 + ], + "score": 1.0, + "content": "for some", + "type": "text" + }, + { + "bbox": [ + 146, + 128, + 167, + 140 + ], + "score": 0.91, + "content": "\\boldsymbol { l } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 127, + 248, + 142 + ], + "score": 1.0, + "content": "which depends on", + "type": "text" + }, + { + "bbox": [ + 248, + 129, + 258, + 138 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 127, + 331, + 142 + ], + "score": 1.0, + "content": ". We then define", + "type": "text" + }, + { + "bbox": [ + 331, + 128, + 372, + 140 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { \\mathrm { p o o l } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 127, + 506, + 142 + ], + "score": 1.0, + "content": "to be the self-interaction layers", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 103, + 136, + 508, + 157 + ], + "spans": [ + { + "bbox": [ + 103, + 136, + 154, + 157 + ], + "score": 1.0, + "content": "which map", + "type": "text" + }, + { + "bbox": [ + 155, + 141, + 181, + 154 + ], + "score": 0.93, + "content": "W _ { l ( D ) } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 136, + 194, + 157 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 195, + 140, + 213, + 153 + ], + "score": 0.92, + "content": "W _ { l _ { T } } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 136, + 282, + 157 + ], + "score": 1.0, + "content": ". This choice of", + "type": "text" + }, + { + "bbox": [ + 282, + 140, + 320, + 152 + ], + "score": 0.93, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 136, + 340, + 157 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 340, + 140, + 380, + 152 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { \\mathrm { p o o l } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 136, + 508, + 157 + ], + "score": 1.0, + "content": ", together with a choice of the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 101, + 147, + 505, + 173 + ], + "spans": [ + { + "bbox": [ + 101, + 147, + 187, + 173 + ], + "score": 1.0, + "content": "number of channels", + "type": "text" + }, + { + "bbox": [ + 188, + 155, + 196, + 164 + ], + "score": 0.81, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 147, + 351, + 173 + ], + "score": 1.0, + "content": "T, defines the final network architecture", + "type": "text" + }, + { + "bbox": [ + 351, + 153, + 493, + 167 + ], + "score": 0.91, + "content": "\\mathcal { F } _ { C , D } ^ { \\mathrm { T F N } } = \\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } ( D ) , \\mathcal { F } _ { \\mathrm { p o o l } } ( D ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 147, + 505, + 173 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 165, + 367, + 177 + ], + "spans": [ + { + "bbox": [ + 105, + 165, + 367, + 177 + ], + "score": 1.0, + "content": "in equation 2. In the appendix we prove the universality of TFN:", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4, + "bbox_fs": [ + 101, + 111, + 508, + 177 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 178, + 258, + 191 + ], + "lines": [ + { + "bbox": [ + 105, + 176, + 255, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 189, + 194 + ], + "score": 1.0, + "content": "Theorem 2. For all", + "type": "text" + }, + { + "bbox": [ + 189, + 178, + 255, + 192 + ], + "score": 0.93, + "content": "n \\in \\mathbb { N } , l _ { T } \\in \\mathbb { N } _ { + } ^ { * }", + "type": "inline_equation" + } + ], + "index": 7 + } + ], + "index": 7, + "bbox_fs": [ + 105, + 176, + 255, + 194 + ] + }, + { + "type": "text", + "bbox": [ + 127, + 199, + 505, + 247 + ], + "lines": [ + { + "bbox": [ + 124, + 192, + 501, + 222 + ], + "spans": [ + { + "bbox": [ + 124, + 192, + 158, + 222 + ], + "score": 1.0, + "content": "1. For", + "type": "text" + }, + { + "bbox": [ + 158, + 201, + 194, + 213 + ], + "score": 0.91, + "content": "D \\in { \\mathbb { N } } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 192, + 221, + 222 + ], + "score": 1.0, + "content": ", every", + "type": "text" + }, + { + "bbox": [ + 221, + 201, + 230, + 211 + ], + "score": 0.78, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 192, + 327, + 222 + ], + "score": 1.0, + "content": "-equivariant polynomial", + "type": "text" + }, + { + "bbox": [ + 327, + 200, + 396, + 213 + ], + "score": 0.93, + "content": "p : \\mathbb { R } ^ { 3 \\times n } W _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 192, + 435, + 222 + ], + "score": 1.0, + "content": "of degree", + "type": "text" + }, + { + "bbox": [ + 435, + 201, + 445, + 211 + ], + "score": 0.78, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 192, + 464, + 222 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 464, + 200, + 501, + 216 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { C ( D ) , D } ^ { \\mathrm { T F N } }", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 128, + 220, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 128, + 220, + 203, + 253 + ], + "score": 1.0, + "content": "2. Every continuoby functions in", + "type": "text" + }, + { + "bbox": [ + 215, + 222, + 224, + 231 + ], + "score": 0.79, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 220, + 505, + 253 + ], + "score": 1.0, + "content": "t function can be approximated uniformly on compact sets", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 204, + 231, + 269, + 247 + ], + "spans": [ + { + "bbox": [ + 204, + 231, + 269, + 247 + ], + "score": 0.91, + "content": "\\cup _ { D \\in \\mathbb { N } _ { + } } \\mathcal { F } _ { C ( D ) , D } ^ { \\mathrm { T F N } }", + "type": "inline_equation" + } + ], + "index": 10 + } + ], + "index": 9, + "bbox_fs": [ + 124, + 192, + 505, + 253 + ] + }, + { + "type": "text", + "bbox": [ + 109, + 254, + 505, + 288 + ], + "lines": [ + { + "bbox": [ + 106, + 254, + 505, + 267 + ], + "spans": [ + { + "bbox": [ + 106, + 254, + 312, + 267 + ], + "score": 1.0, + "content": "As discussed previously, the linear universality of", + "type": "text" + }, + { + "bbox": [ + 313, + 255, + 337, + 266 + ], + "score": 0.91, + "content": "\\mathcal { F } _ { \\mathrm { p o o l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 337, + 254, + 505, + 267 + ], + "score": 1.0, + "content": "is guaranteed. Thus proving Theorem 2", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 265, + 505, + 277 + ], + "spans": [ + { + "bbox": [ + 106, + 265, + 209, + 277 + ], + "score": 1.0, + "content": "amounts to showing that", + "type": "text" + }, + { + "bbox": [ + 210, + 265, + 248, + 277 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 265, + 260, + 277 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 260, + 266, + 270, + 275 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 265, + 505, + 277 + ], + "score": 1.0, + "content": "-spanning. This is done using the sufficient condition for", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 107, + 276, + 243, + 289 + ], + "spans": [ + { + "bbox": [ + 107, + 277, + 116, + 286 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 276, + 243, + 289 + ], + "score": 1.0, + "content": "-spanning defined in Lemma 4.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12, + "bbox_fs": [ + 106, + 254, + 505, + 289 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 299, + 505, + 345 + ], + "lines": [ + { + "bbox": [ + 106, + 300, + 504, + 312 + ], + "spans": [ + { + "bbox": [ + 106, + 300, + 504, + 312 + ], + "score": 1.0, + "content": "Proof idea. The proof is rather technical and involved. A useful observation (see Dai & Xu (2013))", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 311, + 505, + 323 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 295, + 323 + ], + "score": 1.0, + "content": "used in the proof is that the filters of orders", + "type": "text" + }, + { + "bbox": [ + 296, + 311, + 365, + 322 + ], + "score": 0.93, + "content": "\\ell = 0 , 1 , \\ldots , D", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 311, + 505, + 323 + ], + "score": 1.0, + "content": ", defined in equation 13, span all", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 104, + 321, + 504, + 335 + ], + "spans": [ + { + "bbox": [ + 104, + 321, + 234, + 335 + ], + "score": 1.0, + "content": "polynomial functions of degree", + "type": "text" + }, + { + "bbox": [ + 234, + 322, + 244, + 332 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 321, + 257, + 335 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 258, + 321, + 270, + 332 + ], + "score": 0.86, + "content": "\\mathbb { R } ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 321, + 487, + 335 + ], + "score": 1.0, + "content": ". This observation is used to show that all functions in", + "type": "text" + }, + { + "bbox": [ + 488, + 322, + 504, + 333 + ], + "score": 0.87, + "content": "\\mathcal { Q } _ { D }", + "type": "inline_equation" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 333, + 504, + 346 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 189, + 346 + ], + "score": 1.0, + "content": "can be expressed by", + "type": "text" + }, + { + "bbox": [ + 189, + 333, + 227, + 344 + ], + "score": 0.93, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 333, + 257, + 346 + ], + "score": 1.0, + "content": "and so", + "type": "text" + }, + { + "bbox": [ + 257, + 333, + 279, + 344 + ], + "score": 0.9, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 333, + 289, + 346 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 289, + 333, + 299, + 343 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 333, + 432, + 346 + ], + "score": 1.0, + "content": "-spanning, as stated in Lemma 2.", + "type": "text" + }, + { + "bbox": [ + 495, + 334, + 504, + 343 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15.5, + "bbox_fs": [ + 104, + 300, + 505, + 346 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 356, + 505, + 390 + ], + "lines": [ + { + "bbox": [ + 106, + 356, + 505, + 369 + ], + "spans": [ + { + "bbox": [ + 106, + 356, + 447, + 369 + ], + "score": 1.0, + "content": "Alternative architecture The complexity of the TFN network used to construct", + "type": "text" + }, + { + "bbox": [ + 447, + 357, + 456, + 367 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 356, + 505, + 369 + ], + "score": 1.0, + "content": "-equivariant", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 367, + 506, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 198, + 380 + ], + "score": 1.0, + "content": "polynomials of degree", + "type": "text" + }, + { + "bbox": [ + 198, + 368, + 208, + 378 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 208, + 367, + 506, + 380 + ], + "score": 1.0, + "content": ", can be reduced using a simple modifications of the convolutional layer in", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 378, + 505, + 391 + ], + "spans": [ + { + "bbox": [ + 105, + 378, + 254, + 391 + ], + "score": 1.0, + "content": "equation 14: We add two parameters", + "type": "text" + }, + { + "bbox": [ + 255, + 379, + 298, + 390 + ], + "score": 0.92, + "content": "\\theta _ { 1 } , \\theta _ { 2 } \\in \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 378, + 505, + 391 + ], + "score": 1.0, + "content": "to the convolutional layer, which is now defined as:", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 356, + 506, + 391 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 146, + 393, + 465, + 427 + ], + "lines": [ + { + "bbox": [ + 146, + 393, + 465, + 427 + ], + "spans": [ + { + "bbox": [ + 146, + 393, + 465, + 427 + ], + "score": 0.93, + "content": "\\tilde { V } _ { a } ( X , V ) = U ( l _ { f } , l _ { i } ) \\left( \\theta _ { 1 } \\sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \\otimes V _ { b } + \\theta _ { 2 } \\sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \\otimes V _ { a } \\right) .", + "type": "interline_equation", + "image_path": "a62cee15c1d813f1fad5e4a74cd43f39c9d1db3184a1e47f48c5df6dc91c07d9.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 146, + 393, + 465, + 404.3333333333333 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 146, + 404.3333333333333, + 465, + 415.66666666666663 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 146, + 415.66666666666663, + 465, + 426.99999999999994 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 435, + 506, + 459 + ], + "lines": [ + { + "bbox": [ + 105, + 435, + 506, + 449 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 283, + 449 + ], + "score": 1.0, + "content": "With this simple change, we can show that", + "type": "text" + }, + { + "bbox": [ + 283, + 436, + 321, + 448 + ], + "score": 0.93, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 435, + 333, + 449 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 333, + 437, + 342, + 446 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 435, + 506, + 449 + ], + "score": 1.0, + "content": "-spanning even if we only take filters of", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 447, + 382, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 382, + 459 + ], + "score": 1.0, + "content": "order 0 and 1 throughout the network. This is shown in Appendix E.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 435, + 506, + 459 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 474, + 196, + 487 + ], + "lines": [ + { + "bbox": [ + 105, + 471, + 198, + 490 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 198, + 490 + ], + "score": 1.0, + "content": "6 CONCLUSION", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 106, + 498, + 505, + 555 + ], + "lines": [ + { + "bbox": [ + 105, + 498, + 505, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 498, + 425, + 512 + ], + "score": 1.0, + "content": "In this paper, we have presented a new framework for proving the universality of", + "type": "text" + }, + { + "bbox": [ + 425, + 500, + 434, + 509 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 498, + 505, + 512 + ], + "score": 1.0, + "content": "-equivariant point", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 509, + 504, + 523 + ], + "spans": [ + { + "bbox": [ + 106, + 509, + 504, + 523 + ], + "score": 1.0, + "content": "cloud networks. We used this framework for proving the universality of the TFN model Thomas et al.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 521, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 106, + 521, + 505, + 533 + ], + "score": 1.0, + "content": "(2018); Fuchs et al. (2020), and for devising two additional novel simple universal architectures. In", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 531, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 388, + 545 + ], + "score": 1.0, + "content": "the future we hope to extend these simple constructions to operational", + "type": "text" + }, + { + "bbox": [ + 388, + 532, + 397, + 542 + ], + "score": 0.81, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 531, + 505, + 545 + ], + "score": 1.0, + "content": "-equivariant networks with", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 543, + 358, + 556 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 358, + 556 + ], + "score": 1.0, + "content": "universality guarantees and competitive practical performance.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 498, + 505, + 556 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 559, + 505, + 648 + ], + "lines": [ + { + "bbox": [ + 106, + 560, + 505, + 572 + ], + "spans": [ + { + "bbox": [ + 106, + 560, + 505, + 572 + ], + "score": 1.0, + "content": "Our universal architectures do not require activation functions, and use a single self-interaction layer.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 570, + 505, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 505, + 583 + ], + "score": 1.0, + "content": "In Appendix H we present an experiment indicating that the performance of TFN is not significantly", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 582, + 505, + 594 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 505, + 594 + ], + "score": 1.0, + "content": "altered by these simplifications. Our architectures also require high order representations, and our", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 593, + 505, + 605 + ], + "spans": [ + { + "bbox": [ + 105, + 593, + 505, + 605 + ], + "score": 1.0, + "content": "experiments show that using increasingly high order representations does indeed improve perfor-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 604, + 505, + 616 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 505, + 616 + ], + "score": 1.0, + "content": "mance. To date, practical TFN implementation included a relatively small amount of layers, and", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 614, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 505, + 627 + ], + "score": 1.0, + "content": "did not use very high order representations. We believe our theoretical results will inspire interest", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 625, + 506, + 638 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 506, + 638 + ], + "score": 1.0, + "content": "in stable implementation of larger architectures. On the other hand, an interesting open problem is", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 636, + 493, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 493, + 649 + ], + "score": 1.0, + "content": "understanding whether universality can be achieved using only low-dimensional representations.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 35.5, + "bbox_fs": [ + 105, + 560, + 506, + 649 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 653, + 505, + 698 + ], + "lines": [ + { + "bbox": [ + 105, + 653, + 504, + 666 + ], + "spans": [ + { + "bbox": [ + 105, + 653, + 504, + 666 + ], + "score": 1.0, + "content": "Finally, we believe that the framework we developed here will be useful for proving the universality", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 664, + 505, + 677 + ], + "spans": [ + { + "bbox": [ + 105, + 664, + 141, + 677 + ], + "score": 1.0, + "content": "of other", + "type": "text" + }, + { + "bbox": [ + 141, + 665, + 150, + 675 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 664, + 505, + 677 + ], + "score": 1.0, + "content": "-equivariant models for point cloud networks, and other related equivariant models. We", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 675, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 675, + 505, + 689 + ], + "score": 1.0, + "content": "note that large parts of our discussion can be easily generalized to symmetry groups of the form", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 107, + 684, + 450, + 700 + ], + "spans": [ + { + "bbox": [ + 107, + 686, + 186, + 697 + ], + "score": 0.91, + "content": "G = \\mathbb { R } ^ { d } \\rtimes \\breve { H } \\times \\bar { S } _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 684, + 226, + 700 + ], + "score": 1.0, + "content": "acting on", + "type": "text" + }, + { + "bbox": [ + 227, + 686, + 250, + 696 + ], + "score": 0.9, + "content": "\\mathbb { R } ^ { d \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 684, + 281, + 700 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 281, + 687, + 291, + 696 + ], + "score": 0.79, + "content": "H", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 684, + 450, + 700 + ], + "score": 1.0, + "content": "can be any compact topological group.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 41.5, + "bbox_fs": [ + 105, + 653, + 505, + 700 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Acknowledgments The authors would like to thank Fabian B. Fuchs for making code available", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 720, + 504, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 504, + 733 + ], + "score": 1.0, + "content": "and Taco Cohen for helpful discussion. N.D. is supported by THEORINET Simons award 814643.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44.5, + "bbox_fs": [ + 105, + 709, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 176, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 176, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 176, + 94 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 100, + 504, + 123 + ], + "lines": [ + { + "bbox": [ + 105, + 99, + 504, + 113 + ], + "spans": [ + { + "bbox": [ + 105, + 99, + 504, + 113 + ], + "score": 1.0, + "content": "Matan Atzmon, Haggai Maron, and Yaron Lipman. Point convolutional neural networks by exten-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 111, + 341, + 123 + ], + "spans": [ + { + "bbox": [ + 116, + 111, + 341, + 123 + ], + "score": 1.0, + "content": "sion operators. arXiv preprint arXiv:1803.10091, 2018.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 106, + 130, + 503, + 164 + ], + "lines": [ + { + "bbox": [ + 106, + 130, + 505, + 142 + ], + "spans": [ + { + "bbox": [ + 106, + 130, + 505, + 142 + ], + "score": 1.0, + "content": "Alexander Bogatskiy, Brandon Anderson, Jan T Offermann, Marwah Roussi, David W Miller, and", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 141, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 115, + 141, + 505, + 155 + ], + "score": 1.0, + "content": "Risi Kondor. Lorentz group equivariant neural network for particle physics. arXiv preprint", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 153, + 219, + 163 + ], + "spans": [ + { + "bbox": [ + 115, + 153, + 219, + 163 + ], + "score": 1.0, + "content": "arXiv:2006.04780, 2020.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 105, + 171, + 504, + 194 + ], + "lines": [ + { + "bbox": [ + 105, + 169, + 505, + 185 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 505, + 185 + ], + "score": 1.0, + "content": "Taco S Cohen, Mario Geiger, Jonas Kohler, and Max Welling. Spherical cnns. ¨ arXiv preprint", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 182, + 219, + 195 + ], + "spans": [ + { + "bbox": [ + 115, + 182, + 219, + 195 + ], + "score": 1.0, + "content": "arXiv:1801.10130, 2018.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 105, + 201, + 504, + 225 + ], + "lines": [ + { + "bbox": [ + 105, + 200, + 505, + 215 + ], + "spans": [ + { + "bbox": [ + 105, + 200, + 505, + 215 + ], + "score": 1.0, + "content": "Feng Dai and Yuan Xu. Approximation theory and harmonic analysis on spheres and balls, vol-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 212, + 217, + 225 + ], + "spans": [ + { + "bbox": [ + 115, + 212, + 217, + 225 + ], + "score": 1.0, + "content": "ume 23. Springer, 2013.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 108, + 231, + 504, + 266 + ], + "lines": [ + { + "bbox": [ + 105, + 230, + 506, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 506, + 245 + ], + "score": 1.0, + "content": "Carlos Esteves, Christine Allen-Blanchette, Ameesh Makadia, and Kostas Daniilidis. Learning so", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 243, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 116, + 243, + 505, + 255 + ], + "score": 1.0, + "content": "(3) equivariant representations with spherical cnns. In Proceedings of the European Conference", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 254, + 306, + 266 + ], + "spans": [ + { + "bbox": [ + 115, + 254, + 306, + 266 + ], + "score": 1.0, + "content": "on Computer Vision (ECCV), pp. 52–68, 2018.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 106, + 272, + 504, + 296 + ], + "lines": [ + { + "bbox": [ + 106, + 273, + 504, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 504, + 285 + ], + "score": 1.0, + "content": "Fabian B Fuchs, Daniel E Worrall, Volker Fischer, and Max Welling. Se (3)-transformers: 3d roto-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 284, + 449, + 296 + ], + "spans": [ + { + "bbox": [ + 115, + 284, + 449, + 296 + ], + "score": 1.0, + "content": "translation equivariant attention networks. arXiv preprint arXiv:2006.10503, 2020.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 106, + 303, + 503, + 326 + ], + "lines": [ + { + "bbox": [ + 106, + 302, + 504, + 316 + ], + "spans": [ + { + "bbox": [ + 106, + 302, + 504, + 316 + ], + "score": 1.0, + "content": "William Fulton and Joe Harris. Representation theory: a first course, volume 129. Springer Science", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 116, + 315, + 220, + 325 + ], + "spans": [ + { + "bbox": [ + 116, + 315, + 220, + 325 + ], + "score": 1.0, + "content": "& Business Media, 2013.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 106, + 333, + 502, + 357 + ], + "lines": [ + { + "bbox": [ + 105, + 333, + 503, + 347 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 503, + 347 + ], + "score": 1.0, + "content": "Justin Gilmer, Samuel S Schoenholz, Patrick F Riley, Oriol Vinyals, and George E Dahl. Neural", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 345, + 444, + 357 + ], + "spans": [ + { + "bbox": [ + 115, + 345, + 444, + 357 + ], + "score": 1.0, + "content": "message passing for quantum chemistry. arXiv preprint arXiv:1704.01212, 2017.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 106, + 363, + 504, + 397 + ], + "lines": [ + { + "bbox": [ + 106, + 362, + 505, + 377 + ], + "spans": [ + { + "bbox": [ + 106, + 362, + 505, + 377 + ], + "score": 1.0, + "content": "Yulan Guo, Hanyun Wang, Qingyong Hu, Hao Liu, Li Liu, and Mohammed Bennamoun. Deep", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 375, + 505, + 387 + ], + "spans": [ + { + "bbox": [ + 116, + 375, + 505, + 387 + ], + "score": 1.0, + "content": "learning for 3d point clouds: A survey. IEEE Transactions on Pattern Analysis and Machine", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 116, + 385, + 193, + 398 + ], + "spans": [ + { + "bbox": [ + 116, + 385, + 193, + 398 + ], + "score": 1.0, + "content": "Intelligence, 2020.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 404, + 503, + 428 + ], + "lines": [ + { + "bbox": [ + 106, + 405, + 504, + 417 + ], + "spans": [ + { + "bbox": [ + 106, + 405, + 504, + 417 + ], + "score": 1.0, + "content": "Nicolas Keriven and Gabriel Peyre. Universal invariant and equivariant graph neural networks. ´", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 116, + 415, + 453, + 428 + ], + "spans": [ + { + "bbox": [ + 116, + 415, + 453, + 428 + ], + "score": 1.0, + "content": "CoRR, abs/1905.04943, 2019. URL http://arxiv.org/abs/1905.04943.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 106, + 435, + 504, + 458 + ], + "lines": [ + { + "bbox": [ + 105, + 433, + 505, + 449 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 505, + 449 + ], + "score": 1.0, + "content": "Risi Kondor. N-body networks: a covariant hierarchical neural network architecture for learning", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 447, + 353, + 457 + ], + "spans": [ + { + "bbox": [ + 115, + 447, + 353, + 457 + ], + "score": 1.0, + "content": "atomic potentials. arXiv preprint arXiv:1803.01588, 2018.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5 + }, + { + "type": "text", + "bbox": [ + 106, + 464, + 504, + 498 + ], + "lines": [ + { + "bbox": [ + 106, + 464, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 464, + 505, + 478 + ], + "score": 1.0, + "content": "Risi Kondor, Zhen Lin, and Shubhendu Trivedi. Clebsch–gordan nets: a fully fourier space spherical", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 476, + 506, + 489 + ], + "spans": [ + { + "bbox": [ + 115, + 476, + 506, + 489 + ], + "score": 1.0, + "content": "convolutional neural network. In Advances in Neural Information Processing Systems, pp. 10117–", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 486, + 176, + 500 + ], + "spans": [ + { + "bbox": [ + 116, + 486, + 176, + 500 + ], + "score": 1.0, + "content": "10126, 2018a.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 105, + 506, + 504, + 529 + ], + "lines": [ + { + "bbox": [ + 106, + 507, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 106, + 507, + 505, + 519 + ], + "score": 1.0, + "content": "Risi Kondor, Hy Truong Son, Horace Pan, Brandon Anderson, and Shubhendu Trivedi. Covariant", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 518, + 463, + 530 + ], + "spans": [ + { + "bbox": [ + 115, + 518, + 463, + 530 + ], + "score": 1.0, + "content": "compositional networks for learning graphs. arXiv preprint arXiv:1801.02144, 2018b.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 105, + 536, + 504, + 560 + ], + "lines": [ + { + "bbox": [ + 105, + 535, + 506, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 506, + 550 + ], + "score": 1.0, + "content": "Hanspeter Kraft and Claudio Procesi. Classical invariant theory, a primer. Lecture Notes, Version,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 114, + 546, + 143, + 560 + ], + "spans": [ + { + "bbox": [ + 114, + 546, + 143, + 560 + ], + "score": 1.0, + "content": "2000.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 107, + 566, + 504, + 600 + ], + "lines": [ + { + "bbox": [ + 105, + 566, + 505, + 580 + ], + "spans": [ + { + "bbox": [ + 105, + 566, + 505, + 580 + ], + "score": 1.0, + "content": "Jiaxin Li, Yingcai Bi, and Gim Hee Lee. Discrete rotation equivariance for point cloud recognition.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 577, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 115, + 577, + 505, + 591 + ], + "score": 1.0, + "content": "In 2019 International Conference on Robotics and Automation (ICRA), pp. 7269–7275. IEEE,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 114, + 587, + 143, + 601 + ], + "spans": [ + { + "bbox": [ + 114, + 587, + 143, + 601 + ], + "score": 1.0, + "content": "2019.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 106, + 607, + 504, + 631 + ], + "lines": [ + { + "bbox": [ + 106, + 607, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 607, + 505, + 621 + ], + "score": 1.0, + "content": "Takanori Maehara and Hoang NT. A simple proof of the universality of invariant/equivariant graph", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 619, + 210, + 630 + ], + "spans": [ + { + "bbox": [ + 115, + 619, + 210, + 630 + ], + "score": 1.0, + "content": "neural networks, 2019.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 106, + 638, + 504, + 661 + ], + "lines": [ + { + "bbox": [ + 105, + 637, + 505, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 651 + ], + "score": 1.0, + "content": "Haggai Maron, Heli Ben-Hamu, Hadar Serviansky, and Yaron Lipman. Provably powerful graph", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 649, + 325, + 661 + ], + "spans": [ + { + "bbox": [ + 116, + 649, + 325, + 661 + ], + "score": 1.0, + "content": "networks. arXiv preprint arXiv:1905.11136, 2019a.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 108, + 668, + 503, + 703 + ], + "lines": [ + { + "bbox": [ + 105, + 667, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 505, + 682 + ], + "score": 1.0, + "content": "Haggai Maron, Heli Ben-Hamu, Nadav Shamir, and Yaron Lipman. Invariant and equivariant graph", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 678, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 115, + 678, + 505, + 693 + ], + "score": 1.0, + "content": "networks. In International Conference on Learning Representations, 2019b. URL https:", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 117, + 690, + 336, + 703 + ], + "spans": [ + { + "bbox": [ + 117, + 690, + 336, + 703 + ], + "score": 1.0, + "content": "//openreview.net/forum?id=Syx72jC9tm.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 502, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Haggai Maron, Ethan Fetaya, Nimrod Segol, and Yaron Lipman. On the universality of invariant", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 720, + 388, + 732 + ], + "spans": [ + { + "bbox": [ + 115, + 720, + 388, + 732 + ], + "score": 1.0, + "content": "networks. In International conference on machine learning, 2019c.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "10", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 176, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 176, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 176, + 94 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 100, + 504, + 123 + ], + "lines": [ + { + "bbox": [ + 105, + 99, + 504, + 113 + ], + "spans": [ + { + "bbox": [ + 105, + 99, + 504, + 113 + ], + "score": 1.0, + "content": "Matan Atzmon, Haggai Maron, and Yaron Lipman. Point convolutional neural networks by exten-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 111, + 341, + 123 + ], + "spans": [ + { + "bbox": [ + 116, + 111, + 341, + 123 + ], + "score": 1.0, + "content": "sion operators. arXiv preprint arXiv:1803.10091, 2018.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 99, + 504, + 123 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 130, + 503, + 164 + ], + "lines": [ + { + "bbox": [ + 106, + 130, + 505, + 142 + ], + "spans": [ + { + "bbox": [ + 106, + 130, + 505, + 142 + ], + "score": 1.0, + "content": "Alexander Bogatskiy, Brandon Anderson, Jan T Offermann, Marwah Roussi, David W Miller, and", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 141, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 115, + 141, + 505, + 155 + ], + "score": 1.0, + "content": "Risi Kondor. Lorentz group equivariant neural network for particle physics. arXiv preprint", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 153, + 219, + 163 + ], + "spans": [ + { + "bbox": [ + 115, + 153, + 219, + 163 + ], + "score": 1.0, + "content": "arXiv:2006.04780, 2020.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4, + "bbox_fs": [ + 106, + 130, + 505, + 163 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 171, + 504, + 194 + ], + "lines": [ + { + "bbox": [ + 105, + 169, + 505, + 185 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 505, + 185 + ], + "score": 1.0, + "content": "Taco S Cohen, Mario Geiger, Jonas Kohler, and Max Welling. Spherical cnns. ¨ arXiv preprint", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 182, + 219, + 195 + ], + "spans": [ + { + "bbox": [ + 115, + 182, + 219, + 195 + ], + "score": 1.0, + "content": "arXiv:1801.10130, 2018.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5, + "bbox_fs": [ + 105, + 169, + 505, + 195 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 201, + 504, + 225 + ], + "lines": [ + { + "bbox": [ + 105, + 200, + 505, + 215 + ], + "spans": [ + { + "bbox": [ + 105, + 200, + 505, + 215 + ], + "score": 1.0, + "content": "Feng Dai and Yuan Xu. Approximation theory and harmonic analysis on spheres and balls, vol-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 212, + 217, + 225 + ], + "spans": [ + { + "bbox": [ + 115, + 212, + 217, + 225 + ], + "score": 1.0, + "content": "ume 23. Springer, 2013.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 200, + 505, + 225 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 231, + 504, + 266 + ], + "lines": [ + { + "bbox": [ + 105, + 230, + 506, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 506, + 245 + ], + "score": 1.0, + "content": "Carlos Esteves, Christine Allen-Blanchette, Ameesh Makadia, and Kostas Daniilidis. Learning so", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 243, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 116, + 243, + 505, + 255 + ], + "score": 1.0, + "content": "(3) equivariant representations with spherical cnns. In Proceedings of the European Conference", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 254, + 306, + 266 + ], + "spans": [ + { + "bbox": [ + 115, + 254, + 306, + 266 + ], + "score": 1.0, + "content": "on Computer Vision (ECCV), pp. 52–68, 2018.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11, + "bbox_fs": [ + 105, + 230, + 506, + 266 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 272, + 504, + 296 + ], + "lines": [ + { + "bbox": [ + 106, + 273, + 504, + 285 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 504, + 285 + ], + "score": 1.0, + "content": "Fabian B Fuchs, Daniel E Worrall, Volker Fischer, and Max Welling. Se (3)-transformers: 3d roto-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 284, + 449, + 296 + ], + "spans": [ + { + "bbox": [ + 115, + 284, + 449, + 296 + ], + "score": 1.0, + "content": "translation equivariant attention networks. arXiv preprint arXiv:2006.10503, 2020.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5, + "bbox_fs": [ + 106, + 273, + 504, + 296 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 303, + 503, + 326 + ], + "lines": [ + { + "bbox": [ + 106, + 302, + 504, + 316 + ], + "spans": [ + { + "bbox": [ + 106, + 302, + 504, + 316 + ], + "score": 1.0, + "content": "William Fulton and Joe Harris. Representation theory: a first course, volume 129. Springer Science", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 116, + 315, + 220, + 325 + ], + "spans": [ + { + "bbox": [ + 116, + 315, + 220, + 325 + ], + "score": 1.0, + "content": "& Business Media, 2013.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 106, + 302, + 504, + 325 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 333, + 502, + 357 + ], + "lines": [ + { + "bbox": [ + 105, + 333, + 503, + 347 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 503, + 347 + ], + "score": 1.0, + "content": "Justin Gilmer, Samuel S Schoenholz, Patrick F Riley, Oriol Vinyals, and George E Dahl. Neural", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 345, + 444, + 357 + ], + "spans": [ + { + "bbox": [ + 115, + 345, + 444, + 357 + ], + "score": 1.0, + "content": "message passing for quantum chemistry. arXiv preprint arXiv:1704.01212, 2017.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 105, + 333, + 503, + 357 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 363, + 504, + 397 + ], + "lines": [ + { + "bbox": [ + 106, + 362, + 505, + 377 + ], + "spans": [ + { + "bbox": [ + 106, + 362, + 505, + 377 + ], + "score": 1.0, + "content": "Yulan Guo, Hanyun Wang, Qingyong Hu, Hao Liu, Li Liu, and Mohammed Bennamoun. Deep", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 116, + 375, + 505, + 387 + ], + "spans": [ + { + "bbox": [ + 116, + 375, + 505, + 387 + ], + "score": 1.0, + "content": "learning for 3d point clouds: A survey. IEEE Transactions on Pattern Analysis and Machine", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 116, + 385, + 193, + 398 + ], + "spans": [ + { + "bbox": [ + 116, + 385, + 193, + 398 + ], + "score": 1.0, + "content": "Intelligence, 2020.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 106, + 362, + 505, + 398 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 404, + 503, + 428 + ], + "lines": [ + { + "bbox": [ + 106, + 405, + 504, + 417 + ], + "spans": [ + { + "bbox": [ + 106, + 405, + 504, + 417 + ], + "score": 1.0, + "content": "Nicolas Keriven and Gabriel Peyre. Universal invariant and equivariant graph neural networks. ´", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 116, + 415, + 453, + 428 + ], + "spans": [ + { + "bbox": [ + 116, + 415, + 453, + 428 + ], + "score": 1.0, + "content": "CoRR, abs/1905.04943, 2019. URL http://arxiv.org/abs/1905.04943.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5, + "bbox_fs": [ + 106, + 405, + 504, + 428 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 435, + 504, + 458 + ], + "lines": [ + { + "bbox": [ + 105, + 433, + 505, + 449 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 505, + 449 + ], + "score": 1.0, + "content": "Risi Kondor. N-body networks: a covariant hierarchical neural network architecture for learning", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 447, + 353, + 457 + ], + "spans": [ + { + "bbox": [ + 115, + 447, + 353, + 457 + ], + "score": 1.0, + "content": "atomic potentials. arXiv preprint arXiv:1803.01588, 2018.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 433, + 505, + 457 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 464, + 504, + 498 + ], + "lines": [ + { + "bbox": [ + 106, + 464, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 464, + 505, + 478 + ], + "score": 1.0, + "content": "Risi Kondor, Zhen Lin, and Shubhendu Trivedi. Clebsch–gordan nets: a fully fourier space spherical", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 476, + 506, + 489 + ], + "spans": [ + { + "bbox": [ + 115, + 476, + 506, + 489 + ], + "score": 1.0, + "content": "convolutional neural network. In Advances in Neural Information Processing Systems, pp. 10117–", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 486, + 176, + 500 + ], + "spans": [ + { + "bbox": [ + 116, + 486, + 176, + 500 + ], + "score": 1.0, + "content": "10126, 2018a.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27, + "bbox_fs": [ + 106, + 464, + 506, + 500 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 506, + 504, + 529 + ], + "lines": [ + { + "bbox": [ + 106, + 507, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 106, + 507, + 505, + 519 + ], + "score": 1.0, + "content": "Risi Kondor, Hy Truong Son, Horace Pan, Brandon Anderson, and Shubhendu Trivedi. Covariant", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 115, + 518, + 463, + 530 + ], + "spans": [ + { + "bbox": [ + 115, + 518, + 463, + 530 + ], + "score": 1.0, + "content": "compositional networks for learning graphs. arXiv preprint arXiv:1801.02144, 2018b.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5, + "bbox_fs": [ + 106, + 507, + 505, + 530 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 536, + 504, + 560 + ], + "lines": [ + { + "bbox": [ + 105, + 535, + 506, + 550 + ], + "spans": [ + { + "bbox": [ + 105, + 535, + 506, + 550 + ], + "score": 1.0, + "content": "Hanspeter Kraft and Claudio Procesi. Classical invariant theory, a primer. Lecture Notes, Version,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 114, + 546, + 143, + 560 + ], + "spans": [ + { + "bbox": [ + 114, + 546, + 143, + 560 + ], + "score": 1.0, + "content": "2000.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 535, + 506, + 560 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 566, + 504, + 600 + ], + "lines": [ + { + "bbox": [ + 105, + 566, + 505, + 580 + ], + "spans": [ + { + "bbox": [ + 105, + 566, + 505, + 580 + ], + "score": 1.0, + "content": "Jiaxin Li, Yingcai Bi, and Gim Hee Lee. Discrete rotation equivariance for point cloud recognition.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 577, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 115, + 577, + 505, + 591 + ], + "score": 1.0, + "content": "In 2019 International Conference on Robotics and Automation (ICRA), pp. 7269–7275. IEEE,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 114, + 587, + 143, + 601 + ], + "spans": [ + { + "bbox": [ + 114, + 587, + 143, + 601 + ], + "score": 1.0, + "content": "2019.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 566, + 505, + 601 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 607, + 504, + 631 + ], + "lines": [ + { + "bbox": [ + 106, + 607, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 607, + 505, + 621 + ], + "score": 1.0, + "content": "Takanori Maehara and Hoang NT. A simple proof of the universality of invariant/equivariant graph", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 619, + 210, + 630 + ], + "spans": [ + { + "bbox": [ + 115, + 619, + 210, + 630 + ], + "score": 1.0, + "content": "neural networks, 2019.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5, + "bbox_fs": [ + 106, + 607, + 505, + 630 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 638, + 504, + 661 + ], + "lines": [ + { + "bbox": [ + 105, + 637, + 505, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 651 + ], + "score": 1.0, + "content": "Haggai Maron, Heli Ben-Hamu, Hadar Serviansky, and Yaron Lipman. Provably powerful graph", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 649, + 325, + 661 + ], + "spans": [ + { + "bbox": [ + 116, + 649, + 325, + 661 + ], + "score": 1.0, + "content": "networks. arXiv preprint arXiv:1905.11136, 2019a.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 105, + 637, + 505, + 661 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 668, + 503, + 703 + ], + "lines": [ + { + "bbox": [ + 105, + 667, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 505, + 682 + ], + "score": 1.0, + "content": "Haggai Maron, Heli Ben-Hamu, Nadav Shamir, and Yaron Lipman. Invariant and equivariant graph", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 678, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 115, + 678, + 505, + 693 + ], + "score": 1.0, + "content": "networks. In International Conference on Learning Representations, 2019b. URL https:", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 117, + 690, + 336, + 703 + ], + "spans": [ + { + "bbox": [ + 117, + 690, + 336, + 703 + ], + "score": 1.0, + "content": "//openreview.net/forum?id=Syx72jC9tm.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41, + "bbox_fs": [ + 105, + 667, + 505, + 703 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 709, + 502, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "Haggai Maron, Ethan Fetaya, Nimrod Segol, and Yaron Lipman. On the universality of invariant", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 720, + 388, + 732 + ], + "spans": [ + { + "bbox": [ + 115, + 720, + 388, + 732 + ], + "score": 1.0, + "content": "networks. In International conference on machine learning, 2019c.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43.5, + "bbox_fs": [ + 106, + 709, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 503, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "Haggai Maron, Or Litany, Gal Chechik, and Ethan Fetaya. On learning sets of symmetric elements.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 93, + 278, + 105 + ], + "spans": [ + { + "bbox": [ + 116, + 93, + 278, + 105 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2002.08599, 2020.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 107, + 111, + 504, + 146 + ], + "lines": [ + { + "bbox": [ + 105, + 111, + 505, + 125 + ], + "spans": [ + { + "bbox": [ + 105, + 111, + 505, + 125 + ], + "score": 1.0, + "content": "Christopher Morris, Martin Ritzert, Matthias Fey, William L Hamilton, Jan Eric Lenssen, Gaurav", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 116, + 124, + 504, + 135 + ], + "spans": [ + { + "bbox": [ + 116, + 124, + 504, + 135 + ], + "score": 1.0, + "content": "Rattan, and Martin Grohe. Weisfeiler and leman go neural: Higher-order graph neural networks.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 116, + 134, + 279, + 146 + ], + "spans": [ + { + "bbox": [ + 116, + 134, + 279, + 146 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1810.02244, 2018.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 107, + 152, + 504, + 186 + ], + "lines": [ + { + "bbox": [ + 105, + 151, + 506, + 166 + ], + "spans": [ + { + "bbox": [ + 105, + 151, + 506, + 166 + ], + "score": 1.0, + "content": "Adam Paszke, Sam Gross, Soumith Chintala, Gregory Chanan, Edward Yang, Zachary DeVito,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 164, + 505, + 176 + ], + "spans": [ + { + "bbox": [ + 116, + 164, + 505, + 176 + ], + "score": 1.0, + "content": "Zeming Lin, Alban Desmaison, Luca Antiga, and Adam Lerer. Automatic differentiation in", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 176, + 178, + 186 + ], + "spans": [ + { + "bbox": [ + 115, + 176, + 178, + 186 + ], + "score": 1.0, + "content": "pytorch. 2017.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 107, + 193, + 505, + 227 + ], + "lines": [ + { + "bbox": [ + 105, + 193, + 504, + 206 + ], + "spans": [ + { + "bbox": [ + 105, + 193, + 504, + 206 + ], + "score": 1.0, + "content": "Adrien Poulenard, Marie-Julie Rakotosaona, Yann Ponty, and Maks Ovsjanikov. Effective rotation-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 117, + 206, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 117, + 206, + 505, + 217 + ], + "score": 1.0, + "content": "invariant point cnn with spherical harmonics kernels. In 2019 International Conference on 3D", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 117, + 216, + 269, + 228 + ], + "spans": [ + { + "bbox": [ + 117, + 216, + 269, + 228 + ], + "score": 1.0, + "content": "Vision (3DV), pp. 47–56. IEEE, 2019.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 106, + 234, + 504, + 268 + ], + "lines": [ + { + "bbox": [ + 105, + 233, + 506, + 248 + ], + "spans": [ + { + "bbox": [ + 105, + 233, + 506, + 248 + ], + "score": 1.0, + "content": "Charles R Qi, Hao Su, Kaichun Mo, and Leonidas J Guibas. Pointnet: Deep learning on point sets", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 245, + 505, + 258 + ], + "spans": [ + { + "bbox": [ + 115, + 245, + 505, + 258 + ], + "score": 1.0, + "content": "for 3d classification and segmentation. Proc. Computer Vision and Pattern Recognition (CVPR),", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 116, + 255, + 201, + 268 + ], + "spans": [ + { + "bbox": [ + 116, + 255, + 201, + 268 + ], + "score": 1.0, + "content": "IEEE, 1(2):4, 2017a.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 275, + 504, + 309 + ], + "lines": [ + { + "bbox": [ + 106, + 275, + 505, + 287 + ], + "spans": [ + { + "bbox": [ + 106, + 275, + 411, + 287 + ], + "score": 1.0, + "content": "Charles Ruizhongtai Qi, Li Yi, Hao Su, and Leonidas J Guibas. Pointne", + "type": "text" + }, + { + "bbox": [ + 411, + 277, + 424, + 286 + ], + "score": 0.33, + "content": "^ { + + }", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 275, + 505, + 287 + ], + "score": 1.0, + "content": ": Deep hierarchical", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 285, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 115, + 285, + 505, + 300 + ], + "score": 1.0, + "content": "feature learning on point sets in a metric space. In Advances in neural information processing", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 298, + 246, + 309 + ], + "spans": [ + { + "bbox": [ + 115, + 298, + 246, + 309 + ], + "score": 1.0, + "content": "systems, pp. 5099–5108, 2017b.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 106, + 316, + 504, + 339 + ], + "lines": [ + { + "bbox": [ + 105, + 316, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 316, + 505, + 329 + ], + "score": 1.0, + "content": "Raghunathan Ramakrishnan, Pavlo O Dral, Matthias Rupp, and O Anatole Von Lilienfeld. Quantum", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 327, + 477, + 340 + ], + "spans": [ + { + "bbox": [ + 116, + 327, + 477, + 340 + ], + "score": 1.0, + "content": "chemistry structures and properties of 134 kilo molecules. Scientific data, 1(1):1–7, 2014.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 105, + 345, + 505, + 369 + ], + "lines": [ + { + "bbox": [ + 106, + 345, + 506, + 359 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 202, + 357 + ], + "score": 1.0, + "content": "Siamak Ravanbakhsh.", + "type": "text" + }, + { + "bbox": [ + 221, + 345, + 422, + 359 + ], + "score": 1.0, + "content": "Universal equivariant multilayer perceptrons.", + "type": "text" + }, + { + "bbox": [ + 438, + 345, + 506, + 359 + ], + "score": 1.0, + "content": "arXiv preprint", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 357, + 219, + 367 + ], + "spans": [ + { + "bbox": [ + 115, + 357, + 219, + 367 + ], + "score": 1.0, + "content": "arXiv:2002.02912, 2020.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 105, + 375, + 505, + 398 + ], + "lines": [ + { + "bbox": [ + 105, + 374, + 506, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 506, + 389 + ], + "score": 1.0, + "content": "Nimrod Segol and Yaron Lipman. On universal equivariant set networks. arXiv preprint", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 386, + 219, + 398 + ], + "spans": [ + { + "bbox": [ + 115, + 386, + 219, + 398 + ], + "score": 1.0, + "content": "arXiv:1910.02421, 2019.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 106, + 405, + 504, + 429 + ], + "lines": [ + { + "bbox": [ + 106, + 405, + 505, + 418 + ], + "spans": [ + { + "bbox": [ + 106, + 405, + 505, + 418 + ], + "score": 1.0, + "content": "Hadar Serviansky, Nimrod Segol, Jonathan Shlomi, Kyle Cranmer, Eilam Gross, Haggai Maron, and", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 117, + 416, + 498, + 429 + ], + "spans": [ + { + "bbox": [ + 117, + 416, + 498, + 429 + ], + "score": 1.0, + "content": "Yaron Lipman. Set2graph: Learning graphs from sets. arXiv preprint arXiv:2002.08772, 2020.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 107, + 434, + 504, + 469 + ], + "lines": [ + { + "bbox": [ + 105, + 435, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 505, + 447 + ], + "score": 1.0, + "content": "Nathaniel Thomas, Tess Smidt, Steven Kearnes, Lusann Yang, Li Li, Kai Kohlhoff, and Patrick", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 116, + 446, + 505, + 458 + ], + "spans": [ + { + "bbox": [ + 116, + 446, + 505, + 458 + ], + "score": 1.0, + "content": "Riley. Tensor field networks: Rotation-and translation-equivariant neural networks for 3d point", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 457, + 310, + 469 + ], + "spans": [ + { + "bbox": [ + 116, + 457, + 310, + 469 + ], + "score": 1.0, + "content": "clouds. arXiv preprint arXiv:1802.08219, 2018.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 108, + 475, + 504, + 510 + ], + "lines": [ + { + "bbox": [ + 105, + 475, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 505, + 489 + ], + "score": 1.0, + "content": "Minjie Wang, Lingfan Yu, Da Zheng, Quan Gan, Yu Gai, Zihao Ye, Mufei Li, Jinjing Zhou,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 116, + 486, + 506, + 501 + ], + "spans": [ + { + "bbox": [ + 116, + 486, + 506, + 501 + ], + "score": 1.0, + "content": "Qi Huang, Chao Ma, et al. Deep graph library: Towards efficient and scalable deep learning", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 116, + 498, + 328, + 510 + ], + "spans": [ + { + "bbox": [ + 116, + 498, + 328, + 510 + ], + "score": 1.0, + "content": "on graphs. arXiv preprint arXiv:1909.01315, 2019a.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 516, + 504, + 550 + ], + "lines": [ + { + "bbox": [ + 105, + 515, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 505, + 529 + ], + "score": 1.0, + "content": "Yue Wang, Yongbin Sun, Ziwei Liu, Sanjay E Sarma, Michael M Bronstein, and Justin M Solomon.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 527, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 115, + 527, + 505, + 541 + ], + "score": 1.0, + "content": "Dynamic graph cnn for learning on point clouds. Acm Transactions On Graphics (tog), 38(5):", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 117, + 539, + 171, + 549 + ], + "spans": [ + { + "bbox": [ + 117, + 539, + 171, + 549 + ], + "score": 1.0, + "content": "1–12, 2019b.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 557, + 504, + 591 + ], + "lines": [ + { + "bbox": [ + 106, + 557, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 557, + 505, + 569 + ], + "score": 1.0, + "content": "Maurice Weiler, Mario Geiger, Max Welling, Wouter Boomsma, and Taco Cohen. 3D Steerable", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 116, + 568, + 504, + 581 + ], + "spans": [ + { + "bbox": [ + 116, + 568, + 504, + 581 + ], + "score": 1.0, + "content": "CNNs: Learning Rotationally Equivariant Features in Volumetric Data. 2018. URL http:", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 117, + 580, + 277, + 591 + ], + "spans": [ + { + "bbox": [ + 117, + 580, + 277, + 591 + ], + "score": 1.0, + "content": "//arxiv.org/abs/1807.02547.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 105, + 597, + 504, + 621 + ], + "lines": [ + { + "bbox": [ + 105, + 597, + 506, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 506, + 611 + ], + "score": 1.0, + "content": "Daniel Worrall and Gabriel Brostow. Cubenet: Equivariance to 3d rotation and translation. In", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 609, + 482, + 622 + ], + "spans": [ + { + "bbox": [ + 116, + 609, + 482, + 622 + ], + "score": 1.0, + "content": "Proceedings of the European Conference on Computer Vision (ECCV), pp. 567–584, 2018.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 108, + 627, + 504, + 662 + ], + "lines": [ + { + "bbox": [ + 105, + 627, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 505, + 641 + ], + "score": 1.0, + "content": "Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 638, + 504, + 651 + ], + "spans": [ + { + "bbox": [ + 115, + 638, + 504, + 651 + ], + "score": 1.0, + "content": "networks? In International Conference on Learning Representations, 2019. URL https:", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 117, + 650, + 337, + 662 + ], + "spans": [ + { + "bbox": [ + 117, + 650, + 266, + 662 + ], + "score": 1.0, + "content": "//openreview.net/forum?id", + "type": "text" + }, + { + "bbox": [ + 267, + 651, + 273, + 659 + ], + "score": 0.55, + "content": "=", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 650, + 337, + 662 + ], + "score": 1.0, + "content": "ryGs6iA5Km.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 104, + 668, + 504, + 691 + ], + "lines": [ + { + "bbox": [ + 105, + 667, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 505, + 682 + ], + "score": 1.0, + "content": "Dmitry Yarotsky. Universal approximations of invariant maps by neural networks. arXiv preprint", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 679, + 219, + 691 + ], + "spans": [ + { + "bbox": [ + 115, + 679, + 219, + 691 + ], + "score": 1.0, + "content": "arXiv:1804.10306, 2018.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42.5 + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "Manzil Zaheer, Satwik Kottur, Siamak Ravanbakhsh, Barnabas Poczos, Ruslan R Salakhutdinov,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 708, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 115, + 708, + 505, + 724 + ], + "score": 1.0, + "content": "and Alexander J Smola. Deep sets. In Advances in neural information processing systems, pp.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 115, + 720, + 192, + 732 + ], + "spans": [ + { + "bbox": [ + 115, + 720, + 192, + 732 + ], + "score": 1.0, + "content": "3391–3401, 2017.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 503, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "Haggai Maron, Or Litany, Gal Chechik, and Ethan Fetaya. On learning sets of symmetric elements.", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 93, + 278, + 105 + ], + "spans": [ + { + "bbox": [ + 116, + 93, + 278, + 105 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:2002.08599, 2020.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 105, + 81, + 505, + 105 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 111, + 504, + 146 + ], + "lines": [ + { + "bbox": [ + 105, + 111, + 505, + 125 + ], + "spans": [ + { + "bbox": [ + 105, + 111, + 505, + 125 + ], + "score": 1.0, + "content": "Christopher Morris, Martin Ritzert, Matthias Fey, William L Hamilton, Jan Eric Lenssen, Gaurav", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 116, + 124, + 504, + 135 + ], + "spans": [ + { + "bbox": [ + 116, + 124, + 504, + 135 + ], + "score": 1.0, + "content": "Rattan, and Martin Grohe. Weisfeiler and leman go neural: Higher-order graph neural networks.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 116, + 134, + 279, + 146 + ], + "spans": [ + { + "bbox": [ + 116, + 134, + 279, + 146 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1810.02244, 2018.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3, + "bbox_fs": [ + 105, + 111, + 505, + 146 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 152, + 504, + 186 + ], + "lines": [ + { + "bbox": [ + 105, + 151, + 506, + 166 + ], + "spans": [ + { + "bbox": [ + 105, + 151, + 506, + 166 + ], + "score": 1.0, + "content": "Adam Paszke, Sam Gross, Soumith Chintala, Gregory Chanan, Edward Yang, Zachary DeVito,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 164, + 505, + 176 + ], + "spans": [ + { + "bbox": [ + 116, + 164, + 505, + 176 + ], + "score": 1.0, + "content": "Zeming Lin, Alban Desmaison, Luca Antiga, and Adam Lerer. Automatic differentiation in", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 176, + 178, + 186 + ], + "spans": [ + { + "bbox": [ + 115, + 176, + 178, + 186 + ], + "score": 1.0, + "content": "pytorch. 2017.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6, + "bbox_fs": [ + 105, + 151, + 506, + 186 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 193, + 505, + 227 + ], + "lines": [ + { + "bbox": [ + 105, + 193, + 504, + 206 + ], + "spans": [ + { + "bbox": [ + 105, + 193, + 504, + 206 + ], + "score": 1.0, + "content": "Adrien Poulenard, Marie-Julie Rakotosaona, Yann Ponty, and Maks Ovsjanikov. Effective rotation-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 117, + 206, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 117, + 206, + 505, + 217 + ], + "score": 1.0, + "content": "invariant point cnn with spherical harmonics kernels. In 2019 International Conference on 3D", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 117, + 216, + 269, + 228 + ], + "spans": [ + { + "bbox": [ + 117, + 216, + 269, + 228 + ], + "score": 1.0, + "content": "Vision (3DV), pp. 47–56. IEEE, 2019.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 193, + 505, + 228 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 234, + 504, + 268 + ], + "lines": [ + { + "bbox": [ + 105, + 233, + 506, + 248 + ], + "spans": [ + { + "bbox": [ + 105, + 233, + 506, + 248 + ], + "score": 1.0, + "content": "Charles R Qi, Hao Su, Kaichun Mo, and Leonidas J Guibas. Pointnet: Deep learning on point sets", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 245, + 505, + 258 + ], + "spans": [ + { + "bbox": [ + 115, + 245, + 505, + 258 + ], + "score": 1.0, + "content": "for 3d classification and segmentation. Proc. Computer Vision and Pattern Recognition (CVPR),", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 116, + 255, + 201, + 268 + ], + "spans": [ + { + "bbox": [ + 116, + 255, + 201, + 268 + ], + "score": 1.0, + "content": "IEEE, 1(2):4, 2017a.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12, + "bbox_fs": [ + 105, + 233, + 506, + 268 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 275, + 504, + 309 + ], + "lines": [ + { + "bbox": [ + 106, + 275, + 505, + 287 + ], + "spans": [ + { + "bbox": [ + 106, + 275, + 411, + 287 + ], + "score": 1.0, + "content": "Charles Ruizhongtai Qi, Li Yi, Hao Su, and Leonidas J Guibas. Pointne", + "type": "text" + }, + { + "bbox": [ + 411, + 277, + 424, + 286 + ], + "score": 0.33, + "content": "^ { + + }", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 275, + 505, + 287 + ], + "score": 1.0, + "content": ": Deep hierarchical", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 285, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 115, + 285, + 505, + 300 + ], + "score": 1.0, + "content": "feature learning on point sets in a metric space. In Advances in neural information processing", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 298, + 246, + 309 + ], + "spans": [ + { + "bbox": [ + 115, + 298, + 246, + 309 + ], + "score": 1.0, + "content": "systems, pp. 5099–5108, 2017b.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15, + "bbox_fs": [ + 106, + 275, + 505, + 309 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 316, + 504, + 339 + ], + "lines": [ + { + "bbox": [ + 105, + 316, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 316, + 505, + 329 + ], + "score": 1.0, + "content": "Raghunathan Ramakrishnan, Pavlo O Dral, Matthias Rupp, and O Anatole Von Lilienfeld. Quantum", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 327, + 477, + 340 + ], + "spans": [ + { + "bbox": [ + 116, + 327, + 477, + 340 + ], + "score": 1.0, + "content": "chemistry structures and properties of 134 kilo molecules. Scientific data, 1(1):1–7, 2014.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 105, + 316, + 505, + 340 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 345, + 505, + 369 + ], + "lines": [ + { + "bbox": [ + 106, + 345, + 506, + 359 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 202, + 357 + ], + "score": 1.0, + "content": "Siamak Ravanbakhsh.", + "type": "text" + }, + { + "bbox": [ + 221, + 345, + 422, + 359 + ], + "score": 1.0, + "content": "Universal equivariant multilayer perceptrons.", + "type": "text" + }, + { + "bbox": [ + 438, + 345, + 506, + 359 + ], + "score": 1.0, + "content": "arXiv preprint", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 357, + 219, + 367 + ], + "spans": [ + { + "bbox": [ + 115, + 357, + 219, + 367 + ], + "score": 1.0, + "content": "arXiv:2002.02912, 2020.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 106, + 345, + 506, + 367 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 375, + 505, + 398 + ], + "lines": [ + { + "bbox": [ + 105, + 374, + 506, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 506, + 389 + ], + "score": 1.0, + "content": "Nimrod Segol and Yaron Lipman. On universal equivariant set networks. arXiv preprint", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 386, + 219, + 398 + ], + "spans": [ + { + "bbox": [ + 115, + 386, + 219, + 398 + ], + "score": 1.0, + "content": "arXiv:1910.02421, 2019.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 374, + 506, + 398 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 405, + 504, + 429 + ], + "lines": [ + { + "bbox": [ + 106, + 405, + 505, + 418 + ], + "spans": [ + { + "bbox": [ + 106, + 405, + 505, + 418 + ], + "score": 1.0, + "content": "Hadar Serviansky, Nimrod Segol, Jonathan Shlomi, Kyle Cranmer, Eilam Gross, Haggai Maron, and", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 117, + 416, + 498, + 429 + ], + "spans": [ + { + "bbox": [ + 117, + 416, + 498, + 429 + ], + "score": 1.0, + "content": "Yaron Lipman. Set2graph: Learning graphs from sets. arXiv preprint arXiv:2002.08772, 2020.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 106, + 405, + 505, + 429 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 434, + 504, + 469 + ], + "lines": [ + { + "bbox": [ + 105, + 435, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 505, + 447 + ], + "score": 1.0, + "content": "Nathaniel Thomas, Tess Smidt, Steven Kearnes, Lusann Yang, Li Li, Kai Kohlhoff, and Patrick", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 116, + 446, + 505, + 458 + ], + "spans": [ + { + "bbox": [ + 116, + 446, + 505, + 458 + ], + "score": 1.0, + "content": "Riley. Tensor field networks: Rotation-and translation-equivariant neural networks for 3d point", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 457, + 310, + 469 + ], + "spans": [ + { + "bbox": [ + 116, + 457, + 310, + 469 + ], + "score": 1.0, + "content": "clouds. arXiv preprint arXiv:1802.08219, 2018.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 435, + 505, + 469 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 475, + 504, + 510 + ], + "lines": [ + { + "bbox": [ + 105, + 475, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 505, + 489 + ], + "score": 1.0, + "content": "Minjie Wang, Lingfan Yu, Da Zheng, Quan Gan, Yu Gai, Zihao Ye, Mufei Li, Jinjing Zhou,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 116, + 486, + 506, + 501 + ], + "spans": [ + { + "bbox": [ + 116, + 486, + 506, + 501 + ], + "score": 1.0, + "content": "Qi Huang, Chao Ma, et al. Deep graph library: Towards efficient and scalable deep learning", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 116, + 498, + 328, + 510 + ], + "spans": [ + { + "bbox": [ + 116, + 498, + 328, + 510 + ], + "score": 1.0, + "content": "on graphs. arXiv preprint arXiv:1909.01315, 2019a.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 475, + 506, + 510 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 516, + 504, + 550 + ], + "lines": [ + { + "bbox": [ + 105, + 515, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 505, + 529 + ], + "score": 1.0, + "content": "Yue Wang, Yongbin Sun, Ziwei Liu, Sanjay E Sarma, Michael M Bronstein, and Justin M Solomon.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 527, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 115, + 527, + 505, + 541 + ], + "score": 1.0, + "content": "Dynamic graph cnn for learning on point clouds. Acm Transactions On Graphics (tog), 38(5):", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 117, + 539, + 171, + 549 + ], + "spans": [ + { + "bbox": [ + 117, + 539, + 171, + 549 + ], + "score": 1.0, + "content": "1–12, 2019b.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32, + "bbox_fs": [ + 105, + 515, + 505, + 549 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 557, + 504, + 591 + ], + "lines": [ + { + "bbox": [ + 106, + 557, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 557, + 505, + 569 + ], + "score": 1.0, + "content": "Maurice Weiler, Mario Geiger, Max Welling, Wouter Boomsma, and Taco Cohen. 3D Steerable", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 116, + 568, + 504, + 581 + ], + "spans": [ + { + "bbox": [ + 116, + 568, + 504, + 581 + ], + "score": 1.0, + "content": "CNNs: Learning Rotationally Equivariant Features in Volumetric Data. 2018. URL http:", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 117, + 580, + 277, + 591 + ], + "spans": [ + { + "bbox": [ + 117, + 580, + 277, + 591 + ], + "score": 1.0, + "content": "//arxiv.org/abs/1807.02547.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35, + "bbox_fs": [ + 106, + 557, + 505, + 591 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 597, + 504, + 621 + ], + "lines": [ + { + "bbox": [ + 105, + 597, + 506, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 506, + 611 + ], + "score": 1.0, + "content": "Daniel Worrall and Gabriel Brostow. Cubenet: Equivariance to 3d rotation and translation. In", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 609, + 482, + 622 + ], + "spans": [ + { + "bbox": [ + 116, + 609, + 482, + 622 + ], + "score": 1.0, + "content": "Proceedings of the European Conference on Computer Vision (ECCV), pp. 567–584, 2018.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 597, + 506, + 622 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 627, + 504, + 662 + ], + "lines": [ + { + "bbox": [ + 105, + 627, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 505, + 641 + ], + "score": 1.0, + "content": "Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 638, + 504, + 651 + ], + "spans": [ + { + "bbox": [ + 115, + 638, + 504, + 651 + ], + "score": 1.0, + "content": "networks? In International Conference on Learning Representations, 2019. URL https:", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 117, + 650, + 337, + 662 + ], + "spans": [ + { + "bbox": [ + 117, + 650, + 266, + 662 + ], + "score": 1.0, + "content": "//openreview.net/forum?id", + "type": "text" + }, + { + "bbox": [ + 267, + 651, + 273, + 659 + ], + "score": 0.55, + "content": "=", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 650, + 337, + 662 + ], + "score": 1.0, + "content": "ryGs6iA5Km.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40, + "bbox_fs": [ + 105, + 627, + 505, + 662 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 668, + 504, + 691 + ], + "lines": [ + { + "bbox": [ + 105, + 667, + 505, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 505, + 682 + ], + "score": 1.0, + "content": "Dmitry Yarotsky. Universal approximations of invariant maps by neural networks. arXiv preprint", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 679, + 219, + 691 + ], + "spans": [ + { + "bbox": [ + 115, + 679, + 219, + 691 + ], + "score": 1.0, + "content": "arXiv:1804.10306, 2018.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42.5, + "bbox_fs": [ + 105, + 667, + 505, + 691 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 698, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "Manzil Zaheer, Satwik Kottur, Siamak Ravanbakhsh, Barnabas Poczos, Ruslan R Salakhutdinov,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 708, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 115, + 708, + 505, + 724 + ], + "score": 1.0, + "content": "and Alexander J Smola. Deep sets. In Advances in neural information processing systems, pp.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 115, + 720, + 192, + 732 + ], + "spans": [ + { + "bbox": [ + 115, + 720, + 192, + 732 + ], + "score": 1.0, + "content": "3391–3401, 2017.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45, + "bbox_fs": [ + 105, + 698, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Yongheng Zhao, Tolga Birdal, Jan Eric Lenssen, Emanuele Menegatti, Leonidas Guibas, and Fed-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "erico Tombari. Quaternion equivariant capsule networks for 3d point clouds. arXiv preprint", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 104, + 220, + 116 + ], + "spans": [ + { + "bbox": [ + 115, + 104, + 220, + 116 + ], + "score": 1.0, + "content": "arXiv:1912.12098, 2019.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "title", + "bbox": [ + 108, + 135, + 183, + 148 + ], + "lines": [ + { + "bbox": [ + 105, + 133, + 185, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 133, + 185, + 151 + ], + "score": 1.0, + "content": "A NOTATION", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 106, + 159, + 505, + 183 + ], + "lines": [ + { + "bbox": [ + 105, + 158, + 505, + 172 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 482, + 172 + ], + "score": 1.0, + "content": "We introduce some notation for the proofs in the appendices. We use the shortened notation", + "type": "text" + }, + { + "bbox": [ + 482, + 159, + 505, + 171 + ], + "score": 0.83, + "content": "\\bar { X } =", + "type": "inline_equation" + } + ], + "index": 4 + }, + { + "bbox": [ + 107, + 169, + 402, + 184 + ], + "spans": [ + { + "bbox": [ + 107, + 170, + 167, + 184 + ], + "score": 0.93, + "content": "\\begin{array} { r } { X - \\frac { 1 } { n } X \\mathbb { 1 } _ { n } \\mathbb { 1 } _ { n } ^ { T } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 169, + 276, + 184 + ], + "score": 1.0, + "content": "and denote the columns of", + "type": "text" + }, + { + "bbox": [ + 276, + 170, + 286, + 181 + ], + "score": 0.86, + "content": "\\bar { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 169, + 300, + 184 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 300, + 171, + 352, + 183 + ], + "score": 0.92, + "content": "\\left( \\hat { x } _ { 1 } , \\ldots , \\hat { x } _ { n } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 169, + 402, + 184 + ], + "score": 1.0, + "content": ". We denote", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "interline_equation", + "bbox": [ + 248, + 185, + 363, + 200 + ], + "lines": [ + { + "bbox": [ + 248, + 185, + 363, + 200 + ], + "spans": [ + { + "bbox": [ + 248, + 185, + 363, + 200 + ], + "score": 0.9, + "content": "\\Sigma _ { T } = \\{ \\vec { r } \\in \\mathbb { N } _ { + } ^ { * } | \\| \\vec { r } \\| _ { 1 } = T \\}", + "type": "interline_equation", + "image_path": "ea97350b1cbf196b3aa1c4c70d02d5268f172baf7f348c41df90846546deb63e.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 248, + 185, + 363, + 200 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "title", + "bbox": [ + 108, + 213, + 251, + 226 + ], + "lines": [ + { + "bbox": [ + 105, + 212, + 252, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 252, + 228 + ], + "score": 1.0, + "content": "B PROOFS FOR SECTION 3", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "title", + "bbox": [ + 107, + 238, + 320, + 249 + ], + "lines": [ + { + "bbox": [ + 106, + 238, + 320, + 251 + ], + "spans": [ + { + "bbox": [ + 106, + 238, + 132, + 251 + ], + "score": 1.0, + "content": "B.1", + "type": "text" + }, + { + "bbox": [ + 132, + 238, + 141, + 248 + ], + "score": 0.75, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 238, + 320, + 251 + ], + "score": 1.0, + "content": "-EQUIVARIANT POLYNOMIALS ARE DENSE", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 107, + 258, + 506, + 292 + ], + "lines": [ + { + "bbox": [ + 106, + 258, + 506, + 271 + ], + "spans": [ + { + "bbox": [ + 106, + 258, + 252, + 271 + ], + "score": 1.0, + "content": "A first step in proving denseness of", + "type": "text" + }, + { + "bbox": [ + 252, + 259, + 261, + 268 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 258, + 506, + 271 + ], + "score": 1.0, + "content": "-equivariance polynomials, and in the proof used in the next", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 270, + 505, + 281 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 505, + 281 + ], + "score": 1.0, + "content": "subsection is the following simple lemma, which shows that translation invariance can be dealt with", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 280, + 262, + 293 + ], + "spans": [ + { + "bbox": [ + 106, + 280, + 262, + 293 + ], + "score": 1.0, + "content": "simply by centralizing the point cloud.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 297, + 503, + 321 + ], + "lines": [ + { + "bbox": [ + 104, + 295, + 501, + 311 + ], + "spans": [ + { + "bbox": [ + 104, + 295, + 174, + 311 + ], + "score": 1.0, + "content": "In the following,", + "type": "text" + }, + { + "bbox": [ + 174, + 300, + 194, + 309 + ], + "score": 0.87, + "content": "\\rho _ { W _ { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 295, + 296, + 311 + ], + "score": 1.0, + "content": "is some representation of", + "type": "text" + }, + { + "bbox": [ + 297, + 297, + 323, + 309 + ], + "score": 0.29, + "content": "\\mathrm { S O ( 3 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 295, + 484, + 311 + ], + "score": 1.0, + "content": "on a finite dimensional real vector space", + "type": "text" + }, + { + "bbox": [ + 484, + 298, + 501, + 309 + ], + "score": 0.89, + "content": "W _ { T }", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 308, + 339, + 322 + ], + "spans": [ + { + "bbox": [ + 105, + 308, + 195, + 322 + ], + "score": 1.0, + "content": "this induces an action", + "type": "text" + }, + { + "bbox": [ + 195, + 311, + 230, + 321 + ], + "score": 0.9, + "content": "\\rho _ { W _ { T } \\times S _ { n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 308, + 243, + 322 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 244, + 308, + 294, + 321 + ], + "score": 0.91, + "content": "\\mathrm { S O } ( 3 ) \\times S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 308, + 308, + 322 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 308, + 309, + 325, + 321 + ], + "score": 0.9, + "content": "W _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 308, + 339, + 322 + ], + "score": 1.0, + "content": "by", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "interline_equation", + "bbox": [ + 231, + 324, + 380, + 339 + ], + "lines": [ + { + "bbox": [ + 231, + 324, + 380, + 339 + ], + "spans": [ + { + "bbox": [ + 231, + 324, + 380, + 339 + ], + "score": 0.92, + "content": "\\rho _ { W _ { T } \\times S _ { n } } ( R , P ) ( Y ) = \\rho _ { W _ { T } } ( R ) Y P ^ { T }", + "type": "interline_equation", + "image_path": "ed30f3a425d843a9f90649371c9135255a4219739b1c0e989ac811c1a575c53e.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 231, + 324, + 380, + 339 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 342, + 504, + 363 + ], + "lines": [ + { + "bbox": [ + 105, + 339, + 505, + 355 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 212, + 355 + ], + "score": 1.0, + "content": "This is also the action of", + "type": "text" + }, + { + "bbox": [ + 213, + 342, + 222, + 351 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 339, + 306, + 355 + ], + "score": 1.0, + "content": "which we consider,", + "type": "text" + }, + { + "bbox": [ + 307, + 344, + 370, + 353 + ], + "score": 0.87, + "content": "\\rho _ { G } ~ = ~ \\rho _ { \\underline { { W } } _ { T } \\times S _ { n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 339, + 505, + 355 + ], + "score": 1.0, + "content": ", where we have invariance with", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 104, + 350, + 457, + 365 + ], + "spans": [ + { + "bbox": [ + 104, + 350, + 310, + 365 + ], + "score": 1.0, + "content": "respect to the translation coordinate. The action of", + "type": "text" + }, + { + "bbox": [ + 310, + 353, + 319, + 362 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 319, + 350, + 333, + 365 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 333, + 353, + 356, + 362 + ], + "score": 0.88, + "content": "\\mathbb { R } ^ { 3 \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 350, + 457, + 365 + ], + "score": 1.0, + "content": "is defined in equation 1.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 106, + 365, + 504, + 389 + ], + "lines": [ + { + "bbox": [ + 104, + 363, + 504, + 380 + ], + "spans": [ + { + "bbox": [ + 104, + 363, + 208, + 380 + ], + "score": 1.0, + "content": "Lemma B.1. A function", + "type": "text" + }, + { + "bbox": [ + 208, + 365, + 279, + 379 + ], + "score": 0.91, + "content": "f : \\mathbb { R } ^ { 3 \\times n } W _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 363, + 290, + 380 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 290, + 366, + 299, + 376 + ], + "score": 0.81, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 363, + 497, + 380 + ], + "score": 1.0, + "content": "-equivariant, if and only if there exists a function", + "type": "text" + }, + { + "bbox": [ + 497, + 367, + 504, + 376 + ], + "score": 0.68, + "content": "h", + "type": "inline_equation" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 375, + 415, + 391 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 303, + 391 + ], + "score": 1.0, + "content": "which is equivariant with respect to the action of", + "type": "text" + }, + { + "bbox": [ + 304, + 377, + 354, + 390 + ], + "score": 0.91, + "content": "\\mathrm { S O } ( 3 ) \\times S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 375, + 368, + 391 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 368, + 377, + 392, + 387 + ], + "score": 0.91, + "content": "\\mathbb { R } ^ { 3 \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 375, + 415, + 391 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "interline_equation", + "bbox": [ + 249, + 392, + 362, + 416 + ], + "lines": [ + { + "bbox": [ + 249, + 392, + 362, + 416 + ], + "spans": [ + { + "bbox": [ + 249, + 392, + 362, + 416 + ], + "score": 0.94, + "content": "f ( X ) = h ( X - \\frac 1 n X 1 _ { n } 1 _ { n } ^ { T } )", + "type": "interline_equation", + "image_path": "24acaf8abdd52513d14c27352e968eb1385247ccf18f19540ce3a97abb1b7030.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 249, + 392, + 362, + 416 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 426, + 505, + 449 + ], + "lines": [ + { + "bbox": [ + 106, + 426, + 505, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 181, + 439 + ], + "score": 1.0, + "content": "Proof. Recall that", + "type": "text" + }, + { + "bbox": [ + 181, + 427, + 190, + 437 + ], + "score": 0.85, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 426, + 272, + 439 + ], + "score": 1.0, + "content": "-equivariance means", + "type": "text" + }, + { + "bbox": [ + 273, + 426, + 321, + 438 + ], + "score": 0.91, + "content": "\\mathrm { S O } ( 3 ) \\times S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 426, + 505, + 439 + ], + "score": 1.0, + "content": "equivariance and translation invariance. Thus", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 437, + 334, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 115, + 450 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 115, + 438, + 122, + 449 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 123, + 437, + 132, + 450 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 133, + 438, + 142, + 447 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 437, + 304, + 450 + ], + "score": 1.0, + "content": "-equivariant then equation 16 holds with", + "type": "text" + }, + { + "bbox": [ + 304, + 438, + 330, + 449 + ], + "score": 0.9, + "content": "h = f", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 437, + 334, + 450 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 108, + 454, + 504, + 478 + ], + "lines": [ + { + "bbox": [ + 105, + 453, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 199, + 466 + ], + "score": 1.0, + "content": "On the other hand, if", + "type": "text" + }, + { + "bbox": [ + 199, + 455, + 206, + 466 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 453, + 378, + 466 + ], + "score": 1.0, + "content": "satisfies equation 16 then we claim it is", + "type": "text" + }, + { + "bbox": [ + 379, + 455, + 387, + 464 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 453, + 505, + 466 + ], + "score": 1.0, + "content": "-equivariant. Indeed, for all", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 460, + 384, + 482 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 230, + 478 + ], + "score": 0.91, + "content": "( t , R , P ) \\in \\mathbb { R } ^ { d } \\rtimes \\mathrm { S O ( 3 ) } \\times S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 460, + 258, + 482 + ], + "score": 1.0, + "content": ", since", + "type": "text" + }, + { + "bbox": [ + 259, + 465, + 379, + 477 + ], + "score": 0.92, + "content": "P ^ { T } 1 _ { n } 1 _ { n } ^ { T } = 1 _ { n } 1 _ { n } ^ { T } = 1 _ { n } 1 _ { n } ^ { T } P ^ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 460, + 384, + 482 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "interline_equation", + "bbox": [ + 119, + 481, + 492, + 575 + ], + "lines": [ + { + "bbox": [ + 119, + 481, + 492, + 575 + ], + "spans": [ + { + "bbox": [ + 119, + 481, + 492, + 575 + ], + "score": 0.94, + "content": "\\begin{array} { l } { f \\left( \\rho _ { G } ( t , R , P ) ( X ) \\right) = f ( R ( X + t \\boldsymbol { 1 } _ { n } ) P ^ { T } ) = h ( R ( X + t \\boldsymbol { 1 } _ { n } ) P ^ { T } - \\displaystyle \\frac { 1 } { n } R ( X + t \\boldsymbol { 1 } ) P ^ { T } \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) } \\\\ { \\displaystyle \\quad \\quad = h ( R ( X - \\frac { 1 } { n } X \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) P ^ { T } ) = h \\left( \\rho _ { \\mathbb { R } ^ { 3 } \\times S _ { n } } ( R , P ) ( X - \\frac { 1 } { n } X \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) \\right) } \\\\ { \\displaystyle \\quad = \\rho _ { W _ { T } \\times S _ { n } } ( R , P ) h \\left( X - \\frac { 1 } { n } X \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } \\right) } \\\\ { \\displaystyle \\quad = \\rho _ { G } ( t , R , P ) f ( X ) . } \\end{array}", + "type": "interline_equation", + "image_path": "bfe1ba1553323a3c64485a603dd3bea449a778962e3611934c34d05529273d92.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 119, + 481, + 492, + 512.3333333333334 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 119, + 512.3333333333334, + 492, + 543.6666666666667 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 119, + 543.6666666666667, + 492, + 575.0000000000001 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 606, + 504, + 628 + ], + "lines": [ + { + "bbox": [ + 106, + 605, + 505, + 619 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 218, + 619 + ], + "score": 1.0, + "content": "We now prove denseness of", + "type": "text" + }, + { + "bbox": [ + 219, + 607, + 227, + 616 + ], + "score": 0.84, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 605, + 387, + 619 + ], + "score": 1.0, + "content": "-equivariant polynomials in the space of", + "type": "text" + }, + { + "bbox": [ + 388, + 607, + 396, + 616 + ], + "score": 0.84, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 605, + 505, + 619 + ], + "score": 1.0, + "content": "-invariant continuous func-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 617, + 178, + 629 + ], + "spans": [ + { + "bbox": [ + 106, + 617, + 178, + 629 + ], + "score": 1.0, + "content": "tions (Lemma 1).", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 106, + 630, + 502, + 654 + ], + "lines": [ + { + "bbox": [ + 104, + 628, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 104, + 628, + 222, + 645 + ], + "score": 1.0, + "content": "Lemma 1. Any continuous", + "type": "text" + }, + { + "bbox": [ + 222, + 632, + 231, + 641 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 628, + 330, + 645 + ], + "score": 1.0, + "content": "-equivariant function in", + "type": "text" + }, + { + "bbox": [ + 331, + 631, + 394, + 642 + ], + "score": 0.82, + "content": "\\mathcal { C } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , W _ { T } ^ { n } )", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 628, + 505, + 645 + ], + "score": 1.0, + "content": "can be approximated uni-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 640, + 404, + 656 + ], + "spans": [ + { + "bbox": [ + 104, + 640, + 213, + 656 + ], + "score": 1.0, + "content": "formly on compact sets by", + "type": "text" + }, + { + "bbox": [ + 213, + 642, + 222, + 652 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 640, + 333, + 656 + ], + "score": 1.0, + "content": "-equivariant polynomials in", + "type": "text" + }, + { + "bbox": [ + 334, + 642, + 399, + 654 + ], + "score": 0.78, + "content": "\\mathcal { P } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , \\dot { W } _ { T } ^ { n } )", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 640, + 404, + 656 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 106, + 664, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 664, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 664, + 204, + 678 + ], + "score": 1.0, + "content": "Proof of Lemma 1. Let", + "type": "text" + }, + { + "bbox": [ + 204, + 665, + 258, + 677 + ], + "score": 0.92, + "content": "K ~ \\subseteq ~ \\mathbb { R } ^ { 3 \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 664, + 491, + 678 + ], + "score": 1.0, + "content": "be a compact set. We need to show that continuous", + "type": "text" + }, + { + "bbox": [ + 492, + 666, + 500, + 676 + ], + "score": 0.81, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 664, + 505, + 678 + ], + "score": 1.0, + "content": "-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 101, + 676, + 504, + 712 + ], + "spans": [ + { + "bbox": [ + 101, + 676, + 291, + 712 + ], + "score": 1.0, + "content": "equivariant functions can be approximated undenote the compact set which is the image of By Lemma B.1, it is sufficient to show that eve", + "type": "text" + }, + { + "bbox": [ + 334, + 677, + 344, + 687 + ], + "score": 0.82, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 676, + 359, + 712 + ], + "score": 1.0, + "content": "y ene", + "type": "text" + }, + { + "bbox": [ + 359, + 677, + 368, + 687 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 676, + 414, + 712 + ], + "score": 1.0, + "content": "-equivarianlizing map ivariant con", + "type": "text" + }, + { + "bbox": [ + 490, + 677, + 504, + 687 + ], + "score": 0.85, + "content": "K _ { 0 }", + "type": "inline_equation" + } + ], + "index": 32 + }, + { + "bbox": [ + 292, + 687, + 501, + 701 + ], + "spans": [ + { + "bbox": [ + 292, + 688, + 302, + 698 + ], + "score": 0.83, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 687, + 501, + 701 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\dot { X } \\stackrel { \\bullet } { \\mapsto } X - \\frac { 1 } { n } X 1 _ { n } 1 _ { n } ^ { T } } \\end{array}", + "type": "inline_equation" + } + ], + "index": 33 + }, + { + "bbox": [ + 302, + 699, + 488, + 710 + ], + "spans": [ + { + "bbox": [ + 302, + 699, + 351, + 710 + ], + "score": 0.9, + "content": "\\mathrm { S O } ( 3 ) \\times S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 481, + 700, + 488, + 710 + ], + "score": 0.83, + "content": "f", + "type": "inline_equation" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 232, + 722 + ], + "score": 1.0, + "content": "be approximated uniformly on", + "type": "text" + }, + { + "bbox": [ + 232, + 710, + 246, + 721 + ], + "score": 0.88, + "content": "K _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 709, + 318, + 722 + ], + "score": 1.0, + "content": "by a sequence of", + "type": "text" + }, + { + "bbox": [ + 319, + 710, + 370, + 722 + ], + "score": 0.91, + "content": "\\mathrm { S O } ( 3 ) \\times S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 709, + 470, + 722 + ], + "score": 1.0, + "content": "equivariant polynomials", + "type": "text" + }, + { + "bbox": [ + 470, + 711, + 481, + 721 + ], + "score": 0.86, + "content": "p _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 482, + 709, + 506, + 722 + ], + "score": 1.0, + "content": ". The", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 721, + 351, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 351, + 734 + ], + "score": 1.0, + "content": "argument is now concluded by the following general lemma:", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 33.5 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 495, + 582, + 504, + 593 + ], + "lines": [ + { + "bbox": [ + 496, + 584, + 504, + 594 + ], + "spans": [ + { + "bbox": [ + 496, + 584, + 504, + 594 + ], + "score": 0.998, + "content": "□", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Yongheng Zhao, Tolga Birdal, Jan Eric Lenssen, Emanuele Menegatti, Leonidas Guibas, and Fed-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "erico Tombari. Quaternion equivariant capsule networks for 3d point clouds. arXiv preprint", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 104, + 220, + 116 + ], + "spans": [ + { + "bbox": [ + 115, + 104, + 220, + 116 + ], + "score": 1.0, + "content": "arXiv:1912.12098, 2019.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 106, + 82, + 505, + 116 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 135, + 183, + 148 + ], + "lines": [ + { + "bbox": [ + 105, + 133, + 185, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 133, + 185, + 151 + ], + "score": 1.0, + "content": "A NOTATION", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 106, + 159, + 505, + 183 + ], + "lines": [ + { + "bbox": [ + 105, + 158, + 505, + 172 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 482, + 172 + ], + "score": 1.0, + "content": "We introduce some notation for the proofs in the appendices. We use the shortened notation", + "type": "text" + }, + { + "bbox": [ + 482, + 159, + 505, + 171 + ], + "score": 0.83, + "content": "\\bar { X } =", + "type": "inline_equation" + } + ], + "index": 4 + }, + { + "bbox": [ + 107, + 169, + 402, + 184 + ], + "spans": [ + { + "bbox": [ + 107, + 170, + 167, + 184 + ], + "score": 0.93, + "content": "\\begin{array} { r } { X - \\frac { 1 } { n } X \\mathbb { 1 } _ { n } \\mathbb { 1 } _ { n } ^ { T } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 169, + 276, + 184 + ], + "score": 1.0, + "content": "and denote the columns of", + "type": "text" + }, + { + "bbox": [ + 276, + 170, + 286, + 181 + ], + "score": 0.86, + "content": "\\bar { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 169, + 300, + 184 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 300, + 171, + 352, + 183 + ], + "score": 0.92, + "content": "\\left( \\hat { x } _ { 1 } , \\ldots , \\hat { x } _ { n } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 169, + 402, + 184 + ], + "score": 1.0, + "content": ". We denote", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 105, + 158, + 505, + 184 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 248, + 185, + 363, + 200 + ], + "lines": [ + { + "bbox": [ + 248, + 185, + 363, + 200 + ], + "spans": [ + { + "bbox": [ + 248, + 185, + 363, + 200 + ], + "score": 0.9, + "content": "\\Sigma _ { T } = \\{ \\vec { r } \\in \\mathbb { N } _ { + } ^ { * } | \\| \\vec { r } \\| _ { 1 } = T \\}", + "type": "interline_equation", + "image_path": "ea97350b1cbf196b3aa1c4c70d02d5268f172baf7f348c41df90846546deb63e.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 248, + 185, + 363, + 200 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "title", + "bbox": [ + 108, + 213, + 251, + 226 + ], + "lines": [ + { + "bbox": [ + 105, + 212, + 252, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 252, + 228 + ], + "score": 1.0, + "content": "B PROOFS FOR SECTION 3", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "title", + "bbox": [ + 107, + 238, + 320, + 249 + ], + "lines": [ + { + "bbox": [ + 106, + 238, + 320, + 251 + ], + "spans": [ + { + "bbox": [ + 106, + 238, + 132, + 251 + ], + "score": 1.0, + "content": "B.1", + "type": "text" + }, + { + "bbox": [ + 132, + 238, + 141, + 248 + ], + "score": 0.75, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 238, + 320, + 251 + ], + "score": 1.0, + "content": "-EQUIVARIANT POLYNOMIALS ARE DENSE", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 107, + 258, + 506, + 292 + ], + "lines": [ + { + "bbox": [ + 106, + 258, + 506, + 271 + ], + "spans": [ + { + "bbox": [ + 106, + 258, + 252, + 271 + ], + "score": 1.0, + "content": "A first step in proving denseness of", + "type": "text" + }, + { + "bbox": [ + 252, + 259, + 261, + 268 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 258, + 506, + 271 + ], + "score": 1.0, + "content": "-equivariance polynomials, and in the proof used in the next", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 270, + 505, + 281 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 505, + 281 + ], + "score": 1.0, + "content": "subsection is the following simple lemma, which shows that translation invariance can be dealt with", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 280, + 262, + 293 + ], + "spans": [ + { + "bbox": [ + 106, + 280, + 262, + 293 + ], + "score": 1.0, + "content": "simply by centralizing the point cloud.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 258, + 506, + 293 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 297, + 503, + 321 + ], + "lines": [ + { + "bbox": [ + 104, + 295, + 501, + 311 + ], + "spans": [ + { + "bbox": [ + 104, + 295, + 174, + 311 + ], + "score": 1.0, + "content": "In the following,", + "type": "text" + }, + { + "bbox": [ + 174, + 300, + 194, + 309 + ], + "score": 0.87, + "content": "\\rho _ { W _ { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 295, + 296, + 311 + ], + "score": 1.0, + "content": "is some representation of", + "type": "text" + }, + { + "bbox": [ + 297, + 297, + 323, + 309 + ], + "score": 0.29, + "content": "\\mathrm { S O ( 3 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 295, + 484, + 311 + ], + "score": 1.0, + "content": "on a finite dimensional real vector space", + "type": "text" + }, + { + "bbox": [ + 484, + 298, + 501, + 309 + ], + "score": 0.89, + "content": "W _ { T }", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 308, + 339, + 322 + ], + "spans": [ + { + "bbox": [ + 105, + 308, + 195, + 322 + ], + "score": 1.0, + "content": "this induces an action", + "type": "text" + }, + { + "bbox": [ + 195, + 311, + 230, + 321 + ], + "score": 0.9, + "content": "\\rho _ { W _ { T } \\times S _ { n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 308, + 243, + 322 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 244, + 308, + 294, + 321 + ], + "score": 0.91, + "content": "\\mathrm { S O } ( 3 ) \\times S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 308, + 308, + 322 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 308, + 309, + 325, + 321 + ], + "score": 0.9, + "content": "W _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 308, + 339, + 322 + ], + "score": 1.0, + "content": "by", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 104, + 295, + 501, + 322 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 231, + 324, + 380, + 339 + ], + "lines": [ + { + "bbox": [ + 231, + 324, + 380, + 339 + ], + "spans": [ + { + "bbox": [ + 231, + 324, + 380, + 339 + ], + "score": 0.92, + "content": "\\rho _ { W _ { T } \\times S _ { n } } ( R , P ) ( Y ) = \\rho _ { W _ { T } } ( R ) Y P ^ { T }", + "type": "interline_equation", + "image_path": "ed30f3a425d843a9f90649371c9135255a4219739b1c0e989ac811c1a575c53e.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 231, + 324, + 380, + 339 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 342, + 504, + 363 + ], + "lines": [ + { + "bbox": [ + 105, + 339, + 505, + 355 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 212, + 355 + ], + "score": 1.0, + "content": "This is also the action of", + "type": "text" + }, + { + "bbox": [ + 213, + 342, + 222, + 351 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 339, + 306, + 355 + ], + "score": 1.0, + "content": "which we consider,", + "type": "text" + }, + { + "bbox": [ + 307, + 344, + 370, + 353 + ], + "score": 0.87, + "content": "\\rho _ { G } ~ = ~ \\rho _ { \\underline { { W } } _ { T } \\times S _ { n } }", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 339, + 505, + 355 + ], + "score": 1.0, + "content": ", where we have invariance with", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 104, + 350, + 457, + 365 + ], + "spans": [ + { + "bbox": [ + 104, + 350, + 310, + 365 + ], + "score": 1.0, + "content": "respect to the translation coordinate. The action of", + "type": "text" + }, + { + "bbox": [ + 310, + 353, + 319, + 362 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 319, + 350, + 333, + 365 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 333, + 353, + 356, + 362 + ], + "score": 0.88, + "content": "\\mathbb { R } ^ { 3 \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 350, + 457, + 365 + ], + "score": 1.0, + "content": "is defined in equation 1.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 104, + 339, + 505, + 365 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 365, + 504, + 389 + ], + "lines": [ + { + "bbox": [ + 104, + 363, + 504, + 380 + ], + "spans": [ + { + "bbox": [ + 104, + 363, + 208, + 380 + ], + "score": 1.0, + "content": "Lemma B.1. A function", + "type": "text" + }, + { + "bbox": [ + 208, + 365, + 279, + 379 + ], + "score": 0.91, + "content": "f : \\mathbb { R } ^ { 3 \\times n } W _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 363, + 290, + 380 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 290, + 366, + 299, + 376 + ], + "score": 0.81, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 363, + 497, + 380 + ], + "score": 1.0, + "content": "-equivariant, if and only if there exists a function", + "type": "text" + }, + { + "bbox": [ + 497, + 367, + 504, + 376 + ], + "score": 0.68, + "content": "h", + "type": "inline_equation" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 375, + 415, + 391 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 303, + 391 + ], + "score": 1.0, + "content": "which is equivariant with respect to the action of", + "type": "text" + }, + { + "bbox": [ + 304, + 377, + 354, + 390 + ], + "score": 0.91, + "content": "\\mathrm { S O } ( 3 ) \\times S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 375, + 368, + 391 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 368, + 377, + 392, + 387 + ], + "score": 0.91, + "content": "\\mathbb { R } ^ { 3 \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 375, + 415, + 391 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 104, + 363, + 504, + 391 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 249, + 392, + 362, + 416 + ], + "lines": [ + { + "bbox": [ + 249, + 392, + 362, + 416 + ], + "spans": [ + { + "bbox": [ + 249, + 392, + 362, + 416 + ], + "score": 0.94, + "content": "f ( X ) = h ( X - \\frac 1 n X 1 _ { n } 1 _ { n } ^ { T } )", + "type": "interline_equation", + "image_path": "24acaf8abdd52513d14c27352e968eb1385247ccf18f19540ce3a97abb1b7030.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 249, + 392, + 362, + 416 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 426, + 505, + 449 + ], + "lines": [ + { + "bbox": [ + 106, + 426, + 505, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 181, + 439 + ], + "score": 1.0, + "content": "Proof. Recall that", + "type": "text" + }, + { + "bbox": [ + 181, + 427, + 190, + 437 + ], + "score": 0.85, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 426, + 272, + 439 + ], + "score": 1.0, + "content": "-equivariance means", + "type": "text" + }, + { + "bbox": [ + 273, + 426, + 321, + 438 + ], + "score": 0.91, + "content": "\\mathrm { S O } ( 3 ) \\times S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 426, + 505, + 439 + ], + "score": 1.0, + "content": "equivariance and translation invariance. Thus", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 437, + 334, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 115, + 450 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 115, + 438, + 122, + 449 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 123, + 437, + 132, + 450 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 133, + 438, + 142, + 447 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 437, + 304, + 450 + ], + "score": 1.0, + "content": "-equivariant then equation 16 holds with", + "type": "text" + }, + { + "bbox": [ + 304, + 438, + 330, + 449 + ], + "score": 0.9, + "content": "h = f", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 437, + 334, + 450 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5, + "bbox_fs": [ + 105, + 426, + 505, + 450 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 454, + 504, + 478 + ], + "lines": [ + { + "bbox": [ + 105, + 453, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 199, + 466 + ], + "score": 1.0, + "content": "On the other hand, if", + "type": "text" + }, + { + "bbox": [ + 199, + 455, + 206, + 466 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 453, + 378, + 466 + ], + "score": 1.0, + "content": "satisfies equation 16 then we claim it is", + "type": "text" + }, + { + "bbox": [ + 379, + 455, + 387, + 464 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 453, + 505, + 466 + ], + "score": 1.0, + "content": "-equivariant. Indeed, for all", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 460, + 384, + 482 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 230, + 478 + ], + "score": 0.91, + "content": "( t , R , P ) \\in \\mathbb { R } ^ { d } \\rtimes \\mathrm { S O ( 3 ) } \\times S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 460, + 258, + 482 + ], + "score": 1.0, + "content": ", since", + "type": "text" + }, + { + "bbox": [ + 259, + 465, + 379, + 477 + ], + "score": 0.92, + "content": "P ^ { T } 1 _ { n } 1 _ { n } ^ { T } = 1 _ { n } 1 _ { n } ^ { T } = 1 _ { n } 1 _ { n } ^ { T } P ^ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 460, + 384, + 482 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 453, + 505, + 482 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 119, + 481, + 492, + 575 + ], + "lines": [ + { + "bbox": [ + 119, + 481, + 492, + 575 + ], + "spans": [ + { + "bbox": [ + 119, + 481, + 492, + 575 + ], + "score": 0.94, + "content": "\\begin{array} { l } { f \\left( \\rho _ { G } ( t , R , P ) ( X ) \\right) = f ( R ( X + t \\boldsymbol { 1 } _ { n } ) P ^ { T } ) = h ( R ( X + t \\boldsymbol { 1 } _ { n } ) P ^ { T } - \\displaystyle \\frac { 1 } { n } R ( X + t \\boldsymbol { 1 } ) P ^ { T } \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) } \\\\ { \\displaystyle \\quad \\quad = h ( R ( X - \\frac { 1 } { n } X \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) P ^ { T } ) = h \\left( \\rho _ { \\mathbb { R } ^ { 3 } \\times S _ { n } } ( R , P ) ( X - \\frac { 1 } { n } X \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) \\right) } \\\\ { \\displaystyle \\quad = \\rho _ { W _ { T } \\times S _ { n } } ( R , P ) h \\left( X - \\frac { 1 } { n } X \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } \\right) } \\\\ { \\displaystyle \\quad = \\rho _ { G } ( t , R , P ) f ( X ) . } \\end{array}", + "type": "interline_equation", + "image_path": "bfe1ba1553323a3c64485a603dd3bea449a778962e3611934c34d05529273d92.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 119, + 481, + 492, + 512.3333333333334 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 119, + 512.3333333333334, + 492, + 543.6666666666667 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 119, + 543.6666666666667, + 492, + 575.0000000000001 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 606, + 504, + 628 + ], + "lines": [ + { + "bbox": [ + 106, + 605, + 505, + 619 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 218, + 619 + ], + "score": 1.0, + "content": "We now prove denseness of", + "type": "text" + }, + { + "bbox": [ + 219, + 607, + 227, + 616 + ], + "score": 0.84, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 605, + 387, + 619 + ], + "score": 1.0, + "content": "-equivariant polynomials in the space of", + "type": "text" + }, + { + "bbox": [ + 388, + 607, + 396, + 616 + ], + "score": 0.84, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 605, + 505, + 619 + ], + "score": 1.0, + "content": "-invariant continuous func-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 617, + 178, + 629 + ], + "spans": [ + { + "bbox": [ + 106, + 617, + 178, + 629 + ], + "score": 1.0, + "content": "tions (Lemma 1).", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5, + "bbox_fs": [ + 106, + 605, + 505, + 629 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 630, + 502, + 654 + ], + "lines": [ + { + "bbox": [ + 104, + 628, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 104, + 628, + 222, + 645 + ], + "score": 1.0, + "content": "Lemma 1. Any continuous", + "type": "text" + }, + { + "bbox": [ + 222, + 632, + 231, + 641 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 628, + 330, + 645 + ], + "score": 1.0, + "content": "-equivariant function in", + "type": "text" + }, + { + "bbox": [ + 331, + 631, + 394, + 642 + ], + "score": 0.82, + "content": "\\mathcal { C } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , W _ { T } ^ { n } )", + "type": "inline_equation" + }, + { + "bbox": [ + 395, + 628, + 505, + 645 + ], + "score": 1.0, + "content": "can be approximated uni-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 640, + 404, + 656 + ], + "spans": [ + { + "bbox": [ + 104, + 640, + 213, + 656 + ], + "score": 1.0, + "content": "formly on compact sets by", + "type": "text" + }, + { + "bbox": [ + 213, + 642, + 222, + 652 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 640, + 333, + 656 + ], + "score": 1.0, + "content": "-equivariant polynomials in", + "type": "text" + }, + { + "bbox": [ + 334, + 642, + 399, + 654 + ], + "score": 0.78, + "content": "\\mathcal { P } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , \\dot { W } _ { T } ^ { n } )", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 640, + 404, + 656 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5, + "bbox_fs": [ + 104, + 628, + 505, + 656 + ] + }, + { + "type": "list", + "bbox": [ + 106, + 664, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 664, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 664, + 204, + 678 + ], + "score": 1.0, + "content": "Proof of Lemma 1. Let", + "type": "text" + }, + { + "bbox": [ + 204, + 665, + 258, + 677 + ], + "score": 0.92, + "content": "K ~ \\subseteq ~ \\mathbb { R } ^ { 3 \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 664, + 491, + 678 + ], + "score": 1.0, + "content": "be a compact set. We need to show that continuous", + "type": "text" + }, + { + "bbox": [ + 492, + 666, + 500, + 676 + ], + "score": 0.81, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 664, + 505, + 678 + ], + "score": 1.0, + "content": "-", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 101, + 676, + 504, + 712 + ], + "spans": [ + { + "bbox": [ + 101, + 676, + 291, + 712 + ], + "score": 1.0, + "content": "equivariant functions can be approximated undenote the compact set which is the image of By Lemma B.1, it is sufficient to show that eve", + "type": "text" + }, + { + "bbox": [ + 334, + 677, + 344, + 687 + ], + "score": 0.82, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 676, + 359, + 712 + ], + "score": 1.0, + "content": "y ene", + "type": "text" + }, + { + "bbox": [ + 359, + 677, + 368, + 687 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 676, + 414, + 712 + ], + "score": 1.0, + "content": "-equivarianlizing map ivariant con", + "type": "text" + }, + { + "bbox": [ + 490, + 677, + 504, + 687 + ], + "score": 0.85, + "content": "K _ { 0 }", + "type": "inline_equation" + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 292, + 687, + 501, + 701 + ], + "spans": [ + { + "bbox": [ + 292, + 688, + 302, + 698 + ], + "score": 0.83, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 687, + 501, + 701 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\dot { X } \\stackrel { \\bullet } { \\mapsto } X - \\frac { 1 } { n } X 1 _ { n } 1 _ { n } ^ { T } } \\end{array}", + "type": "inline_equation" + } + ], + "index": 33 + }, + { + "bbox": [ + 302, + 699, + 488, + 710 + ], + "spans": [ + { + "bbox": [ + 302, + 699, + 351, + 710 + ], + "score": 0.9, + "content": "\\mathrm { S O } ( 3 ) \\times S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 481, + 700, + 488, + 710 + ], + "score": 0.83, + "content": "f", + "type": "inline_equation" + } + ], + "index": 34, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 232, + 722 + ], + "score": 1.0, + "content": "be approximated uniformly on", + "type": "text" + }, + { + "bbox": [ + 232, + 710, + 246, + 721 + ], + "score": 0.88, + "content": "K _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 709, + 318, + 722 + ], + "score": 1.0, + "content": "by a sequence of", + "type": "text" + }, + { + "bbox": [ + 319, + 710, + 370, + 722 + ], + "score": 0.91, + "content": "\\mathrm { S O } ( 3 ) \\times S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 709, + 470, + 722 + ], + "score": 1.0, + "content": "equivariant polynomials", + "type": "text" + }, + { + "bbox": [ + 470, + 711, + 481, + 721 + ], + "score": 0.86, + "content": "p _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 482, + 709, + 506, + 722 + ], + "score": 1.0, + "content": ". The", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 106, + 721, + 351, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 351, + 734 + ], + "score": 1.0, + "content": "argument is now concluded by the following general lemma:", + "type": "text" + } + ], + "index": 36, + "is_list_start_line": true, + "is_list_end_line": true + } + ], + "index": 33.5, + "bbox_fs": [ + 101, + 664, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 505, + 127 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 179, + 96 + ], + "score": 1.0, + "content": "Lemma B.2. Let", + "type": "text" + }, + { + "bbox": [ + 180, + 83, + 188, + 93 + ], + "score": 0.73, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 82, + 290, + 96 + ], + "score": 1.0, + "content": "be a compact group, Let", + "type": "text" + }, + { + "bbox": [ + 290, + 84, + 301, + 94 + ], + "score": 0.85, + "content": "\\rho _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 301, + 82, + 320, + 96 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 320, + 84, + 331, + 94 + ], + "score": 0.83, + "content": "\\rho _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 82, + 467, + 96 + ], + "score": 1.0, + "content": "be continuous2representations of", + "type": "text" + }, + { + "bbox": [ + 467, + 83, + 477, + 93 + ], + "score": 0.48, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "on the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 180, + 107 + ], + "score": 1.0, + "content": "Euclidean spaces", + "type": "text" + }, + { + "bbox": [ + 180, + 94, + 195, + 105 + ], + "score": 0.88, + "content": "W _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 93, + 215, + 107 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 215, + 94, + 230, + 105 + ], + "score": 0.91, + "content": "W _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 93, + 253, + 107 + ], + "score": 1.0, + "content": ". Let", + "type": "text" + }, + { + "bbox": [ + 253, + 94, + 293, + 105 + ], + "score": 0.91, + "content": "K \\subseteq W _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 93, + 505, + 107 + ], + "score": 1.0, + "content": "be a compact set. Then every equivariant function", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 107, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 107, + 105, + 172, + 116 + ], + "score": 0.9, + "content": "f : W _ { 1 } \\to W _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 104, + 319, + 117 + ], + "score": 1.0, + "content": "can be approximated uniformly on", + "type": "text" + }, + { + "bbox": [ + 320, + 105, + 330, + 114 + ], + "score": 0.8, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "by a sequence of equivariant polynomials", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 115, + 173, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 116, + 169, + 127 + ], + "score": 0.89, + "content": "p _ { k } : W _ { 1 } \\mapsto W _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 115, + 173, + 128 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "text", + "bbox": [ + 106, + 138, + 505, + 161 + ], + "lines": [ + { + "bbox": [ + 105, + 138, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 138, + 123, + 151 + ], + "score": 1.0, + "content": "Let", + "type": "text" + }, + { + "bbox": [ + 123, + 140, + 131, + 150 + ], + "score": 0.81, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 131, + 138, + 411, + 151 + ], + "score": 1.0, + "content": "be the Haar probability measure associated with the compact group", + "type": "text" + }, + { + "bbox": [ + 412, + 139, + 421, + 148 + ], + "score": 0.78, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 138, + 444, + 151 + ], + "score": 1.0, + "content": ". Let", + "type": "text" + }, + { + "bbox": [ + 444, + 139, + 458, + 149 + ], + "score": 0.89, + "content": "K _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 138, + 505, + 151 + ], + "score": 1.0, + "content": "denote the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 147, + 473, + 163 + ], + "spans": [ + { + "bbox": [ + 105, + 147, + 318, + 163 + ], + "score": 1.0, + "content": "compact set obtained as an image of the compact set", + "type": "text" + }, + { + "bbox": [ + 318, + 150, + 348, + 159 + ], + "score": 0.92, + "content": "G \\times K", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 147, + 473, + 163 + ], + "score": 1.0, + "content": "under the continuous mapping", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "interline_equation", + "bbox": [ + 267, + 164, + 344, + 177 + ], + "lines": [ + { + "bbox": [ + 267, + 164, + 344, + 177 + ], + "spans": [ + { + "bbox": [ + 267, + 164, + 344, + 177 + ], + "score": 0.9, + "content": "( g , X ) \\mapsto \\rho _ { 1 } ( g ) X .", + "type": "interline_equation", + "image_path": "5f4461c64095fc77cdd890372dae6fd44c4eb438541773c5e7fd11abefadc925.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 267, + 164, + 344, + 177 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 180, + 504, + 213 + ], + "lines": [ + { + "bbox": [ + 105, + 179, + 504, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 179, + 272, + 194 + ], + "score": 1.0, + "content": "Using the Stone-Weierstrass theorem, let", + "type": "text" + }, + { + "bbox": [ + 273, + 182, + 284, + 192 + ], + "score": 0.86, + "content": "p _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 179, + 504, + 194 + ], + "score": 1.0, + "content": "be a sequence of (not necessarily equivariant) polyno-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 191, + 506, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 209, + 204 + ], + "score": 1.0, + "content": "mials which approximate", + "type": "text" + }, + { + "bbox": [ + 210, + 192, + 217, + 203 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 191, + 273, + 204 + ], + "score": 1.0, + "content": "uniformly on", + "type": "text" + }, + { + "bbox": [ + 273, + 192, + 287, + 203 + ], + "score": 0.9, + "content": "K _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 191, + 347, + 204 + ], + "score": 1.0, + "content": ". Every degree", + "type": "text" + }, + { + "bbox": [ + 347, + 192, + 357, + 201 + ], + "score": 0.84, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 191, + 406, + 204 + ], + "score": 1.0, + "content": "polynomial", + "type": "text" + }, + { + "bbox": [ + 406, + 191, + 464, + 203 + ], + "score": 0.93, + "content": "p : W _ { 1 } \\to W _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 191, + 506, + 204 + ], + "score": 1.0, + "content": "induces a", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 107, + 201, + 200, + 215 + ], + "spans": [ + { + "bbox": [ + 107, + 203, + 115, + 212 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 116, + 201, + 200, + 215 + ], + "score": 1.0, + "content": "-equivariant function", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8 + }, + { + "type": "interline_equation", + "bbox": [ + 224, + 212, + 387, + 238 + ], + "lines": [ + { + "bbox": [ + 224, + 212, + 387, + 238 + ], + "spans": [ + { + "bbox": [ + 224, + 212, + 387, + 238 + ], + "score": 0.94, + "content": "\\langle p \\rangle ( X ) = \\int _ { G } \\rho _ { 2 } ( g ^ { - 1 } ) p ( \\rho _ { 1 } ( g ) X ) d \\mu ( g ) .", + "type": "interline_equation", + "image_path": "15eb986de5f78bc842430871066793bfc7ffb96f9d9b47216630c87a00f73dce.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 224, + 212, + 387, + 238 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 239, + 505, + 277 + ], + "lines": [ + { + "bbox": [ + 101, + 238, + 494, + 273 + ], + "spans": [ + { + "bbox": [ + 101, + 238, + 149, + 273 + ], + "score": 1.0, + "content": "This functformly on", + "type": "text" + }, + { + "bbox": [ + 164, + 239, + 177, + 251 + ], + "score": 0.91, + "content": "\\langle p \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 238, + 225, + 273 + ], + "score": 1.0, + "content": "is a degree “Riemann", + "type": "text" + }, + { + "bbox": [ + 225, + 240, + 235, + 249 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 238, + 300, + 273 + ], + "score": 1.0, + "content": "polynomial as ms” of the form", + "type": "text" + }, + { + "bbox": [ + 385, + 239, + 398, + 251 + ], + "score": 0.89, + "content": "\\langle p \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 238, + 494, + 273 + ], + "score": 1.0, + "content": "e approximated uwhich are degree", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 150, + 250, + 504, + 266 + ], + "spans": [ + { + "bbox": [ + 150, + 253, + 164, + 264 + ], + "score": 0.89, + "content": "K _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 250, + 421, + 266 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\sum _ { j = 1 } ^ { N } w _ { j } \\rho _ { 2 } ( g _ { j } ^ { - 1 } ) p ( \\rho _ { 1 } ( g _ { j } ) X ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 253, + 504, + 262 + ], + "score": 0.79, + "content": "D", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 264, + 390, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 240, + 279 + ], + "score": 1.0, + "content": "polynomials, and because degree", + "type": "text" + }, + { + "bbox": [ + 250, + 264, + 356, + 279 + ], + "score": 1.0, + "content": "polynomials are closed in", + "type": "text" + }, + { + "bbox": [ + 357, + 266, + 385, + 277 + ], + "score": 0.9, + "content": "C ( K _ { 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 264, + 390, + 279 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 105, + 281, + 504, + 305 + ], + "lines": [ + { + "bbox": [ + 105, + 281, + 506, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 157, + 295 + ], + "score": 1.0, + "content": "Now for all", + "type": "text" + }, + { + "bbox": [ + 157, + 282, + 195, + 293 + ], + "score": 0.89, + "content": "X \\ \\in \\ K _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 281, + 307, + 295 + ], + "score": 1.0, + "content": ", continuity of the function", + "type": "text" + }, + { + "bbox": [ + 307, + 281, + 365, + 294 + ], + "score": 0.92, + "content": "g \\mapsto \\rho _ { 2 } ( g ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 281, + 506, + 295 + ], + "score": 1.0, + "content": "implies that the operator norm of", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 292, + 364, + 307 + ], + "spans": [ + { + "bbox": [ + 106, + 293, + 141, + 305 + ], + "score": 0.92, + "content": "\\rho _ { 2 } ( g ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 292, + 302, + 307 + ], + "score": 1.0, + "content": "is bounded uniformly by some constant", + "type": "text" + }, + { + "bbox": [ + 303, + 294, + 330, + 303 + ], + "score": 0.89, + "content": "N > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 292, + 364, + 307 + ], + "score": 1.0, + "content": ", and so", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 308, + 501, + 364 + ], + "lines": [ + { + "bbox": [ + 111, + 308, + 501, + 364 + ], + "spans": [ + { + "bbox": [ + 111, + 308, + 501, + 364 + ], + "score": 0.93, + "content": "\\begin{array} { l } { \\displaystyle { | \\langle p _ { k } \\rangle ( X ) - f ( X ) | = \\left| \\int _ { G } \\rho _ { 2 } ( g ^ { - 1 } ) p _ { k } ( \\rho _ { 1 } ( g ) X ) - \\rho _ { 2 } ( g ^ { - 1 } ) f ( \\rho _ { 1 } ( g ) X ) d \\mu ( g ) \\right| } } \\\\ { \\displaystyle { \\phantom { \\rho _ { k } ( \\rho _ { 1 } ) ( } = \\left| \\int _ { G } \\rho _ { 2 } ( g ^ { - 1 } ) \\left[ p _ { k } ( \\rho _ { 1 } ( g ) X ) - f ( \\rho _ { 1 } ( g ) X ) \\right] d \\mu ( g ) \\right| \\leq N \\| f - p _ { k } \\| _ { C ( K _ { 1 } ) } \\to 0 } } \\end{array}", + "type": "interline_equation", + "image_path": "667bbdb38be9394d312e35d3e43bdc61e9e0b663f47435897f5af36d3106aead.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 111, + 308, + 501, + 326.6666666666667 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 111, + 326.6666666666667, + 501, + 345.33333333333337 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 111, + 345.33333333333337, + 501, + 364.00000000000006 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "title", + "bbox": [ + 107, + 389, + 229, + 401 + ], + "lines": [ + { + "bbox": [ + 106, + 389, + 230, + 402 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 230, + 402 + ], + "score": 1.0, + "content": "B.2 PROOF OF THEOREM 1", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 410, + 505, + 444 + ], + "lines": [ + { + "bbox": [ + 105, + 408, + 505, + 424 + ], + "spans": [ + { + "bbox": [ + 105, + 408, + 169, + 424 + ], + "score": 1.0, + "content": "Theorem 1. If", + "type": "text" + }, + { + "bbox": [ + 169, + 411, + 190, + 421 + ], + "score": 0.88, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 408, + 201, + 424 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 201, + 411, + 210, + 420 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 408, + 269, + 424 + ], + "score": 1.0, + "content": "-spanning and", + "type": "text" + }, + { + "bbox": [ + 270, + 411, + 293, + 423 + ], + "score": 0.91, + "content": "\\mathcal { F } _ { \\mathrm { p o o l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 408, + 469, + 424 + ], + "score": 1.0, + "content": "is linearly universal, then there exists some", + "type": "text" + }, + { + "bbox": [ + 469, + 411, + 505, + 423 + ], + "score": 0.9, + "content": "C ( D ) \\in", + "type": "inline_equation" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 420, + 506, + 435 + ], + "spans": [ + { + "bbox": [ + 106, + 422, + 116, + 432 + ], + "score": 0.54, + "content": "\\mathbb { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 116, + 420, + 188, + 435 + ], + "score": 1.0, + "content": "such that for all", + "type": "text" + }, + { + "bbox": [ + 189, + 422, + 239, + 434 + ], + "score": 0.92, + "content": "C \\geq C ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 420, + 320, + 435 + ], + "score": 1.0, + "content": "the function space", + "type": "text" + }, + { + "bbox": [ + 321, + 421, + 390, + 434 + ], + "score": 0.9, + "content": "\\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } , \\mathcal { F } _ { \\mathrm { p o o l } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 420, + 446, + 435 + ], + "score": 1.0, + "content": "contains all", + "type": "text" + }, + { + "bbox": [ + 446, + 422, + 455, + 432 + ], + "score": 0.77, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 420, + 506, + 435 + ], + "score": 1.0, + "content": "-equivariant", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 432, + 220, + 445 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 196, + 445 + ], + "score": 1.0, + "content": "polynomials of degree", + "type": "text" + }, + { + "bbox": [ + 197, + 433, + 217, + 443 + ], + "score": 0.83, + "content": "\\le D", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 432, + 220, + 445 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 456, + 504, + 479 + ], + "lines": [ + { + "bbox": [ + 106, + 456, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 106, + 456, + 165, + 469 + ], + "score": 1.0, + "content": "Proof. By the", + "type": "text" + }, + { + "bbox": [ + 166, + 457, + 176, + 466 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 456, + 312, + 469 + ], + "score": 1.0, + "content": "-spanning assumption, there exist", + "type": "text" + }, + { + "bbox": [ + 312, + 456, + 391, + 468 + ], + "score": 0.92, + "content": "f _ { 1 } , \\dots , f _ { K } \\in \\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 456, + 506, + 469 + ], + "score": 1.0, + "content": "such that any vector valued", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 103, + 464, + 504, + 482 + ], + "spans": [ + { + "bbox": [ + 103, + 464, + 154, + 482 + ], + "score": 1.0, + "content": "polynomial", + "type": "text" + }, + { + "bbox": [ + 155, + 467, + 219, + 479 + ], + "score": 0.9, + "content": "p : \\mathbb { R } ^ { 3 \\bar { \\times } n } \\bar { \\mathbb { R } } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 464, + 504, + 482 + ], + "score": 1.0, + "content": "invariant to translations and equivariant to permutations is of the form", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "interline_equation", + "bbox": [ + 254, + 482, + 356, + 516 + ], + "lines": [ + { + "bbox": [ + 254, + 482, + 356, + 516 + ], + "spans": [ + { + "bbox": [ + 254, + 482, + 356, + 516 + ], + "score": 0.95, + "content": "p ( X ) = \\sum _ { k = 1 } ^ { K } \\hat { \\Lambda } _ { k } ( f _ { k } ( X ) ) ,", + "type": "interline_equation", + "image_path": "2463d691fa6fda41539a973d733e2689860ca5ec0c6ddd0d49de84349dd24050.jpg" + } + ] + } + ], + "index": 25.5, + "virtual_lines": [ + { + "bbox": [ + 254, + 482, + 356, + 499.0 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 254, + 499.0, + 356, + 516.0 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 519, + 505, + 575 + ], + "lines": [ + { + "bbox": [ + 105, + 517, + 507, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 517, + 136, + 533 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 136, + 520, + 149, + 531 + ], + "score": 0.87, + "content": "\\Lambda _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 517, + 252, + 533 + ], + "score": 1.0, + "content": "are linear functions to", + "type": "text" + }, + { + "bbox": [ + 252, + 520, + 260, + 530 + ], + "score": 0.82, + "content": "\\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 517, + 283, + 533 + ], + "score": 1.0, + "content": ". If", + "type": "text" + }, + { + "bbox": [ + 284, + 522, + 290, + 532 + ], + "score": 0.82, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 517, + 466, + 533 + ], + "score": 1.0, + "content": "is a matrix valued polynomial mapping", + "type": "text" + }, + { + "bbox": [ + 467, + 519, + 491, + 530 + ], + "score": 0.9, + "content": "\\mathbb { R } ^ { 3 \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 491, + 517, + 507, + 533 + ], + "score": 1.0, + "content": "to", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 107, + 528, + 507, + 544 + ], + "spans": [ + { + "bbox": [ + 107, + 531, + 162, + 543 + ], + "score": 0.88, + "content": "W _ { T } ^ { n } = \\mathbb { R } ^ { t \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 528, + 507, + 544 + ], + "score": 1.0, + "content": ", which is invariant to translations and equivariant to permutations, then it is of the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 540, + 507, + 557 + ], + "spans": [ + { + "bbox": [ + 104, + 540, + 128, + 557 + ], + "score": 1.0, + "content": "form", + "type": "text" + }, + { + "bbox": [ + 129, + 542, + 205, + 555 + ], + "score": 0.93, + "content": "p = ( p _ { i j } ) _ { i \\in [ t ] , j \\in [ n ] }", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 540, + 246, + 557 + ], + "score": 1.0, + "content": ", and each", + "type": "text" + }, + { + "bbox": [ + 247, + 542, + 308, + 555 + ], + "score": 0.93, + "content": "p _ { i } = ( p _ { i j } ) _ { j \\in [ n ] }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 540, + 507, + 557 + ], + "score": 1.0, + "content": "is itself invariant to translations and permutation", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 551, + 505, + 566 + ], + "spans": [ + { + "bbox": [ + 104, + 551, + 278, + 566 + ], + "score": 1.0, + "content": "equivariant. It follows that matrix valued", + "type": "text" + }, + { + "bbox": [ + 278, + 555, + 284, + 564 + ], + "score": 0.8, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 551, + 505, + 566 + ], + "score": 1.0, + "content": "can also be written in the form equation 17, the only", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 563, + 381, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 323, + 576 + ], + "score": 1.0, + "content": "difference being that the image of the linear functions", + "type": "text" + }, + { + "bbox": [ + 323, + 564, + 336, + 574 + ], + "score": 0.89, + "content": "\\Lambda _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 563, + 366, + 576 + ], + "score": 1.0, + "content": "is now", + "type": "text" + }, + { + "bbox": [ + 366, + 563, + 378, + 573 + ], + "score": 0.88, + "content": "\\mathbb { R } ^ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 563, + 381, + 576 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 106, + 579, + 504, + 642 + ], + "lines": [ + { + "bbox": [ + 104, + 578, + 506, + 594 + ], + "spans": [ + { + "bbox": [ + 104, + 578, + 141, + 594 + ], + "score": 1.0, + "content": "Now let", + "type": "text" + }, + { + "bbox": [ + 142, + 579, + 213, + 592 + ], + "score": 0.9, + "content": "p : \\mathbb { R } ^ { 3 \\times n } W _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 578, + 235, + 594 + ], + "score": 1.0, + "content": "be a", + "type": "text" + }, + { + "bbox": [ + 236, + 581, + 245, + 591 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 578, + 382, + 594 + ], + "score": 1.0, + "content": "-equivariant polynomial of degree", + "type": "text" + }, + { + "bbox": [ + 383, + 581, + 403, + 591 + ], + "score": 0.87, + "content": "\\le D", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 578, + 506, + 594 + ], + "score": 1.0, + "content": ". It remains to show that", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 591, + 506, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 591, + 168, + 604 + ], + "score": 1.0, + "content": "we can choose", + "type": "text" + }, + { + "bbox": [ + 168, + 592, + 181, + 603 + ], + "score": 0.87, + "content": "\\Lambda _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 591, + 506, + 604 + ], + "score": 1.0, + "content": "to be SO(3) equivariant. We do this by a symmetrization argument: denote the", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 602, + 504, + 615 + ], + "spans": [ + { + "bbox": [ + 106, + 602, + 224, + 615 + ], + "score": 1.0, + "content": "Haar probability measure on", + "type": "text" + }, + { + "bbox": [ + 224, + 603, + 251, + 614 + ], + "score": 0.31, + "content": "\\mathrm { S O ( 3 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 602, + 266, + 615 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 266, + 604, + 272, + 612 + ], + "score": 0.75, + "content": "\\nu", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 602, + 388, + 615 + ], + "score": 1.0, + "content": ", and the action of SO(3) on", + "type": "text" + }, + { + "bbox": [ + 389, + 603, + 413, + 614 + ], + "score": 0.88, + "content": "W _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 602, + 432, + 615 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 432, + 603, + 449, + 614 + ], + "score": 0.89, + "content": "W _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 602, + 463, + 615 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 463, + 604, + 474, + 614 + ], + "score": 0.84, + "content": "\\rho _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 602, + 492, + 615 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 493, + 604, + 504, + 614 + ], + "score": 0.82, + "content": "\\rho _ { 2 }", + "type": "inline_equation" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 613, + 505, + 629 + ], + "spans": [ + { + "bbox": [ + 104, + 614, + 190, + 629 + ], + "score": 1.0, + "content": "respectively Denote", + "type": "text" + }, + { + "bbox": [ + 190, + 615, + 243, + 629 + ], + "score": 0.93, + "content": "p = ( p _ { j } ) _ { j = 1 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 614, + 263, + 629 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 263, + 613, + 321, + 629 + ], + "score": 0.94, + "content": "f _ { k } = ( f _ { k } ^ { j } ) _ { j = 1 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 614, + 370, + 629 + ], + "score": 1.0, + "content": ". For every", + "type": "text" + }, + { + "bbox": [ + 370, + 615, + 425, + 627 + ], + "score": 0.91, + "content": "j = 1 , \\dotsc , n", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 614, + 505, + 629 + ], + "score": 1.0, + "content": ", we use the SO(3)", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 628, + 249, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 629, + 170, + 642 + ], + "score": 1.0, + "content": "equivariance of", + "type": "text" + }, + { + "bbox": [ + 170, + 631, + 180, + 642 + ], + "score": 0.86, + "content": "p _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 629, + 199, + 642 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 199, + 628, + 210, + 642 + ], + "score": 0.9, + "content": "f _ { k } ^ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 629, + 249, + 642 + ], + "score": 1.0, + "content": "to obtain", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34 + }, + { + "type": "interline_equation", + "bbox": [ + 124, + 645, + 487, + 715 + ], + "lines": [ + { + "bbox": [ + 124, + 645, + 487, + 715 + ], + "spans": [ + { + "bbox": [ + 124, + 645, + 487, + 715 + ], + "score": 0.94, + "content": "\\begin{array} { l } { { \\displaystyle p _ { j } ( X ) = \\int _ { \\mathrm { S O } ( 3 ) } \\rho _ { 2 } ( R ^ { - 1 } ) \\circ p _ { j } ( R X ) d \\nu ( R ) = \\sum _ { k = 1 } ^ { K } \\int _ { \\mathrm { S O } ( 3 ) } \\rho _ { 2 } ( R ^ { - 1 } ) \\circ \\Lambda _ { k } \\circ f _ { j } ^ { k } ( R X ) d \\nu ( R ) } } \\\\ { { \\displaystyle \\qquad = \\sum _ { k = 1 } ^ { K } \\int _ { \\mathrm { S O } ( 3 ) } \\rho _ { 2 } ( R ^ { - 1 } ) \\circ \\Lambda _ { k } \\left( \\rho _ { 1 } ( R ) \\circ f _ { k } ^ { j } ( X ) \\right) d \\nu ( R ) = \\sum _ { k = 1 } ^ { K } \\tilde { \\Lambda } _ { k } \\circ f _ { k } ^ { j } ( X ) , } } \\end{array}", + "type": "interline_equation", + "image_path": "4bc18a56fd3a2e5947357c447088fa0a0015cc7008f44aeacfb7112cf513128c.jpg" + } + ] + } + ], + "index": 38, + "virtual_lines": [ + { + "bbox": [ + 124, + 645, + 487, + 668.3333333333334 + ], + "spans": [], + "index": 37 + }, + { + "bbox": [ + 124, + 668.3333333333334, + 487, + 691.6666666666667 + ], + "spans": [], + "index": 38 + }, + { + "bbox": [ + 124, + 691.6666666666667, + 487, + 715.0000000000001 + ], + "spans": [], + "index": 39 + } + ] + }, + { + "type": "text", + "bbox": [ + 117, + 720, + 419, + 733 + ], + "lines": [ + { + "bbox": [ + 118, + 719, + 419, + 734 + ], + "spans": [ + { + "bbox": [ + 118, + 719, + 235, + 734 + ], + "score": 1.0, + "content": "2By this we mean that the maps", + "type": "text" + }, + { + "bbox": [ + 235, + 721, + 337, + 732 + ], + "score": 0.93, + "content": "( g , X ) \\mapsto \\rho _ { j } ( g ) X , j = 1 , 2", + "type": "inline_equation" + }, + { + "bbox": [ + 337, + 719, + 419, + 734 + ], + "score": 1.0, + "content": "are jointly continuous", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "13", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 366, + 505, + 376 + ], + "lines": [ + { + "bbox": [ + 495, + 367, + 505, + 377 + ], + "spans": [ + { + "bbox": [ + 495, + 367, + 505, + 377 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 505, + 127 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 179, + 96 + ], + "score": 1.0, + "content": "Lemma B.2. Let", + "type": "text" + }, + { + "bbox": [ + 180, + 83, + 188, + 93 + ], + "score": 0.73, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 82, + 290, + 96 + ], + "score": 1.0, + "content": "be a compact group, Let", + "type": "text" + }, + { + "bbox": [ + 290, + 84, + 301, + 94 + ], + "score": 0.85, + "content": "\\rho _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 301, + 82, + 320, + 96 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 320, + 84, + 331, + 94 + ], + "score": 0.83, + "content": "\\rho _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 82, + 467, + 96 + ], + "score": 1.0, + "content": "be continuous2representations of", + "type": "text" + }, + { + "bbox": [ + 467, + 83, + 477, + 93 + ], + "score": 0.48, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "on the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 505, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 180, + 107 + ], + "score": 1.0, + "content": "Euclidean spaces", + "type": "text" + }, + { + "bbox": [ + 180, + 94, + 195, + 105 + ], + "score": 0.88, + "content": "W _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 93, + 215, + 107 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 215, + 94, + 230, + 105 + ], + "score": 0.91, + "content": "W _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 93, + 253, + 107 + ], + "score": 1.0, + "content": ". Let", + "type": "text" + }, + { + "bbox": [ + 253, + 94, + 293, + 105 + ], + "score": 0.91, + "content": "K \\subseteq W _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 93, + 505, + 107 + ], + "score": 1.0, + "content": "be a compact set. Then every equivariant function", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 107, + 104, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 107, + 105, + 172, + 116 + ], + "score": 0.9, + "content": "f : W _ { 1 } \\to W _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 104, + 319, + 117 + ], + "score": 1.0, + "content": "can be approximated uniformly on", + "type": "text" + }, + { + "bbox": [ + 320, + 105, + 330, + 114 + ], + "score": 0.8, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 104, + 505, + 117 + ], + "score": 1.0, + "content": "by a sequence of equivariant polynomials", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 115, + 173, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 116, + 169, + 127 + ], + "score": 0.89, + "content": "p _ { k } : W _ { 1 } \\mapsto W _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 115, + 173, + 128 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5, + "bbox_fs": [ + 105, + 82, + 506, + 128 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 138, + 505, + 161 + ], + "lines": [ + { + "bbox": [ + 105, + 138, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 138, + 123, + 151 + ], + "score": 1.0, + "content": "Let", + "type": "text" + }, + { + "bbox": [ + 123, + 140, + 131, + 150 + ], + "score": 0.81, + "content": "\\mu", + "type": "inline_equation" + }, + { + "bbox": [ + 131, + 138, + 411, + 151 + ], + "score": 1.0, + "content": "be the Haar probability measure associated with the compact group", + "type": "text" + }, + { + "bbox": [ + 412, + 139, + 421, + 148 + ], + "score": 0.78, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 138, + 444, + 151 + ], + "score": 1.0, + "content": ". Let", + "type": "text" + }, + { + "bbox": [ + 444, + 139, + 458, + 149 + ], + "score": 0.89, + "content": "K _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 138, + 505, + 151 + ], + "score": 1.0, + "content": "denote the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 147, + 473, + 163 + ], + "spans": [ + { + "bbox": [ + 105, + 147, + 318, + 163 + ], + "score": 1.0, + "content": "compact set obtained as an image of the compact set", + "type": "text" + }, + { + "bbox": [ + 318, + 150, + 348, + 159 + ], + "score": 0.92, + "content": "G \\times K", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 147, + 473, + 163 + ], + "score": 1.0, + "content": "under the continuous mapping", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 105, + 138, + 505, + 163 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 267, + 164, + 344, + 177 + ], + "lines": [ + { + "bbox": [ + 267, + 164, + 344, + 177 + ], + "spans": [ + { + "bbox": [ + 267, + 164, + 344, + 177 + ], + "score": 0.9, + "content": "( g , X ) \\mapsto \\rho _ { 1 } ( g ) X .", + "type": "interline_equation", + "image_path": "5f4461c64095fc77cdd890372dae6fd44c4eb438541773c5e7fd11abefadc925.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 267, + 164, + 344, + 177 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 180, + 504, + 213 + ], + "lines": [ + { + "bbox": [ + 105, + 179, + 504, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 179, + 272, + 194 + ], + "score": 1.0, + "content": "Using the Stone-Weierstrass theorem, let", + "type": "text" + }, + { + "bbox": [ + 273, + 182, + 284, + 192 + ], + "score": 0.86, + "content": "p _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 179, + 504, + 194 + ], + "score": 1.0, + "content": "be a sequence of (not necessarily equivariant) polyno-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 191, + 506, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 209, + 204 + ], + "score": 1.0, + "content": "mials which approximate", + "type": "text" + }, + { + "bbox": [ + 210, + 192, + 217, + 203 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 191, + 273, + 204 + ], + "score": 1.0, + "content": "uniformly on", + "type": "text" + }, + { + "bbox": [ + 273, + 192, + 287, + 203 + ], + "score": 0.9, + "content": "K _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 191, + 347, + 204 + ], + "score": 1.0, + "content": ". Every degree", + "type": "text" + }, + { + "bbox": [ + 347, + 192, + 357, + 201 + ], + "score": 0.84, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 191, + 406, + 204 + ], + "score": 1.0, + "content": "polynomial", + "type": "text" + }, + { + "bbox": [ + 406, + 191, + 464, + 203 + ], + "score": 0.93, + "content": "p : W _ { 1 } \\to W _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 191, + 506, + 204 + ], + "score": 1.0, + "content": "induces a", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 107, + 201, + 200, + 215 + ], + "spans": [ + { + "bbox": [ + 107, + 203, + 115, + 212 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 116, + 201, + 200, + 215 + ], + "score": 1.0, + "content": "-equivariant function", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8, + "bbox_fs": [ + 105, + 179, + 506, + 215 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 224, + 212, + 387, + 238 + ], + "lines": [ + { + "bbox": [ + 224, + 212, + 387, + 238 + ], + "spans": [ + { + "bbox": [ + 224, + 212, + 387, + 238 + ], + "score": 0.94, + "content": "\\langle p \\rangle ( X ) = \\int _ { G } \\rho _ { 2 } ( g ^ { - 1 } ) p ( \\rho _ { 1 } ( g ) X ) d \\mu ( g ) .", + "type": "interline_equation", + "image_path": "15eb986de5f78bc842430871066793bfc7ffb96f9d9b47216630c87a00f73dce.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 224, + 212, + 387, + 238 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 239, + 505, + 277 + ], + "lines": [ + { + "bbox": [ + 101, + 238, + 494, + 273 + ], + "spans": [ + { + "bbox": [ + 101, + 238, + 149, + 273 + ], + "score": 1.0, + "content": "This functformly on", + "type": "text" + }, + { + "bbox": [ + 164, + 239, + 177, + 251 + ], + "score": 0.91, + "content": "\\langle p \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 238, + 225, + 273 + ], + "score": 1.0, + "content": "is a degree “Riemann", + "type": "text" + }, + { + "bbox": [ + 225, + 240, + 235, + 249 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 238, + 300, + 273 + ], + "score": 1.0, + "content": "polynomial as ms” of the form", + "type": "text" + }, + { + "bbox": [ + 385, + 239, + 398, + 251 + ], + "score": 0.89, + "content": "\\langle p \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 238, + 494, + 273 + ], + "score": 1.0, + "content": "e approximated uwhich are degree", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 150, + 250, + 504, + 266 + ], + "spans": [ + { + "bbox": [ + 150, + 253, + 164, + 264 + ], + "score": 0.89, + "content": "K _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 250, + 421, + 266 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\sum _ { j = 1 } ^ { N } w _ { j } \\rho _ { 2 } ( g _ { j } ^ { - 1 } ) p ( \\rho _ { 1 } ( g _ { j } ) X ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 253, + 504, + 262 + ], + "score": 0.79, + "content": "D", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 264, + 390, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 240, + 279 + ], + "score": 1.0, + "content": "polynomials, and because degree", + "type": "text" + }, + { + "bbox": [ + 250, + 264, + 356, + 279 + ], + "score": 1.0, + "content": "polynomials are closed in", + "type": "text" + }, + { + "bbox": [ + 357, + 266, + 385, + 277 + ], + "score": 0.9, + "content": "C ( K _ { 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 264, + 390, + 279 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12, + "bbox_fs": [ + 101, + 238, + 504, + 279 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 281, + 504, + 305 + ], + "lines": [ + { + "bbox": [ + 105, + 281, + 506, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 157, + 295 + ], + "score": 1.0, + "content": "Now for all", + "type": "text" + }, + { + "bbox": [ + 157, + 282, + 195, + 293 + ], + "score": 0.89, + "content": "X \\ \\in \\ K _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 281, + 307, + 295 + ], + "score": 1.0, + "content": ", continuity of the function", + "type": "text" + }, + { + "bbox": [ + 307, + 281, + 365, + 294 + ], + "score": 0.92, + "content": "g \\mapsto \\rho _ { 2 } ( g ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 281, + 506, + 295 + ], + "score": 1.0, + "content": "implies that the operator norm of", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 292, + 364, + 307 + ], + "spans": [ + { + "bbox": [ + 106, + 293, + 141, + 305 + ], + "score": 0.92, + "content": "\\rho _ { 2 } ( g ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 292, + 302, + 307 + ], + "score": 1.0, + "content": "is bounded uniformly by some constant", + "type": "text" + }, + { + "bbox": [ + 303, + 294, + 330, + 303 + ], + "score": 0.89, + "content": "N > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 292, + 364, + 307 + ], + "score": 1.0, + "content": ", and so", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 281, + 506, + 307 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 308, + 501, + 364 + ], + "lines": [ + { + "bbox": [ + 111, + 308, + 501, + 364 + ], + "spans": [ + { + "bbox": [ + 111, + 308, + 501, + 364 + ], + "score": 0.93, + "content": "\\begin{array} { l } { \\displaystyle { | \\langle p _ { k } \\rangle ( X ) - f ( X ) | = \\left| \\int _ { G } \\rho _ { 2 } ( g ^ { - 1 } ) p _ { k } ( \\rho _ { 1 } ( g ) X ) - \\rho _ { 2 } ( g ^ { - 1 } ) f ( \\rho _ { 1 } ( g ) X ) d \\mu ( g ) \\right| } } \\\\ { \\displaystyle { \\phantom { \\rho _ { k } ( \\rho _ { 1 } ) ( } = \\left| \\int _ { G } \\rho _ { 2 } ( g ^ { - 1 } ) \\left[ p _ { k } ( \\rho _ { 1 } ( g ) X ) - f ( \\rho _ { 1 } ( g ) X ) \\right] d \\mu ( g ) \\right| \\leq N \\| f - p _ { k } \\| _ { C ( K _ { 1 } ) } \\to 0 } } \\end{array}", + "type": "interline_equation", + "image_path": "667bbdb38be9394d312e35d3e43bdc61e9e0b663f47435897f5af36d3106aead.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 111, + 308, + 501, + 326.6666666666667 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 111, + 326.6666666666667, + 501, + 345.33333333333337 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 111, + 345.33333333333337, + 501, + 364.00000000000006 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "title", + "bbox": [ + 107, + 389, + 229, + 401 + ], + "lines": [ + { + "bbox": [ + 106, + 389, + 230, + 402 + ], + "spans": [ + { + "bbox": [ + 106, + 389, + 230, + 402 + ], + "score": 1.0, + "content": "B.2 PROOF OF THEOREM 1", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 410, + 505, + 444 + ], + "lines": [ + { + "bbox": [ + 105, + 408, + 505, + 424 + ], + "spans": [ + { + "bbox": [ + 105, + 408, + 169, + 424 + ], + "score": 1.0, + "content": "Theorem 1. If", + "type": "text" + }, + { + "bbox": [ + 169, + 411, + 190, + 421 + ], + "score": 0.88, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 408, + 201, + 424 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 201, + 411, + 210, + 420 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 408, + 269, + 424 + ], + "score": 1.0, + "content": "-spanning and", + "type": "text" + }, + { + "bbox": [ + 270, + 411, + 293, + 423 + ], + "score": 0.91, + "content": "\\mathcal { F } _ { \\mathrm { p o o l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 408, + 469, + 424 + ], + "score": 1.0, + "content": "is linearly universal, then there exists some", + "type": "text" + }, + { + "bbox": [ + 469, + 411, + 505, + 423 + ], + "score": 0.9, + "content": "C ( D ) \\in", + "type": "inline_equation" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 420, + 506, + 435 + ], + "spans": [ + { + "bbox": [ + 106, + 422, + 116, + 432 + ], + "score": 0.54, + "content": "\\mathbb { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 116, + 420, + 188, + 435 + ], + "score": 1.0, + "content": "such that for all", + "type": "text" + }, + { + "bbox": [ + 189, + 422, + 239, + 434 + ], + "score": 0.92, + "content": "C \\geq C ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 420, + 320, + 435 + ], + "score": 1.0, + "content": "the function space", + "type": "text" + }, + { + "bbox": [ + 321, + 421, + 390, + 434 + ], + "score": 0.9, + "content": "\\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } , \\mathcal { F } _ { \\mathrm { p o o l } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 420, + 446, + 435 + ], + "score": 1.0, + "content": "contains all", + "type": "text" + }, + { + "bbox": [ + 446, + 422, + 455, + 432 + ], + "score": 0.77, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 420, + 506, + 435 + ], + "score": 1.0, + "content": "-equivariant", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 432, + 220, + 445 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 196, + 445 + ], + "score": 1.0, + "content": "polynomials of degree", + "type": "text" + }, + { + "bbox": [ + 197, + 433, + 217, + 443 + ], + "score": 0.83, + "content": "\\le D", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 432, + 220, + 445 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21, + "bbox_fs": [ + 105, + 408, + 506, + 445 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 456, + 504, + 479 + ], + "lines": [ + { + "bbox": [ + 106, + 456, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 106, + 456, + 165, + 469 + ], + "score": 1.0, + "content": "Proof. By the", + "type": "text" + }, + { + "bbox": [ + 166, + 457, + 176, + 466 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 456, + 312, + 469 + ], + "score": 1.0, + "content": "-spanning assumption, there exist", + "type": "text" + }, + { + "bbox": [ + 312, + 456, + 391, + 468 + ], + "score": 0.92, + "content": "f _ { 1 } , \\dots , f _ { K } \\in \\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 456, + 506, + 469 + ], + "score": 1.0, + "content": "such that any vector valued", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 103, + 464, + 504, + 482 + ], + "spans": [ + { + "bbox": [ + 103, + 464, + 154, + 482 + ], + "score": 1.0, + "content": "polynomial", + "type": "text" + }, + { + "bbox": [ + 155, + 467, + 219, + 479 + ], + "score": 0.9, + "content": "p : \\mathbb { R } ^ { 3 \\bar { \\times } n } \\bar { \\mathbb { R } } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 464, + 504, + 482 + ], + "score": 1.0, + "content": "invariant to translations and equivariant to permutations is of the form", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 103, + 456, + 506, + 482 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 254, + 482, + 356, + 516 + ], + "lines": [ + { + "bbox": [ + 254, + 482, + 356, + 516 + ], + "spans": [ + { + "bbox": [ + 254, + 482, + 356, + 516 + ], + "score": 0.95, + "content": "p ( X ) = \\sum _ { k = 1 } ^ { K } \\hat { \\Lambda } _ { k } ( f _ { k } ( X ) ) ,", + "type": "interline_equation", + "image_path": "2463d691fa6fda41539a973d733e2689860ca5ec0c6ddd0d49de84349dd24050.jpg" + } + ] + } + ], + "index": 25.5, + "virtual_lines": [ + { + "bbox": [ + 254, + 482, + 356, + 499.0 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 254, + 499.0, + 356, + 516.0 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 519, + 505, + 575 + ], + "lines": [ + { + "bbox": [ + 105, + 517, + 507, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 517, + 136, + 533 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 136, + 520, + 149, + 531 + ], + "score": 0.87, + "content": "\\Lambda _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 517, + 252, + 533 + ], + "score": 1.0, + "content": "are linear functions to", + "type": "text" + }, + { + "bbox": [ + 252, + 520, + 260, + 530 + ], + "score": 0.82, + "content": "\\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 517, + 283, + 533 + ], + "score": 1.0, + "content": ". If", + "type": "text" + }, + { + "bbox": [ + 284, + 522, + 290, + 532 + ], + "score": 0.82, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 517, + 466, + 533 + ], + "score": 1.0, + "content": "is a matrix valued polynomial mapping", + "type": "text" + }, + { + "bbox": [ + 467, + 519, + 491, + 530 + ], + "score": 0.9, + "content": "\\mathbb { R } ^ { 3 \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 491, + 517, + 507, + 533 + ], + "score": 1.0, + "content": "to", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 107, + 528, + 507, + 544 + ], + "spans": [ + { + "bbox": [ + 107, + 531, + 162, + 543 + ], + "score": 0.88, + "content": "W _ { T } ^ { n } = \\mathbb { R } ^ { t \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 528, + 507, + 544 + ], + "score": 1.0, + "content": ", which is invariant to translations and equivariant to permutations, then it is of the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 540, + 507, + 557 + ], + "spans": [ + { + "bbox": [ + 104, + 540, + 128, + 557 + ], + "score": 1.0, + "content": "form", + "type": "text" + }, + { + "bbox": [ + 129, + 542, + 205, + 555 + ], + "score": 0.93, + "content": "p = ( p _ { i j } ) _ { i \\in [ t ] , j \\in [ n ] }", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 540, + 246, + 557 + ], + "score": 1.0, + "content": ", and each", + "type": "text" + }, + { + "bbox": [ + 247, + 542, + 308, + 555 + ], + "score": 0.93, + "content": "p _ { i } = ( p _ { i j } ) _ { j \\in [ n ] }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 540, + 507, + 557 + ], + "score": 1.0, + "content": "is itself invariant to translations and permutation", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 104, + 551, + 505, + 566 + ], + "spans": [ + { + "bbox": [ + 104, + 551, + 278, + 566 + ], + "score": 1.0, + "content": "equivariant. It follows that matrix valued", + "type": "text" + }, + { + "bbox": [ + 278, + 555, + 284, + 564 + ], + "score": 0.8, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 551, + 505, + 566 + ], + "score": 1.0, + "content": "can also be written in the form equation 17, the only", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 563, + 381, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 323, + 576 + ], + "score": 1.0, + "content": "difference being that the image of the linear functions", + "type": "text" + }, + { + "bbox": [ + 323, + 564, + 336, + 574 + ], + "score": 0.89, + "content": "\\Lambda _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 563, + 366, + 576 + ], + "score": 1.0, + "content": "is now", + "type": "text" + }, + { + "bbox": [ + 366, + 563, + 378, + 573 + ], + "score": 0.88, + "content": "\\mathbb { R } ^ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 563, + 381, + 576 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29, + "bbox_fs": [ + 104, + 517, + 507, + 576 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 579, + 504, + 642 + ], + "lines": [ + { + "bbox": [ + 104, + 578, + 506, + 594 + ], + "spans": [ + { + "bbox": [ + 104, + 578, + 141, + 594 + ], + "score": 1.0, + "content": "Now let", + "type": "text" + }, + { + "bbox": [ + 142, + 579, + 213, + 592 + ], + "score": 0.9, + "content": "p : \\mathbb { R } ^ { 3 \\times n } W _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 578, + 235, + 594 + ], + "score": 1.0, + "content": "be a", + "type": "text" + }, + { + "bbox": [ + 236, + 581, + 245, + 591 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 578, + 382, + 594 + ], + "score": 1.0, + "content": "-equivariant polynomial of degree", + "type": "text" + }, + { + "bbox": [ + 383, + 581, + 403, + 591 + ], + "score": 0.87, + "content": "\\le D", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 578, + 506, + 594 + ], + "score": 1.0, + "content": ". It remains to show that", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 591, + 506, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 591, + 168, + 604 + ], + "score": 1.0, + "content": "we can choose", + "type": "text" + }, + { + "bbox": [ + 168, + 592, + 181, + 603 + ], + "score": 0.87, + "content": "\\Lambda _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 591, + 506, + 604 + ], + "score": 1.0, + "content": "to be SO(3) equivariant. We do this by a symmetrization argument: denote the", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 602, + 504, + 615 + ], + "spans": [ + { + "bbox": [ + 106, + 602, + 224, + 615 + ], + "score": 1.0, + "content": "Haar probability measure on", + "type": "text" + }, + { + "bbox": [ + 224, + 603, + 251, + 614 + ], + "score": 0.31, + "content": "\\mathrm { S O ( 3 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 602, + 266, + 615 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 266, + 604, + 272, + 612 + ], + "score": 0.75, + "content": "\\nu", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 602, + 388, + 615 + ], + "score": 1.0, + "content": ", and the action of SO(3) on", + "type": "text" + }, + { + "bbox": [ + 389, + 603, + 413, + 614 + ], + "score": 0.88, + "content": "W _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 602, + 432, + 615 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 432, + 603, + 449, + 614 + ], + "score": 0.89, + "content": "W _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 602, + 463, + 615 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 463, + 604, + 474, + 614 + ], + "score": 0.84, + "content": "\\rho _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 602, + 492, + 615 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 493, + 604, + 504, + 614 + ], + "score": 0.82, + "content": "\\rho _ { 2 }", + "type": "inline_equation" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 613, + 505, + 629 + ], + "spans": [ + { + "bbox": [ + 104, + 614, + 190, + 629 + ], + "score": 1.0, + "content": "respectively Denote", + "type": "text" + }, + { + "bbox": [ + 190, + 615, + 243, + 629 + ], + "score": 0.93, + "content": "p = ( p _ { j } ) _ { j = 1 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 614, + 263, + 629 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 263, + 613, + 321, + 629 + ], + "score": 0.94, + "content": "f _ { k } = ( f _ { k } ^ { j } ) _ { j = 1 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 322, + 614, + 370, + 629 + ], + "score": 1.0, + "content": ". For every", + "type": "text" + }, + { + "bbox": [ + 370, + 615, + 425, + 627 + ], + "score": 0.91, + "content": "j = 1 , \\dotsc , n", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 614, + 505, + 629 + ], + "score": 1.0, + "content": ", we use the SO(3)", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 628, + 249, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 629, + 170, + 642 + ], + "score": 1.0, + "content": "equivariance of", + "type": "text" + }, + { + "bbox": [ + 170, + 631, + 180, + 642 + ], + "score": 0.86, + "content": "p _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 629, + 199, + 642 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 199, + 628, + 210, + 642 + ], + "score": 0.9, + "content": "f _ { k } ^ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 629, + 249, + 642 + ], + "score": 1.0, + "content": "to obtain", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34, + "bbox_fs": [ + 104, + 578, + 506, + 642 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 124, + 645, + 487, + 715 + ], + "lines": [ + { + "bbox": [ + 124, + 645, + 487, + 715 + ], + "spans": [ + { + "bbox": [ + 124, + 645, + 487, + 715 + ], + "score": 0.94, + "content": "\\begin{array} { l } { { \\displaystyle p _ { j } ( X ) = \\int _ { \\mathrm { S O } ( 3 ) } \\rho _ { 2 } ( R ^ { - 1 } ) \\circ p _ { j } ( R X ) d \\nu ( R ) = \\sum _ { k = 1 } ^ { K } \\int _ { \\mathrm { S O } ( 3 ) } \\rho _ { 2 } ( R ^ { - 1 } ) \\circ \\Lambda _ { k } \\circ f _ { j } ^ { k } ( R X ) d \\nu ( R ) } } \\\\ { { \\displaystyle \\qquad = \\sum _ { k = 1 } ^ { K } \\int _ { \\mathrm { S O } ( 3 ) } \\rho _ { 2 } ( R ^ { - 1 } ) \\circ \\Lambda _ { k } \\left( \\rho _ { 1 } ( R ) \\circ f _ { k } ^ { j } ( X ) \\right) d \\nu ( R ) = \\sum _ { k = 1 } ^ { K } \\tilde { \\Lambda } _ { k } \\circ f _ { k } ^ { j } ( X ) , } } \\end{array}", + "type": "interline_equation", + "image_path": "4bc18a56fd3a2e5947357c447088fa0a0015cc7008f44aeacfb7112cf513128c.jpg" + } + ] + } + ], + "index": 38, + "virtual_lines": [ + { + "bbox": [ + 124, + 645, + 487, + 668.3333333333334 + ], + "spans": [], + "index": 37 + }, + { + "bbox": [ + 124, + 668.3333333333334, + 487, + 691.6666666666667 + ], + "spans": [], + "index": 38 + }, + { + "bbox": [ + 124, + 691.6666666666667, + 487, + 715.0000000000001 + ], + "spans": [], + "index": 39 + } + ] + }, + { + "type": "text", + "bbox": [ + 117, + 720, + 419, + 733 + ], + "lines": [ + { + "bbox": [ + 118, + 719, + 419, + 734 + ], + "spans": [ + { + "bbox": [ + 118, + 719, + 235, + 734 + ], + "score": 1.0, + "content": "2By this we mean that the maps", + "type": "text" + }, + { + "bbox": [ + 235, + 721, + 337, + 732 + ], + "score": 0.93, + "content": "( g , X ) \\mapsto \\rho _ { j } ( g ) X , j = 1 , 2", + "type": "inline_equation" + }, + { + "bbox": [ + 337, + 719, + 419, + 734 + ], + "score": 1.0, + "content": "are jointly continuous", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40, + "bbox_fs": [ + 118, + 719, + 419, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 81, + 499, + 95 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 501, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 133, + 97 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 81, + 146, + 93 + ], + "score": 0.9, + "content": "\\tilde { \\Lambda } _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 80, + 340, + 97 + ], + "score": 1.0, + "content": "stands for the equivariant linear functional from", + "type": "text" + }, + { + "bbox": [ + 341, + 83, + 365, + 94 + ], + "score": 0.9, + "content": "W _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 80, + 376, + 97 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 376, + 83, + 393, + 94 + ], + "score": 0.89, + "content": "W _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 80, + 443, + 97 + ], + "score": 1.0, + "content": ", defined for", + "type": "text" + }, + { + "bbox": [ + 443, + 83, + 487, + 94 + ], + "score": 0.91, + "content": "w \\in W _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 80, + 501, + 97 + ], + "score": 1.0, + "content": "by", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "interline_equation", + "bbox": [ + 205, + 97, + 406, + 124 + ], + "lines": [ + { + "bbox": [ + 205, + 97, + 406, + 124 + ], + "spans": [ + { + "bbox": [ + 205, + 97, + 406, + 124 + ], + "score": 0.92, + "content": "\\tilde { \\Lambda } _ { k } ( w ) = \\int _ { \\mathrm { S O ( 3 ) } } \\rho _ { 2 } ( R ^ { - 1 } ) \\circ \\Lambda _ { k } \\left( \\rho _ { 1 } ( R ) w \\right) d \\nu ( R ) .", + "type": "interline_equation", + "image_path": "7ed28a552a728ceae7ebf55ba5d40395d91da3c7d42661bf2dedb6f9d6431448.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 205, + 97, + 406, + 124 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 128, + 406, + 141 + ], + "lines": [ + { + "bbox": [ + 106, + 127, + 406, + 142 + ], + "spans": [ + { + "bbox": [ + 106, + 127, + 209, + 142 + ], + "score": 1.0, + "content": "Thus we have shown that", + "type": "text" + }, + { + "bbox": [ + 210, + 130, + 216, + 140 + ], + "score": 0.82, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 127, + 236, + 142 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 237, + 128, + 307, + 141 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } , \\mathcal { F } _ { \\mathrm { p o o l } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 127, + 322, + 142 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 323, + 129, + 354, + 139 + ], + "score": 0.9, + "content": "C = K", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 127, + 406, + 142 + ], + "score": 1.0, + "content": ", as required.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "title", + "bbox": [ + 107, + 155, + 252, + 169 + ], + "lines": [ + { + "bbox": [ + 106, + 154, + 252, + 170 + ], + "spans": [ + { + "bbox": [ + 106, + 154, + 252, + 170 + ], + "score": 1.0, + "content": "C PROOFS FOR SECTION 4", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 107, + 180, + 187, + 191 + ], + "lines": [ + { + "bbox": [ + 106, + 179, + 188, + 194 + ], + "spans": [ + { + "bbox": [ + 106, + 179, + 188, + 194 + ], + "score": 1.0, + "content": "We prove Lemma 2", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 106, + 194, + 341, + 207 + ], + "lines": [ + { + "bbox": [ + 105, + 192, + 342, + 209 + ], + "spans": [ + { + "bbox": [ + 105, + 192, + 194, + 209 + ], + "score": 1.0, + "content": "Lemma 2. For every", + "type": "text" + }, + { + "bbox": [ + 195, + 195, + 230, + 206 + ], + "score": 0.92, + "content": "D \\in { \\mathbb { N } } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 192, + 262, + 209 + ], + "score": 1.0, + "content": ", the set", + "type": "text" + }, + { + "bbox": [ + 263, + 195, + 278, + 205 + ], + "score": 0.89, + "content": "\\mathcal { Q } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 192, + 289, + 209 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 289, + 195, + 299, + 204 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 192, + 342, + 209 + ], + "score": 1.0, + "content": "-spanning.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 105, + 217, + 504, + 243 + ], + "lines": [ + { + "bbox": [ + 105, + 217, + 506, + 231 + ], + "spans": [ + { + "bbox": [ + 105, + 217, + 409, + 231 + ], + "score": 1.0, + "content": "Proof. It is known (Segol & Lipman, 2019) (Theorem 2) that polynomials", + "type": "text" + }, + { + "bbox": [ + 409, + 217, + 476, + 230 + ], + "score": 0.9, + "content": "p : \\mathbb { R } ^ { 3 \\times n } \\mathbb { R } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 217, + 506, + 231 + ], + "score": 1.0, + "content": "which", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 228, + 452, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 121, + 245 + ], + "score": 1.0, + "content": "are", + "type": "text" + }, + { + "bbox": [ + 121, + 231, + 134, + 242 + ], + "score": 0.88, + "content": "S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 228, + 347, + 245 + ], + "score": 1.0, + "content": "-equivariant, are spanned by polynomials of the form", + "type": "text" + }, + { + "bbox": [ + 347, + 229, + 404, + 244 + ], + "score": 0.95, + "content": "p _ { \\vec { \\alpha } } = \\stackrel { \\cdot } { ( } p _ { \\vec { \\alpha } } ^ { j } ) _ { j = 1 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 228, + 452, + 245 + ], + "score": 1.0, + "content": ", defined as", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "interline_equation", + "bbox": [ + 233, + 246, + 377, + 281 + ], + "lines": [ + { + "bbox": [ + 233, + 246, + 377, + 281 + ], + "spans": [ + { + "bbox": [ + 233, + 246, + 377, + 281 + ], + "score": 0.95, + "content": "p _ { \\vec { \\alpha } } ^ { j } ( X ) = \\sum _ { i _ { 2 } , \\ldots , i _ { K } = 1 } ^ { n } x _ { j } ^ { \\alpha _ { 1 } } x _ { i _ { 2 } } ^ { \\alpha _ { 2 } } \\cdot \\cdot \\cdot x _ { i _ { k } } ^ { \\alpha _ { k } }", + "type": "interline_equation", + "image_path": "71ac45911fef7a6e9f0f1f15d080c1024f7fd811c2059571fc556dfbcd4b1c9b.jpg" + } + ] + } + ], + "index": 8.5, + "virtual_lines": [ + { + "bbox": [ + 233, + 246, + 377, + 263.5 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 233, + 263.5, + 377, + 281.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 284, + 505, + 335 + ], + "lines": [ + { + "bbox": [ + 106, + 284, + 506, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 284, + 135, + 298 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 285, + 216, + 297 + ], + "score": 0.92, + "content": "\\vec { \\alpha } \\ : = \\ : ( \\alpha _ { 1 } , \\ldots , \\alpha _ { K } )", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 284, + 259, + 298 + ], + "score": 1.0, + "content": "and each", + "type": "text" + }, + { + "bbox": [ + 259, + 284, + 303, + 298 + ], + "score": 0.93, + "content": "\\alpha _ { k } ~ \\in ~ \\mathbb { N } _ { + } ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 284, + 443, + 298 + ], + "score": 1.0, + "content": "is a multi-index. It follows that", + "type": "text" + }, + { + "bbox": [ + 443, + 286, + 456, + 296 + ], + "score": 0.89, + "content": "S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 284, + 506, + 298 + ], + "score": 1.0, + "content": "-equivariant", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 104, + 295, + 506, + 315 + ], + "spans": [ + { + "bbox": [ + 104, + 295, + 200, + 315 + ], + "score": 1.0, + "content": "polynomials of degree", + "type": "text" + }, + { + "bbox": [ + 200, + 299, + 222, + 310 + ], + "score": 0.87, + "content": "\\leq D", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 295, + 390, + 315 + ], + "score": 1.0, + "content": "are spanned by polynomials of the form", + "type": "text" + }, + { + "bbox": [ + 390, + 297, + 402, + 311 + ], + "score": 0.89, + "content": "p _ { \\vec { \\alpha } } ^ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 295, + 432, + 315 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 432, + 297, + 501, + 312 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\sum _ { k = 1 } ^ { K } \\left| \\alpha _ { k } \\right| \\leq D } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 295, + 506, + 315 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 103, + 307, + 508, + 328 + ], + "spans": [ + { + "bbox": [ + 103, + 307, + 146, + 328 + ], + "score": 1.0, + "content": "Denoting", + "type": "text" + }, + { + "bbox": [ + 146, + 311, + 242, + 324 + ], + "score": 0.86, + "content": "r _ { k } = | \\alpha _ { k } | , k = 1 , \\ldots K", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 307, + 302, + 328 + ], + "score": 1.0, + "content": ", the sum of all", + "type": "text" + }, + { + "bbox": [ + 303, + 313, + 313, + 323 + ], + "score": 0.85, + "content": "r _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 307, + 327, + 328 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 327, + 312, + 335, + 322 + ], + "score": 0.82, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 307, + 356, + 328 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 356, + 311, + 407, + 324 + ], + "score": 0.92, + "content": "\\vec { r } = ( r _ { k } ) _ { k = 1 } ^ { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 307, + 508, + 328 + ], + "score": 1.0, + "content": ", we see that there exists", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 322, + 284, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 181, + 335 + ], + "score": 1.0, + "content": "a linear functional", + "type": "text" + }, + { + "bbox": [ + 182, + 324, + 244, + 335 + ], + "score": 0.91, + "content": "\\Lambda _ { \\vec { \\alpha } , \\vec { r } } : \\mathcal { T } _ { T } \\to \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 322, + 284, + 335 + ], + "score": 1.0, + "content": "such that", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11.5 + }, + { + "type": "interline_equation", + "bbox": [ + 254, + 338, + 356, + 355 + ], + "lines": [ + { + "bbox": [ + 254, + 338, + 356, + 355 + ], + "spans": [ + { + "bbox": [ + 254, + 338, + 356, + 355 + ], + "score": 0.92, + "content": "p _ { \\vec { \\alpha } } ^ { j } ( X ) = \\Lambda _ { \\vec { \\alpha } , \\vec { r } ^ { } ^ { \\mathrm { ~ O ~ } } } Q _ { j } ^ { \\vec { r } } ( X )", + "type": "interline_equation", + "image_path": "20b00c92a9be1f81a246671a464bcc308502fd9e032dbed0308d3ee0ca025373.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 254, + 338, + 356, + 355 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 358, + 385, + 380 + ], + "lines": [ + { + "bbox": [ + 101, + 352, + 388, + 383 + ], + "spans": [ + { + "bbox": [ + 101, + 352, + 190, + 383 + ], + "score": 1.0, + "content": "where we recall that", + "type": "text" + }, + { + "bbox": [ + 190, + 358, + 277, + 380 + ], + "score": 0.91, + "content": "Q ^ { \\vec { r } } = \\left( Q _ { j } ^ { ( \\vec { r } ) } ( X ) \\right) _ { j = 1 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 277, + 352, + 388, + 383 + ], + "score": 1.0, + "content": "is defined in equation 6 as", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "interline_equation", + "bbox": [ + 186, + 383, + 424, + 418 + ], + "lines": [ + { + "bbox": [ + 186, + 383, + 424, + 418 + ], + "spans": [ + { + "bbox": [ + 186, + 383, + 424, + 418 + ], + "score": 0.92, + "content": "Q _ { j } ^ { ( \\vec { r } ) } ( X ) = \\sum _ { i _ { 2 } , \\ldots , i _ { K } = 1 } ^ { n } x _ { j } ^ { \\otimes r _ { 1 } } \\otimes x _ { i _ { 2 } } ^ { \\otimes r _ { 2 } } \\otimes x _ { i _ { 3 } } ^ { \\otimes r _ { 3 } } \\otimes \\ldots \\otimes x _ { i _ { K } } ^ { \\otimes r _ { K } } .", + "type": "interline_equation", + "image_path": "c77b67b34053601544f6bcf7516644f3b241b0b16af382236e465577f65a6f52.jpg" + } + ] + } + ], + "index": 16.5, + "virtual_lines": [ + { + "bbox": [ + 186, + 383, + 424, + 400.5 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 186, + 400.5, + 424, + 418.0 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 426, + 501, + 439 + ], + "lines": [ + { + "bbox": [ + 105, + 424, + 503, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 181, + 442 + ], + "score": 1.0, + "content": "Thus polynomials", + "type": "text" + }, + { + "bbox": [ + 181, + 427, + 231, + 441 + ], + "score": 0.94, + "content": "p = ( p _ { j } ) _ { j = 1 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 424, + 313, + 442 + ], + "score": 1.0, + "content": "which are of degree", + "type": "text" + }, + { + "bbox": [ + 314, + 428, + 334, + 438 + ], + "score": 0.9, + "content": "\\leq D", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 424, + 369, + 442 + ], + "score": 1.0, + "content": ", and are", + "type": "text" + }, + { + "bbox": [ + 370, + 428, + 382, + 438 + ], + "score": 0.9, + "content": "S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 382, + 424, + 503, + 442 + ], + "score": 1.0, + "content": "equivariant, can be written as", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "interline_equation", + "bbox": [ + 115, + 443, + 493, + 473 + ], + "lines": [ + { + "bbox": [ + 115, + 443, + 493, + 473 + ], + "spans": [ + { + "bbox": [ + 115, + 443, + 493, + 473 + ], + "score": 0.93, + "content": "p _ { j } ( X ) = \\sum _ { T \\leq D } \\sum _ { \\vec { r } \\in \\Sigma _ { T } } \\sum _ { \\vec { \\alpha } | | \\alpha _ { k } | = r _ { k } } \\Lambda _ { \\vec { \\alpha } , \\vec { r } } \\left( Q _ { j } ^ { ( \\vec { r } ) } ( X ) \\right) = \\sum _ { T \\leq D } \\sum _ { \\vec { r } \\in \\Sigma _ { T } } \\Lambda _ { \\vec { r } } \\left( \\iota \\circ Q _ { j } ^ { ( \\vec { r } ) } ( X ) \\right) , j = 1 , \\dots , n ,", + "type": "interline_equation", + "image_path": "a850ab1e1e90d06a22cc3b7a3205b0605b223fbfd89301a4c88f8a588642481c.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 115, + 443, + 493, + 473 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 477, + 505, + 502 + ], + "lines": [ + { + "bbox": [ + 103, + 473, + 508, + 497 + ], + "spans": [ + { + "bbox": [ + 103, + 473, + 134, + 497 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 477, + 256, + 492 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\Lambda _ { \\vec { r } } = \\sum _ { \\vec { \\alpha } | | \\alpha _ { k } | = r _ { k } } \\Lambda _ { \\vec { \\alpha } , \\vec { r } } \\circ \\iota _ { T } ^ { - 1 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 473, + 279, + 497 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 279, + 477, + 294, + 491 + ], + "score": 0.92, + "content": "\\iota _ { T } ^ { - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 473, + 446, + 497 + ], + "score": 1.0, + "content": "is the left inverse of the embedding", + "type": "text" + }, + { + "bbox": [ + 446, + 480, + 451, + 488 + ], + "score": 0.7, + "content": "\\iota", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 473, + 468, + 497 + ], + "score": 1.0, + "content": ". If", + "type": "text" + }, + { + "bbox": [ + 468, + 480, + 474, + 490 + ], + "score": 0.81, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 473, + 508, + 497 + ], + "score": 1.0, + "content": "is also", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 490, + 211, + 502 + ], + "spans": [ + { + "bbox": [ + 106, + 490, + 211, + 502 + ], + "score": 1.0, + "content": "translation invariant, then", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5 + }, + { + "type": "interline_equation", + "bbox": [ + 156, + 504, + 455, + 537 + ], + "lines": [ + { + "bbox": [ + 156, + 504, + 455, + 537 + ], + "spans": [ + { + "bbox": [ + 156, + 504, + 455, + 537 + ], + "score": 0.93, + "content": "p ( X ) = p ( X - \\frac { 1 } { n } X 1 _ { n } \\mathbf { 1 } _ { n } ^ { T } ) = \\sum _ { T \\leq D } \\sum _ { \\vec { r } \\in \\Sigma _ { T } } \\hat { \\Lambda } _ { \\vec { r } } \\left( \\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac { 1 } { n } X 1 _ { n } \\mathbf { 1 } _ { n } ^ { T } ) \\right) .", + "type": "interline_equation", + "image_path": "72fdaf5398588aaec2a0b484faf92aa54c621a5dada65c8b759b1ebe5787c18e.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 156, + 504, + 455, + 515.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 156, + 515.0, + 455, + 526.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 156, + 526.0, + 455, + 537.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 538, + 208, + 551 + ], + "lines": [ + { + "bbox": [ + 105, + 536, + 209, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 536, + 129, + 554 + ], + "score": 1.0, + "content": "Thus", + "type": "text" + }, + { + "bbox": [ + 129, + 540, + 145, + 551 + ], + "score": 0.9, + "content": "\\mathcal { Q } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 536, + 156, + 554 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 156, + 540, + 165, + 549 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 536, + 209, + 554 + ], + "score": 1.0, + "content": "-spanning.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 563, + 187, + 574 + ], + "lines": [ + { + "bbox": [ + 106, + 562, + 188, + 576 + ], + "spans": [ + { + "bbox": [ + 106, + 562, + 188, + 576 + ], + "score": 1.0, + "content": "We prove Lemma 3", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 105, + 577, + 299, + 589 + ], + "lines": [ + { + "bbox": [ + 106, + 576, + 299, + 591 + ], + "spans": [ + { + "bbox": [ + 106, + 576, + 221, + 591 + ], + "score": 1.0, + "content": "Lemma 3. The function set", + "type": "text" + }, + { + "bbox": [ + 221, + 578, + 236, + 588 + ], + "score": 0.88, + "content": "\\mathcal { Q } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 576, + 299, + 591 + ], + "score": 1.0, + "content": "is contained in", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "interline_equation", + "bbox": [ + 140, + 592, + 471, + 615 + ], + "lines": [ + { + "bbox": [ + 140, + 592, + 471, + 615 + ], + "spans": [ + { + "bbox": [ + 140, + 592, + 471, + 615 + ], + "score": 0.9, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } } = \\big \\{ \\iota \\circ \\pi _ { V } \\circ f ^ { 1 } \\circ f ^ { 2 } \\circ \\ldots \\circ f ^ { T } \\circ \\mathrm { e x t } ( X - \\frac { 1 } { n } X \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) | f ^ { j } \\in \\mathcal { F } _ { m i n } , T \\leq D \\big \\} .", + "type": "interline_equation", + "image_path": "d9975ca5b0f8b179121a196b20c5c59424456b8b001824e241861043d7bbc3aa.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 140, + 592, + 471, + 615 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 618, + 212, + 630 + ], + "lines": [ + { + "bbox": [ + 105, + 616, + 214, + 633 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 128, + 633 + ], + "score": 1.0, + "content": "Thus", + "type": "text" + }, + { + "bbox": [ + 129, + 619, + 150, + 629 + ], + "score": 0.89, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 616, + 160, + 633 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 161, + 619, + 171, + 628 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 616, + 214, + 633 + ], + "score": 1.0, + "content": "-spanning.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 105, + 641, + 468, + 654 + ], + "lines": [ + { + "bbox": [ + 105, + 640, + 469, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 301, + 656 + ], + "score": 1.0, + "content": "Proof. In this proof we make the dependence of", + "type": "text" + }, + { + "bbox": [ + 301, + 642, + 323, + 653 + ], + "score": 0.91, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 640, + 336, + 656 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 337, + 642, + 347, + 652 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 640, + 426, + 656 + ], + "score": 1.0, + "content": "explicit and denote", + "type": "text" + }, + { + "bbox": [ + 426, + 642, + 464, + 654 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 640, + 469, + 656 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 658, + 503, + 682 + ], + "lines": [ + { + "bbox": [ + 106, + 658, + 505, + 671 + ], + "spans": [ + { + "bbox": [ + 106, + 658, + 248, + 671 + ], + "score": 1.0, + "content": "We prove the claim by induction on", + "type": "text" + }, + { + "bbox": [ + 248, + 659, + 257, + 669 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 658, + 295, + 671 + ], + "score": 1.0, + "content": ". Assume", + "type": "text" + }, + { + "bbox": [ + 296, + 659, + 324, + 669 + ], + "score": 0.89, + "content": "D = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 658, + 351, + 671 + ], + "score": 1.0, + "content": ". Then", + "type": "text" + }, + { + "bbox": [ + 351, + 659, + 364, + 670 + ], + "score": 0.87, + "content": "\\mathcal { Q } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 658, + 505, + 671 + ], + "score": 1.0, + "content": "contains only the constant function", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 669, + 389, + 683 + ], + "spans": [ + { + "bbox": [ + 106, + 670, + 168, + 682 + ], + "score": 0.92, + "content": "X \\mapsto 1 _ { n } \\in \\mathcal { T } _ { 0 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 669, + 303, + 683 + ], + "score": 1.0, + "content": ", and this is precisely the function", + "type": "text" + }, + { + "bbox": [ + 304, + 670, + 386, + 682 + ], + "score": 0.9, + "content": "\\pi _ { V } \\circ \\mathrm { e x t } \\in \\mathcal { F } _ { \\mathrm { f e a t } } ( 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 669, + 389, + 683 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 106, + 685, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 685, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 105, + 685, + 251, + 699 + ], + "score": 1.0, + "content": "Now assume the claim holds for all", + "type": "text" + }, + { + "bbox": [ + 252, + 687, + 264, + 697 + ], + "score": 0.88, + "content": "D ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 685, + 286, + 699 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 286, + 686, + 357, + 698 + ], + "score": 0.93, + "content": "D - 1 \\geq D ^ { \\prime } \\geq 0", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 685, + 458, + 699 + ], + "score": 1.0, + "content": ", and prove the claim for", + "type": "text" + }, + { + "bbox": [ + 458, + 687, + 467, + 696 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 685, + 505, + 699 + ], + "score": 1.0, + "content": ". Choose", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 107, + 697, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 107, + 698, + 203, + 710 + ], + "score": 0.91, + "content": "\\vec { r } = ( r _ { 1 } , \\ldots , r _ { k } ) \\in \\Sigma _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 697, + 244, + 712 + ], + "score": 1.0, + "content": "for some", + "type": "text" + }, + { + "bbox": [ + 244, + 699, + 276, + 710 + ], + "score": 0.92, + "content": "T \\leq D", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 697, + 421, + 712 + ], + "score": 1.0, + "content": ", we need to show that the function", + "type": "text" + }, + { + "bbox": [ + 421, + 697, + 441, + 710 + ], + "score": 0.94, + "content": "Q ^ { ( \\vec { r } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 697, + 462, + 712 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 463, + 698, + 501, + 710 + ], + "score": 0.89, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 697, + 505, + 712 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 131, + 723 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 132, + 710, + 237, + 721 + ], + "score": 0.9, + "content": "{ \\mathcal { F } } _ { \\mathrm { f e a t } } ( D - 1 ) \\subseteq { \\mathcal { F } } _ { \\mathrm { f e a t } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 708, + 470, + 723 + ], + "score": 1.0, + "content": "we know from the induction hypothesis that this is true if", + "type": "text" + }, + { + "bbox": [ + 471, + 711, + 501, + 720 + ], + "score": 0.88, + "content": "T < D", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 708, + 506, + 723 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 720, + 290, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 160, + 732 + ], + "score": 1.0, + "content": "Now assume", + "type": "text" + }, + { + "bbox": [ + 160, + 721, + 190, + 730 + ], + "score": 0.89, + "content": "T = D", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 720, + 290, + 732 + ], + "score": 1.0, + "content": ". We consider two cases:", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "14", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 540, + 505, + 550 + ], + "lines": [ + { + "bbox": [ + 496, + 541, + 504, + 550 + ], + "spans": [ + { + "bbox": [ + 496, + 541, + 504, + 550 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 128, + 505, + 139 + ], + "lines": [ + { + "bbox": [ + 496, + 130, + 505, + 140 + ], + "spans": [ + { + "bbox": [ + 496, + 130, + 505, + 140 + ], + "score": 0.998, + "content": "□", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 81, + 499, + 95 + ], + "lines": [ + { + "bbox": [ + 105, + 80, + 501, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 80, + 133, + 97 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 81, + 146, + 93 + ], + "score": 0.9, + "content": "\\tilde { \\Lambda } _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 80, + 340, + 97 + ], + "score": 1.0, + "content": "stands for the equivariant linear functional from", + "type": "text" + }, + { + "bbox": [ + 341, + 83, + 365, + 94 + ], + "score": 0.9, + "content": "W _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 80, + 376, + 97 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 376, + 83, + 393, + 94 + ], + "score": 0.89, + "content": "W _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 80, + 443, + 97 + ], + "score": 1.0, + "content": ", defined for", + "type": "text" + }, + { + "bbox": [ + 443, + 83, + 487, + 94 + ], + "score": 0.91, + "content": "w \\in W _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 80, + 501, + 97 + ], + "score": 1.0, + "content": "by", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 105, + 80, + 501, + 97 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 205, + 97, + 406, + 124 + ], + "lines": [ + { + "bbox": [ + 205, + 97, + 406, + 124 + ], + "spans": [ + { + "bbox": [ + 205, + 97, + 406, + 124 + ], + "score": 0.92, + "content": "\\tilde { \\Lambda } _ { k } ( w ) = \\int _ { \\mathrm { S O ( 3 ) } } \\rho _ { 2 } ( R ^ { - 1 } ) \\circ \\Lambda _ { k } \\left( \\rho _ { 1 } ( R ) w \\right) d \\nu ( R ) .", + "type": "interline_equation", + "image_path": "7ed28a552a728ceae7ebf55ba5d40395d91da3c7d42661bf2dedb6f9d6431448.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 205, + 97, + 406, + 124 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 128, + 406, + 141 + ], + "lines": [ + { + "bbox": [ + 106, + 127, + 406, + 142 + ], + "spans": [ + { + "bbox": [ + 106, + 127, + 209, + 142 + ], + "score": 1.0, + "content": "Thus we have shown that", + "type": "text" + }, + { + "bbox": [ + 210, + 130, + 216, + 140 + ], + "score": 0.82, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 127, + 236, + 142 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 237, + 128, + 307, + 141 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } , \\mathcal { F } _ { \\mathrm { p o o l } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 127, + 322, + 142 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 323, + 129, + 354, + 139 + ], + "score": 0.9, + "content": "C = K", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 127, + 406, + 142 + ], + "score": 1.0, + "content": ", as required.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2, + "bbox_fs": [ + 106, + 127, + 406, + 142 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 155, + 252, + 169 + ], + "lines": [ + { + "bbox": [ + 106, + 154, + 252, + 170 + ], + "spans": [ + { + "bbox": [ + 106, + 154, + 252, + 170 + ], + "score": 1.0, + "content": "C PROOFS FOR SECTION 4", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 107, + 180, + 187, + 191 + ], + "lines": [ + { + "bbox": [ + 106, + 179, + 188, + 194 + ], + "spans": [ + { + "bbox": [ + 106, + 179, + 188, + 194 + ], + "score": 1.0, + "content": "We prove Lemma 2", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4, + "bbox_fs": [ + 106, + 179, + 188, + 194 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 194, + 341, + 207 + ], + "lines": [ + { + "bbox": [ + 105, + 192, + 342, + 209 + ], + "spans": [ + { + "bbox": [ + 105, + 192, + 194, + 209 + ], + "score": 1.0, + "content": "Lemma 2. For every", + "type": "text" + }, + { + "bbox": [ + 195, + 195, + 230, + 206 + ], + "score": 0.92, + "content": "D \\in { \\mathbb { N } } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 192, + 262, + 209 + ], + "score": 1.0, + "content": ", the set", + "type": "text" + }, + { + "bbox": [ + 263, + 195, + 278, + 205 + ], + "score": 0.89, + "content": "\\mathcal { Q } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 192, + 289, + 209 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 289, + 195, + 299, + 204 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 192, + 342, + 209 + ], + "score": 1.0, + "content": "-spanning.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 192, + 342, + 209 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 217, + 504, + 243 + ], + "lines": [ + { + "bbox": [ + 105, + 217, + 506, + 231 + ], + "spans": [ + { + "bbox": [ + 105, + 217, + 409, + 231 + ], + "score": 1.0, + "content": "Proof. It is known (Segol & Lipman, 2019) (Theorem 2) that polynomials", + "type": "text" + }, + { + "bbox": [ + 409, + 217, + 476, + 230 + ], + "score": 0.9, + "content": "p : \\mathbb { R } ^ { 3 \\times n } \\mathbb { R } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 217, + 506, + 231 + ], + "score": 1.0, + "content": "which", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 228, + 452, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 228, + 121, + 245 + ], + "score": 1.0, + "content": "are", + "type": "text" + }, + { + "bbox": [ + 121, + 231, + 134, + 242 + ], + "score": 0.88, + "content": "S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 228, + 347, + 245 + ], + "score": 1.0, + "content": "-equivariant, are spanned by polynomials of the form", + "type": "text" + }, + { + "bbox": [ + 347, + 229, + 404, + 244 + ], + "score": 0.95, + "content": "p _ { \\vec { \\alpha } } = \\stackrel { \\cdot } { ( } p _ { \\vec { \\alpha } } ^ { j } ) _ { j = 1 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 228, + 452, + 245 + ], + "score": 1.0, + "content": ", defined as", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5, + "bbox_fs": [ + 105, + 217, + 506, + 245 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 233, + 246, + 377, + 281 + ], + "lines": [ + { + "bbox": [ + 233, + 246, + 377, + 281 + ], + "spans": [ + { + "bbox": [ + 233, + 246, + 377, + 281 + ], + "score": 0.95, + "content": "p _ { \\vec { \\alpha } } ^ { j } ( X ) = \\sum _ { i _ { 2 } , \\ldots , i _ { K } = 1 } ^ { n } x _ { j } ^ { \\alpha _ { 1 } } x _ { i _ { 2 } } ^ { \\alpha _ { 2 } } \\cdot \\cdot \\cdot x _ { i _ { k } } ^ { \\alpha _ { k } }", + "type": "interline_equation", + "image_path": "71ac45911fef7a6e9f0f1f15d080c1024f7fd811c2059571fc556dfbcd4b1c9b.jpg" + } + ] + } + ], + "index": 8.5, + "virtual_lines": [ + { + "bbox": [ + 233, + 246, + 377, + 263.5 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 233, + 263.5, + 377, + 281.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 284, + 505, + 335 + ], + "lines": [ + { + "bbox": [ + 106, + 284, + 506, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 284, + 135, + 298 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 135, + 285, + 216, + 297 + ], + "score": 0.92, + "content": "\\vec { \\alpha } \\ : = \\ : ( \\alpha _ { 1 } , \\ldots , \\alpha _ { K } )", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 284, + 259, + 298 + ], + "score": 1.0, + "content": "and each", + "type": "text" + }, + { + "bbox": [ + 259, + 284, + 303, + 298 + ], + "score": 0.93, + "content": "\\alpha _ { k } ~ \\in ~ \\mathbb { N } _ { + } ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 284, + 443, + 298 + ], + "score": 1.0, + "content": "is a multi-index. It follows that", + "type": "text" + }, + { + "bbox": [ + 443, + 286, + 456, + 296 + ], + "score": 0.89, + "content": "S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 284, + 506, + 298 + ], + "score": 1.0, + "content": "-equivariant", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 104, + 295, + 506, + 315 + ], + "spans": [ + { + "bbox": [ + 104, + 295, + 200, + 315 + ], + "score": 1.0, + "content": "polynomials of degree", + "type": "text" + }, + { + "bbox": [ + 200, + 299, + 222, + 310 + ], + "score": 0.87, + "content": "\\leq D", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 295, + 390, + 315 + ], + "score": 1.0, + "content": "are spanned by polynomials of the form", + "type": "text" + }, + { + "bbox": [ + 390, + 297, + 402, + 311 + ], + "score": 0.89, + "content": "p _ { \\vec { \\alpha } } ^ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 295, + 432, + 315 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 432, + 297, + 501, + 312 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\sum _ { k = 1 } ^ { K } \\left| \\alpha _ { k } \\right| \\leq D } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 295, + 506, + 315 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 103, + 307, + 508, + 328 + ], + "spans": [ + { + "bbox": [ + 103, + 307, + 146, + 328 + ], + "score": 1.0, + "content": "Denoting", + "type": "text" + }, + { + "bbox": [ + 146, + 311, + 242, + 324 + ], + "score": 0.86, + "content": "r _ { k } = | \\alpha _ { k } | , k = 1 , \\ldots K", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 307, + 302, + 328 + ], + "score": 1.0, + "content": ", the sum of all", + "type": "text" + }, + { + "bbox": [ + 303, + 313, + 313, + 323 + ], + "score": 0.85, + "content": "r _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 307, + 327, + 328 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 327, + 312, + 335, + 322 + ], + "score": 0.82, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 307, + 356, + 328 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 356, + 311, + 407, + 324 + ], + "score": 0.92, + "content": "\\vec { r } = ( r _ { k } ) _ { k = 1 } ^ { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 307, + 508, + 328 + ], + "score": 1.0, + "content": ", we see that there exists", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 322, + 284, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 181, + 335 + ], + "score": 1.0, + "content": "a linear functional", + "type": "text" + }, + { + "bbox": [ + 182, + 324, + 244, + 335 + ], + "score": 0.91, + "content": "\\Lambda _ { \\vec { \\alpha } , \\vec { r } } : \\mathcal { T } _ { T } \\to \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 322, + 284, + 335 + ], + "score": 1.0, + "content": "such that", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11.5, + "bbox_fs": [ + 103, + 284, + 508, + 335 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 254, + 338, + 356, + 355 + ], + "lines": [ + { + "bbox": [ + 254, + 338, + 356, + 355 + ], + "spans": [ + { + "bbox": [ + 254, + 338, + 356, + 355 + ], + "score": 0.92, + "content": "p _ { \\vec { \\alpha } } ^ { j } ( X ) = \\Lambda _ { \\vec { \\alpha } , \\vec { r } ^ { } ^ { \\mathrm { ~ O ~ } } } Q _ { j } ^ { \\vec { r } } ( X )", + "type": "interline_equation", + "image_path": "20b00c92a9be1f81a246671a464bcc308502fd9e032dbed0308d3ee0ca025373.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 254, + 338, + 356, + 355 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 358, + 385, + 380 + ], + "lines": [ + { + "bbox": [ + 101, + 352, + 388, + 383 + ], + "spans": [ + { + "bbox": [ + 101, + 352, + 190, + 383 + ], + "score": 1.0, + "content": "where we recall that", + "type": "text" + }, + { + "bbox": [ + 190, + 358, + 277, + 380 + ], + "score": 0.91, + "content": "Q ^ { \\vec { r } } = \\left( Q _ { j } ^ { ( \\vec { r } ) } ( X ) \\right) _ { j = 1 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 277, + 352, + 388, + 383 + ], + "score": 1.0, + "content": "is defined in equation 6 as", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15, + "bbox_fs": [ + 101, + 352, + 388, + 383 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 186, + 383, + 424, + 418 + ], + "lines": [ + { + "bbox": [ + 186, + 383, + 424, + 418 + ], + "spans": [ + { + "bbox": [ + 186, + 383, + 424, + 418 + ], + "score": 0.92, + "content": "Q _ { j } ^ { ( \\vec { r } ) } ( X ) = \\sum _ { i _ { 2 } , \\ldots , i _ { K } = 1 } ^ { n } x _ { j } ^ { \\otimes r _ { 1 } } \\otimes x _ { i _ { 2 } } ^ { \\otimes r _ { 2 } } \\otimes x _ { i _ { 3 } } ^ { \\otimes r _ { 3 } } \\otimes \\ldots \\otimes x _ { i _ { K } } ^ { \\otimes r _ { K } } .", + "type": "interline_equation", + "image_path": "c77b67b34053601544f6bcf7516644f3b241b0b16af382236e465577f65a6f52.jpg" + } + ] + } + ], + "index": 16.5, + "virtual_lines": [ + { + "bbox": [ + 186, + 383, + 424, + 400.5 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 186, + 400.5, + 424, + 418.0 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 426, + 501, + 439 + ], + "lines": [ + { + "bbox": [ + 105, + 424, + 503, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 181, + 442 + ], + "score": 1.0, + "content": "Thus polynomials", + "type": "text" + }, + { + "bbox": [ + 181, + 427, + 231, + 441 + ], + "score": 0.94, + "content": "p = ( p _ { j } ) _ { j = 1 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 424, + 313, + 442 + ], + "score": 1.0, + "content": "which are of degree", + "type": "text" + }, + { + "bbox": [ + 314, + 428, + 334, + 438 + ], + "score": 0.9, + "content": "\\leq D", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 424, + 369, + 442 + ], + "score": 1.0, + "content": ", and are", + "type": "text" + }, + { + "bbox": [ + 370, + 428, + 382, + 438 + ], + "score": 0.9, + "content": "S _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 382, + 424, + 503, + 442 + ], + "score": 1.0, + "content": "equivariant, can be written as", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 424, + 503, + 442 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 115, + 443, + 493, + 473 + ], + "lines": [ + { + "bbox": [ + 115, + 443, + 493, + 473 + ], + "spans": [ + { + "bbox": [ + 115, + 443, + 493, + 473 + ], + "score": 0.93, + "content": "p _ { j } ( X ) = \\sum _ { T \\leq D } \\sum _ { \\vec { r } \\in \\Sigma _ { T } } \\sum _ { \\vec { \\alpha } | | \\alpha _ { k } | = r _ { k } } \\Lambda _ { \\vec { \\alpha } , \\vec { r } } \\left( Q _ { j } ^ { ( \\vec { r } ) } ( X ) \\right) = \\sum _ { T \\leq D } \\sum _ { \\vec { r } \\in \\Sigma _ { T } } \\Lambda _ { \\vec { r } } \\left( \\iota \\circ Q _ { j } ^ { ( \\vec { r } ) } ( X ) \\right) , j = 1 , \\dots , n ,", + "type": "interline_equation", + "image_path": "a850ab1e1e90d06a22cc3b7a3205b0605b223fbfd89301a4c88f8a588642481c.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 115, + 443, + 493, + 473 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 104, + 477, + 505, + 502 + ], + "lines": [ + { + "bbox": [ + 103, + 473, + 508, + 497 + ], + "spans": [ + { + "bbox": [ + 103, + 473, + 134, + 497 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 477, + 256, + 492 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\Lambda _ { \\vec { r } } = \\sum _ { \\vec { \\alpha } | | \\alpha _ { k } | = r _ { k } } \\Lambda _ { \\vec { \\alpha } , \\vec { r } } \\circ \\iota _ { T } ^ { - 1 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 473, + 279, + 497 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 279, + 477, + 294, + 491 + ], + "score": 0.92, + "content": "\\iota _ { T } ^ { - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 473, + 446, + 497 + ], + "score": 1.0, + "content": "is the left inverse of the embedding", + "type": "text" + }, + { + "bbox": [ + 446, + 480, + 451, + 488 + ], + "score": 0.7, + "content": "\\iota", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 473, + 468, + 497 + ], + "score": 1.0, + "content": ". If", + "type": "text" + }, + { + "bbox": [ + 468, + 480, + 474, + 490 + ], + "score": 0.81, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 473, + 508, + 497 + ], + "score": 1.0, + "content": "is also", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 490, + 211, + 502 + ], + "spans": [ + { + "bbox": [ + 106, + 490, + 211, + 502 + ], + "score": 1.0, + "content": "translation invariant, then", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5, + "bbox_fs": [ + 103, + 473, + 508, + 502 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 156, + 504, + 455, + 537 + ], + "lines": [ + { + "bbox": [ + 156, + 504, + 455, + 537 + ], + "spans": [ + { + "bbox": [ + 156, + 504, + 455, + 537 + ], + "score": 0.93, + "content": "p ( X ) = p ( X - \\frac { 1 } { n } X 1 _ { n } \\mathbf { 1 } _ { n } ^ { T } ) = \\sum _ { T \\leq D } \\sum _ { \\vec { r } \\in \\Sigma _ { T } } \\hat { \\Lambda } _ { \\vec { r } } \\left( \\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac { 1 } { n } X 1 _ { n } \\mathbf { 1 } _ { n } ^ { T } ) \\right) .", + "type": "interline_equation", + "image_path": "72fdaf5398588aaec2a0b484faf92aa54c621a5dada65c8b759b1ebe5787c18e.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 156, + 504, + 455, + 515.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 156, + 515.0, + 455, + 526.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 156, + 526.0, + 455, + 537.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 538, + 208, + 551 + ], + "lines": [ + { + "bbox": [ + 105, + 536, + 209, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 536, + 129, + 554 + ], + "score": 1.0, + "content": "Thus", + "type": "text" + }, + { + "bbox": [ + 129, + 540, + 145, + 551 + ], + "score": 0.9, + "content": "\\mathcal { Q } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 536, + 156, + 554 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 156, + 540, + 165, + 549 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 536, + 209, + 554 + ], + "score": 1.0, + "content": "-spanning.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25, + "bbox_fs": [ + 105, + 536, + 209, + 554 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 563, + 187, + 574 + ], + "lines": [ + { + "bbox": [ + 106, + 562, + 188, + 576 + ], + "spans": [ + { + "bbox": [ + 106, + 562, + 188, + 576 + ], + "score": 1.0, + "content": "We prove Lemma 3", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26, + "bbox_fs": [ + 106, + 562, + 188, + 576 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 577, + 299, + 589 + ], + "lines": [ + { + "bbox": [ + 106, + 576, + 299, + 591 + ], + "spans": [ + { + "bbox": [ + 106, + 576, + 221, + 591 + ], + "score": 1.0, + "content": "Lemma 3. The function set", + "type": "text" + }, + { + "bbox": [ + 221, + 578, + 236, + 588 + ], + "score": 0.88, + "content": "\\mathcal { Q } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 576, + 299, + 591 + ], + "score": 1.0, + "content": "is contained in", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27, + "bbox_fs": [ + 106, + 576, + 299, + 591 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 140, + 592, + 471, + 615 + ], + "lines": [ + { + "bbox": [ + 140, + 592, + 471, + 615 + ], + "spans": [ + { + "bbox": [ + 140, + 592, + 471, + 615 + ], + "score": 0.9, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } } = \\big \\{ \\iota \\circ \\pi _ { V } \\circ f ^ { 1 } \\circ f ^ { 2 } \\circ \\ldots \\circ f ^ { T } \\circ \\mathrm { e x t } ( X - \\frac { 1 } { n } X \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) | f ^ { j } \\in \\mathcal { F } _ { m i n } , T \\leq D \\big \\} .", + "type": "interline_equation", + "image_path": "d9975ca5b0f8b179121a196b20c5c59424456b8b001824e241861043d7bbc3aa.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 140, + 592, + 471, + 615 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 618, + 212, + 630 + ], + "lines": [ + { + "bbox": [ + 105, + 616, + 214, + 633 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 128, + 633 + ], + "score": 1.0, + "content": "Thus", + "type": "text" + }, + { + "bbox": [ + 129, + 619, + 150, + 629 + ], + "score": 0.89, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 616, + 160, + 633 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 161, + 619, + 171, + 628 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 616, + 214, + 633 + ], + "score": 1.0, + "content": "-spanning.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 616, + 214, + 633 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 641, + 468, + 654 + ], + "lines": [ + { + "bbox": [ + 105, + 640, + 469, + 656 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 301, + 656 + ], + "score": 1.0, + "content": "Proof. In this proof we make the dependence of", + "type": "text" + }, + { + "bbox": [ + 301, + 642, + 323, + 653 + ], + "score": 0.91, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 640, + 336, + 656 + ], + "score": 1.0, + "content": "on", + "type": "text" + }, + { + "bbox": [ + 337, + 642, + 347, + 652 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 640, + 426, + 656 + ], + "score": 1.0, + "content": "explicit and denote", + "type": "text" + }, + { + "bbox": [ + 426, + 642, + 464, + 654 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 640, + 469, + 656 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 640, + 469, + 656 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 658, + 503, + 682 + ], + "lines": [ + { + "bbox": [ + 106, + 658, + 505, + 671 + ], + "spans": [ + { + "bbox": [ + 106, + 658, + 248, + 671 + ], + "score": 1.0, + "content": "We prove the claim by induction on", + "type": "text" + }, + { + "bbox": [ + 248, + 659, + 257, + 669 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 658, + 295, + 671 + ], + "score": 1.0, + "content": ". Assume", + "type": "text" + }, + { + "bbox": [ + 296, + 659, + 324, + 669 + ], + "score": 0.89, + "content": "D = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 658, + 351, + 671 + ], + "score": 1.0, + "content": ". Then", + "type": "text" + }, + { + "bbox": [ + 351, + 659, + 364, + 670 + ], + "score": 0.87, + "content": "\\mathcal { Q } _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 658, + 505, + 671 + ], + "score": 1.0, + "content": "contains only the constant function", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 669, + 389, + 683 + ], + "spans": [ + { + "bbox": [ + 106, + 670, + 168, + 682 + ], + "score": 0.92, + "content": "X \\mapsto 1 _ { n } \\in \\mathcal { T } _ { 0 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 669, + 303, + 683 + ], + "score": 1.0, + "content": ", and this is precisely the function", + "type": "text" + }, + { + "bbox": [ + 304, + 670, + 386, + 682 + ], + "score": 0.9, + "content": "\\pi _ { V } \\circ \\mathrm { e x t } \\in \\mathcal { F } _ { \\mathrm { f e a t } } ( 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 669, + 389, + 683 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5, + "bbox_fs": [ + 106, + 658, + 505, + 683 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 685, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 685, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 105, + 685, + 251, + 699 + ], + "score": 1.0, + "content": "Now assume the claim holds for all", + "type": "text" + }, + { + "bbox": [ + 252, + 687, + 264, + 697 + ], + "score": 0.88, + "content": "D ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 685, + 286, + 699 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 286, + 686, + 357, + 698 + ], + "score": 0.93, + "content": "D - 1 \\geq D ^ { \\prime } \\geq 0", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 685, + 458, + 699 + ], + "score": 1.0, + "content": ", and prove the claim for", + "type": "text" + }, + { + "bbox": [ + 458, + 687, + 467, + 696 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 685, + 505, + 699 + ], + "score": 1.0, + "content": ". Choose", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 107, + 697, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 107, + 698, + 203, + 710 + ], + "score": 0.91, + "content": "\\vec { r } = ( r _ { 1 } , \\ldots , r _ { k } ) \\in \\Sigma _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 697, + 244, + 712 + ], + "score": 1.0, + "content": "for some", + "type": "text" + }, + { + "bbox": [ + 244, + 699, + 276, + 710 + ], + "score": 0.92, + "content": "T \\leq D", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 697, + 421, + 712 + ], + "score": 1.0, + "content": ", we need to show that the function", + "type": "text" + }, + { + "bbox": [ + 421, + 697, + 441, + 710 + ], + "score": 0.94, + "content": "Q ^ { ( \\vec { r } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 697, + 462, + 712 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 463, + 698, + 501, + 710 + ], + "score": 0.89, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 697, + 505, + 712 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 131, + 723 + ], + "score": 1.0, + "content": "Since", + "type": "text" + }, + { + "bbox": [ + 132, + 710, + 237, + 721 + ], + "score": 0.9, + "content": "{ \\mathcal { F } } _ { \\mathrm { f e a t } } ( D - 1 ) \\subseteq { \\mathcal { F } } _ { \\mathrm { f e a t } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 708, + 470, + 723 + ], + "score": 1.0, + "content": "we know from the induction hypothesis that this is true if", + "type": "text" + }, + { + "bbox": [ + 471, + 711, + 501, + 720 + ], + "score": 0.88, + "content": "T < D", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 708, + 506, + 723 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 720, + 290, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 160, + 732 + ], + "score": 1.0, + "content": "Now assume", + "type": "text" + }, + { + "bbox": [ + 160, + 721, + 190, + 730 + ], + "score": 0.89, + "content": "T = D", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 720, + 290, + 732 + ], + "score": 1.0, + "content": ". We consider two cases:", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5, + "bbox_fs": [ + 105, + 685, + 506, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 130, + 81, + 504, + 106 + ], + "lines": [ + { + "bbox": [ + 128, + 79, + 505, + 97 + ], + "spans": [ + { + "bbox": [ + 128, + 79, + 151, + 97 + ], + "score": 1.0, + "content": "1. If", + "type": "text" + }, + { + "bbox": [ + 151, + 83, + 180, + 93 + ], + "score": 0.91, + "content": "r _ { 1 } > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 79, + 212, + 97 + ], + "score": 1.0, + "content": ", we set", + "type": "text" + }, + { + "bbox": [ + 212, + 83, + 313, + 95 + ], + "score": 0.92, + "content": "\\tilde { r } = ( r _ { 1 } - 1 , r _ { 2 } , \\dots , r _ { K } )", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 79, + 375, + 97 + ], + "score": 1.0, + "content": ". We know that", + "type": "text" + }, + { + "bbox": [ + 375, + 81, + 491, + 95 + ], + "score": 0.92, + "content": "\\iota \\circ Q ^ { ( \\tilde { r } ) } ( \\bar { X } ) \\in \\mathcal { F } _ { \\mathrm { f e a t } } ( D - 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 491, + 79, + 505, + 97 + ], + "score": 1.0, + "content": "by", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 141, + 93, + 387, + 106 + ], + "spans": [ + { + "bbox": [ + 141, + 93, + 302, + 106 + ], + "score": 1.0, + "content": "the induction hypothesis. So there exist", + "type": "text" + }, + { + "bbox": [ + 302, + 94, + 347, + 105 + ], + "score": 0.93, + "content": "f _ { 2 } , \\ldots , f _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 93, + 387, + 106 + ], + "score": 1.0, + "content": "such that", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "interline_equation", + "bbox": [ + 226, + 111, + 420, + 127 + ], + "lines": [ + { + "bbox": [ + 226, + 111, + 420, + 127 + ], + "spans": [ + { + "bbox": [ + 226, + 111, + 420, + 127 + ], + "score": 0.91, + "content": "\\iota \\circ \\pi _ { V } \\circ f _ { 2 } \\circ . . . \\circ f _ { D } \\circ \\mathrm { e x t } ( \\bar { X } ) = \\iota \\circ Q ^ { ( \\tilde { r } ) } ( \\bar { X } ) .", + "type": "interline_equation", + "image_path": "40b995011f66678151cd2807461802b1c8bf5061fa4a256c935fa6cd4368a147.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 226, + 111, + 420, + 127 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 142, + 133, + 506, + 162 + ], + "lines": [ + { + "bbox": [ + 140, + 131, + 506, + 150 + ], + "spans": [ + { + "bbox": [ + 140, + 131, + 195, + 150 + ], + "score": 1.0, + "content": "Now choose", + "type": "text" + }, + { + "bbox": [ + 195, + 135, + 240, + 146 + ], + "score": 0.91, + "content": "f _ { 1 } \\in \\mathcal { F } _ { m i n }", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 131, + 343, + 150 + ], + "score": 1.0, + "content": "to be the function whose", + "type": "text" + }, + { + "bbox": [ + 344, + 135, + 353, + 145 + ], + "score": 0.82, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 131, + 399, + 150 + ], + "score": 1.0, + "content": "coordinate", + "type": "text" + }, + { + "bbox": [ + 399, + 133, + 454, + 148 + ], + "score": 0.94, + "content": "\\tilde { V } = ( \\tilde { V } _ { j } ) _ { j = 1 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 131, + 506, + 150 + ], + "score": 1.0, + "content": ", is given by", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 142, + 147, + 493, + 162 + ], + "spans": [ + { + "bbox": [ + 142, + 147, + 228, + 162 + ], + "score": 0.93, + "content": "\\tilde { V } _ { j } ( X , V ) = x _ { j } \\otimes V _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 147, + 309, + 162 + ], + "score": 1.0, + "content": ", obtained by setting", + "type": "text" + }, + { + "bbox": [ + 310, + 149, + 370, + 160 + ], + "score": 0.93, + "content": "\\theta _ { 1 } = 1 , \\theta _ { 2 } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 147, + 493, + 162 + ], + "score": 1.0, + "content": "in equation 9. Then , we have", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "interline_equation", + "bbox": [ + 187, + 167, + 457, + 222 + ], + "lines": [ + { + "bbox": [ + 187, + 167, + 457, + 222 + ], + "spans": [ + { + "bbox": [ + 187, + 167, + 457, + 222 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { \\tilde { V } _ { j } ( \\bar { X } , Q ^ { ( \\tilde { r } ) } ( \\bar { X } ) ) = \\displaystyle \\sum _ { i _ { 2 } , \\ldots , i _ { K } = 1 } ^ { n } \\bar { x } _ { j } \\otimes \\bar { x } _ { j } ^ { \\otimes ( r _ { 1 } - 1 ) } \\otimes \\bar { x } _ { i _ { 2 } } ^ { \\otimes r _ { 2 } } \\otimes \\ldots \\otimes \\bar { x } _ { i _ { K } } ^ { \\otimes r _ { K } } } \\\\ & { \\qquad = Q _ { j } ^ { ( \\tilde { r } ) } ( \\bar { X } ) . } \\end{array}", + "type": "interline_equation", + "image_path": "ba9a7d56acf3f3ea11547843107770808066270253a13811a1572470db75093e.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 187, + 167, + 457, + 185.33333333333334 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 187, + 185.33333333333334, + 457, + 203.66666666666669 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 187, + 203.66666666666669, + 457, + 222.00000000000003 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 142, + 225, + 171, + 236 + ], + "lines": [ + { + "bbox": [ + 141, + 224, + 171, + 236 + ], + "spans": [ + { + "bbox": [ + 141, + 224, + 171, + 236 + ], + "score": 1.0, + "content": "and so", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "interline_equation", + "bbox": [ + 142, + 234, + 455, + 278 + ], + "lines": [ + { + "bbox": [ + 157, + 234, + 455, + 278 + ], + "spans": [ + { + "bbox": [ + 157, + 234, + 455, + 278 + ], + "score": 0.66, + "content": "\\begin{array} { c } { \\displaystyle { \\imath \\circ \\pi _ { V } \\circ f _ { 1 } \\circ f _ { 2 } \\circ . . . \\circ f _ { D } \\circ \\mathrm { e x t } ( X - \\frac 1 n X 1 _ { n } \\mathbf { 1 } _ { n } ^ { T } ) = \\imath \\circ Q ^ { ( \\vec { r } ) } ( \\bar { X } ) . } } \\\\ { \\displaystyle { \\mathrm { } } } \\\\ { \\displaystyle { \\mathrm { } \\mathrm { } } \\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac 1 n X 1 _ { n } \\mathbf { 1 } _ { n } ^ { T } ) \\in \\mathcal { F } _ { \\mathrm { f e a t } } ( D ) . } \\end{array}", + "type": "interline_equation", + "image_path": "15d4df6e6d3c363ae960037b26f79b35c7236d0bad9887e341c2287dac072080.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 142, + 234, + 455, + 248.66666666666666 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 142, + 248.66666666666666, + 455, + 263.3333333333333 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 142, + 263.3333333333333, + 455, + 278.0 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "text", + "bbox": [ + 131, + 282, + 505, + 344 + ], + "lines": [ + { + "bbox": [ + 128, + 281, + 505, + 296 + ], + "spans": [ + { + "bbox": [ + 128, + 281, + 151, + 296 + ], + "score": 1.0, + "content": "2. If", + "type": "text" + }, + { + "bbox": [ + 151, + 283, + 180, + 294 + ], + "score": 0.91, + "content": "r _ { 1 } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 281, + 352, + 296 + ], + "score": 1.0, + "content": ". We assume without loss of generality that", + "type": "text" + }, + { + "bbox": [ + 353, + 283, + 381, + 294 + ], + "score": 0.89, + "content": "r _ { 2 } > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 281, + 401, + 296 + ], + "score": 1.0, + "content": ". Set", + "type": "text" + }, + { + "bbox": [ + 401, + 283, + 501, + 295 + ], + "score": 0.9, + "content": "\\tilde { r } = ( r _ { 2 } - 1 , r _ { 3 } , \\dots , r _ { K } )", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 281, + 505, + 296 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 292, + 506, + 307 + ], + "spans": [ + { + "bbox": [ + 141, + 292, + 348, + 307 + ], + "score": 1.0, + "content": "As before by the induction hypothesis there exist", + "type": "text" + }, + { + "bbox": [ + 348, + 294, + 393, + 305 + ], + "score": 0.92, + "content": "f _ { 2 } , \\ldots , f _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 292, + 506, + 307 + ], + "score": 1.0, + "content": "which satisfy equation 19.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 304, + 507, + 320 + ], + "spans": [ + { + "bbox": [ + 141, + 304, + 228, + 320 + ], + "score": 1.0, + "content": "This time we choose", + "type": "text" + }, + { + "bbox": [ + 229, + 306, + 276, + 317 + ], + "score": 0.92, + "content": "f _ { 1 } \\in \\mathcal { F } _ { m i n }", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 304, + 379, + 320 + ], + "score": 1.0, + "content": "to be the function whose", + "type": "text" + }, + { + "bbox": [ + 380, + 306, + 389, + 316 + ], + "score": 0.79, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 304, + 435, + 320 + ], + "score": 1.0, + "content": "coordinate", + "type": "text" + }, + { + "bbox": [ + 436, + 304, + 491, + 319 + ], + "score": 0.93, + "content": "\\tilde { V } = \\tilde { ( } \\tilde { V } _ { j } ) _ { j = 1 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 492, + 304, + 507, + 320 + ], + "score": 1.0, + "content": ", is", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 142, + 318, + 506, + 334 + ], + "spans": [ + { + "bbox": [ + 142, + 318, + 179, + 334 + ], + "score": 1.0, + "content": "given by", + "type": "text" + }, + { + "bbox": [ + 179, + 318, + 280, + 334 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\tilde { V } _ { j } ( X , V ) = \\sum _ { j } x _ { j } \\otimes V _ { j } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 318, + 362, + 334 + ], + "score": 1.0, + "content": ", obtained by setting", + "type": "text" + }, + { + "bbox": [ + 363, + 320, + 423, + 331 + ], + "score": 0.92, + "content": "\\theta _ { 1 } = 0 , \\theta _ { 2 } = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 318, + 506, + 334 + ], + "score": 1.0, + "content": "in equation 9. Then", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 331, + 178, + 344 + ], + "spans": [ + { + "bbox": [ + 141, + 331, + 178, + 344 + ], + "score": 1.0, + "content": "we have", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14 + }, + { + "type": "interline_equation", + "bbox": [ + 178, + 347, + 466, + 437 + ], + "lines": [ + { + "bbox": [ + 178, + 347, + 466, + 437 + ], + "spans": [ + { + "bbox": [ + 178, + 347, + 466, + 437 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { \\tilde { V } _ { j } ( \\bar { X } , Q ^ { ( \\tilde { r } ) } ( \\bar { X } ) ) = \\displaystyle \\sum _ { j = 1 } ^ { n } \\displaystyle \\sum _ { i _ { 3 } , \\ldots , i _ { K } = 1 } ^ { n } \\bar { x } _ { j } \\otimes \\bar { x } _ { j } ^ { \\otimes ( r _ { 2 } - 1 ) } \\otimes \\bar { x } _ { i _ { 3 } } ^ { \\otimes r _ { 2 } } \\otimes \\ldots \\otimes \\bar { x } _ { i _ { K } } ^ { \\otimes r _ { K } } } \\\\ & { \\qquad = \\displaystyle \\sum _ { i _ { 2 } , i _ { 3 } , \\ldots , i _ { K } = 1 } ^ { n } \\bar { x } _ { i _ { 2 } } ^ { \\otimes r _ { 2 } } \\otimes \\bar { x } _ { i _ { 3 } } ^ { \\otimes r _ { 2 } } \\otimes \\ldots \\otimes \\bar { x } _ { i _ { K } } ^ { \\otimes r _ { K } } } \\\\ & { \\qquad = Q _ { j } ^ { ( \\tilde { r } ) } ( \\bar { X } ) . } \\end{array}", + "type": "interline_equation", + "image_path": "06ce0b180180e15501631eabed7866ecd38a9b64670f91d2718ba54d91ad1b0d.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 178, + 347, + 466, + 377.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 178, + 377.0, + 466, + 407.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 178, + 407.0, + 466, + 437.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 142, + 441, + 496, + 456 + ], + "lines": [ + { + "bbox": [ + 140, + 439, + 495, + 458 + ], + "spans": [ + { + "bbox": [ + 140, + 439, + 345, + 458 + ], + "score": 1.0, + "content": "Thus equation 20 holds, and so again we have that", + "type": "text" + }, + { + "bbox": [ + 346, + 442, + 495, + 456 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac { 1 } { n } X 1 _ { n } \\mathbf { 1 } _ { n } ^ { T } ) \\in \\mathcal { F } _ { \\mathrm { f e a t } } ( D ) . } \\end{array}", + "type": "inline_equation" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 496, + 216, + 508 + ], + "lines": [ + { + "bbox": [ + 106, + 496, + 217, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 217, + 510 + ], + "score": 1.0, + "content": "Finally we prove Lemma 4", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 511, + 314, + 524 + ], + "lines": [ + { + "bbox": [ + 105, + 511, + 314, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 225, + 525 + ], + "score": 1.0, + "content": "Lemma 4. If all functions in", + "type": "text" + }, + { + "bbox": [ + 225, + 512, + 241, + 523 + ], + "score": 0.88, + "content": "\\mathcal { Q } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 511, + 314, + 525 + ], + "score": 1.0, + "content": "can be written as", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "interline_equation", + "bbox": [ + 219, + 529, + 391, + 564 + ], + "lines": [ + { + "bbox": [ + 219, + 529, + 391, + 564 + ], + "spans": [ + { + "bbox": [ + 219, + 529, + 391, + 564 + ], + "score": 0.93, + "content": "\\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac { 1 } { n } X 1 _ { n } 1 _ { n } ^ { T } ) = \\sum _ { k = 1 } ^ { K } \\hat { A } _ { k } f _ { k } ( X ) ,", + "type": "interline_equation", + "image_path": "493a0cc1da7307942808fb8357e09c06774e8240571893dcab10a98a54fa1810.jpg" + } + ] + } + ], + "index": 23.5, + "virtual_lines": [ + { + "bbox": [ + 219, + 529, + 391, + 546.5 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 219, + 546.5, + 391, + 564.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 570, + 504, + 596 + ], + "lines": [ + { + "bbox": [ + 102, + 567, + 508, + 591 + ], + "spans": [ + { + "bbox": [ + 102, + 567, + 134, + 591 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 573, + 180, + 584 + ], + "score": 0.73, + "content": "f _ { k } \\in \\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 567, + 185, + 591 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 185, + 572, + 274, + 585 + ], + "score": 0.88, + "content": "A _ { k } : W _ { \\mathrm { f e a t } } \\to W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 567, + 294, + 591 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 295, + 570, + 397, + 586 + ], + "score": 0.92, + "content": "{ \\hat { A } } _ { k } : W _ { \\mathrm { f e a t } } ^ { n } \\to ( W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } } ) ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 567, + 508, + 591 + ], + "score": 1.0, + "content": "is defined by elementwise", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 582, + 286, + 597 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 164, + 597 + ], + "score": 1.0, + "content": "application of", + "type": "text" + }, + { + "bbox": [ + 165, + 585, + 178, + 595 + ], + "score": 0.89, + "content": "A _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 582, + 201, + 597 + ], + "score": 1.0, + "content": ", then", + "type": "text" + }, + { + "bbox": [ + 202, + 585, + 223, + 595 + ], + "score": 0.71, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 582, + 234, + 597 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 234, + 585, + 244, + 594 + ], + "score": 0.56, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 582, + 286, + 597 + ], + "score": 1.0, + "content": "-spanning.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 106, + 610, + 504, + 634 + ], + "lines": [ + { + "bbox": [ + 106, + 611, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 106, + 611, + 320, + 624 + ], + "score": 1.0, + "content": "Proof. If the conditions in Lemma 4 hold, then since", + "type": "text" + }, + { + "bbox": [ + 320, + 612, + 336, + 622 + ], + "score": 0.9, + "content": "\\mathcal { Q } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 611, + 346, + 624 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 347, + 612, + 356, + 621 + ], + "score": 0.84, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 611, + 505, + 624 + ], + "score": 1.0, + "content": "-spanning, every translation invariant", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 622, + 401, + 635 + ], + "spans": [ + { + "bbox": [ + 106, + 622, + 272, + 635 + ], + "score": 1.0, + "content": "and permutation equivariant polynomials", + "type": "text" + }, + { + "bbox": [ + 273, + 624, + 279, + 634 + ], + "score": 0.82, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 622, + 320, + 635 + ], + "score": 1.0, + "content": "of degree", + "type": "text" + }, + { + "bbox": [ + 320, + 623, + 330, + 632 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 622, + 401, + 635 + ], + "score": 1.0, + "content": "can be written as", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5 + }, + { + "type": "interline_equation", + "bbox": [ + 125, + 640, + 487, + 715 + ], + "lines": [ + { + "bbox": [ + 125, + 640, + 487, + 715 + ], + "spans": [ + { + "bbox": [ + 125, + 640, + 487, + 715 + ], + "score": 0.94, + "content": "\\begin{array} { l } { \\displaystyle p ( X ) = \\sum _ { \\vec { r } ^ { \\prime } | | \\vec { r } | | _ { 1 } \\le D } \\hat { \\Lambda } _ { \\vec { r } } \\bigg ( \\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac 1 n X \\mathbf { 1 } _ { n } \\mathbf { 1 } _ { n } ^ { T } ) \\bigg ) = \\sum _ { \\vec { r } ^ { \\prime } | | \\vec { r } ^ { \\prime } | | _ { 1 } \\le D } \\hat { \\Lambda } _ { \\vec { r } } \\bigg ( \\sum _ { k = 1 } ^ { K _ { \\vec { r } } } \\iota \\circ \\hat { A } _ { k , \\vec { r } } f _ { k , \\vec { r } } ( X ) \\bigg ) } \\\\ { = \\sum _ { \\vec { r } ^ { \\prime } | | \\vec { r } | | _ { 1 } \\le D } \\sum _ { k = 1 } ^ { K _ { \\vec { r } } } \\hat { \\Lambda } _ { k , \\vec { r } } ( f _ { k , \\vec { r } } ( X ) ) } \\end{array}", + "type": "interline_equation", + "image_path": "7ad44bc5e953bd5e76d62a99ae401620be4a17c0feaaf999df5c1d81b653db59.jpg" + } + ] + } + ], + "index": 30, + "virtual_lines": [ + { + "bbox": [ + 125, + 640, + 487, + 665.0 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 125, + 665.0, + 487, + 690.0 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 125, + 690.0, + 487, + 715.0 + ], + "spans": [], + "index": 31 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 720, + 417, + 733 + ], + "lines": [ + { + "bbox": [ + 105, + 717, + 418, + 735 + ], + "spans": [ + { + "bbox": [ + 105, + 717, + 177, + 735 + ], + "score": 1.0, + "content": "where we denote", + "type": "text" + }, + { + "bbox": [ + 177, + 721, + 262, + 733 + ], + "score": 0.92, + "content": "\\Lambda _ { k , \\vec { r } } = \\Lambda _ { \\vec { r } } \\circ \\iota \\circ A _ { k , \\vec { r } } .", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 717, + 332, + 735 + ], + "score": 1.0, + "content": "Thus we proved", + "type": "text" + }, + { + "bbox": [ + 333, + 721, + 354, + 732 + ], + "score": 0.9, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 717, + 365, + 735 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 365, + 721, + 374, + 730 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 717, + 418, + 735 + ], + "score": 1.0, + "content": "-spanning.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + } + ], + "page_idx": 14, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "15", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 721, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 496, + 722, + 504, + 731 + ], + "spans": [ + { + "bbox": [ + 496, + 722, + 504, + 731 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 469, + 505, + 480 + ], + "lines": [ + { + "bbox": [ + 495, + 470, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 495, + 470, + 505, + 480 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 130, + 81, + 504, + 106 + ], + "lines": [ + { + "bbox": [ + 128, + 79, + 505, + 97 + ], + "spans": [ + { + "bbox": [ + 128, + 79, + 151, + 97 + ], + "score": 1.0, + "content": "1. If", + "type": "text" + }, + { + "bbox": [ + 151, + 83, + 180, + 93 + ], + "score": 0.91, + "content": "r _ { 1 } > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 79, + 212, + 97 + ], + "score": 1.0, + "content": ", we set", + "type": "text" + }, + { + "bbox": [ + 212, + 83, + 313, + 95 + ], + "score": 0.92, + "content": "\\tilde { r } = ( r _ { 1 } - 1 , r _ { 2 } , \\dots , r _ { K } )", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 79, + 375, + 97 + ], + "score": 1.0, + "content": ". We know that", + "type": "text" + }, + { + "bbox": [ + 375, + 81, + 491, + 95 + ], + "score": 0.92, + "content": "\\iota \\circ Q ^ { ( \\tilde { r } ) } ( \\bar { X } ) \\in \\mathcal { F } _ { \\mathrm { f e a t } } ( D - 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 491, + 79, + 505, + 97 + ], + "score": 1.0, + "content": "by", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 141, + 93, + 387, + 106 + ], + "spans": [ + { + "bbox": [ + 141, + 93, + 302, + 106 + ], + "score": 1.0, + "content": "the induction hypothesis. So there exist", + "type": "text" + }, + { + "bbox": [ + 302, + 94, + 347, + 105 + ], + "score": 0.93, + "content": "f _ { 2 } , \\ldots , f _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 93, + 387, + 106 + ], + "score": 1.0, + "content": "such that", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 128, + 79, + 505, + 106 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 226, + 111, + 420, + 127 + ], + "lines": [ + { + "bbox": [ + 226, + 111, + 420, + 127 + ], + "spans": [ + { + "bbox": [ + 226, + 111, + 420, + 127 + ], + "score": 0.91, + "content": "\\iota \\circ \\pi _ { V } \\circ f _ { 2 } \\circ . . . \\circ f _ { D } \\circ \\mathrm { e x t } ( \\bar { X } ) = \\iota \\circ Q ^ { ( \\tilde { r } ) } ( \\bar { X } ) .", + "type": "interline_equation", + "image_path": "40b995011f66678151cd2807461802b1c8bf5061fa4a256c935fa6cd4368a147.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 226, + 111, + 420, + 127 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 142, + 133, + 506, + 162 + ], + "lines": [ + { + "bbox": [ + 140, + 131, + 506, + 150 + ], + "spans": [ + { + "bbox": [ + 140, + 131, + 195, + 150 + ], + "score": 1.0, + "content": "Now choose", + "type": "text" + }, + { + "bbox": [ + 195, + 135, + 240, + 146 + ], + "score": 0.91, + "content": "f _ { 1 } \\in \\mathcal { F } _ { m i n }", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 131, + 343, + 150 + ], + "score": 1.0, + "content": "to be the function whose", + "type": "text" + }, + { + "bbox": [ + 344, + 135, + 353, + 145 + ], + "score": 0.82, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 131, + 399, + 150 + ], + "score": 1.0, + "content": "coordinate", + "type": "text" + }, + { + "bbox": [ + 399, + 133, + 454, + 148 + ], + "score": 0.94, + "content": "\\tilde { V } = ( \\tilde { V } _ { j } ) _ { j = 1 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 131, + 506, + 150 + ], + "score": 1.0, + "content": ", is given by", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 142, + 147, + 493, + 162 + ], + "spans": [ + { + "bbox": [ + 142, + 147, + 228, + 162 + ], + "score": 0.93, + "content": "\\tilde { V } _ { j } ( X , V ) = x _ { j } \\otimes V _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 147, + 309, + 162 + ], + "score": 1.0, + "content": ", obtained by setting", + "type": "text" + }, + { + "bbox": [ + 310, + 149, + 370, + 160 + ], + "score": 0.93, + "content": "\\theta _ { 1 } = 1 , \\theta _ { 2 } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 147, + 493, + 162 + ], + "score": 1.0, + "content": "in equation 9. Then , we have", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 140, + 131, + 506, + 162 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 187, + 167, + 457, + 222 + ], + "lines": [ + { + "bbox": [ + 187, + 167, + 457, + 222 + ], + "spans": [ + { + "bbox": [ + 187, + 167, + 457, + 222 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { \\tilde { V } _ { j } ( \\bar { X } , Q ^ { ( \\tilde { r } ) } ( \\bar { X } ) ) = \\displaystyle \\sum _ { i _ { 2 } , \\ldots , i _ { K } = 1 } ^ { n } \\bar { x } _ { j } \\otimes \\bar { x } _ { j } ^ { \\otimes ( r _ { 1 } - 1 ) } \\otimes \\bar { x } _ { i _ { 2 } } ^ { \\otimes r _ { 2 } } \\otimes \\ldots \\otimes \\bar { x } _ { i _ { K } } ^ { \\otimes r _ { K } } } \\\\ & { \\qquad = Q _ { j } ^ { ( \\tilde { r } ) } ( \\bar { X } ) . } \\end{array}", + "type": "interline_equation", + "image_path": "ba9a7d56acf3f3ea11547843107770808066270253a13811a1572470db75093e.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 187, + 167, + 457, + 185.33333333333334 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 187, + 185.33333333333334, + 457, + 203.66666666666669 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 187, + 203.66666666666669, + 457, + 222.00000000000003 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 142, + 225, + 171, + 236 + ], + "lines": [ + { + "bbox": [ + 141, + 224, + 171, + 236 + ], + "spans": [ + { + "bbox": [ + 141, + 224, + 171, + 236 + ], + "score": 1.0, + "content": "and so", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 141, + 224, + 171, + 236 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 142, + 234, + 455, + 278 + ], + "lines": [ + { + "bbox": [ + 157, + 234, + 455, + 278 + ], + "spans": [ + { + "bbox": [ + 157, + 234, + 455, + 278 + ], + "score": 0.66, + "content": "\\begin{array} { c } { \\displaystyle { \\imath \\circ \\pi _ { V } \\circ f _ { 1 } \\circ f _ { 2 } \\circ . . . \\circ f _ { D } \\circ \\mathrm { e x t } ( X - \\frac 1 n X 1 _ { n } \\mathbf { 1 } _ { n } ^ { T } ) = \\imath \\circ Q ^ { ( \\vec { r } ) } ( \\bar { X } ) . } } \\\\ { \\displaystyle { \\mathrm { } } } \\\\ { \\displaystyle { \\mathrm { } \\mathrm { } } \\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac 1 n X 1 _ { n } \\mathbf { 1 } _ { n } ^ { T } ) \\in \\mathcal { F } _ { \\mathrm { f e a t } } ( D ) . } \\end{array}", + "type": "interline_equation", + "image_path": "15d4df6e6d3c363ae960037b26f79b35c7236d0bad9887e341c2287dac072080.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 142, + 234, + 455, + 248.66666666666666 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 142, + 248.66666666666666, + 455, + 263.3333333333333 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 142, + 263.3333333333333, + 455, + 278.0 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "text", + "bbox": [ + 131, + 282, + 505, + 344 + ], + "lines": [ + { + "bbox": [ + 128, + 281, + 505, + 296 + ], + "spans": [ + { + "bbox": [ + 128, + 281, + 151, + 296 + ], + "score": 1.0, + "content": "2. If", + "type": "text" + }, + { + "bbox": [ + 151, + 283, + 180, + 294 + ], + "score": 0.91, + "content": "r _ { 1 } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 281, + 352, + 296 + ], + "score": 1.0, + "content": ". We assume without loss of generality that", + "type": "text" + }, + { + "bbox": [ + 353, + 283, + 381, + 294 + ], + "score": 0.89, + "content": "r _ { 2 } > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 281, + 401, + 296 + ], + "score": 1.0, + "content": ". Set", + "type": "text" + }, + { + "bbox": [ + 401, + 283, + 501, + 295 + ], + "score": 0.9, + "content": "\\tilde { r } = ( r _ { 2 } - 1 , r _ { 3 } , \\dots , r _ { K } )", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 281, + 505, + 296 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 292, + 506, + 307 + ], + "spans": [ + { + "bbox": [ + 141, + 292, + 348, + 307 + ], + "score": 1.0, + "content": "As before by the induction hypothesis there exist", + "type": "text" + }, + { + "bbox": [ + 348, + 294, + 393, + 305 + ], + "score": 0.92, + "content": "f _ { 2 } , \\ldots , f _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 292, + 506, + 307 + ], + "score": 1.0, + "content": "which satisfy equation 19.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 304, + 507, + 320 + ], + "spans": [ + { + "bbox": [ + 141, + 304, + 228, + 320 + ], + "score": 1.0, + "content": "This time we choose", + "type": "text" + }, + { + "bbox": [ + 229, + 306, + 276, + 317 + ], + "score": 0.92, + "content": "f _ { 1 } \\in \\mathcal { F } _ { m i n }", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 304, + 379, + 320 + ], + "score": 1.0, + "content": "to be the function whose", + "type": "text" + }, + { + "bbox": [ + 380, + 306, + 389, + 316 + ], + "score": 0.79, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 304, + 435, + 320 + ], + "score": 1.0, + "content": "coordinate", + "type": "text" + }, + { + "bbox": [ + 436, + 304, + 491, + 319 + ], + "score": 0.93, + "content": "\\tilde { V } = \\tilde { ( } \\tilde { V } _ { j } ) _ { j = 1 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 492, + 304, + 507, + 320 + ], + "score": 1.0, + "content": ", is", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 142, + 318, + 506, + 334 + ], + "spans": [ + { + "bbox": [ + 142, + 318, + 179, + 334 + ], + "score": 1.0, + "content": "given by", + "type": "text" + }, + { + "bbox": [ + 179, + 318, + 280, + 334 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\tilde { V } _ { j } ( X , V ) = \\sum _ { j } x _ { j } \\otimes V _ { j } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 318, + 362, + 334 + ], + "score": 1.0, + "content": ", obtained by setting", + "type": "text" + }, + { + "bbox": [ + 363, + 320, + 423, + 331 + ], + "score": 0.92, + "content": "\\theta _ { 1 } = 0 , \\theta _ { 2 } = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 318, + 506, + 334 + ], + "score": 1.0, + "content": "in equation 9. Then", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 331, + 178, + 344 + ], + "spans": [ + { + "bbox": [ + 141, + 331, + 178, + 344 + ], + "score": 1.0, + "content": "we have", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14, + "bbox_fs": [ + 128, + 281, + 507, + 344 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 178, + 347, + 466, + 437 + ], + "lines": [ + { + "bbox": [ + 178, + 347, + 466, + 437 + ], + "spans": [ + { + "bbox": [ + 178, + 347, + 466, + 437 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { \\tilde { V } _ { j } ( \\bar { X } , Q ^ { ( \\tilde { r } ) } ( \\bar { X } ) ) = \\displaystyle \\sum _ { j = 1 } ^ { n } \\displaystyle \\sum _ { i _ { 3 } , \\ldots , i _ { K } = 1 } ^ { n } \\bar { x } _ { j } \\otimes \\bar { x } _ { j } ^ { \\otimes ( r _ { 2 } - 1 ) } \\otimes \\bar { x } _ { i _ { 3 } } ^ { \\otimes r _ { 2 } } \\otimes \\ldots \\otimes \\bar { x } _ { i _ { K } } ^ { \\otimes r _ { K } } } \\\\ & { \\qquad = \\displaystyle \\sum _ { i _ { 2 } , i _ { 3 } , \\ldots , i _ { K } = 1 } ^ { n } \\bar { x } _ { i _ { 2 } } ^ { \\otimes r _ { 2 } } \\otimes \\bar { x } _ { i _ { 3 } } ^ { \\otimes r _ { 2 } } \\otimes \\ldots \\otimes \\bar { x } _ { i _ { K } } ^ { \\otimes r _ { K } } } \\\\ & { \\qquad = Q _ { j } ^ { ( \\tilde { r } ) } ( \\bar { X } ) . } \\end{array}", + "type": "interline_equation", + "image_path": "06ce0b180180e15501631eabed7866ecd38a9b64670f91d2718ba54d91ad1b0d.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 178, + 347, + 466, + 377.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 178, + 377.0, + 466, + 407.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 178, + 407.0, + 466, + 437.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 142, + 441, + 496, + 456 + ], + "lines": [ + { + "bbox": [ + 140, + 439, + 495, + 458 + ], + "spans": [ + { + "bbox": [ + 140, + 439, + 345, + 458 + ], + "score": 1.0, + "content": "Thus equation 20 holds, and so again we have that", + "type": "text" + }, + { + "bbox": [ + 346, + 442, + 495, + 456 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac { 1 } { n } X 1 _ { n } \\mathbf { 1 } _ { n } ^ { T } ) \\in \\mathcal { F } _ { \\mathrm { f e a t } } ( D ) . } \\end{array}", + "type": "inline_equation" + } + ], + "index": 20 + } + ], + "index": 20, + "bbox_fs": [ + 140, + 439, + 495, + 458 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 496, + 216, + 508 + ], + "lines": [ + { + "bbox": [ + 106, + 496, + 217, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 217, + 510 + ], + "score": 1.0, + "content": "Finally we prove Lemma 4", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21, + "bbox_fs": [ + 106, + 496, + 217, + 510 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 511, + 314, + 524 + ], + "lines": [ + { + "bbox": [ + 105, + 511, + 314, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 225, + 525 + ], + "score": 1.0, + "content": "Lemma 4. If all functions in", + "type": "text" + }, + { + "bbox": [ + 225, + 512, + 241, + 523 + ], + "score": 0.88, + "content": "\\mathcal { Q } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 511, + 314, + 525 + ], + "score": 1.0, + "content": "can be written as", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 511, + 314, + 525 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 219, + 529, + 391, + 564 + ], + "lines": [ + { + "bbox": [ + 219, + 529, + 391, + 564 + ], + "spans": [ + { + "bbox": [ + 219, + 529, + 391, + 564 + ], + "score": 0.93, + "content": "\\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac { 1 } { n } X 1 _ { n } 1 _ { n } ^ { T } ) = \\sum _ { k = 1 } ^ { K } \\hat { A } _ { k } f _ { k } ( X ) ,", + "type": "interline_equation", + "image_path": "493a0cc1da7307942808fb8357e09c06774e8240571893dcab10a98a54fa1810.jpg" + } + ] + } + ], + "index": 23.5, + "virtual_lines": [ + { + "bbox": [ + 219, + 529, + 391, + 546.5 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 219, + 546.5, + 391, + 564.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 570, + 504, + 596 + ], + "lines": [ + { + "bbox": [ + 102, + 567, + 508, + 591 + ], + "spans": [ + { + "bbox": [ + 102, + 567, + 134, + 591 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 573, + 180, + 584 + ], + "score": 0.73, + "content": "f _ { k } \\in \\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 567, + 185, + 591 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 185, + 572, + 274, + 585 + ], + "score": 0.88, + "content": "A _ { k } : W _ { \\mathrm { f e a t } } \\to W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 567, + 294, + 591 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 295, + 570, + 397, + 586 + ], + "score": 0.92, + "content": "{ \\hat { A } } _ { k } : W _ { \\mathrm { f e a t } } ^ { n } \\to ( W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } } ) ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 567, + 508, + 591 + ], + "score": 1.0, + "content": "is defined by elementwise", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 582, + 286, + 597 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 164, + 597 + ], + "score": 1.0, + "content": "application of", + "type": "text" + }, + { + "bbox": [ + 165, + 585, + 178, + 595 + ], + "score": 0.89, + "content": "A _ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 582, + 201, + 597 + ], + "score": 1.0, + "content": ", then", + "type": "text" + }, + { + "bbox": [ + 202, + 585, + 223, + 595 + ], + "score": 0.71, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 582, + 234, + 597 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 234, + 585, + 244, + 594 + ], + "score": 0.56, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 582, + 286, + 597 + ], + "score": 1.0, + "content": "-spanning.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 102, + 567, + 508, + 597 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 610, + 504, + 634 + ], + "lines": [ + { + "bbox": [ + 106, + 611, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 106, + 611, + 320, + 624 + ], + "score": 1.0, + "content": "Proof. If the conditions in Lemma 4 hold, then since", + "type": "text" + }, + { + "bbox": [ + 320, + 612, + 336, + 622 + ], + "score": 0.9, + "content": "\\mathcal { Q } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 611, + 346, + 624 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 347, + 612, + 356, + 621 + ], + "score": 0.84, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 611, + 505, + 624 + ], + "score": 1.0, + "content": "-spanning, every translation invariant", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 622, + 401, + 635 + ], + "spans": [ + { + "bbox": [ + 106, + 622, + 272, + 635 + ], + "score": 1.0, + "content": "and permutation equivariant polynomials", + "type": "text" + }, + { + "bbox": [ + 273, + 624, + 279, + 634 + ], + "score": 0.82, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 622, + 320, + 635 + ], + "score": 1.0, + "content": "of degree", + "type": "text" + }, + { + "bbox": [ + 320, + 623, + 330, + 632 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 622, + 401, + 635 + ], + "score": 1.0, + "content": "can be written as", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5, + "bbox_fs": [ + 106, + 611, + 505, + 635 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 125, + 640, + 487, + 715 + ], + "lines": [ + { + "bbox": [ + 125, + 640, + 487, + 715 + ], + "spans": [ + { + "bbox": [ + 125, + 640, + 487, + 715 + ], + "score": 0.94, + "content": "\\begin{array} { l } { \\displaystyle p ( X ) = \\sum _ { \\vec { r } ^ { \\prime } | | \\vec { r } | | _ { 1 } \\le D } \\hat { \\Lambda } _ { \\vec { r } } \\bigg ( \\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac 1 n X \\mathbf { 1 } _ { n } \\mathbf { 1 } _ { n } ^ { T } ) \\bigg ) = \\sum _ { \\vec { r } ^ { \\prime } | | \\vec { r } ^ { \\prime } | | _ { 1 } \\le D } \\hat { \\Lambda } _ { \\vec { r } } \\bigg ( \\sum _ { k = 1 } ^ { K _ { \\vec { r } } } \\iota \\circ \\hat { A } _ { k , \\vec { r } } f _ { k , \\vec { r } } ( X ) \\bigg ) } \\\\ { = \\sum _ { \\vec { r } ^ { \\prime } | | \\vec { r } | | _ { 1 } \\le D } \\sum _ { k = 1 } ^ { K _ { \\vec { r } } } \\hat { \\Lambda } _ { k , \\vec { r } } ( f _ { k , \\vec { r } } ( X ) ) } \\end{array}", + "type": "interline_equation", + "image_path": "7ad44bc5e953bd5e76d62a99ae401620be4a17c0feaaf999df5c1d81b653db59.jpg" + } + ] + } + ], + "index": 30, + "virtual_lines": [ + { + "bbox": [ + 125, + 640, + 487, + 665.0 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 125, + 665.0, + 487, + 690.0 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 125, + 690.0, + 487, + 715.0 + ], + "spans": [], + "index": 31 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 720, + 417, + 733 + ], + "lines": [ + { + "bbox": [ + 105, + 717, + 418, + 735 + ], + "spans": [ + { + "bbox": [ + 105, + 717, + 177, + 735 + ], + "score": 1.0, + "content": "where we denote", + "type": "text" + }, + { + "bbox": [ + 177, + 721, + 262, + 733 + ], + "score": 0.92, + "content": "\\Lambda _ { k , \\vec { r } } = \\Lambda _ { \\vec { r } } \\circ \\iota \\circ A _ { k , \\vec { r } } .", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 717, + 332, + 735 + ], + "score": 1.0, + "content": "Thus we proved", + "type": "text" + }, + { + "bbox": [ + 333, + 721, + 354, + 732 + ], + "score": 0.9, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 717, + 365, + 735 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 365, + 721, + 374, + 730 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 717, + 418, + 735 + ], + "score": 1.0, + "content": "-spanning.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32, + "bbox_fs": [ + 105, + 717, + 418, + 735 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 80, + 252, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 80, + 252, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 80, + 252, + 96 + ], + "score": 1.0, + "content": "D PROOFS FOR SECTION 5", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 106, + 187, + 118 + ], + "lines": [ + { + "bbox": [ + 106, + 105, + 188, + 119 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 188, + 119 + ], + "score": 1.0, + "content": "We prove Lemma 5", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 106, + 121, + 505, + 162 + ], + "lines": [ + { + "bbox": [ + 103, + 115, + 494, + 145 + ], + "spans": [ + { + "bbox": [ + 103, + 115, + 169, + 145 + ], + "score": 1.0, + "content": "Lemma 5. Let", + "type": "text" + }, + { + "bbox": [ + 169, + 121, + 258, + 137 + ], + "score": 0.9, + "content": "l ^ { ( 1 ) } = ( \\ell _ { 1 } ^ { ( 1 ) } , \\dots , \\ell _ { K _ { 1 } } ^ { ( 1 ) } )", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 115, + 261, + 145 + ], + "score": 0.0, + "content": "", + "type": "text" + }, + { + "bbox": [ + 277, + 122, + 366, + 137 + ], + "score": 0.89, + "content": "l ^ { ( 2 ) } = ( \\ell _ { 1 } ^ { ( 2 ) } , \\dots , \\ell _ { K _ { 2 } } ^ { ( 2 ) } )", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 115, + 414, + 145 + ], + "score": 1.0, + "content": ". A function", + "type": "text" + }, + { + "bbox": [ + 414, + 124, + 494, + 136 + ], + "score": 0.9, + "content": "\\boldsymbol { \\Lambda } = \\left( \\Lambda _ { 1 } , \\ldots , \\Lambda _ { K _ { 2 } } \\right)", + "type": "inline_equation" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 135, + 506, + 148 + ], + "spans": [ + { + "bbox": [ + 105, + 135, + 261, + 148 + ], + "score": 1.0, + "content": "a linear equivariant mapping between", + "type": "text" + }, + { + "bbox": [ + 261, + 136, + 283, + 147 + ], + "score": 0.88, + "content": "W _ { l ^ { ( 1 ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 283, + 135, + 303, + 148 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 303, + 137, + 325, + 147 + ], + "score": 0.88, + "content": "W _ { l ^ { ( 2 ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 135, + 436, + 148 + ], + "score": 1.0, + "content": ", if and only if there exists a", + "type": "text" + }, + { + "bbox": [ + 437, + 137, + 475, + 146 + ], + "score": 0.89, + "content": "K _ { 1 } \\times K _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 135, + 506, + 148 + ], + "score": 1.0, + "content": "matrix", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 107, + 143, + 307, + 165 + ], + "spans": [ + { + "bbox": [ + 107, + 149, + 119, + 160 + ], + "score": 0.69, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 143, + 139, + 165 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 140, + 149, + 176, + 162 + ], + "score": 0.91, + "content": "M _ { i j } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 143, + 217, + 165 + ], + "score": 1.0, + "content": "whenever", + "type": "text" + }, + { + "bbox": [ + 218, + 146, + 261, + 163 + ], + "score": 0.94, + "content": "\\ell _ { i } ^ { ( 1 ) } \\neq \\ell _ { j } ^ { ( 2 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 143, + 307, + 165 + ], + "score": 1.0, + "content": ", such that", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3 + }, + { + "type": "interline_equation", + "bbox": [ + 263, + 169, + 348, + 203 + ], + "lines": [ + { + "bbox": [ + 263, + 169, + 348, + 203 + ], + "spans": [ + { + "bbox": [ + 263, + 169, + 348, + 203 + ], + "score": 0.94, + "content": "\\Lambda _ { j } ( V ) = \\sum _ { i = 1 } ^ { K _ { 1 } } M _ { i j } V _ { i }", + "type": "interline_equation", + "image_path": "c86fa12e12e78bc2dff340de63d341d041103c9ad97279aa1d5349fbec1ba3c2.jpg" + } + ] + } + ], + "index": 5.5, + "virtual_lines": [ + { + "bbox": [ + 263, + 169, + 348, + 186.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 263, + 186.0, + 348, + 203.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 208, + 338, + 225 + ], + "lines": [ + { + "bbox": [ + 104, + 206, + 335, + 227 + ], + "spans": [ + { + "bbox": [ + 104, + 206, + 133, + 226 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 209, + 185, + 223 + ], + "score": 0.93, + "content": "V = ( V _ { i } ) _ { i = 1 } ^ { K _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 206, + 204, + 226 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 205, + 210, + 248, + 227 + ], + "score": 0.92, + "content": "V _ { i } \\in W _ { \\ell _ { i } ^ { ( 1 ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 206, + 278, + 226 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 278, + 210, + 335, + 222 + ], + "score": 0.91, + "content": "i = 1 , \\ldots , K _ { 1 }", + "type": "inline_equation" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 106, + 239, + 505, + 273 + ], + "lines": [ + { + "bbox": [ + 105, + 239, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 505, + 253 + ], + "score": 1.0, + "content": "Proof. As mentioned in the main text, this lemma is based on Schur’s lemma. This lemma is typi-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 251, + 504, + 263 + ], + "spans": [ + { + "bbox": [ + 106, + 251, + 504, + 263 + ], + "score": 1.0, + "content": "cally stated for complex representations, but holds for odd dimensional real representation as well.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 261, + 482, + 274 + ], + "spans": [ + { + "bbox": [ + 106, + 261, + 482, + 274 + ], + "score": 1.0, + "content": "We recount the lemma and its proof here for completeness (see also (Fulton & Harris, 2013)).", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 104, + 280, + 505, + 304 + ], + "lines": [ + { + "bbox": [ + 105, + 278, + 507, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 278, + 299, + 295 + ], + "score": 1.0, + "content": "Lemma D.1 (Schur’s Lemma for SO(3)). Let", + "type": "text" + }, + { + "bbox": [ + 299, + 281, + 371, + 293 + ], + "score": 0.92, + "content": "\\Lambda : W _ { \\ell _ { 1 } } \\to W _ { \\ell _ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 278, + 507, + 295 + ], + "score": 1.0, + "content": "be a linear equivariant map. If", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 291, + 381, + 304 + ], + "spans": [ + { + "bbox": [ + 106, + 291, + 138, + 303 + ], + "score": 0.92, + "content": "\\ell _ { 1 } \\neq \\ell _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 291, + 159, + 304 + ], + "score": 1.0, + "content": "then", + "type": "text" + }, + { + "bbox": [ + 159, + 292, + 185, + 302 + ], + "score": 0.88, + "content": "\\Lambda = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 291, + 233, + 304 + ], + "score": 1.0, + "content": ". Otherwise", + "type": "text" + }, + { + "bbox": [ + 234, + 292, + 242, + 302 + ], + "score": 0.43, + "content": "\\Lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 291, + 381, + 304 + ], + "score": 1.0, + "content": "is a scalar multiply of the identity.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 106, + 316, + 506, + 373 + ], + "lines": [ + { + "bbox": [ + 106, + 317, + 506, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 152, + 330 + ], + "score": 1.0, + "content": "Proof. Let", + "type": "text" + }, + { + "bbox": [ + 153, + 318, + 222, + 329 + ], + "score": 0.92, + "content": "\\Lambda : W _ { \\ell _ { 1 } } \\to W _ { \\ell _ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 317, + 442, + 330 + ], + "score": 1.0, + "content": "be a linear equivariant map. The image and kernel of", + "type": "text" + }, + { + "bbox": [ + 443, + 318, + 451, + 327 + ], + "score": 0.75, + "content": "\\Lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 317, + 506, + 330 + ], + "score": 1.0, + "content": "are invariant", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 328, + 506, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 161, + 342 + ], + "score": 1.0, + "content": "subspaces of", + "type": "text" + }, + { + "bbox": [ + 161, + 329, + 179, + 340 + ], + "score": 0.9, + "content": "W _ { \\ell _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 328, + 199, + 342 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 199, + 330, + 217, + 340 + ], + "score": 0.9, + "content": "W _ { \\ell _ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 328, + 346, + 342 + ], + "score": 1.0, + "content": ", respectively. It follows that if", + "type": "text" + }, + { + "bbox": [ + 346, + 329, + 375, + 340 + ], + "score": 0.92, + "content": "\\Lambda \\neq 0", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 328, + 397, + 342 + ], + "score": 1.0, + "content": "then", + "type": "text" + }, + { + "bbox": [ + 397, + 329, + 405, + 339 + ], + "score": 0.79, + "content": "\\Lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 328, + 506, + 342 + ], + "score": 1.0, + "content": "is a linear isomorphism", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 339, + 506, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 166, + 352 + ], + "score": 1.0, + "content": "so necessarily", + "type": "text" + }, + { + "bbox": [ + 166, + 340, + 201, + 351 + ], + "score": 0.92, + "content": "\\ell _ { 1 } = \\ell _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 339, + 262, + 352 + ], + "score": 1.0, + "content": ". Now assume", + "type": "text" + }, + { + "bbox": [ + 263, + 340, + 297, + 351 + ], + "score": 0.92, + "content": "\\ell _ { 1 } = \\ell _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 339, + 402, + 352 + ], + "score": 1.0, + "content": ". Since the dimension of", + "type": "text" + }, + { + "bbox": [ + 402, + 340, + 419, + 352 + ], + "score": 0.91, + "content": "W _ { \\ell _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 339, + 453, + 352 + ], + "score": 1.0, + "content": "is odd,", + "type": "text" + }, + { + "bbox": [ + 453, + 340, + 461, + 350 + ], + "score": 0.74, + "content": "\\Lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 339, + 506, + 352 + ], + "score": 1.0, + "content": "has a real", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 350, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 151, + 363 + ], + "score": 1.0, + "content": "eigenvalue", + "type": "text" + }, + { + "bbox": [ + 152, + 351, + 159, + 361 + ], + "score": 0.75, + "content": "\\lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 350, + 244, + 363 + ], + "score": 1.0, + "content": ". The linear function", + "type": "text" + }, + { + "bbox": [ + 244, + 351, + 276, + 361 + ], + "score": 0.91, + "content": "\\Lambda - \\lambda I", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 350, + 461, + 363 + ], + "score": 1.0, + "content": "is equivariant and has a non-trivial kernel, so", + "type": "text" + }, + { + "bbox": [ + 461, + 351, + 505, + 361 + ], + "score": 0.88, + "content": "\\Lambda - \\lambda I =", + "type": "inline_equation" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 361, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 117, + 374 + ], + "score": 1.0, + "content": "0.", + "type": "text" + }, + { + "bbox": [ + 494, + 362, + 505, + 372 + ], + "score": 1.0, + "content": "□", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 107, + 386, + 504, + 417 + ], + "lines": [ + { + "bbox": [ + 105, + 386, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 105, + 386, + 338, + 401 + ], + "score": 1.0, + "content": "We now return to the proof of Lemma 5. Note that each", + "type": "text" + }, + { + "bbox": [ + 338, + 388, + 421, + 401 + ], + "score": 0.88, + "content": "\\Lambda _ { j } : W _ { l ^ { ( 1 ) } } \\to W _ { \\ell _ { \\vec { \\cdot } } ^ { ( 2 ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 386, + 505, + 401 + ], + "score": 1.0, + "content": "is linear and SO(3)", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 402, + 496, + 418 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 300, + 416 + ], + "score": 1.0, + "content": "equivariant. Next denote the restrictions of each", + "type": "text" + }, + { + "bbox": [ + 300, + 403, + 312, + 415 + ], + "score": 0.87, + "content": "\\Lambda _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 402, + 324, + 416 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 325, + 402, + 409, + 418 + ], + "score": 0.89, + "content": "W _ { \\ell _ { i } ^ { ( 1 ) } } , i = 1 , \\dots , K _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 402, + 423, + 416 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 423, + 403, + 437, + 415 + ], + "score": 0.88, + "content": "\\Lambda _ { i j }", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 402, + 496, + 416 + ], + "score": 1.0, + "content": ", and note that", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "interline_equation", + "bbox": [ + 239, + 424, + 372, + 459 + ], + "lines": [ + { + "bbox": [ + 239, + 424, + 372, + 459 + ], + "spans": [ + { + "bbox": [ + 239, + 424, + 372, + 459 + ], + "score": 0.95, + "content": "\\Lambda _ { j } ( V _ { 1 } , \\dots , V _ { K _ { 1 } } ) = \\sum _ { i = 1 } ^ { K _ { 1 } } \\Lambda _ { i j } ( V _ { i } ) .", + "type": "interline_equation", + "image_path": "1ba83e118e1d6d86913cdc6af9f0324cb8056d6b0463bb01413adffe4b0e3059.jpg" + } + ] + } + ], + "index": 20.5, + "virtual_lines": [ + { + "bbox": [ + 239, + 424, + 372, + 441.5 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 239, + 441.5, + 372, + 459.0 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 463, + 505, + 510 + ], + "lines": [ + { + "bbox": [ + 104, + 460, + 505, + 481 + ], + "spans": [ + { + "bbox": [ + 104, + 460, + 212, + 481 + ], + "score": 1.0, + "content": "By considering vectors in", + "type": "text" + }, + { + "bbox": [ + 212, + 465, + 235, + 477 + ], + "score": 0.92, + "content": "W _ { l ^ { ( 1 ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 460, + 285, + 481 + ], + "score": 1.0, + "content": "of the form", + "type": "text" + }, + { + "bbox": [ + 286, + 464, + 373, + 477 + ], + "score": 0.91, + "content": "( 0 , \\ldots , 0 , V _ { i } , 0 \\ldots , 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 460, + 443, + 481 + ], + "score": 1.0, + "content": "we see that each", + "type": "text" + }, + { + "bbox": [ + 444, + 464, + 505, + 480 + ], + "score": 0.91, + "content": "\\Lambda _ { i j } : W _ { \\ell _ { i } ^ { ( 1 ) } } \\to", + "type": "inline_equation" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 476, + 504, + 499 + ], + "spans": [ + { + "bbox": [ + 106, + 482, + 129, + 499 + ], + "score": 0.91, + "content": "W _ { \\ell _ { j } ^ { ( 2 ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 130, + 476, + 368, + 499 + ], + "score": 1.0, + "content": "is linear and SO(3)-equivariant. Thus by Schur’s lemma, if", + "type": "text" + }, + { + "bbox": [ + 368, + 479, + 412, + 496 + ], + "score": 0.94, + "content": "\\ell _ { i } ^ { ( 1 ) } = \\ell _ { j } ^ { ( 2 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 476, + 433, + 499 + ], + "score": 1.0, + "content": "then", + "type": "text" + }, + { + "bbox": [ + 433, + 481, + 504, + 495 + ], + "score": 0.91, + "content": "\\Lambda _ { i j } ( V _ { i } ) = M _ { i j } V _ { i }", + "type": "inline_equation" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 496, + 498, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 161, + 511 + ], + "score": 1.0, + "content": "for some real", + "type": "text" + }, + { + "bbox": [ + 162, + 497, + 179, + 510 + ], + "score": 0.91, + "content": "M _ { i j }", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 496, + 241, + 511 + ], + "score": 1.0, + "content": ", and otherwise", + "type": "text" + }, + { + "bbox": [ + 241, + 497, + 277, + 510 + ], + "score": 0.93, + "content": "M _ { i j } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 496, + 498, + 511 + ], + "score": 1.0, + "content": ". Plugging this into equation 21 we obtain equation 12.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 105, + 539, + 491, + 552 + ], + "lines": [ + { + "bbox": [ + 106, + 539, + 491, + 552 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 491, + 552 + ], + "score": 1.0, + "content": "We prove Theorem 2 which shows that the TFN network described in the main text is universal:", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 108, + 555, + 258, + 567 + ], + "lines": [ + { + "bbox": [ + 106, + 553, + 259, + 570 + ], + "spans": [ + { + "bbox": [ + 106, + 553, + 189, + 570 + ], + "score": 1.0, + "content": "Theorem 2. For all", + "type": "text" + }, + { + "bbox": [ + 190, + 555, + 255, + 568 + ], + "score": 0.92, + "content": "n \\in \\mathbb { N } , l _ { T } \\in \\mathbb { N } _ { + } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 553, + 259, + 570 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 128, + 576, + 505, + 625 + ], + "lines": [ + { + "bbox": [ + 125, + 570, + 501, + 599 + ], + "spans": [ + { + "bbox": [ + 125, + 570, + 158, + 599 + ], + "score": 1.0, + "content": "1. For", + "type": "text" + }, + { + "bbox": [ + 159, + 578, + 194, + 590 + ], + "score": 0.9, + "content": "D \\in { \\mathbb { N } } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 570, + 221, + 599 + ], + "score": 1.0, + "content": ", every", + "type": "text" + }, + { + "bbox": [ + 221, + 579, + 230, + 588 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 570, + 327, + 599 + ], + "score": 1.0, + "content": "-equivariant polynomial", + "type": "text" + }, + { + "bbox": [ + 327, + 578, + 396, + 591 + ], + "score": 0.92, + "content": "p : \\mathbb { R } ^ { 3 \\times n } W _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 570, + 435, + 599 + ], + "score": 1.0, + "content": "of degree", + "type": "text" + }, + { + "bbox": [ + 435, + 579, + 445, + 588 + ], + "score": 0.78, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 570, + 464, + 599 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 464, + 577, + 501, + 593 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { C ( D ) , D } ^ { \\mathrm { T F N } }", + "type": "inline_equation" + } + ], + "index": 27 + }, + { + "bbox": [ + 128, + 598, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 128, + 598, + 215, + 613 + ], + "score": 1.0, + "content": "2. Every continuous", + "type": "text" + }, + { + "bbox": [ + 215, + 600, + 224, + 609 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 598, + 505, + 613 + ], + "score": 1.0, + "content": "-equivariant function can be approximated uniformly on compact sets", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 140, + 609, + 272, + 626 + ], + "spans": [ + { + "bbox": [ + 140, + 609, + 272, + 626 + ], + "score": 1.0, + "content": "by functions in ∪D∈N+ FTFNC(D),D", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 638, + 506, + 673 + ], + "lines": [ + { + "bbox": [ + 105, + 637, + 506, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 455, + 652 + ], + "score": 1.0, + "content": "Proof. As mentioned in the main text, we only need to show that the function space", + "type": "text" + }, + { + "bbox": [ + 456, + 639, + 494, + 651 + ], + "score": 0.94, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 637, + 506, + 652 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 107, + 649, + 506, + 663 + ], + "spans": [ + { + "bbox": [ + 107, + 650, + 116, + 660 + ], + "score": 0.76, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 649, + 209, + 663 + ], + "score": 1.0, + "content": "-spanning. Recall that", + "type": "text" + }, + { + "bbox": [ + 209, + 650, + 247, + 662 + ], + "score": 0.93, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 649, + 310, + 663 + ], + "score": 1.0, + "content": "is obtained by", + "type": "text" + }, + { + "bbox": [ + 310, + 650, + 325, + 660 + ], + "score": 0.81, + "content": "2 D", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 649, + 452, + 663 + ], + "score": 1.0, + "content": "consecutive convolutions with", + "type": "text" + }, + { + "bbox": [ + 452, + 650, + 461, + 660 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 649, + 506, + 663 + ], + "score": 1.0, + "content": "-filters. In", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 661, + 505, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 355, + 675 + ], + "score": 1.0, + "content": "general, we denote the space of functions defined by applying", + "type": "text" + }, + { + "bbox": [ + 356, + 661, + 363, + 671 + ], + "score": 0.81, + "content": "J", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 661, + 479, + 675 + ], + "score": 1.0, + "content": "consecutive convolutions by", + "type": "text" + }, + { + "bbox": [ + 479, + 661, + 500, + 673 + ], + "score": 0.91, + "content": "\\mathcal { G } _ { J , D }", + "type": "inline_equation" + }, + { + "bbox": [ + 500, + 661, + 505, + 675 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 106, + 677, + 504, + 701 + ], + "lines": [ + { + "bbox": [ + 104, + 676, + 506, + 693 + ], + "spans": [ + { + "bbox": [ + 104, + 676, + 116, + 693 + ], + "score": 1.0, + "content": "If", + "type": "text" + }, + { + "bbox": [ + 117, + 679, + 126, + 689 + ], + "score": 0.84, + "content": "\\mathcal { V }", + "type": "inline_equation" + }, + { + "bbox": [ + 126, + 676, + 245, + 693 + ], + "score": 1.0, + "content": "is a space of functions from", + "type": "text" + }, + { + "bbox": [ + 246, + 678, + 302, + 689 + ], + "score": 0.9, + "content": "\\mathbb { R } ^ { 3 \\times n } Y ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 676, + 365, + 693 + ], + "score": 1.0, + "content": ", we denote by", + "type": "text" + }, + { + "bbox": [ + 365, + 678, + 397, + 691 + ], + "score": 0.92, + "content": "\\langle \\mathcal { V } , \\mathcal { T } _ { T } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 676, + 506, + 693 + ], + "score": 1.0, + "content": "the space of all functions", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 107, + 686, + 223, + 703 + ], + "spans": [ + { + "bbox": [ + 107, + 689, + 173, + 702 + ], + "score": 0.92, + "content": "p : \\mathbb { R } ^ { 3 \\times n } \\overset { \\cdot } { } \\mathcal { T } _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 686, + 223, + 703 + ], + "score": 1.0, + "content": "of the form", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + }, + { + "type": "interline_equation", + "bbox": [ + 258, + 701, + 353, + 735 + ], + "lines": [ + { + "bbox": [ + 258, + 701, + 353, + 735 + ], + "spans": [ + { + "bbox": [ + 258, + 701, + 353, + 735 + ], + "score": 0.95, + "content": "p ( X ) = \\sum _ { k = 1 } ^ { K } \\hat { A } _ { k } f _ { k } ( X ) ,", + "type": "interline_equation", + "image_path": "5b7d61c5e1f6bacf58ed590936872904db01b04c9207febcb1fb370a2b2e1cce.jpg" + } + ] + } + ], + "index": 35.5, + "virtual_lines": [ + { + "bbox": [ + 258, + 701, + 353, + 718.0 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 258, + 718.0, + 353, + 735.0 + ], + "spans": [], + "index": 36 + } + ] + } + ], + "page_idx": 15, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 14 + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 514, + 504, + 525 + ], + "lines": [ + { + "bbox": [ + 496, + 516, + 504, + 524 + ], + "spans": [ + { + "bbox": [ + 496, + 516, + 504, + 524 + ], + "score": 0.997, + "content": "□", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 80, + 252, + 94 + ], + "lines": [ + { + "bbox": [ + 106, + 80, + 252, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 80, + 252, + 96 + ], + "score": 1.0, + "content": "D PROOFS FOR SECTION 5", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 106, + 187, + 118 + ], + "lines": [ + { + "bbox": [ + 106, + 105, + 188, + 119 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 188, + 119 + ], + "score": 1.0, + "content": "We prove Lemma 5", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1, + "bbox_fs": [ + 106, + 105, + 188, + 119 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 121, + 505, + 162 + ], + "lines": [ + { + "bbox": [ + 103, + 115, + 494, + 145 + ], + "spans": [ + { + "bbox": [ + 103, + 115, + 169, + 145 + ], + "score": 1.0, + "content": "Lemma 5. Let", + "type": "text" + }, + { + "bbox": [ + 169, + 121, + 258, + 137 + ], + "score": 0.9, + "content": "l ^ { ( 1 ) } = ( \\ell _ { 1 } ^ { ( 1 ) } , \\dots , \\ell _ { K _ { 1 } } ^ { ( 1 ) } )", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 115, + 261, + 145 + ], + "score": 0.0, + "content": "", + "type": "text" + }, + { + "bbox": [ + 277, + 122, + 366, + 137 + ], + "score": 0.89, + "content": "l ^ { ( 2 ) } = ( \\ell _ { 1 } ^ { ( 2 ) } , \\dots , \\ell _ { K _ { 2 } } ^ { ( 2 ) } )", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 115, + 414, + 145 + ], + "score": 1.0, + "content": ". A function", + "type": "text" + }, + { + "bbox": [ + 414, + 124, + 494, + 136 + ], + "score": 0.9, + "content": "\\boldsymbol { \\Lambda } = \\left( \\Lambda _ { 1 } , \\ldots , \\Lambda _ { K _ { 2 } } \\right)", + "type": "inline_equation" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 135, + 506, + 148 + ], + "spans": [ + { + "bbox": [ + 105, + 135, + 261, + 148 + ], + "score": 1.0, + "content": "a linear equivariant mapping between", + "type": "text" + }, + { + "bbox": [ + 261, + 136, + 283, + 147 + ], + "score": 0.88, + "content": "W _ { l ^ { ( 1 ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 283, + 135, + 303, + 148 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 303, + 137, + 325, + 147 + ], + "score": 0.88, + "content": "W _ { l ^ { ( 2 ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 135, + 436, + 148 + ], + "score": 1.0, + "content": ", if and only if there exists a", + "type": "text" + }, + { + "bbox": [ + 437, + 137, + 475, + 146 + ], + "score": 0.89, + "content": "K _ { 1 } \\times K _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 135, + 506, + 148 + ], + "score": 1.0, + "content": "matrix", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 107, + 143, + 307, + 165 + ], + "spans": [ + { + "bbox": [ + 107, + 149, + 119, + 160 + ], + "score": 0.69, + "content": "M", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 143, + 139, + 165 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 140, + 149, + 176, + 162 + ], + "score": 0.91, + "content": "M _ { i j } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 143, + 217, + 165 + ], + "score": 1.0, + "content": "whenever", + "type": "text" + }, + { + "bbox": [ + 218, + 146, + 261, + 163 + ], + "score": 0.94, + "content": "\\ell _ { i } ^ { ( 1 ) } \\neq \\ell _ { j } ^ { ( 2 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 143, + 307, + 165 + ], + "score": 1.0, + "content": ", such that", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3, + "bbox_fs": [ + 103, + 115, + 506, + 165 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 263, + 169, + 348, + 203 + ], + "lines": [ + { + "bbox": [ + 263, + 169, + 348, + 203 + ], + "spans": [ + { + "bbox": [ + 263, + 169, + 348, + 203 + ], + "score": 0.94, + "content": "\\Lambda _ { j } ( V ) = \\sum _ { i = 1 } ^ { K _ { 1 } } M _ { i j } V _ { i }", + "type": "interline_equation", + "image_path": "c86fa12e12e78bc2dff340de63d341d041103c9ad97279aa1d5349fbec1ba3c2.jpg" + } + ] + } + ], + "index": 5.5, + "virtual_lines": [ + { + "bbox": [ + 263, + 169, + 348, + 186.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 263, + 186.0, + 348, + 203.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 208, + 338, + 225 + ], + "lines": [ + { + "bbox": [ + 104, + 206, + 335, + 227 + ], + "spans": [ + { + "bbox": [ + 104, + 206, + 133, + 226 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 209, + 185, + 223 + ], + "score": 0.93, + "content": "V = ( V _ { i } ) _ { i = 1 } ^ { K _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 206, + 204, + 226 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 205, + 210, + 248, + 227 + ], + "score": 0.92, + "content": "V _ { i } \\in W _ { \\ell _ { i } ^ { ( 1 ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 206, + 278, + 226 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 278, + 210, + 335, + 222 + ], + "score": 0.91, + "content": "i = 1 , \\ldots , K _ { 1 }", + "type": "inline_equation" + } + ], + "index": 7 + } + ], + "index": 7, + "bbox_fs": [ + 104, + 206, + 335, + 227 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 239, + 505, + 273 + ], + "lines": [ + { + "bbox": [ + 105, + 239, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 505, + 253 + ], + "score": 1.0, + "content": "Proof. As mentioned in the main text, this lemma is based on Schur’s lemma. This lemma is typi-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 251, + 504, + 263 + ], + "spans": [ + { + "bbox": [ + 106, + 251, + 504, + 263 + ], + "score": 1.0, + "content": "cally stated for complex representations, but holds for odd dimensional real representation as well.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 261, + 482, + 274 + ], + "spans": [ + { + "bbox": [ + 106, + 261, + 482, + 274 + ], + "score": 1.0, + "content": "We recount the lemma and its proof here for completeness (see also (Fulton & Harris, 2013)).", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 239, + 505, + 274 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 280, + 505, + 304 + ], + "lines": [ + { + "bbox": [ + 105, + 278, + 507, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 278, + 299, + 295 + ], + "score": 1.0, + "content": "Lemma D.1 (Schur’s Lemma for SO(3)). Let", + "type": "text" + }, + { + "bbox": [ + 299, + 281, + 371, + 293 + ], + "score": 0.92, + "content": "\\Lambda : W _ { \\ell _ { 1 } } \\to W _ { \\ell _ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 278, + 507, + 295 + ], + "score": 1.0, + "content": "be a linear equivariant map. If", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 291, + 381, + 304 + ], + "spans": [ + { + "bbox": [ + 106, + 291, + 138, + 303 + ], + "score": 0.92, + "content": "\\ell _ { 1 } \\neq \\ell _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 291, + 159, + 304 + ], + "score": 1.0, + "content": "then", + "type": "text" + }, + { + "bbox": [ + 159, + 292, + 185, + 302 + ], + "score": 0.88, + "content": "\\Lambda = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 291, + 233, + 304 + ], + "score": 1.0, + "content": ". Otherwise", + "type": "text" + }, + { + "bbox": [ + 234, + 292, + 242, + 302 + ], + "score": 0.43, + "content": "\\Lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 291, + 381, + 304 + ], + "score": 1.0, + "content": "is a scalar multiply of the identity.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 278, + 507, + 304 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 316, + 506, + 373 + ], + "lines": [ + { + "bbox": [ + 106, + 317, + 506, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 152, + 330 + ], + "score": 1.0, + "content": "Proof. Let", + "type": "text" + }, + { + "bbox": [ + 153, + 318, + 222, + 329 + ], + "score": 0.92, + "content": "\\Lambda : W _ { \\ell _ { 1 } } \\to W _ { \\ell _ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 317, + 442, + 330 + ], + "score": 1.0, + "content": "be a linear equivariant map. The image and kernel of", + "type": "text" + }, + { + "bbox": [ + 443, + 318, + 451, + 327 + ], + "score": 0.75, + "content": "\\Lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 317, + 506, + 330 + ], + "score": 1.0, + "content": "are invariant", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 328, + 506, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 161, + 342 + ], + "score": 1.0, + "content": "subspaces of", + "type": "text" + }, + { + "bbox": [ + 161, + 329, + 179, + 340 + ], + "score": 0.9, + "content": "W _ { \\ell _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 328, + 199, + 342 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 199, + 330, + 217, + 340 + ], + "score": 0.9, + "content": "W _ { \\ell _ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 328, + 346, + 342 + ], + "score": 1.0, + "content": ", respectively. It follows that if", + "type": "text" + }, + { + "bbox": [ + 346, + 329, + 375, + 340 + ], + "score": 0.92, + "content": "\\Lambda \\neq 0", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 328, + 397, + 342 + ], + "score": 1.0, + "content": "then", + "type": "text" + }, + { + "bbox": [ + 397, + 329, + 405, + 339 + ], + "score": 0.79, + "content": "\\Lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 328, + 506, + 342 + ], + "score": 1.0, + "content": "is a linear isomorphism", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 339, + 506, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 166, + 352 + ], + "score": 1.0, + "content": "so necessarily", + "type": "text" + }, + { + "bbox": [ + 166, + 340, + 201, + 351 + ], + "score": 0.92, + "content": "\\ell _ { 1 } = \\ell _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 339, + 262, + 352 + ], + "score": 1.0, + "content": ". Now assume", + "type": "text" + }, + { + "bbox": [ + 263, + 340, + 297, + 351 + ], + "score": 0.92, + "content": "\\ell _ { 1 } = \\ell _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 339, + 402, + 352 + ], + "score": 1.0, + "content": ". Since the dimension of", + "type": "text" + }, + { + "bbox": [ + 402, + 340, + 419, + 352 + ], + "score": 0.91, + "content": "W _ { \\ell _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 339, + 453, + 352 + ], + "score": 1.0, + "content": "is odd,", + "type": "text" + }, + { + "bbox": [ + 453, + 340, + 461, + 350 + ], + "score": 0.74, + "content": "\\Lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 339, + 506, + 352 + ], + "score": 1.0, + "content": "has a real", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 350, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 151, + 363 + ], + "score": 1.0, + "content": "eigenvalue", + "type": "text" + }, + { + "bbox": [ + 152, + 351, + 159, + 361 + ], + "score": 0.75, + "content": "\\lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 350, + 244, + 363 + ], + "score": 1.0, + "content": ". The linear function", + "type": "text" + }, + { + "bbox": [ + 244, + 351, + 276, + 361 + ], + "score": 0.91, + "content": "\\Lambda - \\lambda I", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 350, + 461, + 363 + ], + "score": 1.0, + "content": "is equivariant and has a non-trivial kernel, so", + "type": "text" + }, + { + "bbox": [ + 461, + 351, + 505, + 361 + ], + "score": 0.88, + "content": "\\Lambda - \\lambda I =", + "type": "inline_equation" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 361, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 117, + 374 + ], + "score": 1.0, + "content": "0.", + "type": "text" + }, + { + "bbox": [ + 494, + 362, + 505, + 372 + ], + "score": 1.0, + "content": "□", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15, + "bbox_fs": [ + 105, + 317, + 506, + 374 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 386, + 504, + 417 + ], + "lines": [ + { + "bbox": [ + 105, + 386, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 105, + 386, + 338, + 401 + ], + "score": 1.0, + "content": "We now return to the proof of Lemma 5. Note that each", + "type": "text" + }, + { + "bbox": [ + 338, + 388, + 421, + 401 + ], + "score": 0.88, + "content": "\\Lambda _ { j } : W _ { l ^ { ( 1 ) } } \\to W _ { \\ell _ { \\vec { \\cdot } } ^ { ( 2 ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 386, + 505, + 401 + ], + "score": 1.0, + "content": "is linear and SO(3)", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 402, + 496, + 418 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 300, + 416 + ], + "score": 1.0, + "content": "equivariant. Next denote the restrictions of each", + "type": "text" + }, + { + "bbox": [ + 300, + 403, + 312, + 415 + ], + "score": 0.87, + "content": "\\Lambda _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 402, + 324, + 416 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 325, + 402, + 409, + 418 + ], + "score": 0.89, + "content": "W _ { \\ell _ { i } ^ { ( 1 ) } } , i = 1 , \\dots , K _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 402, + 423, + 416 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 423, + 403, + 437, + 415 + ], + "score": 0.88, + "content": "\\Lambda _ { i j }", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 402, + 496, + 416 + ], + "score": 1.0, + "content": ", and note that", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 386, + 505, + 418 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 239, + 424, + 372, + 459 + ], + "lines": [ + { + "bbox": [ + 239, + 424, + 372, + 459 + ], + "spans": [ + { + "bbox": [ + 239, + 424, + 372, + 459 + ], + "score": 0.95, + "content": "\\Lambda _ { j } ( V _ { 1 } , \\dots , V _ { K _ { 1 } } ) = \\sum _ { i = 1 } ^ { K _ { 1 } } \\Lambda _ { i j } ( V _ { i } ) .", + "type": "interline_equation", + "image_path": "1ba83e118e1d6d86913cdc6af9f0324cb8056d6b0463bb01413adffe4b0e3059.jpg" + } + ] + } + ], + "index": 20.5, + "virtual_lines": [ + { + "bbox": [ + 239, + 424, + 372, + 441.5 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 239, + 441.5, + 372, + 459.0 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 463, + 505, + 510 + ], + "lines": [ + { + "bbox": [ + 104, + 460, + 505, + 481 + ], + "spans": [ + { + "bbox": [ + 104, + 460, + 212, + 481 + ], + "score": 1.0, + "content": "By considering vectors in", + "type": "text" + }, + { + "bbox": [ + 212, + 465, + 235, + 477 + ], + "score": 0.92, + "content": "W _ { l ^ { ( 1 ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 460, + 285, + 481 + ], + "score": 1.0, + "content": "of the form", + "type": "text" + }, + { + "bbox": [ + 286, + 464, + 373, + 477 + ], + "score": 0.91, + "content": "( 0 , \\ldots , 0 , V _ { i } , 0 \\ldots , 0 )", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 460, + 443, + 481 + ], + "score": 1.0, + "content": "we see that each", + "type": "text" + }, + { + "bbox": [ + 444, + 464, + 505, + 480 + ], + "score": 0.91, + "content": "\\Lambda _ { i j } : W _ { \\ell _ { i } ^ { ( 1 ) } } \\to", + "type": "inline_equation" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 476, + 504, + 499 + ], + "spans": [ + { + "bbox": [ + 106, + 482, + 129, + 499 + ], + "score": 0.91, + "content": "W _ { \\ell _ { j } ^ { ( 2 ) } }", + "type": "inline_equation" + }, + { + "bbox": [ + 130, + 476, + 368, + 499 + ], + "score": 1.0, + "content": "is linear and SO(3)-equivariant. Thus by Schur’s lemma, if", + "type": "text" + }, + { + "bbox": [ + 368, + 479, + 412, + 496 + ], + "score": 0.94, + "content": "\\ell _ { i } ^ { ( 1 ) } = \\ell _ { j } ^ { ( 2 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 476, + 433, + 499 + ], + "score": 1.0, + "content": "then", + "type": "text" + }, + { + "bbox": [ + 433, + 481, + 504, + 495 + ], + "score": 0.91, + "content": "\\Lambda _ { i j } ( V _ { i } ) = M _ { i j } V _ { i }", + "type": "inline_equation" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 496, + 498, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 161, + 511 + ], + "score": 1.0, + "content": "for some real", + "type": "text" + }, + { + "bbox": [ + 162, + 497, + 179, + 510 + ], + "score": 0.91, + "content": "M _ { i j }", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 496, + 241, + 511 + ], + "score": 1.0, + "content": ", and otherwise", + "type": "text" + }, + { + "bbox": [ + 241, + 497, + 277, + 510 + ], + "score": 0.93, + "content": "M _ { i j } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 496, + 498, + 511 + ], + "score": 1.0, + "content": ". Plugging this into equation 21 we obtain equation 12.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23, + "bbox_fs": [ + 104, + 460, + 505, + 511 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 539, + 491, + 552 + ], + "lines": [ + { + "bbox": [ + 106, + 539, + 491, + 552 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 491, + 552 + ], + "score": 1.0, + "content": "We prove Theorem 2 which shows that the TFN network described in the main text is universal:", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25, + "bbox_fs": [ + 106, + 539, + 491, + 552 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 555, + 258, + 567 + ], + "lines": [ + { + "bbox": [ + 106, + 553, + 259, + 570 + ], + "spans": [ + { + "bbox": [ + 106, + 553, + 189, + 570 + ], + "score": 1.0, + "content": "Theorem 2. For all", + "type": "text" + }, + { + "bbox": [ + 190, + 555, + 255, + 568 + ], + "score": 0.92, + "content": "n \\in \\mathbb { N } , l _ { T } \\in \\mathbb { N } _ { + } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 553, + 259, + 570 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26, + "bbox_fs": [ + 106, + 553, + 259, + 570 + ] + }, + { + "type": "text", + "bbox": [ + 128, + 576, + 505, + 625 + ], + "lines": [ + { + "bbox": [ + 125, + 570, + 501, + 599 + ], + "spans": [ + { + "bbox": [ + 125, + 570, + 158, + 599 + ], + "score": 1.0, + "content": "1. For", + "type": "text" + }, + { + "bbox": [ + 159, + 578, + 194, + 590 + ], + "score": 0.9, + "content": "D \\in { \\mathbb { N } } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 570, + 221, + 599 + ], + "score": 1.0, + "content": ", every", + "type": "text" + }, + { + "bbox": [ + 221, + 579, + 230, + 588 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 570, + 327, + 599 + ], + "score": 1.0, + "content": "-equivariant polynomial", + "type": "text" + }, + { + "bbox": [ + 327, + 578, + 396, + 591 + ], + "score": 0.92, + "content": "p : \\mathbb { R } ^ { 3 \\times n } W _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 570, + 435, + 599 + ], + "score": 1.0, + "content": "of degree", + "type": "text" + }, + { + "bbox": [ + 435, + 579, + 445, + 588 + ], + "score": 0.78, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 570, + 464, + 599 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 464, + 577, + 501, + 593 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { C ( D ) , D } ^ { \\mathrm { T F N } }", + "type": "inline_equation" + } + ], + "index": 27 + }, + { + "bbox": [ + 128, + 598, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 128, + 598, + 215, + 613 + ], + "score": 1.0, + "content": "2. Every continuous", + "type": "text" + }, + { + "bbox": [ + 215, + 600, + 224, + 609 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 598, + 505, + 613 + ], + "score": 1.0, + "content": "-equivariant function can be approximated uniformly on compact sets", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 140, + 609, + 272, + 626 + ], + "spans": [ + { + "bbox": [ + 140, + 609, + 272, + 626 + ], + "score": 1.0, + "content": "by functions in ∪D∈N+ FTFNC(D),D", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28, + "bbox_fs": [ + 125, + 570, + 505, + 626 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 638, + 506, + 673 + ], + "lines": [ + { + "bbox": [ + 105, + 637, + 506, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 455, + 652 + ], + "score": 1.0, + "content": "Proof. As mentioned in the main text, we only need to show that the function space", + "type": "text" + }, + { + "bbox": [ + 456, + 639, + 494, + 651 + ], + "score": 0.94, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 637, + 506, + 652 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 107, + 649, + 506, + 663 + ], + "spans": [ + { + "bbox": [ + 107, + 650, + 116, + 660 + ], + "score": 0.76, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 649, + 209, + 663 + ], + "score": 1.0, + "content": "-spanning. Recall that", + "type": "text" + }, + { + "bbox": [ + 209, + 650, + 247, + 662 + ], + "score": 0.93, + "content": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 649, + 310, + 663 + ], + "score": 1.0, + "content": "is obtained by", + "type": "text" + }, + { + "bbox": [ + 310, + 650, + 325, + 660 + ], + "score": 0.81, + "content": "2 D", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 649, + 452, + 663 + ], + "score": 1.0, + "content": "consecutive convolutions with", + "type": "text" + }, + { + "bbox": [ + 452, + 650, + 461, + 660 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 649, + 506, + 663 + ], + "score": 1.0, + "content": "-filters. In", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 661, + 505, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 355, + 675 + ], + "score": 1.0, + "content": "general, we denote the space of functions defined by applying", + "type": "text" + }, + { + "bbox": [ + 356, + 661, + 363, + 671 + ], + "score": 0.81, + "content": "J", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 661, + 479, + 675 + ], + "score": 1.0, + "content": "consecutive convolutions by", + "type": "text" + }, + { + "bbox": [ + 479, + 661, + 500, + 673 + ], + "score": 0.91, + "content": "\\mathcal { G } _ { J , D }", + "type": "inline_equation" + }, + { + "bbox": [ + 500, + 661, + 505, + 675 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 637, + 506, + 675 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 677, + 504, + 701 + ], + "lines": [ + { + "bbox": [ + 104, + 676, + 506, + 693 + ], + "spans": [ + { + "bbox": [ + 104, + 676, + 116, + 693 + ], + "score": 1.0, + "content": "If", + "type": "text" + }, + { + "bbox": [ + 117, + 679, + 126, + 689 + ], + "score": 0.84, + "content": "\\mathcal { V }", + "type": "inline_equation" + }, + { + "bbox": [ + 126, + 676, + 245, + 693 + ], + "score": 1.0, + "content": "is a space of functions from", + "type": "text" + }, + { + "bbox": [ + 246, + 678, + 302, + 689 + ], + "score": 0.9, + "content": "\\mathbb { R } ^ { 3 \\times n } Y ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 676, + 365, + 693 + ], + "score": 1.0, + "content": ", we denote by", + "type": "text" + }, + { + "bbox": [ + 365, + 678, + 397, + 691 + ], + "score": 0.92, + "content": "\\langle \\mathcal { V } , \\mathcal { T } _ { T } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 676, + 506, + 693 + ], + "score": 1.0, + "content": "the space of all functions", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 107, + 686, + 223, + 703 + ], + "spans": [ + { + "bbox": [ + 107, + 689, + 173, + 702 + ], + "score": 0.92, + "content": "p : \\mathbb { R } ^ { 3 \\times n } \\overset { \\cdot } { } \\mathcal { T } _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 686, + 223, + 703 + ], + "score": 1.0, + "content": "of the form", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5, + "bbox_fs": [ + 104, + 676, + 506, + 703 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 258, + 701, + 353, + 735 + ], + "lines": [ + { + "bbox": [ + 258, + 701, + 353, + 735 + ], + "spans": [ + { + "bbox": [ + 258, + 701, + 353, + 735 + ], + "score": 0.95, + "content": "p ( X ) = \\sum _ { k = 1 } ^ { K } \\hat { A } _ { k } f _ { k } ( X ) ,", + "type": "interline_equation", + "image_path": "5b7d61c5e1f6bacf58ed590936872904db01b04c9207febcb1fb370a2b2e1cce.jpg" + } + ] + } + ], + "index": 35.5, + "virtual_lines": [ + { + "bbox": [ + 258, + 701, + 353, + 718.0 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 258, + 718.0, + 353, + 735.0 + ], + "spans": [], + "index": 36 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 80, + 505, + 165 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 133, + 96 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 83, + 191, + 94 + ], + "score": 0.92, + "content": "A _ { k } : Y \\mathcal { T } _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 81, + 275, + 96 + ], + "score": 1.0, + "content": "are linear functions,", + "type": "text" + }, + { + "bbox": [ + 276, + 81, + 340, + 95 + ], + "score": 0.93, + "content": "\\hat { A } _ { k } : Y ^ { n } \\mathcal { T } _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "are induced by elementwise application,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 104, + 92, + 504, + 109 + ], + "spans": [ + { + "bbox": [ + 104, + 92, + 123, + 109 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 95, + 155, + 106 + ], + "score": 0.91, + "content": "f _ { k } \\in \\mathcal { V }", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 92, + 467, + 109 + ], + "score": 1.0, + "content": ". This notation is useful because: (i) by Lemma 4 it is sufficient to show that", + "type": "text" + }, + { + "bbox": [ + 467, + 93, + 504, + 107 + ], + "score": 0.93, + "content": "Q ^ { ( \\vec { r } ) } ( \\bar { X } )", + "type": "inline_equation" + } + ], + "index": 1 + }, + { + "bbox": [ + 104, + 105, + 506, + 120 + ], + "spans": [ + { + "bbox": [ + 104, + 105, + 128, + 120 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 128, + 106, + 179, + 118 + ], + "score": 0.93, + "content": "\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { T } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 105, + 210, + 120 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 210, + 106, + 245, + 117 + ], + "score": 0.92, + "content": "\\vec { r } \\in \\Sigma _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 105, + 278, + 120 + ], + "score": 1.0, + "content": "and all", + "type": "text" + }, + { + "bbox": [ + 279, + 106, + 313, + 117 + ], + "score": 0.92, + "content": "T \\leq D", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 105, + 506, + 120 + ], + "score": 1.0, + "content": ", and because (ii) it enables comparison of the", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 104, + 114, + 507, + 132 + ], + "spans": [ + { + "bbox": [ + 104, + 114, + 253, + 132 + ], + "score": 1.0, + "content": "expressive power of function spaces", + "type": "text" + }, + { + "bbox": [ + 254, + 117, + 282, + 128 + ], + "score": 0.92, + "content": "\\mathcal { \\scriptsize { N } } _ { 1 } , \\mathcal { \\scriptsize { N } } _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 114, + 445, + 132 + ], + "score": 1.0, + "content": "whose elements map to different spaces", + "type": "text" + }, + { + "bbox": [ + 445, + 117, + 478, + 129 + ], + "score": 0.63, + "content": "Y _ { 1 } ^ { n } , Y _ { 2 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 114, + 507, + 132 + ], + "score": 1.0, + "content": ", since", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 126, + 505, + 142 + ], + "spans": [ + { + "bbox": [ + 104, + 126, + 171, + 142 + ], + "score": 1.0, + "content": "the elements in", + "type": "text" + }, + { + "bbox": [ + 171, + 128, + 243, + 140 + ], + "score": 0.91, + "content": "\\langle y _ { i } , \\mathcal { T } _ { T } \\rangle , i = \\bar { 1 } , 2", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 126, + 505, + 142 + ], + "score": 1.0, + "content": "both map to the same space. In particular, note that if for every", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 107, + 139, + 506, + 154 + ], + "spans": [ + { + "bbox": [ + 107, + 140, + 141, + 152 + ], + "score": 0.9, + "content": "f \\in \\mathcal { V } _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 140, + 185, + 154 + ], + "score": 1.0, + "content": "there is a", + "type": "text" + }, + { + "bbox": [ + 185, + 141, + 219, + 152 + ], + "score": 0.9, + "content": "g \\in \\mathcal { V } 1", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 140, + 294, + 154 + ], + "score": 1.0, + "content": "and a linear map", + "type": "text" + }, + { + "bbox": [ + 295, + 140, + 356, + 152 + ], + "score": 0.92, + "content": "A : Y _ { 1 } Y _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 140, + 398, + 154 + ], + "score": 1.0, + "content": "such that", + "type": "text" + }, + { + "bbox": [ + 399, + 139, + 480, + 153 + ], + "score": 0.94, + "content": "f ( X ) \\ = \\ \\hat { A } \\circ g ( X )", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 140, + 506, + 154 + ], + "score": 1.0, + "content": ", then", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 107, + 151, + 195, + 165 + ], + "spans": [ + { + "bbox": [ + 107, + 152, + 191, + 164 + ], + "score": 0.9, + "content": "\\langle { \\mathcal { V } } _ { 2 } , { \\mathcal { T } } _ { T } \\rangle \\subseteq \\langle { \\mathcal { V } } _ { 1 } , { \\mathcal { T } } _ { T } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 151, + 195, + 165 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 106, + 168, + 505, + 252 + ], + "lines": [ + { + "bbox": [ + 105, + 168, + 506, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 168, + 506, + 182 + ], + "score": 1.0, + "content": "We now use this abstract discussion to prove some useful results: the first is that for the purpose", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 179, + 506, + 192 + ], + "spans": [ + { + "bbox": [ + 105, + 179, + 506, + 192 + ], + "score": 1.0, + "content": "of this lemma, we can ‘forget about’ the multiplication by a unitary matrix in equation 14, used", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 190, + 506, + 205 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 406, + 205 + ], + "score": 1.0, + "content": "for decomposition into irreducible representations: To see this, denote by", + "type": "text" + }, + { + "bbox": [ + 406, + 190, + 427, + 204 + ], + "score": 0.92, + "content": "\\tilde { \\mathcal { G } } _ { J , D }", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 191, + 506, + 205 + ], + "score": 1.0, + "content": "the function space", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 104, + 201, + 506, + 217 + ], + "spans": [ + { + "bbox": [ + 104, + 201, + 181, + 217 + ], + "score": 1.0, + "content": "obtained by taking", + "type": "text" + }, + { + "bbox": [ + 182, + 204, + 190, + 213 + ], + "score": 0.75, + "content": "J", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 201, + 311, + 217 + ], + "score": 1.0, + "content": "consecutive convolutions with", + "type": "text" + }, + { + "bbox": [ + 311, + 204, + 320, + 213 + ], + "score": 0.79, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 201, + 506, + 217 + ], + "score": 1.0, + "content": "-filters without multiplying by a unitary matrix", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 213, + 505, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 505, + 227 + ], + "score": 1.0, + "content": "in equation 14. Since Kronecker products of unitary matrices are unitary matrices, we obtain that the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 225, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 226, + 154, + 240 + ], + "score": 1.0, + "content": "elements in", + "type": "text" + }, + { + "bbox": [ + 154, + 226, + 176, + 239 + ], + "score": 0.91, + "content": "\\mathcal { G } _ { J , D }", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 226, + 193, + 240 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 194, + 225, + 214, + 239 + ], + "score": 0.92, + "content": "\\tilde { \\mathcal { G } } _ { J , D }", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 226, + 449, + 240 + ], + "score": 1.0, + "content": "differ only by multiplication by a unitary matrix, and thus", + "type": "text" + }, + { + "bbox": [ + 449, + 225, + 505, + 239 + ], + "score": 0.92, + "content": "\\langle \\tilde { \\mathcal { G } } _ { J , D } , \\mathcal { T } _ { T } \\rangle \\subseteq", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 107, + 238, + 365, + 252 + ], + "spans": [ + { + "bbox": [ + 107, + 239, + 151, + 252 + ], + "score": 0.92, + "content": "\\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { T } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 239, + 170, + 252 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 170, + 238, + 271, + 252 + ], + "score": 0.94, + "content": "\\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { T } \\rangle \\subseteq \\langle \\tilde { \\mathcal { G } } _ { J , D } , \\mathcal { T } _ { T } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 239, + 365, + 252 + ], + "score": 1.0, + "content": ", so both sets are equal.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 107, + 256, + 503, + 279 + ], + "lines": [ + { + "bbox": [ + 105, + 254, + 505, + 270 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 344, + 270 + ], + "score": 1.0, + "content": "Next, we prove that adding convolutional layers (enlarging", + "type": "text" + }, + { + "bbox": [ + 344, + 257, + 351, + 267 + ], + "score": 0.74, + "content": "J", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 254, + 505, + 270 + ], + "score": 1.0, + "content": ") or taking higher order filters (enlarg-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 267, + 349, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 267, + 122, + 279 + ], + "score": 1.0, + "content": "ing", + "type": "text" + }, + { + "bbox": [ + 122, + 268, + 132, + 277 + ], + "score": 0.75, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 267, + 349, + 279 + ], + "score": 1.0, + "content": ") can only increase the expressive power of a network.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5 + }, + { + "type": "text", + "bbox": [ + 106, + 283, + 253, + 296 + ], + "lines": [ + { + "bbox": [ + 105, + 282, + 253, + 298 + ], + "spans": [ + { + "bbox": [ + 105, + 282, + 193, + 298 + ], + "score": 1.0, + "content": "Lemma D.2. For all", + "type": "text" + }, + { + "bbox": [ + 194, + 284, + 250, + 296 + ], + "score": 0.93, + "content": "J , D , T \\in \\mathbb { N } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 282, + 253, + 298 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 129, + 306, + 258, + 339 + ], + "lines": [ + { + "bbox": [ + 132, + 306, + 256, + 320 + ], + "spans": [ + { + "bbox": [ + 132, + 306, + 256, + 320 + ], + "score": 0.75, + "content": "l . \\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { T } \\rangle \\subseteq \\langle \\mathcal { G } _ { J + 1 , D } , \\mathcal { T } _ { T } \\rangle .", + "type": "inline_equation" + } + ], + "index": 17 + }, + { + "bbox": [ + 128, + 325, + 257, + 339 + ], + "spans": [ + { + "bbox": [ + 128, + 325, + 136, + 339 + ], + "score": 1.0, + "content": "2", + "type": "text" + }, + { + "bbox": [ + 142, + 325, + 255, + 338 + ], + "score": 0.76, + "content": "\\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { T } \\rangle \\subseteq \\langle \\mathcal { G } _ { J , D + 1 } , \\mathcal { T } _ { T } \\rangle .", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 325, + 257, + 339 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 108, + 348, + 505, + 383 + ], + "lines": [ + { + "bbox": [ + 105, + 348, + 506, + 362 + ], + "spans": [ + { + "bbox": [ + 105, + 348, + 367, + 362 + ], + "score": 1.0, + "content": "Proof. The first claim follows from the fact that every function", + "type": "text" + }, + { + "bbox": [ + 368, + 350, + 375, + 361 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 348, + 388, + 362 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 389, + 349, + 433, + 361 + ], + "score": 0.93, + "content": "\\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { T } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 348, + 506, + 362 + ], + "score": 1.0, + "content": "can be identified", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 359, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 359, + 181, + 374 + ], + "score": 1.0, + "content": "with a function in", + "type": "text" + }, + { + "bbox": [ + 182, + 360, + 236, + 372 + ], + "score": 0.93, + "content": "\\langle \\mathcal { G } _ { J + 1 , D } , \\mathcal { T } _ { T } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 359, + 294, + 374 + ], + "score": 1.0, + "content": "by taking the", + "type": "text" + }, + { + "bbox": [ + 294, + 361, + 320, + 371 + ], + "score": 0.91, + "content": "J + 1", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 359, + 482, + 374 + ], + "score": 1.0, + "content": "convolutional layer in equation 14 with", + "type": "text" + }, + { + "bbox": [ + 482, + 361, + 505, + 372 + ], + "score": 0.87, + "content": "\\theta _ { 0 } =", + "type": "inline_equation" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 370, + 148, + 383 + ], + "spans": [ + { + "bbox": [ + 106, + 371, + 143, + 383 + ], + "score": 0.9, + "content": "1 , F = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 370, + 148, + 383 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 387, + 505, + 411 + ], + "lines": [ + { + "bbox": [ + 105, + 387, + 505, + 400 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 290, + 400 + ], + "score": 1.0, + "content": "The second claim follows from the fact that", + "type": "text" + }, + { + "bbox": [ + 290, + 389, + 300, + 398 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 387, + 420, + 400 + ], + "score": 1.0, + "content": "-filters can be identified with", + "type": "text" + }, + { + "bbox": [ + 421, + 388, + 449, + 399 + ], + "score": 0.9, + "content": "D + 1", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 387, + 505, + 400 + ], + "score": 1.0, + "content": "-filters whose", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 107, + 398, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 107, + 399, + 133, + 410 + ], + "score": 0.88, + "content": "D + 1", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 398, + 178, + 411 + ], + "score": 1.0, + "content": "-th entry is", + "type": "text" + }, + { + "bbox": [ + 178, + 400, + 184, + 409 + ], + "score": 0.26, + "content": "0", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 398, + 187, + 411 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 495, + 400, + 505, + 409 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 107, + 422, + 282, + 434 + ], + "lines": [ + { + "bbox": [ + 106, + 422, + 282, + 435 + ], + "spans": [ + { + "bbox": [ + 106, + 422, + 282, + 435 + ], + "score": 1.0, + "content": "The last preliminary lemma we will need is", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 106, + 439, + 504, + 461 + ], + "lines": [ + { + "bbox": [ + 104, + 437, + 504, + 454 + ], + "spans": [ + { + "bbox": [ + 104, + 437, + 207, + 454 + ], + "score": 1.0, + "content": "Lemma D.3. For every", + "type": "text" + }, + { + "bbox": [ + 207, + 440, + 255, + 451 + ], + "score": 0.92, + "content": "J , D \\in \\mathbb { N } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 437, + 302, + 454 + ], + "score": 1.0, + "content": ", and every", + "type": "text" + }, + { + "bbox": [ + 303, + 440, + 345, + 451 + ], + "score": 0.9, + "content": "t , s \\in \\mathbb { N } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 437, + 357, + 454 + ], + "score": 1.0, + "content": ", if", + "type": "text" + }, + { + "bbox": [ + 357, + 439, + 421, + 452 + ], + "score": 0.9, + "content": "\\dot { \\boldsymbol { p } } \\in \\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { t } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 437, + 497, + 454 + ], + "score": 1.0, + "content": ", then the function", + "type": "text" + }, + { + "bbox": [ + 498, + 442, + 504, + 451 + ], + "score": 0.65, + "content": "q", + "type": "inline_equation" + } + ], + "index": 25 + }, + { + "bbox": [ + 104, + 450, + 151, + 465 + ], + "spans": [ + { + "bbox": [ + 104, + 450, + 151, + 465 + ], + "score": 1.0, + "content": "defined by", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "interline_equation", + "bbox": [ + 232, + 460, + 378, + 493 + ], + "lines": [ + { + "bbox": [ + 232, + 460, + 378, + 493 + ], + "spans": [ + { + "bbox": [ + 232, + 460, + 378, + 493 + ], + "score": 0.95, + "content": "q _ { a } ( X ) = \\sum _ { b = 1 } ^ { n } ( { \\bar { x } } _ { a } - { \\bar { x } } _ { b } ) ^ { \\otimes s } \\otimes p _ { b } ( X )", + "type": "interline_equation", + "image_path": "157b4eb8d42596b21df90c7c20da2c4e448ee8aa03cc6efccf2f970ca8ef510b.jpg" + } + ] + } + ], + "index": 27.5, + "virtual_lines": [ + { + "bbox": [ + 232, + 460, + 378, + 476.5 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 232, + 476.5, + 378, + 493.0 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 493, + 191, + 506 + ], + "lines": [ + { + "bbox": [ + 104, + 491, + 188, + 508 + ], + "spans": [ + { + "bbox": [ + 104, + 491, + 126, + 508 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 127, + 493, + 188, + 506 + ], + "score": 0.9, + "content": "\\langle \\mathcal { G } _ { J + 1 , D } , \\mathcal { T } _ { t + s } \\rangle", + "type": "inline_equation" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 106, + 517, + 505, + 565 + ], + "lines": [ + { + "bbox": [ + 105, + 517, + 504, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 517, + 356, + 530 + ], + "score": 1.0, + "content": "Proof. This lemma is based on the fact that the space of", + "type": "text" + }, + { + "bbox": [ + 357, + 521, + 363, + 528 + ], + "score": 0.73, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 517, + 491, + 530 + ], + "score": 1.0, + "content": "homogeneous polynomial on", + "type": "text" + }, + { + "bbox": [ + 491, + 517, + 504, + 528 + ], + "score": 0.86, + "content": "\\mathbb { R } ^ { 3 }", + "type": "inline_equation" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 527, + 506, + 543 + ], + "spans": [ + { + "bbox": [ + 104, + 527, + 274, + 543 + ], + "score": 1.0, + "content": "is spanned by polynomials of the form", + "type": "text" + }, + { + "bbox": [ + 275, + 528, + 332, + 541 + ], + "score": 0.93, + "content": "\\| x \\| ^ { s - \\ell } Y _ { m } ^ { \\ell } ( \\bar { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 527, + 351, + 543 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 351, + 529, + 451, + 541 + ], + "score": 0.89, + "content": "\\ell = s , s - 2 , s - 4 \\ldots", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 527, + 487, + 543 + ], + "score": 1.0, + "content": "(Dai &", + "type": "text" + }, + { + "bbox": [ + 487, + 529, + 501, + 540 + ], + "score": 0.33, + "content": "\\mathrm { X u }", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 527, + 506, + 543 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 538, + 506, + 556 + ], + "spans": [ + { + "bbox": [ + 104, + 538, + 198, + 556 + ], + "score": 1.0, + "content": "2013). For each such", + "type": "text" + }, + { + "bbox": [ + 198, + 542, + 203, + 551 + ], + "score": 0.68, + "content": "\\ell", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 538, + 225, + 556 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 226, + 542, + 256, + 552 + ], + "score": 0.92, + "content": "s \\leq D", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 538, + 441, + 556 + ], + "score": 1.0, + "content": ", these polynomials can be realized by filters", + "type": "text" + }, + { + "bbox": [ + 442, + 540, + 460, + 551 + ], + "score": 0.9, + "content": "F ^ { ( \\ell ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 538, + 506, + 556 + ], + "score": 1.0, + "content": "by setting", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 107, + 551, + 213, + 568 + ], + "spans": [ + { + "bbox": [ + 107, + 552, + 182, + 565 + ], + "score": 0.93, + "content": "R ^ { ( \\ell ) } ( \\| x \\| ) = \\| x \\| ^ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 551, + 213, + 568 + ], + "score": 1.0, + "content": "so that", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31.5 + }, + { + "type": "interline_equation", + "bbox": [ + 222, + 567, + 388, + 583 + ], + "lines": [ + { + "bbox": [ + 222, + 567, + 388, + 583 + ], + "spans": [ + { + "bbox": [ + 222, + 567, + 388, + 583 + ], + "score": 0.92, + "content": "F _ { m } ^ { ( \\ell ) } ( x ) = \\| x \\| ^ { s } Y _ { m } ^ { \\ell } ( \\hat { x } ) = \\| x \\| ^ { s - \\ell } Y _ { m } ^ { \\ell } ( x ) .", + "type": "interline_equation", + "image_path": "fe191eb6b097a60008f6c16047d29cdf855c55919dd9b7baf237fd7baa214758.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 222, + 567, + 388, + 583 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 586, + 505, + 632 + ], + "lines": [ + { + "bbox": [ + 104, + 583, + 507, + 600 + ], + "spans": [ + { + "bbox": [ + 104, + 583, + 150, + 600 + ], + "score": 1.0, + "content": "For every", + "type": "text" + }, + { + "bbox": [ + 150, + 587, + 186, + 597 + ], + "score": 0.91, + "content": "D \\ \\in \\ \\mathbb { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 583, + 207, + 600 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 207, + 587, + 240, + 597 + ], + "score": 0.91, + "content": "s \\leq D", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 583, + 329, + 600 + ], + "score": 1.0, + "content": ", we can construct a", + "type": "text" + }, + { + "bbox": [ + 329, + 587, + 339, + 596 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 583, + 364, + 600 + ], + "score": 1.0, + "content": "-filter", + "type": "text" + }, + { + "bbox": [ + 364, + 585, + 475, + 599 + ], + "score": 0.92, + "content": "F ^ { s , D } = ( F ^ { ( 0 ) } , \\ldots , F ^ { ( D ) } )", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 583, + 507, + 600 + ], + "score": 1.0, + "content": "where", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 107, + 595, + 507, + 613 + ], + "spans": [ + { + "bbox": [ + 107, + 597, + 174, + 609 + ], + "score": 0.84, + "content": "F ^ { ( s ) } , F ^ { ( s - 2 ) } , \\ldots .", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 595, + 507, + 613 + ], + "score": 1.0, + "content": "are as defined above and the other filters are zero. Since both the entries of", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 606, + 506, + 624 + ], + "spans": [ + { + "bbox": [ + 106, + 609, + 142, + 621 + ], + "score": 0.9, + "content": "F ^ { s , D } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 606, + 220, + 624 + ], + "score": 1.0, + "content": ", and the entries of", + "type": "text" + }, + { + "bbox": [ + 221, + 609, + 237, + 620 + ], + "score": 0.89, + "content": "x ^ { \\otimes s }", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 606, + 315, + 624 + ], + "score": 1.0, + "content": ", span the space of", + "type": "text" + }, + { + "bbox": [ + 315, + 612, + 321, + 619 + ], + "score": 0.78, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 606, + 447, + 624 + ], + "score": 1.0, + "content": "-homogeneous polynomials on", + "type": "text" + }, + { + "bbox": [ + 447, + 609, + 460, + 619 + ], + "score": 0.88, + "content": "\\mathbb { R } ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 460, + 606, + 506, + 624 + ], + "score": 1.0, + "content": ", it follows", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 620, + 337, + 633 + ], + "spans": [ + { + "bbox": [ + 106, + 620, + 241, + 633 + ], + "score": 1.0, + "content": "that there exists a linear mapping", + "type": "text" + }, + { + "bbox": [ + 241, + 621, + 306, + 633 + ], + "score": 0.92, + "content": "B _ { s } : W _ { l _ { D } } T _ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 620, + 337, + 633 + ], + "score": 1.0, + "content": "so that", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36.5 + }, + { + "type": "interline_equation", + "bbox": [ + 241, + 635, + 370, + 650 + ], + "lines": [ + { + "bbox": [ + 241, + 635, + 370, + 650 + ], + "spans": [ + { + "bbox": [ + 241, + 635, + 370, + 650 + ], + "score": 0.91, + "content": "x ^ { \\otimes s } = B _ { s } ( F ^ { s , D } ( { \\boldsymbol { x } } ) ) , \\forall { \\boldsymbol { x } } \\in \\mathbb { R } ^ { 3 } .", + "type": "interline_equation", + "image_path": "8fc3c35c727a8ae16fa51601ee4c0c1e54ea890358760d99bf70a12f5446c458.jpg" + } + ] + } + ], + "index": 39, + "virtual_lines": [ + { + "bbox": [ + 241, + 635, + 370, + 650 + ], + "spans": [], + "index": 39 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 657, + 506, + 689 + ], + "lines": [ + { + "bbox": [ + 105, + 657, + 506, + 670 + ], + "spans": [ + { + "bbox": [ + 105, + 657, + 154, + 670 + ], + "score": 1.0, + "content": "Thus, since", + "type": "text" + }, + { + "bbox": [ + 154, + 660, + 161, + 669 + ], + "score": 0.81, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 657, + 472, + 670 + ], + "score": 1.0, + "content": "can be written as a sum of compositions of linear mappings with functions in", + "type": "text" + }, + { + "bbox": [ + 472, + 658, + 492, + 669 + ], + "score": 0.91, + "content": "\\mathcal { G } _ { J , D }", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 657, + 506, + 670 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 667, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 104, + 667, + 221, + 681 + ], + "score": 1.0, + "content": "in equation 22, and similarly", + "type": "text" + }, + { + "bbox": [ + 222, + 668, + 238, + 678 + ], + "score": 0.89, + "content": "x ^ { \\otimes s }", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 667, + 412, + 681 + ], + "score": 1.0, + "content": "is obtained as a linear image of functions in", + "type": "text" + }, + { + "bbox": [ + 412, + 669, + 432, + 681 + ], + "score": 0.92, + "content": "\\mathcal { G } _ { 1 , D }", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 667, + 506, + 681 + ], + "score": 1.0, + "content": "as in equation 23,", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 679, + 169, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 679, + 169, + 691 + ], + "score": 1.0, + "content": "we deduce that", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41 + }, + { + "type": "interline_equation", + "bbox": [ + 203, + 687, + 406, + 720 + ], + "lines": [ + { + "bbox": [ + 203, + 687, + 406, + 720 + ], + "spans": [ + { + "bbox": [ + 203, + 687, + 406, + 720 + ], + "score": 0.93, + "content": "\\sum _ { b = 1 } ^ { n } ( x _ { a } - x _ { b } ) \\otimes p _ { b } ( X ) = \\sum _ { b = 1 } ^ { n } ( \\bar { x } _ { a } - \\bar { x } _ { b } ) \\otimes p _ { b } ( X )", + "type": "interline_equation", + "image_path": "fa10a55443e5f7a520ffd122229e37d3368fba368ccfb6597ac7da43ab02cec6.jpg" + } + ] + } + ], + "index": 43.5, + "virtual_lines": [ + { + "bbox": [ + 203, + 687, + 406, + 703.5 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 203, + 703.5, + 406, + 720.0 + ], + "spans": [], + "index": 44 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 720, + 189, + 733 + ], + "lines": [ + { + "bbox": [ + 104, + 717, + 189, + 736 + ], + "spans": [ + { + "bbox": [ + 104, + 717, + 126, + 736 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 127, + 720, + 189, + 733 + ], + "score": 0.89, + "content": "\\langle \\mathcal { G } _ { J + 1 , D } , \\mathcal { T } _ { t + s } \\rangle", + "type": "inline_equation" + } + ], + "index": 45 + } + ], + "index": 45 + } + ], + "page_idx": 16, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "17", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 721, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 496, + 722, + 505, + 731 + ], + "spans": [ + { + "bbox": [ + 496, + 722, + 505, + 731 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 80, + 505, + 165 + ], + "lines": [ + { + "bbox": [ + 106, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 81, + 133, + 96 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 83, + 191, + 94 + ], + "score": 0.92, + "content": "A _ { k } : Y \\mathcal { T } _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 81, + 275, + 96 + ], + "score": 1.0, + "content": "are linear functions,", + "type": "text" + }, + { + "bbox": [ + 276, + 81, + 340, + 95 + ], + "score": 0.93, + "content": "\\hat { A } _ { k } : Y ^ { n } \\mathcal { T } _ { T } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "are induced by elementwise application,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 104, + 92, + 504, + 109 + ], + "spans": [ + { + "bbox": [ + 104, + 92, + 123, + 109 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 95, + 155, + 106 + ], + "score": 0.91, + "content": "f _ { k } \\in \\mathcal { V }", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 92, + 467, + 109 + ], + "score": 1.0, + "content": ". This notation is useful because: (i) by Lemma 4 it is sufficient to show that", + "type": "text" + }, + { + "bbox": [ + 467, + 93, + 504, + 107 + ], + "score": 0.93, + "content": "Q ^ { ( \\vec { r } ) } ( \\bar { X } )", + "type": "inline_equation" + } + ], + "index": 1 + }, + { + "bbox": [ + 104, + 105, + 506, + 120 + ], + "spans": [ + { + "bbox": [ + 104, + 105, + 128, + 120 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 128, + 106, + 179, + 118 + ], + "score": 0.93, + "content": "\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { T } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 105, + 210, + 120 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 210, + 106, + 245, + 117 + ], + "score": 0.92, + "content": "\\vec { r } \\in \\Sigma _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 105, + 278, + 120 + ], + "score": 1.0, + "content": "and all", + "type": "text" + }, + { + "bbox": [ + 279, + 106, + 313, + 117 + ], + "score": 0.92, + "content": "T \\leq D", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 105, + 506, + 120 + ], + "score": 1.0, + "content": ", and because (ii) it enables comparison of the", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 104, + 114, + 507, + 132 + ], + "spans": [ + { + "bbox": [ + 104, + 114, + 253, + 132 + ], + "score": 1.0, + "content": "expressive power of function spaces", + "type": "text" + }, + { + "bbox": [ + 254, + 117, + 282, + 128 + ], + "score": 0.92, + "content": "\\mathcal { \\scriptsize { N } } _ { 1 } , \\mathcal { \\scriptsize { N } } _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 114, + 445, + 132 + ], + "score": 1.0, + "content": "whose elements map to different spaces", + "type": "text" + }, + { + "bbox": [ + 445, + 117, + 478, + 129 + ], + "score": 0.63, + "content": "Y _ { 1 } ^ { n } , Y _ { 2 } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 114, + 507, + 132 + ], + "score": 1.0, + "content": ", since", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 126, + 505, + 142 + ], + "spans": [ + { + "bbox": [ + 104, + 126, + 171, + 142 + ], + "score": 1.0, + "content": "the elements in", + "type": "text" + }, + { + "bbox": [ + 171, + 128, + 243, + 140 + ], + "score": 0.91, + "content": "\\langle y _ { i } , \\mathcal { T } _ { T } \\rangle , i = \\bar { 1 } , 2", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 126, + 505, + 142 + ], + "score": 1.0, + "content": "both map to the same space. In particular, note that if for every", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 107, + 139, + 506, + 154 + ], + "spans": [ + { + "bbox": [ + 107, + 140, + 141, + 152 + ], + "score": 0.9, + "content": "f \\in \\mathcal { V } _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 140, + 185, + 154 + ], + "score": 1.0, + "content": "there is a", + "type": "text" + }, + { + "bbox": [ + 185, + 141, + 219, + 152 + ], + "score": 0.9, + "content": "g \\in \\mathcal { V } 1", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 140, + 294, + 154 + ], + "score": 1.0, + "content": "and a linear map", + "type": "text" + }, + { + "bbox": [ + 295, + 140, + 356, + 152 + ], + "score": 0.92, + "content": "A : Y _ { 1 } Y _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 140, + 398, + 154 + ], + "score": 1.0, + "content": "such that", + "type": "text" + }, + { + "bbox": [ + 399, + 139, + 480, + 153 + ], + "score": 0.94, + "content": "f ( X ) \\ = \\ \\hat { A } \\circ g ( X )", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 140, + 506, + 154 + ], + "score": 1.0, + "content": ", then", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 107, + 151, + 195, + 165 + ], + "spans": [ + { + "bbox": [ + 107, + 152, + 191, + 164 + ], + "score": 0.9, + "content": "\\langle { \\mathcal { V } } _ { 2 } , { \\mathcal { T } } _ { T } \\rangle \\subseteq \\langle { \\mathcal { V } } _ { 1 } , { \\mathcal { T } } _ { T } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 151, + 195, + 165 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3, + "bbox_fs": [ + 104, + 81, + 507, + 165 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 168, + 505, + 252 + ], + "lines": [ + { + "bbox": [ + 105, + 168, + 506, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 168, + 506, + 182 + ], + "score": 1.0, + "content": "We now use this abstract discussion to prove some useful results: the first is that for the purpose", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 179, + 506, + 192 + ], + "spans": [ + { + "bbox": [ + 105, + 179, + 506, + 192 + ], + "score": 1.0, + "content": "of this lemma, we can ‘forget about’ the multiplication by a unitary matrix in equation 14, used", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 190, + 506, + 205 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 406, + 205 + ], + "score": 1.0, + "content": "for decomposition into irreducible representations: To see this, denote by", + "type": "text" + }, + { + "bbox": [ + 406, + 190, + 427, + 204 + ], + "score": 0.92, + "content": "\\tilde { \\mathcal { G } } _ { J , D }", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 191, + 506, + 205 + ], + "score": 1.0, + "content": "the function space", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 104, + 201, + 506, + 217 + ], + "spans": [ + { + "bbox": [ + 104, + 201, + 181, + 217 + ], + "score": 1.0, + "content": "obtained by taking", + "type": "text" + }, + { + "bbox": [ + 182, + 204, + 190, + 213 + ], + "score": 0.75, + "content": "J", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 201, + 311, + 217 + ], + "score": 1.0, + "content": "consecutive convolutions with", + "type": "text" + }, + { + "bbox": [ + 311, + 204, + 320, + 213 + ], + "score": 0.79, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 201, + 506, + 217 + ], + "score": 1.0, + "content": "-filters without multiplying by a unitary matrix", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 213, + 505, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 505, + 227 + ], + "score": 1.0, + "content": "in equation 14. Since Kronecker products of unitary matrices are unitary matrices, we obtain that the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 225, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 226, + 154, + 240 + ], + "score": 1.0, + "content": "elements in", + "type": "text" + }, + { + "bbox": [ + 154, + 226, + 176, + 239 + ], + "score": 0.91, + "content": "\\mathcal { G } _ { J , D }", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 226, + 193, + 240 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 194, + 225, + 214, + 239 + ], + "score": 0.92, + "content": "\\tilde { \\mathcal { G } } _ { J , D }", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 226, + 449, + 240 + ], + "score": 1.0, + "content": "differ only by multiplication by a unitary matrix, and thus", + "type": "text" + }, + { + "bbox": [ + 449, + 225, + 505, + 239 + ], + "score": 0.92, + "content": "\\langle \\tilde { \\mathcal { G } } _ { J , D } , \\mathcal { T } _ { T } \\rangle \\subseteq", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 107, + 238, + 365, + 252 + ], + "spans": [ + { + "bbox": [ + 107, + 239, + 151, + 252 + ], + "score": 0.92, + "content": "\\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { T } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 239, + 170, + 252 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 170, + 238, + 271, + 252 + ], + "score": 0.94, + "content": "\\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { T } \\rangle \\subseteq \\langle \\tilde { \\mathcal { G } } _ { J , D } , \\mathcal { T } _ { T } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 239, + 365, + 252 + ], + "score": 1.0, + "content": ", so both sets are equal.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 10, + "bbox_fs": [ + 104, + 168, + 506, + 252 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 256, + 503, + 279 + ], + "lines": [ + { + "bbox": [ + 105, + 254, + 505, + 270 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 344, + 270 + ], + "score": 1.0, + "content": "Next, we prove that adding convolutional layers (enlarging", + "type": "text" + }, + { + "bbox": [ + 344, + 257, + 351, + 267 + ], + "score": 0.74, + "content": "J", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 254, + 505, + 270 + ], + "score": 1.0, + "content": ") or taking higher order filters (enlarg-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 267, + 349, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 267, + 122, + 279 + ], + "score": 1.0, + "content": "ing", + "type": "text" + }, + { + "bbox": [ + 122, + 268, + 132, + 277 + ], + "score": 0.75, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 267, + 349, + 279 + ], + "score": 1.0, + "content": ") can only increase the expressive power of a network.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 254, + 505, + 279 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 283, + 253, + 296 + ], + "lines": [ + { + "bbox": [ + 105, + 282, + 253, + 298 + ], + "spans": [ + { + "bbox": [ + 105, + 282, + 193, + 298 + ], + "score": 1.0, + "content": "Lemma D.2. For all", + "type": "text" + }, + { + "bbox": [ + 194, + 284, + 250, + 296 + ], + "score": 0.93, + "content": "J , D , T \\in \\mathbb { N } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 282, + 253, + 298 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 132, + 306, + 256, + 320 + ], + "spans": [ + { + "bbox": [ + 132, + 306, + 256, + 320 + ], + "score": 0.75, + "content": "l . \\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { T } \\rangle \\subseteq \\langle \\mathcal { G } _ { J + 1 , D } , \\mathcal { T } _ { T } \\rangle .", + "type": "inline_equation" + } + ], + "index": 17 + }, + { + "bbox": [ + 128, + 325, + 257, + 339 + ], + "spans": [ + { + "bbox": [ + 128, + 325, + 136, + 339 + ], + "score": 1.0, + "content": "2", + "type": "text" + }, + { + "bbox": [ + 142, + 325, + 255, + 338 + ], + "score": 0.76, + "content": "\\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { T } \\rangle \\subseteq \\langle \\mathcal { G } _ { J , D + 1 } , \\mathcal { T } _ { T } \\rangle .", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 325, + 257, + 339 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 282, + 253, + 298 + ] + }, + { + "type": "text", + "bbox": [ + 129, + 306, + 258, + 339 + ], + "lines": [], + "index": 17.5, + "bbox_fs": [ + 128, + 306, + 257, + 339 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 108, + 348, + 505, + 383 + ], + "lines": [ + { + "bbox": [ + 105, + 348, + 506, + 362 + ], + "spans": [ + { + "bbox": [ + 105, + 348, + 367, + 362 + ], + "score": 1.0, + "content": "Proof. The first claim follows from the fact that every function", + "type": "text" + }, + { + "bbox": [ + 368, + 350, + 375, + 361 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 348, + 388, + 362 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 389, + 349, + 433, + 361 + ], + "score": 0.93, + "content": "\\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { T } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 348, + 506, + 362 + ], + "score": 1.0, + "content": "can be identified", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 359, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 359, + 181, + 374 + ], + "score": 1.0, + "content": "with a function in", + "type": "text" + }, + { + "bbox": [ + 182, + 360, + 236, + 372 + ], + "score": 0.93, + "content": "\\langle \\mathcal { G } _ { J + 1 , D } , \\mathcal { T } _ { T } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 359, + 294, + 374 + ], + "score": 1.0, + "content": "by taking the", + "type": "text" + }, + { + "bbox": [ + 294, + 361, + 320, + 371 + ], + "score": 0.91, + "content": "J + 1", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 359, + 482, + 374 + ], + "score": 1.0, + "content": "convolutional layer in equation 14 with", + "type": "text" + }, + { + "bbox": [ + 482, + 361, + 505, + 372 + ], + "score": 0.87, + "content": "\\theta _ { 0 } =", + "type": "inline_equation" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 370, + 148, + 383 + ], + "spans": [ + { + "bbox": [ + 106, + 371, + 143, + 383 + ], + "score": 0.9, + "content": "1 , F = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 370, + 148, + 383 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 348, + 506, + 383 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 387, + 505, + 411 + ], + "lines": [ + { + "bbox": [ + 105, + 387, + 505, + 400 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 290, + 400 + ], + "score": 1.0, + "content": "The second claim follows from the fact that", + "type": "text" + }, + { + "bbox": [ + 290, + 389, + 300, + 398 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 387, + 420, + 400 + ], + "score": 1.0, + "content": "-filters can be identified with", + "type": "text" + }, + { + "bbox": [ + 421, + 388, + 449, + 399 + ], + "score": 0.9, + "content": "D + 1", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 387, + 505, + 400 + ], + "score": 1.0, + "content": "-filters whose", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 107, + 398, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 107, + 399, + 133, + 410 + ], + "score": 0.88, + "content": "D + 1", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 398, + 178, + 411 + ], + "score": 1.0, + "content": "-th entry is", + "type": "text" + }, + { + "bbox": [ + 178, + 400, + 184, + 409 + ], + "score": 0.26, + "content": "0", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 398, + 187, + 411 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 495, + 400, + 505, + 409 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 387, + 505, + 411 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 422, + 282, + 434 + ], + "lines": [ + { + "bbox": [ + 106, + 422, + 282, + 435 + ], + "spans": [ + { + "bbox": [ + 106, + 422, + 282, + 435 + ], + "score": 1.0, + "content": "The last preliminary lemma we will need is", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24, + "bbox_fs": [ + 106, + 422, + 282, + 435 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 439, + 504, + 461 + ], + "lines": [ + { + "bbox": [ + 104, + 437, + 504, + 454 + ], + "spans": [ + { + "bbox": [ + 104, + 437, + 207, + 454 + ], + "score": 1.0, + "content": "Lemma D.3. For every", + "type": "text" + }, + { + "bbox": [ + 207, + 440, + 255, + 451 + ], + "score": 0.92, + "content": "J , D \\in \\mathbb { N } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 437, + 302, + 454 + ], + "score": 1.0, + "content": ", and every", + "type": "text" + }, + { + "bbox": [ + 303, + 440, + 345, + 451 + ], + "score": 0.9, + "content": "t , s \\in \\mathbb { N } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 437, + 357, + 454 + ], + "score": 1.0, + "content": ", if", + "type": "text" + }, + { + "bbox": [ + 357, + 439, + 421, + 452 + ], + "score": 0.9, + "content": "\\dot { \\boldsymbol { p } } \\in \\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { t } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 437, + 497, + 454 + ], + "score": 1.0, + "content": ", then the function", + "type": "text" + }, + { + "bbox": [ + 498, + 442, + 504, + 451 + ], + "score": 0.65, + "content": "q", + "type": "inline_equation" + } + ], + "index": 25 + }, + { + "bbox": [ + 104, + 450, + 151, + 465 + ], + "spans": [ + { + "bbox": [ + 104, + 450, + 151, + 465 + ], + "score": 1.0, + "content": "defined by", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 104, + 437, + 504, + 465 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 232, + 460, + 378, + 493 + ], + "lines": [ + { + "bbox": [ + 232, + 460, + 378, + 493 + ], + "spans": [ + { + "bbox": [ + 232, + 460, + 378, + 493 + ], + "score": 0.95, + "content": "q _ { a } ( X ) = \\sum _ { b = 1 } ^ { n } ( { \\bar { x } } _ { a } - { \\bar { x } } _ { b } ) ^ { \\otimes s } \\otimes p _ { b } ( X )", + "type": "interline_equation", + "image_path": "157b4eb8d42596b21df90c7c20da2c4e448ee8aa03cc6efccf2f970ca8ef510b.jpg" + } + ] + } + ], + "index": 27.5, + "virtual_lines": [ + { + "bbox": [ + 232, + 460, + 378, + 476.5 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 232, + 476.5, + 378, + 493.0 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 493, + 191, + 506 + ], + "lines": [ + { + "bbox": [ + 104, + 491, + 188, + 508 + ], + "spans": [ + { + "bbox": [ + 104, + 491, + 126, + 508 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 127, + 493, + 188, + 506 + ], + "score": 0.9, + "content": "\\langle \\mathcal { G } _ { J + 1 , D } , \\mathcal { T } _ { t + s } \\rangle", + "type": "inline_equation" + } + ], + "index": 29 + } + ], + "index": 29, + "bbox_fs": [ + 104, + 491, + 188, + 508 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 517, + 505, + 565 + ], + "lines": [ + { + "bbox": [ + 105, + 517, + 504, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 517, + 356, + 530 + ], + "score": 1.0, + "content": "Proof. This lemma is based on the fact that the space of", + "type": "text" + }, + { + "bbox": [ + 357, + 521, + 363, + 528 + ], + "score": 0.73, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 517, + 491, + 530 + ], + "score": 1.0, + "content": "homogeneous polynomial on", + "type": "text" + }, + { + "bbox": [ + 491, + 517, + 504, + 528 + ], + "score": 0.86, + "content": "\\mathbb { R } ^ { 3 }", + "type": "inline_equation" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 527, + 506, + 543 + ], + "spans": [ + { + "bbox": [ + 104, + 527, + 274, + 543 + ], + "score": 1.0, + "content": "is spanned by polynomials of the form", + "type": "text" + }, + { + "bbox": [ + 275, + 528, + 332, + 541 + ], + "score": 0.93, + "content": "\\| x \\| ^ { s - \\ell } Y _ { m } ^ { \\ell } ( \\bar { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 527, + 351, + 543 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 351, + 529, + 451, + 541 + ], + "score": 0.89, + "content": "\\ell = s , s - 2 , s - 4 \\ldots", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 527, + 487, + 543 + ], + "score": 1.0, + "content": "(Dai &", + "type": "text" + }, + { + "bbox": [ + 487, + 529, + 501, + 540 + ], + "score": 0.33, + "content": "\\mathrm { X u }", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 527, + 506, + 543 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 538, + 506, + 556 + ], + "spans": [ + { + "bbox": [ + 104, + 538, + 198, + 556 + ], + "score": 1.0, + "content": "2013). For each such", + "type": "text" + }, + { + "bbox": [ + 198, + 542, + 203, + 551 + ], + "score": 0.68, + "content": "\\ell", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 538, + 225, + 556 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 226, + 542, + 256, + 552 + ], + "score": 0.92, + "content": "s \\leq D", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 538, + 441, + 556 + ], + "score": 1.0, + "content": ", these polynomials can be realized by filters", + "type": "text" + }, + { + "bbox": [ + 442, + 540, + 460, + 551 + ], + "score": 0.9, + "content": "F ^ { ( \\ell ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 538, + 506, + 556 + ], + "score": 1.0, + "content": "by setting", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 107, + 551, + 213, + 568 + ], + "spans": [ + { + "bbox": [ + 107, + 552, + 182, + 565 + ], + "score": 0.93, + "content": "R ^ { ( \\ell ) } ( \\| x \\| ) = \\| x \\| ^ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 551, + 213, + 568 + ], + "score": 1.0, + "content": "so that", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31.5, + "bbox_fs": [ + 104, + 517, + 506, + 568 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 222, + 567, + 388, + 583 + ], + "lines": [ + { + "bbox": [ + 222, + 567, + 388, + 583 + ], + "spans": [ + { + "bbox": [ + 222, + 567, + 388, + 583 + ], + "score": 0.92, + "content": "F _ { m } ^ { ( \\ell ) } ( x ) = \\| x \\| ^ { s } Y _ { m } ^ { \\ell } ( \\hat { x } ) = \\| x \\| ^ { s - \\ell } Y _ { m } ^ { \\ell } ( x ) .", + "type": "interline_equation", + "image_path": "fe191eb6b097a60008f6c16047d29cdf855c55919dd9b7baf237fd7baa214758.jpg" + } + ] + } + ], + "index": 34, + "virtual_lines": [ + { + "bbox": [ + 222, + 567, + 388, + 583 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 586, + 505, + 632 + ], + "lines": [ + { + "bbox": [ + 104, + 583, + 507, + 600 + ], + "spans": [ + { + "bbox": [ + 104, + 583, + 150, + 600 + ], + "score": 1.0, + "content": "For every", + "type": "text" + }, + { + "bbox": [ + 150, + 587, + 186, + 597 + ], + "score": 0.91, + "content": "D \\ \\in \\ \\mathbb { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 583, + 207, + 600 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 207, + 587, + 240, + 597 + ], + "score": 0.91, + "content": "s \\leq D", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 583, + 329, + 600 + ], + "score": 1.0, + "content": ", we can construct a", + "type": "text" + }, + { + "bbox": [ + 329, + 587, + 339, + 596 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 583, + 364, + 600 + ], + "score": 1.0, + "content": "-filter", + "type": "text" + }, + { + "bbox": [ + 364, + 585, + 475, + 599 + ], + "score": 0.92, + "content": "F ^ { s , D } = ( F ^ { ( 0 ) } , \\ldots , F ^ { ( D ) } )", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 583, + 507, + 600 + ], + "score": 1.0, + "content": "where", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 107, + 595, + 507, + 613 + ], + "spans": [ + { + "bbox": [ + 107, + 597, + 174, + 609 + ], + "score": 0.84, + "content": "F ^ { ( s ) } , F ^ { ( s - 2 ) } , \\ldots .", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 595, + 507, + 613 + ], + "score": 1.0, + "content": "are as defined above and the other filters are zero. Since both the entries of", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 606, + 506, + 624 + ], + "spans": [ + { + "bbox": [ + 106, + 609, + 142, + 621 + ], + "score": 0.9, + "content": "F ^ { s , D } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 606, + 220, + 624 + ], + "score": 1.0, + "content": ", and the entries of", + "type": "text" + }, + { + "bbox": [ + 221, + 609, + 237, + 620 + ], + "score": 0.89, + "content": "x ^ { \\otimes s }", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 606, + 315, + 624 + ], + "score": 1.0, + "content": ", span the space of", + "type": "text" + }, + { + "bbox": [ + 315, + 612, + 321, + 619 + ], + "score": 0.78, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 606, + 447, + 624 + ], + "score": 1.0, + "content": "-homogeneous polynomials on", + "type": "text" + }, + { + "bbox": [ + 447, + 609, + 460, + 619 + ], + "score": 0.88, + "content": "\\mathbb { R } ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 460, + 606, + 506, + 624 + ], + "score": 1.0, + "content": ", it follows", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 620, + 337, + 633 + ], + "spans": [ + { + "bbox": [ + 106, + 620, + 241, + 633 + ], + "score": 1.0, + "content": "that there exists a linear mapping", + "type": "text" + }, + { + "bbox": [ + 241, + 621, + 306, + 633 + ], + "score": 0.92, + "content": "B _ { s } : W _ { l _ { D } } T _ { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 620, + 337, + 633 + ], + "score": 1.0, + "content": "so that", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36.5, + "bbox_fs": [ + 104, + 583, + 507, + 633 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 241, + 635, + 370, + 650 + ], + "lines": [ + { + "bbox": [ + 241, + 635, + 370, + 650 + ], + "spans": [ + { + "bbox": [ + 241, + 635, + 370, + 650 + ], + "score": 0.91, + "content": "x ^ { \\otimes s } = B _ { s } ( F ^ { s , D } ( { \\boldsymbol { x } } ) ) , \\forall { \\boldsymbol { x } } \\in \\mathbb { R } ^ { 3 } .", + "type": "interline_equation", + "image_path": "8fc3c35c727a8ae16fa51601ee4c0c1e54ea890358760d99bf70a12f5446c458.jpg" + } + ] + } + ], + "index": 39, + "virtual_lines": [ + { + "bbox": [ + 241, + 635, + 370, + 650 + ], + "spans": [], + "index": 39 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 657, + 506, + 689 + ], + "lines": [ + { + "bbox": [ + 105, + 657, + 506, + 670 + ], + "spans": [ + { + "bbox": [ + 105, + 657, + 154, + 670 + ], + "score": 1.0, + "content": "Thus, since", + "type": "text" + }, + { + "bbox": [ + 154, + 660, + 161, + 669 + ], + "score": 0.81, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 657, + 472, + 670 + ], + "score": 1.0, + "content": "can be written as a sum of compositions of linear mappings with functions in", + "type": "text" + }, + { + "bbox": [ + 472, + 658, + 492, + 669 + ], + "score": 0.91, + "content": "\\mathcal { G } _ { J , D }", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 657, + 506, + 670 + ], + "score": 1.0, + "content": "as", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 667, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 104, + 667, + 221, + 681 + ], + "score": 1.0, + "content": "in equation 22, and similarly", + "type": "text" + }, + { + "bbox": [ + 222, + 668, + 238, + 678 + ], + "score": 0.89, + "content": "x ^ { \\otimes s }", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 667, + 412, + 681 + ], + "score": 1.0, + "content": "is obtained as a linear image of functions in", + "type": "text" + }, + { + "bbox": [ + 412, + 669, + 432, + 681 + ], + "score": 0.92, + "content": "\\mathcal { G } _ { 1 , D }", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 667, + 506, + 681 + ], + "score": 1.0, + "content": "as in equation 23,", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 679, + 169, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 679, + 169, + 691 + ], + "score": 1.0, + "content": "we deduce that", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41, + "bbox_fs": [ + 104, + 657, + 506, + 691 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 203, + 687, + 406, + 720 + ], + "lines": [ + { + "bbox": [ + 203, + 687, + 406, + 720 + ], + "spans": [ + { + "bbox": [ + 203, + 687, + 406, + 720 + ], + "score": 0.93, + "content": "\\sum _ { b = 1 } ^ { n } ( x _ { a } - x _ { b } ) \\otimes p _ { b } ( X ) = \\sum _ { b = 1 } ^ { n } ( \\bar { x } _ { a } - \\bar { x } _ { b } ) \\otimes p _ { b } ( X )", + "type": "interline_equation", + "image_path": "fa10a55443e5f7a520ffd122229e37d3368fba368ccfb6597ac7da43ab02cec6.jpg" + } + ] + } + ], + "index": 43.5, + "virtual_lines": [ + { + "bbox": [ + 203, + 687, + 406, + 703.5 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 203, + 703.5, + 406, + 720.0 + ], + "spans": [], + "index": 44 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 720, + 189, + 733 + ], + "lines": [ + { + "bbox": [ + 104, + 717, + 189, + 736 + ], + "spans": [ + { + "bbox": [ + 104, + 717, + 126, + 736 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 127, + 720, + 189, + 733 + ], + "score": 0.89, + "content": "\\langle \\mathcal { G } _ { J + 1 , D } , \\mathcal { T } _ { t + s } \\rangle", + "type": "inline_equation" + } + ], + "index": 45 + } + ], + "index": 45, + "bbox_fs": [ + 104, + 717, + 189, + 736 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 81, + 505, + 106 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 504, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 246, + 95 + ], + "score": 1.0, + "content": "As a final preliminary, we note that", + "type": "text" + }, + { + "bbox": [ + 246, + 83, + 256, + 92 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 81, + 466, + 95 + ], + "score": 1.0, + "content": "-filters can perform an averaging operation by setting", + "type": "text" + }, + { + "bbox": [ + 466, + 81, + 504, + 93 + ], + "score": 0.9, + "content": "R ^ { ( 0 ) } = 1", + "type": "inline_equation" + } + ], + "index": 0 + }, + { + "bbox": [ + 104, + 91, + 507, + 109 + ], + "spans": [ + { + "bbox": [ + 104, + 91, + 123, + 109 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 123, + 93, + 218, + 106 + ], + "score": 0.91, + "content": "\\theta _ { 0 } , R ^ { ( \\mathrm { 1 } ) } , \\ldots , R ^ { ( \\dot { D } ) } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 91, + 395, + 109 + ], + "score": 1.0, + "content": "in equation 13 and equation 14 . We call this", + "type": "text" + }, + { + "bbox": [ + 395, + 95, + 405, + 104 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 91, + 507, + 109 + ], + "score": 1.0, + "content": "-filter an averaging filter.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 106, + 110, + 505, + 157 + ], + "lines": [ + { + "bbox": [ + 105, + 110, + 505, + 124 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 391, + 124 + ], + "score": 1.0, + "content": "We are now ready to prove our claim: we need to show that for every", + "type": "text" + }, + { + "bbox": [ + 392, + 111, + 441, + 123 + ], + "score": 0.92, + "content": "D , T \\in \\mathbb { N } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 110, + 469, + 124 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 469, + 111, + 501, + 123 + ], + "score": 0.89, + "content": "T \\leq D", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 110, + 505, + 124 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 104, + 121, + 506, + 137 + ], + "spans": [ + { + "bbox": [ + 104, + 121, + 146, + 137 + ], + "score": 1.0, + "content": "for every", + "type": "text" + }, + { + "bbox": [ + 146, + 123, + 179, + 135 + ], + "score": 0.93, + "content": "\\vec { r } \\in \\Sigma _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 121, + 235, + 137 + ], + "score": 1.0, + "content": ", the function", + "type": "text" + }, + { + "bbox": [ + 235, + 122, + 254, + 135 + ], + "score": 0.92, + "content": "Q ^ { ( \\vec { r } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 121, + 277, + 137 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 278, + 123, + 328, + 136 + ], + "score": 0.93, + "content": "\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { T } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 121, + 506, + 137 + ], + "score": 1.0, + "content": ". Note that due to the inclusion relations in", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 134, + 505, + 147 + ], + "spans": [ + { + "bbox": [ + 105, + 134, + 315, + 147 + ], + "score": 1.0, + "content": "Lemma D.2 it is sufficient to prove this for the case", + "type": "text" + }, + { + "bbox": [ + 315, + 135, + 346, + 145 + ], + "score": 0.91, + "content": "T = D", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 134, + 474, + 147 + ], + "score": 1.0, + "content": ". We prove this by induction on", + "type": "text" + }, + { + "bbox": [ + 475, + 135, + 484, + 144 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 134, + 505, + 147 + ], + "score": 1.0, + "content": ". For", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 107, + 145, + 308, + 158 + ], + "spans": [ + { + "bbox": [ + 107, + 146, + 135, + 156 + ], + "score": 0.87, + "content": "D = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 145, + 169, + 158 + ], + "score": 1.0, + "content": ", vectors", + "type": "text" + }, + { + "bbox": [ + 170, + 146, + 199, + 156 + ], + "score": 0.92, + "content": "\\vec { r } \\in \\Sigma _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 145, + 308, + 158 + ], + "score": 1.0, + "content": "contains only zeros and so", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3.5 + }, + { + "type": "interline_equation", + "bbox": [ + 215, + 160, + 396, + 176 + ], + "lines": [ + { + "bbox": [ + 215, + 160, + 396, + 176 + ], + "spans": [ + { + "bbox": [ + 215, + 160, + 396, + 176 + ], + "score": 0.91, + "content": "Q ^ { ( \\vec { r } ) } ( \\bar { X } ) = 1 _ { n } = \\pi _ { V } \\circ \\mathrm { e x t } ( X ) \\in \\langle \\mathcal { G } _ { 0 , 0 } , \\mathcal { T } _ { 0 } \\rangle .", + "type": "interline_equation", + "image_path": "6d9daefbbffc124f7a4c1212f098402ea43a3d5fd908e464d9176fcace8e1de8.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 215, + 160, + 396, + 176 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 184, + 505, + 230 + ], + "lines": [ + { + "bbox": [ + 106, + 184, + 506, + 198 + ], + "spans": [ + { + "bbox": [ + 106, + 184, + 266, + 198 + ], + "score": 1.0, + "content": "We now assume the claim is true for all", + "type": "text" + }, + { + "bbox": [ + 266, + 185, + 278, + 195 + ], + "score": 0.88, + "content": "D ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 184, + 299, + 198 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 300, + 185, + 353, + 196 + ], + "score": 0.92, + "content": "D > D ^ { \\prime } \\geq 0", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 184, + 475, + 198 + ], + "score": 1.0, + "content": "and prove the claim is true for", + "type": "text" + }, + { + "bbox": [ + 476, + 186, + 485, + 195 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 184, + 506, + 198 + ], + "score": 1.0, + "content": ". We", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 195, + 506, + 211 + ], + "spans": [ + { + "bbox": [ + 104, + 195, + 216, + 211 + ], + "score": 1.0, + "content": "need to show that for every", + "type": "text" + }, + { + "bbox": [ + 217, + 197, + 249, + 208 + ], + "score": 0.92, + "content": "\\vec { r } \\in \\Sigma _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 195, + 300, + 211 + ], + "score": 1.0, + "content": "the function", + "type": "text" + }, + { + "bbox": [ + 301, + 196, + 320, + 208 + ], + "score": 0.92, + "content": "Q ^ { ( \\vec { r } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 195, + 341, + 211 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 341, + 197, + 393, + 209 + ], + "score": 0.91, + "content": "\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { D } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 195, + 506, + 211 + ], + "score": 1.0, + "content": ". We prove this yet again by", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 207, + 505, + 220 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 249, + 220 + ], + "score": 1.0, + "content": "induction, this time on the value of", + "type": "text" + }, + { + "bbox": [ + 249, + 209, + 259, + 219 + ], + "score": 0.84, + "content": "r _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 207, + 313, + 220 + ], + "score": 1.0, + "content": ": assume that", + "type": "text" + }, + { + "bbox": [ + 314, + 208, + 346, + 219 + ], + "score": 0.91, + "content": "\\vec { r } \\in \\Sigma _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 207, + 365, + 220 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 365, + 209, + 396, + 219 + ], + "score": 0.85, + "content": "r _ { 1 } = 0 .", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 207, + 445, + 220 + ], + "score": 1.0, + "content": ". Denote by", + "type": "text" + }, + { + "bbox": [ + 445, + 208, + 451, + 218 + ], + "score": 0.81, + "content": "\\tilde { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 452, + 207, + 505, + 220 + ], + "score": 1.0, + "content": "the vector in", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 218, + 178, + 232 + ], + "spans": [ + { + "bbox": [ + 107, + 219, + 132, + 231 + ], + "score": 0.9, + "content": "\\Sigma _ { D - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 218, + 178, + 232 + ], + "score": 1.0, + "content": "defined by", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8.5 + }, + { + "type": "interline_equation", + "bbox": [ + 253, + 229, + 357, + 242 + ], + "lines": [ + { + "bbox": [ + 253, + 229, + 357, + 242 + ], + "spans": [ + { + "bbox": [ + 253, + 229, + 357, + 242 + ], + "score": 0.89, + "content": "\\tilde { r } = ( r _ { 2 } - 1 , r _ { 3 } , \\ldots , r _ { K } ) .", + "type": "interline_equation", + "image_path": "46613d7f0da0ef0fce92ad630c0254ebf96380aefc66498e73438c4f46845f8a.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 253, + 229, + 357, + 242 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 244, + 455, + 258 + ], + "lines": [ + { + "bbox": [ + 104, + 241, + 457, + 261 + ], + "spans": [ + { + "bbox": [ + 104, + 241, + 236, + 261 + ], + "score": 1.0, + "content": "By the induction assumption on", + "type": "text" + }, + { + "bbox": [ + 236, + 245, + 245, + 255 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 241, + 305, + 261 + ], + "score": 1.0, + "content": ", we know that", + "type": "text" + }, + { + "bbox": [ + 306, + 244, + 426, + 258 + ], + "score": 0.9, + "content": "Q ^ { ( \\tilde { r } ) } ( \\bar { X } ) \\in \\mathcal { G } _ { 2 ( D - 1 ) , D - 1 , D - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 241, + 457, + 261 + ], + "score": 1.0, + "content": "and so", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "interline_equation", + "bbox": [ + 112, + 261, + 497, + 330 + ], + "lines": [ + { + "bbox": [ + 112, + 261, + 497, + 330 + ], + "spans": [ + { + "bbox": [ + 112, + 261, + 497, + 330 + ], + "score": 0.93, + "content": "\\begin{array} { l } { \\displaystyle q _ { a } ( X ) = \\sum _ { b = 1 } ^ { n } ( \\bar { x } _ { a } - \\bar { x } _ { b } ) \\otimes Q _ { b } ^ { ( \\bar { r } ) } ( \\bar { X } ) = \\sum _ { b = 1 } ^ { n } ( \\bar { x } _ { a } - \\bar { x } _ { b } ) \\otimes \\bar { x } _ { b } ^ { \\otimes r _ { 2 } - 1 } \\otimes \\sum _ { i _ { 3 } , \\dots , i _ { K } = 1 } ^ { n } \\bar { x } _ { i _ { 3 } } ^ { \\otimes r _ { 3 } } \\otimes \\dots \\otimes \\bar { x } _ { i _ { K } } ^ { \\otimes r _ { K } } } \\\\ { \\displaystyle \\quad = \\left( \\bar { x } _ { a } \\otimes \\sum _ { b = 1 } ^ { n } Q _ { b } ^ { ( \\bar { r } ) } ( \\bar { X } ) \\right) - Q ^ { ( \\bar { r } ) } ( \\bar { X } ) } \\end{array}", + "type": "interline_equation", + "image_path": "c752720d6430d8f7ff000a662ea0032fc8b642b3343d74d0bf331ac24b8629ac.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 112, + 261, + 497, + 284.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 112, + 284.0, + 497, + 307.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 112, + 307.0, + 497, + 330.0 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 332, + 505, + 385 + ], + "lines": [ + { + "bbox": [ + 104, + 330, + 506, + 346 + ], + "spans": [ + { + "bbox": [ + 104, + 330, + 126, + 346 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 126, + 332, + 198, + 345 + ], + "score": 0.93, + "content": "\\langle \\mathcal { G } _ { 2 D - 1 , D - 1 } , \\mathcal { T } _ { D } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 330, + 352, + 346 + ], + "score": 1.0, + "content": "by Lemma D.3, which is contained in", + "type": "text" + }, + { + "bbox": [ + 352, + 333, + 414, + 345 + ], + "score": 0.92, + "content": "\\langle \\mathcal { G } _ { 2 D - 1 , D } , \\mathcal { T } _ { D } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 330, + 506, + 346 + ], + "score": 1.0, + "content": "by Lemma D.2. Since", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 344, + 506, + 361 + ], + "spans": [ + { + "bbox": [ + 106, + 348, + 118, + 358 + ], + "score": 0.84, + "content": "\\textstyle { \\bar { x } } _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 344, + 213, + 361 + ], + "score": 1.0, + "content": "has zero mean, while", + "type": "text" + }, + { + "bbox": [ + 214, + 344, + 250, + 359 + ], + "score": 0.94, + "content": "Q _ { a } ^ { ( \\vec { r } ) } ( \\bar { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 344, + 338, + 361 + ], + "score": 1.0, + "content": "does not depend on", + "type": "text" + }, + { + "bbox": [ + 338, + 349, + 345, + 357 + ], + "score": 0.72, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 344, + 371, + 361 + ], + "score": 1.0, + "content": "since", + "type": "text" + }, + { + "bbox": [ + 372, + 348, + 406, + 358 + ], + "score": 0.9, + "content": "r _ { 1 } ~ = ~ 0", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 344, + 506, + 361 + ], + "score": 1.0, + "content": ", applying an averaging", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 104, + 358, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 104, + 358, + 139, + 375 + ], + "score": 1.0, + "content": "filter to", + "type": "text" + }, + { + "bbox": [ + 139, + 363, + 150, + 372 + ], + "score": 0.82, + "content": "q _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 358, + 256, + 375 + ], + "score": 1.0, + "content": "gives us a constant value", + "type": "text" + }, + { + "bbox": [ + 256, + 358, + 300, + 373 + ], + "score": 0.94, + "content": "- Q _ { a } ^ { ( \\vec { r } ) } ( \\bar { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 358, + 379, + 375 + ], + "score": 1.0, + "content": "in each coordinate", + "type": "text" + }, + { + "bbox": [ + 380, + 361, + 412, + 373 + ], + "score": 0.92, + "content": "a \\in [ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 358, + 446, + 375 + ], + "score": 1.0, + "content": ", and so", + "type": "text" + }, + { + "bbox": [ + 447, + 360, + 483, + 373 + ], + "score": 0.93, + "content": "Q ^ { ( \\vec { r } ) } ( \\bar { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 358, + 506, + 375 + ], + "score": 1.0, + "content": "is in", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 107, + 371, + 163, + 385 + ], + "spans": [ + { + "bbox": [ + 107, + 372, + 159, + 384 + ], + "score": 0.91, + "content": "\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { D } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 371, + 163, + 385 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 107, + 388, + 505, + 447 + ], + "lines": [ + { + "bbox": [ + 105, + 389, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 253, + 401 + ], + "score": 1.0, + "content": "Now assume the claim is true for all", + "type": "text" + }, + { + "bbox": [ + 253, + 389, + 285, + 400 + ], + "score": 0.92, + "content": "\\vec { r } \\in \\Sigma _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 389, + 343, + 401 + ], + "score": 1.0, + "content": "which sum to", + "type": "text" + }, + { + "bbox": [ + 343, + 389, + 353, + 399 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 389, + 505, + 401 + ], + "score": 1.0, + "content": ", and whose first coordinate is smaller", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 399, + 506, + 413 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 150, + 413 + ], + "score": 1.0, + "content": "than some", + "type": "text" + }, + { + "bbox": [ + 150, + 400, + 178, + 412 + ], + "score": 0.9, + "content": "r _ { 1 } ^ { \\prime } \\geq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 399, + 421, + 413 + ], + "score": 1.0, + "content": ", we now prove the claim is true when the first coordinate of", + "type": "text" + }, + { + "bbox": [ + 421, + 401, + 428, + 410 + ], + "score": 0.83, + "content": "\\vec { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 399, + 472, + 413 + ], + "score": 1.0, + "content": "is equal to", + "type": "text" + }, + { + "bbox": [ + 472, + 400, + 482, + 411 + ], + "score": 0.86, + "content": "r _ { 1 } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 482, + 399, + 506, + 413 + ], + "score": 1.0, + "content": ". The", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 411, + 505, + 424 + ], + "spans": [ + { + "bbox": [ + 105, + 411, + 134, + 424 + ], + "score": 1.0, + "content": "vector", + "type": "text" + }, + { + "bbox": [ + 134, + 412, + 204, + 423 + ], + "score": 0.91, + "content": "\\tilde { \\boldsymbol { r } } = \\left( r _ { 2 } , \\ldots , r _ { K } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 411, + 264, + 424 + ], + "score": 1.0, + "content": "obtained from", + "type": "text" + }, + { + "bbox": [ + 264, + 411, + 271, + 421 + ], + "score": 0.81, + "content": "\\vec { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 411, + 438, + 424 + ], + "score": 1.0, + "content": "by removing the first coordinate, sums to", + "type": "text" + }, + { + "bbox": [ + 438, + 411, + 505, + 423 + ], + "score": 0.91, + "content": "D ^ { \\prime } = D - r _ { 1 } ^ { \\prime } <", + "type": "inline_equation" + } + ], + "index": 22 + }, + { + "bbox": [ + 107, + 422, + 506, + 436 + ], + "spans": [ + { + "bbox": [ + 107, + 423, + 116, + 433 + ], + "score": 0.79, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 422, + 275, + 436 + ], + "score": 1.0, + "content": ", and so by the induction hypothesis on", + "type": "text" + }, + { + "bbox": [ + 275, + 424, + 285, + 433 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 422, + 343, + 436 + ], + "score": 1.0, + "content": "we know that", + "type": "text" + }, + { + "bbox": [ + 344, + 422, + 435, + 435 + ], + "score": 0.92, + "content": "Q ^ { ( \\tilde { r } ) } \\in \\langle { \\mathcal { G } _ { 2 D ^ { \\prime } , D ^ { \\prime } } } , { \\mathcal { T } _ { D ^ { \\prime } } } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 422, + 506, + 436 + ], + "score": 1.0, + "content": ". By Lemma D.3", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 434, + 390, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 190, + 447 + ], + "score": 1.0, + "content": "we obtain a function", + "type": "text" + }, + { + "bbox": [ + 191, + 434, + 344, + 447 + ], + "score": 0.93, + "content": "q _ { a } \\in \\langle \\mathcal { G } _ { 2 D ^ { \\prime } + 1 , D ^ { \\prime } } , \\mathcal { T } _ { D } \\rangle \\subseteq \\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { D } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 434, + 390, + 447 + ], + "score": 1.0, + "content": "defined by", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22 + }, + { + "type": "interline_equation", + "bbox": [ + 168, + 451, + 442, + 534 + ], + "lines": [ + { + "bbox": [ + 168, + 451, + 442, + 534 + ], + "spans": [ + { + "bbox": [ + 168, + 451, + 442, + 534 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { q _ { a } ( X ) = \\displaystyle \\sum _ { b = 1 } ^ { n } ( \\bar { x } _ { a } - \\bar { x } _ { b } ) ^ { \\otimes r _ { 1 } } \\otimes Q _ { b } ^ { ( \\bar { r } ) } ( \\bar { X } ) } \\\\ & { \\qquad = \\displaystyle \\sum _ { b = 1 } ^ { n } ( \\bar { x } _ { a } - \\bar { x } _ { b } ) ^ { \\otimes r _ { 1 } } \\otimes \\bar { x } _ { b } ^ { \\otimes r _ { 2 } } \\otimes \\displaystyle \\sum _ { i _ { 3 } , \\ldots , i _ { K } = 1 } ^ { n } \\bar { x } _ { i _ { 3 } } ^ { \\otimes r _ { 3 } } \\otimes \\ldots \\otimes \\bar { x } _ { i _ { K } } ^ { \\otimes r _ { K } } } \\\\ & { \\qquad = Q _ { a } ^ { ( \\bar { r } ) } ( \\bar { X } ) + \\mathrm { a d d i t i o n a l ~ t e r m s } } \\end{array}", + "type": "interline_equation", + "image_path": "34ee578e2afa8951963148d2fdb7aac5d4c4b3f96827ae3eda4207445f2efbad.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 168, + 451, + 442, + 478.6666666666667 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 168, + 478.6666666666667, + 442, + 506.33333333333337 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 168, + 506.33333333333337, + 442, + 534.0 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 539, + 505, + 591 + ], + "lines": [ + { + "bbox": [ + 105, + 537, + 506, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 537, + 423, + 554 + ], + "score": 1.0, + "content": "where the additional terms are linear combinations of functions of the form", + "type": "text" + }, + { + "bbox": [ + 423, + 537, + 476, + 552 + ], + "score": 0.94, + "content": "P _ { D } Q _ { a } ^ { ( r ^ { \\prime } ) } ( \\bar { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 476, + 537, + 506, + 554 + ], + "score": 1.0, + "content": "where", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 107, + 550, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 107, + 551, + 143, + 563 + ], + "score": 0.91, + "content": "r ^ { \\prime } \\in \\Sigma _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 550, + 246, + 565 + ], + "score": 1.0, + "content": "and their first coordinate", + "type": "text" + }, + { + "bbox": [ + 246, + 553, + 257, + 563 + ], + "score": 0.84, + "content": "r _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 550, + 319, + 565 + ], + "score": 1.0, + "content": "is smaller than", + "type": "text" + }, + { + "bbox": [ + 320, + 552, + 330, + 563 + ], + "score": 0.89, + "content": "r _ { 1 } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 550, + 351, + 565 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 352, + 552, + 417, + 563 + ], + "score": 0.92, + "content": "P _ { D } : \\mathcal { T } _ { D } \\mathcal { T } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 550, + 505, + 565 + ], + "score": 1.0, + "content": "is a permutation. By", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 563, + 505, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 218, + 581 + ], + "score": 1.0, + "content": "the induction hypothesis on", + "type": "text" + }, + { + "bbox": [ + 218, + 568, + 228, + 577 + ], + "score": 0.84, + "content": "r _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 563, + 273, + 581 + ], + "score": 1.0, + "content": ", each such", + "type": "text" + }, + { + "bbox": [ + 274, + 564, + 295, + 578 + ], + "score": 0.92, + "content": "Q ^ { ( r ^ { \\prime } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 563, + 317, + 581 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 317, + 565, + 368, + 578 + ], + "score": 0.93, + "content": "\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { D } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 563, + 431, + 581 + ], + "score": 1.0, + "content": ". It follows that", + "type": "text" + }, + { + "bbox": [ + 431, + 563, + 505, + 578 + ], + "score": 0.93, + "content": "P _ { D } Q _ { a } ^ { ( r ^ { \\prime } ) } ( \\bar { X } ) , a =", + "type": "inline_equation" + } + ], + "index": 30 + }, + { + "bbox": [ + 107, + 577, + 494, + 591 + ], + "spans": [ + { + "bbox": [ + 107, + 579, + 141, + 590 + ], + "score": 0.89, + "content": "1 , \\ldots , n", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 577, + 181, + 591 + ], + "score": 1.0, + "content": ", and thus", + "type": "text" + }, + { + "bbox": [ + 181, + 577, + 217, + 591 + ], + "score": 0.93, + "content": "Q ^ { ( \\vec { r } ) } ( \\bar { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 577, + 246, + 591 + ], + "score": 1.0, + "content": ", are in", + "type": "text" + }, + { + "bbox": [ + 246, + 578, + 298, + 591 + ], + "score": 0.93, + "content": "\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { D } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 577, + 494, + 591 + ], + "score": 1.0, + "content": "as well. This concludes the proof of Theorem 2.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29.5 + }, + { + "type": "title", + "bbox": [ + 106, + 622, + 311, + 636 + ], + "lines": [ + { + "bbox": [ + 105, + 622, + 312, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 312, + 637 + ], + "score": 1.0, + "content": "E ALTERNATIVE TFN ARCHITECTURE", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 105, + 646, + 505, + 670 + ], + "lines": [ + { + "bbox": [ + 105, + 646, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 105, + 646, + 505, + 659 + ], + "score": 1.0, + "content": "In this appendix we show that replacing the standard TFN convolutional layer with the layer defined", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 658, + 169, + 670 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 169, + 670 + ], + "score": 1.0, + "content": "in equation 15:", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + }, + { + "type": "interline_equation", + "bbox": [ + 145, + 673, + 464, + 707 + ], + "lines": [ + { + "bbox": [ + 145, + 673, + 464, + 707 + ], + "spans": [ + { + "bbox": [ + 145, + 673, + 464, + 707 + ], + "score": 0.93, + "content": "\\tilde { V } _ { a } ( X , V ) = U ( l _ { f } , l _ { i } ) \\left( \\theta _ { 1 } \\sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \\otimes V _ { b } + \\theta _ { 2 } \\sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \\otimes V _ { a } \\right) ,", + "type": "interline_equation", + "image_path": "eef53c7de4865ff017c01000e5c67857f897f3177ae29f9e8392e839781cbb2c.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 145, + 673, + 464, + 684.3333333333334 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 145, + 684.3333333333334, + 464, + 695.6666666666667 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 145, + 695.6666666666667, + 464, + 707.0000000000001 + ], + "spans": [], + "index": 37 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 164, + 722 + ], + "score": 1.0, + "content": "we can obtain", + "type": "text" + }, + { + "bbox": [ + 164, + 710, + 173, + 720 + ], + "score": 0.84, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 709, + 277, + 722 + ], + "score": 1.0, + "content": "-spanning networks using", + "type": "text" + }, + { + "bbox": [ + 277, + 712, + 291, + 720 + ], + "score": 0.85, + "content": "2 D", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "consecutive convolutions with 1-filters (that is, filters", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 116, + 733 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 117, + 721, + 133, + 732 + ], + "score": 0.91, + "content": "W _ { l _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 720, + 164, + 733 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 164, + 720, + 213, + 732 + ], + "score": 0.93, + "content": "l _ { 1 } \\stackrel { \\cdot } { = } [ 0 , 1 ] ^ { \\frac { \\cdot } { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "). Our discussion here is somewhat informal, meant to provide the general", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + } + ], + "page_idx": 17, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 39 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 39 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "18", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 494, + 596, + 505, + 606 + ], + "lines": [ + { + "bbox": [ + 495, + 597, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 495, + 597, + 505, + 607 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 81, + 505, + 106 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 504, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 246, + 95 + ], + "score": 1.0, + "content": "As a final preliminary, we note that", + "type": "text" + }, + { + "bbox": [ + 246, + 83, + 256, + 92 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 81, + 466, + 95 + ], + "score": 1.0, + "content": "-filters can perform an averaging operation by setting", + "type": "text" + }, + { + "bbox": [ + 466, + 81, + 504, + 93 + ], + "score": 0.9, + "content": "R ^ { ( 0 ) } = 1", + "type": "inline_equation" + } + ], + "index": 0 + }, + { + "bbox": [ + 104, + 91, + 507, + 109 + ], + "spans": [ + { + "bbox": [ + 104, + 91, + 123, + 109 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 123, + 93, + 218, + 106 + ], + "score": 0.91, + "content": "\\theta _ { 0 } , R ^ { ( \\mathrm { 1 } ) } , \\ldots , R ^ { ( \\dot { D } ) } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 91, + 395, + 109 + ], + "score": 1.0, + "content": "in equation 13 and equation 14 . We call this", + "type": "text" + }, + { + "bbox": [ + 395, + 95, + 405, + 104 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 91, + 507, + 109 + ], + "score": 1.0, + "content": "-filter an averaging filter.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 104, + 81, + 507, + 109 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 110, + 505, + 157 + ], + "lines": [ + { + "bbox": [ + 105, + 110, + 505, + 124 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 391, + 124 + ], + "score": 1.0, + "content": "We are now ready to prove our claim: we need to show that for every", + "type": "text" + }, + { + "bbox": [ + 392, + 111, + 441, + 123 + ], + "score": 0.92, + "content": "D , T \\in \\mathbb { N } _ { + }", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 110, + 469, + 124 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 469, + 111, + 501, + 123 + ], + "score": 0.89, + "content": "T \\leq D", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 110, + 505, + 124 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 104, + 121, + 506, + 137 + ], + "spans": [ + { + "bbox": [ + 104, + 121, + 146, + 137 + ], + "score": 1.0, + "content": "for every", + "type": "text" + }, + { + "bbox": [ + 146, + 123, + 179, + 135 + ], + "score": 0.93, + "content": "\\vec { r } \\in \\Sigma _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 121, + 235, + 137 + ], + "score": 1.0, + "content": ", the function", + "type": "text" + }, + { + "bbox": [ + 235, + 122, + 254, + 135 + ], + "score": 0.92, + "content": "Q ^ { ( \\vec { r } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 121, + 277, + 137 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 278, + 123, + 328, + 136 + ], + "score": 0.93, + "content": "\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { T } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 121, + 506, + 137 + ], + "score": 1.0, + "content": ". Note that due to the inclusion relations in", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 134, + 505, + 147 + ], + "spans": [ + { + "bbox": [ + 105, + 134, + 315, + 147 + ], + "score": 1.0, + "content": "Lemma D.2 it is sufficient to prove this for the case", + "type": "text" + }, + { + "bbox": [ + 315, + 135, + 346, + 145 + ], + "score": 0.91, + "content": "T = D", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 134, + 474, + 147 + ], + "score": 1.0, + "content": ". We prove this by induction on", + "type": "text" + }, + { + "bbox": [ + 475, + 135, + 484, + 144 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 134, + 505, + 147 + ], + "score": 1.0, + "content": ". For", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 107, + 145, + 308, + 158 + ], + "spans": [ + { + "bbox": [ + 107, + 146, + 135, + 156 + ], + "score": 0.87, + "content": "D = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 145, + 169, + 158 + ], + "score": 1.0, + "content": ", vectors", + "type": "text" + }, + { + "bbox": [ + 170, + 146, + 199, + 156 + ], + "score": 0.92, + "content": "\\vec { r } \\in \\Sigma _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 145, + 308, + 158 + ], + "score": 1.0, + "content": "contains only zeros and so", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3.5, + "bbox_fs": [ + 104, + 110, + 506, + 158 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 215, + 160, + 396, + 176 + ], + "lines": [ + { + "bbox": [ + 215, + 160, + 396, + 176 + ], + "spans": [ + { + "bbox": [ + 215, + 160, + 396, + 176 + ], + "score": 0.91, + "content": "Q ^ { ( \\vec { r } ) } ( \\bar { X } ) = 1 _ { n } = \\pi _ { V } \\circ \\mathrm { e x t } ( X ) \\in \\langle \\mathcal { G } _ { 0 , 0 } , \\mathcal { T } _ { 0 } \\rangle .", + "type": "interline_equation", + "image_path": "6d9daefbbffc124f7a4c1212f098402ea43a3d5fd908e464d9176fcace8e1de8.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 215, + 160, + 396, + 176 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 184, + 505, + 230 + ], + "lines": [ + { + "bbox": [ + 106, + 184, + 506, + 198 + ], + "spans": [ + { + "bbox": [ + 106, + 184, + 266, + 198 + ], + "score": 1.0, + "content": "We now assume the claim is true for all", + "type": "text" + }, + { + "bbox": [ + 266, + 185, + 278, + 195 + ], + "score": 0.88, + "content": "D ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 184, + 299, + 198 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 300, + 185, + 353, + 196 + ], + "score": 0.92, + "content": "D > D ^ { \\prime } \\geq 0", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 184, + 475, + 198 + ], + "score": 1.0, + "content": "and prove the claim is true for", + "type": "text" + }, + { + "bbox": [ + 476, + 186, + 485, + 195 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 184, + 506, + 198 + ], + "score": 1.0, + "content": ". We", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 195, + 506, + 211 + ], + "spans": [ + { + "bbox": [ + 104, + 195, + 216, + 211 + ], + "score": 1.0, + "content": "need to show that for every", + "type": "text" + }, + { + "bbox": [ + 217, + 197, + 249, + 208 + ], + "score": 0.92, + "content": "\\vec { r } \\in \\Sigma _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 195, + 300, + 211 + ], + "score": 1.0, + "content": "the function", + "type": "text" + }, + { + "bbox": [ + 301, + 196, + 320, + 208 + ], + "score": 0.92, + "content": "Q ^ { ( \\vec { r } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 195, + 341, + 211 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 341, + 197, + 393, + 209 + ], + "score": 0.91, + "content": "\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { D } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 195, + 506, + 211 + ], + "score": 1.0, + "content": ". We prove this yet again by", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 207, + 505, + 220 + ], + "spans": [ + { + "bbox": [ + 105, + 207, + 249, + 220 + ], + "score": 1.0, + "content": "induction, this time on the value of", + "type": "text" + }, + { + "bbox": [ + 249, + 209, + 259, + 219 + ], + "score": 0.84, + "content": "r _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 207, + 313, + 220 + ], + "score": 1.0, + "content": ": assume that", + "type": "text" + }, + { + "bbox": [ + 314, + 208, + 346, + 219 + ], + "score": 0.91, + "content": "\\vec { r } \\in \\Sigma _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 207, + 365, + 220 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 365, + 209, + 396, + 219 + ], + "score": 0.85, + "content": "r _ { 1 } = 0 .", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 207, + 445, + 220 + ], + "score": 1.0, + "content": ". Denote by", + "type": "text" + }, + { + "bbox": [ + 445, + 208, + 451, + 218 + ], + "score": 0.81, + "content": "\\tilde { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 452, + 207, + 505, + 220 + ], + "score": 1.0, + "content": "the vector in", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 218, + 178, + 232 + ], + "spans": [ + { + "bbox": [ + 107, + 219, + 132, + 231 + ], + "score": 0.9, + "content": "\\Sigma _ { D - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 218, + 178, + 232 + ], + "score": 1.0, + "content": "defined by", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8.5, + "bbox_fs": [ + 104, + 184, + 506, + 232 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 253, + 229, + 357, + 242 + ], + "lines": [ + { + "bbox": [ + 253, + 229, + 357, + 242 + ], + "spans": [ + { + "bbox": [ + 253, + 229, + 357, + 242 + ], + "score": 0.89, + "content": "\\tilde { r } = ( r _ { 2 } - 1 , r _ { 3 } , \\ldots , r _ { K } ) .", + "type": "interline_equation", + "image_path": "46613d7f0da0ef0fce92ad630c0254ebf96380aefc66498e73438c4f46845f8a.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 253, + 229, + 357, + 242 + ], + "spans": [], + "index": 11 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 244, + 455, + 258 + ], + "lines": [ + { + "bbox": [ + 104, + 241, + 457, + 261 + ], + "spans": [ + { + "bbox": [ + 104, + 241, + 236, + 261 + ], + "score": 1.0, + "content": "By the induction assumption on", + "type": "text" + }, + { + "bbox": [ + 236, + 245, + 245, + 255 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 241, + 305, + 261 + ], + "score": 1.0, + "content": ", we know that", + "type": "text" + }, + { + "bbox": [ + 306, + 244, + 426, + 258 + ], + "score": 0.9, + "content": "Q ^ { ( \\tilde { r } ) } ( \\bar { X } ) \\in \\mathcal { G } _ { 2 ( D - 1 ) , D - 1 , D - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 241, + 457, + 261 + ], + "score": 1.0, + "content": "and so", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12, + "bbox_fs": [ + 104, + 241, + 457, + 261 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 112, + 261, + 497, + 330 + ], + "lines": [ + { + "bbox": [ + 112, + 261, + 497, + 330 + ], + "spans": [ + { + "bbox": [ + 112, + 261, + 497, + 330 + ], + "score": 0.93, + "content": "\\begin{array} { l } { \\displaystyle q _ { a } ( X ) = \\sum _ { b = 1 } ^ { n } ( \\bar { x } _ { a } - \\bar { x } _ { b } ) \\otimes Q _ { b } ^ { ( \\bar { r } ) } ( \\bar { X } ) = \\sum _ { b = 1 } ^ { n } ( \\bar { x } _ { a } - \\bar { x } _ { b } ) \\otimes \\bar { x } _ { b } ^ { \\otimes r _ { 2 } - 1 } \\otimes \\sum _ { i _ { 3 } , \\dots , i _ { K } = 1 } ^ { n } \\bar { x } _ { i _ { 3 } } ^ { \\otimes r _ { 3 } } \\otimes \\dots \\otimes \\bar { x } _ { i _ { K } } ^ { \\otimes r _ { K } } } \\\\ { \\displaystyle \\quad = \\left( \\bar { x } _ { a } \\otimes \\sum _ { b = 1 } ^ { n } Q _ { b } ^ { ( \\bar { r } ) } ( \\bar { X } ) \\right) - Q ^ { ( \\bar { r } ) } ( \\bar { X } ) } \\end{array}", + "type": "interline_equation", + "image_path": "c752720d6430d8f7ff000a662ea0032fc8b642b3343d74d0bf331ac24b8629ac.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 112, + 261, + 497, + 284.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 112, + 284.0, + 497, + 307.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 112, + 307.0, + 497, + 330.0 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 332, + 505, + 385 + ], + "lines": [ + { + "bbox": [ + 104, + 330, + 506, + 346 + ], + "spans": [ + { + "bbox": [ + 104, + 330, + 126, + 346 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 126, + 332, + 198, + 345 + ], + "score": 0.93, + "content": "\\langle \\mathcal { G } _ { 2 D - 1 , D - 1 } , \\mathcal { T } _ { D } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 330, + 352, + 346 + ], + "score": 1.0, + "content": "by Lemma D.3, which is contained in", + "type": "text" + }, + { + "bbox": [ + 352, + 333, + 414, + 345 + ], + "score": 0.92, + "content": "\\langle \\mathcal { G } _ { 2 D - 1 , D } , \\mathcal { T } _ { D } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 330, + 506, + 346 + ], + "score": 1.0, + "content": "by Lemma D.2. Since", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 344, + 506, + 361 + ], + "spans": [ + { + "bbox": [ + 106, + 348, + 118, + 358 + ], + "score": 0.84, + "content": "\\textstyle { \\bar { x } } _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 344, + 213, + 361 + ], + "score": 1.0, + "content": "has zero mean, while", + "type": "text" + }, + { + "bbox": [ + 214, + 344, + 250, + 359 + ], + "score": 0.94, + "content": "Q _ { a } ^ { ( \\vec { r } ) } ( \\bar { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 250, + 344, + 338, + 361 + ], + "score": 1.0, + "content": "does not depend on", + "type": "text" + }, + { + "bbox": [ + 338, + 349, + 345, + 357 + ], + "score": 0.72, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 344, + 371, + 361 + ], + "score": 1.0, + "content": "since", + "type": "text" + }, + { + "bbox": [ + 372, + 348, + 406, + 358 + ], + "score": 0.9, + "content": "r _ { 1 } ~ = ~ 0", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 344, + 506, + 361 + ], + "score": 1.0, + "content": ", applying an averaging", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 104, + 358, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 104, + 358, + 139, + 375 + ], + "score": 1.0, + "content": "filter to", + "type": "text" + }, + { + "bbox": [ + 139, + 363, + 150, + 372 + ], + "score": 0.82, + "content": "q _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 358, + 256, + 375 + ], + "score": 1.0, + "content": "gives us a constant value", + "type": "text" + }, + { + "bbox": [ + 256, + 358, + 300, + 373 + ], + "score": 0.94, + "content": "- Q _ { a } ^ { ( \\vec { r } ) } ( \\bar { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 358, + 379, + 375 + ], + "score": 1.0, + "content": "in each coordinate", + "type": "text" + }, + { + "bbox": [ + 380, + 361, + 412, + 373 + ], + "score": 0.92, + "content": "a \\in [ n ]", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 358, + 446, + 375 + ], + "score": 1.0, + "content": ", and so", + "type": "text" + }, + { + "bbox": [ + 447, + 360, + 483, + 373 + ], + "score": 0.93, + "content": "Q ^ { ( \\vec { r } ) } ( \\bar { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 358, + 506, + 375 + ], + "score": 1.0, + "content": "is in", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 107, + 371, + 163, + 385 + ], + "spans": [ + { + "bbox": [ + 107, + 372, + 159, + 384 + ], + "score": 0.91, + "content": "\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { D } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 371, + 163, + 385 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17.5, + "bbox_fs": [ + 104, + 330, + 506, + 385 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 388, + 505, + 447 + ], + "lines": [ + { + "bbox": [ + 105, + 389, + 505, + 401 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 253, + 401 + ], + "score": 1.0, + "content": "Now assume the claim is true for all", + "type": "text" + }, + { + "bbox": [ + 253, + 389, + 285, + 400 + ], + "score": 0.92, + "content": "\\vec { r } \\in \\Sigma _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 389, + 343, + 401 + ], + "score": 1.0, + "content": "which sum to", + "type": "text" + }, + { + "bbox": [ + 343, + 389, + 353, + 399 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 389, + 505, + 401 + ], + "score": 1.0, + "content": ", and whose first coordinate is smaller", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 399, + 506, + 413 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 150, + 413 + ], + "score": 1.0, + "content": "than some", + "type": "text" + }, + { + "bbox": [ + 150, + 400, + 178, + 412 + ], + "score": 0.9, + "content": "r _ { 1 } ^ { \\prime } \\geq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 399, + 421, + 413 + ], + "score": 1.0, + "content": ", we now prove the claim is true when the first coordinate of", + "type": "text" + }, + { + "bbox": [ + 421, + 401, + 428, + 410 + ], + "score": 0.83, + "content": "\\vec { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 399, + 472, + 413 + ], + "score": 1.0, + "content": "is equal to", + "type": "text" + }, + { + "bbox": [ + 472, + 400, + 482, + 411 + ], + "score": 0.86, + "content": "r _ { 1 } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 482, + 399, + 506, + 413 + ], + "score": 1.0, + "content": ". The", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 411, + 505, + 424 + ], + "spans": [ + { + "bbox": [ + 105, + 411, + 134, + 424 + ], + "score": 1.0, + "content": "vector", + "type": "text" + }, + { + "bbox": [ + 134, + 412, + 204, + 423 + ], + "score": 0.91, + "content": "\\tilde { \\boldsymbol { r } } = \\left( r _ { 2 } , \\ldots , r _ { K } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 411, + 264, + 424 + ], + "score": 1.0, + "content": "obtained from", + "type": "text" + }, + { + "bbox": [ + 264, + 411, + 271, + 421 + ], + "score": 0.81, + "content": "\\vec { r }", + "type": "inline_equation" + }, + { + "bbox": [ + 271, + 411, + 438, + 424 + ], + "score": 1.0, + "content": "by removing the first coordinate, sums to", + "type": "text" + }, + { + "bbox": [ + 438, + 411, + 505, + 423 + ], + "score": 0.91, + "content": "D ^ { \\prime } = D - r _ { 1 } ^ { \\prime } <", + "type": "inline_equation" + } + ], + "index": 22 + }, + { + "bbox": [ + 107, + 422, + 506, + 436 + ], + "spans": [ + { + "bbox": [ + 107, + 423, + 116, + 433 + ], + "score": 0.79, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 422, + 275, + 436 + ], + "score": 1.0, + "content": ", and so by the induction hypothesis on", + "type": "text" + }, + { + "bbox": [ + 275, + 424, + 285, + 433 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 422, + 343, + 436 + ], + "score": 1.0, + "content": "we know that", + "type": "text" + }, + { + "bbox": [ + 344, + 422, + 435, + 435 + ], + "score": 0.92, + "content": "Q ^ { ( \\tilde { r } ) } \\in \\langle { \\mathcal { G } _ { 2 D ^ { \\prime } , D ^ { \\prime } } } , { \\mathcal { T } _ { D ^ { \\prime } } } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 422, + 506, + 436 + ], + "score": 1.0, + "content": ". By Lemma D.3", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 434, + 390, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 190, + 447 + ], + "score": 1.0, + "content": "we obtain a function", + "type": "text" + }, + { + "bbox": [ + 191, + 434, + 344, + 447 + ], + "score": 0.93, + "content": "q _ { a } \\in \\langle \\mathcal { G } _ { 2 D ^ { \\prime } + 1 , D ^ { \\prime } } , \\mathcal { T } _ { D } \\rangle \\subseteq \\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { D } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 434, + 390, + 447 + ], + "score": 1.0, + "content": "defined by", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 389, + 506, + 447 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 168, + 451, + 442, + 534 + ], + "lines": [ + { + "bbox": [ + 168, + 451, + 442, + 534 + ], + "spans": [ + { + "bbox": [ + 168, + 451, + 442, + 534 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { q _ { a } ( X ) = \\displaystyle \\sum _ { b = 1 } ^ { n } ( \\bar { x } _ { a } - \\bar { x } _ { b } ) ^ { \\otimes r _ { 1 } } \\otimes Q _ { b } ^ { ( \\bar { r } ) } ( \\bar { X } ) } \\\\ & { \\qquad = \\displaystyle \\sum _ { b = 1 } ^ { n } ( \\bar { x } _ { a } - \\bar { x } _ { b } ) ^ { \\otimes r _ { 1 } } \\otimes \\bar { x } _ { b } ^ { \\otimes r _ { 2 } } \\otimes \\displaystyle \\sum _ { i _ { 3 } , \\ldots , i _ { K } = 1 } ^ { n } \\bar { x } _ { i _ { 3 } } ^ { \\otimes r _ { 3 } } \\otimes \\ldots \\otimes \\bar { x } _ { i _ { K } } ^ { \\otimes r _ { K } } } \\\\ & { \\qquad = Q _ { a } ^ { ( \\bar { r } ) } ( \\bar { X } ) + \\mathrm { a d d i t i o n a l ~ t e r m s } } \\end{array}", + "type": "interline_equation", + "image_path": "34ee578e2afa8951963148d2fdb7aac5d4c4b3f96827ae3eda4207445f2efbad.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 168, + 451, + 442, + 478.6666666666667 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 168, + 478.6666666666667, + 442, + 506.33333333333337 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 168, + 506.33333333333337, + 442, + 534.0 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 539, + 505, + 591 + ], + "lines": [ + { + "bbox": [ + 105, + 537, + 506, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 537, + 423, + 554 + ], + "score": 1.0, + "content": "where the additional terms are linear combinations of functions of the form", + "type": "text" + }, + { + "bbox": [ + 423, + 537, + 476, + 552 + ], + "score": 0.94, + "content": "P _ { D } Q _ { a } ^ { ( r ^ { \\prime } ) } ( \\bar { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 476, + 537, + 506, + 554 + ], + "score": 1.0, + "content": "where", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 107, + 550, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 107, + 551, + 143, + 563 + ], + "score": 0.91, + "content": "r ^ { \\prime } \\in \\Sigma _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 550, + 246, + 565 + ], + "score": 1.0, + "content": "and their first coordinate", + "type": "text" + }, + { + "bbox": [ + 246, + 553, + 257, + 563 + ], + "score": 0.84, + "content": "r _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 550, + 319, + 565 + ], + "score": 1.0, + "content": "is smaller than", + "type": "text" + }, + { + "bbox": [ + 320, + 552, + 330, + 563 + ], + "score": 0.89, + "content": "r _ { 1 } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 550, + 351, + 565 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 352, + 552, + 417, + 563 + ], + "score": 0.92, + "content": "P _ { D } : \\mathcal { T } _ { D } \\mathcal { T } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 417, + 550, + 505, + 565 + ], + "score": 1.0, + "content": "is a permutation. By", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 563, + 505, + 581 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 218, + 581 + ], + "score": 1.0, + "content": "the induction hypothesis on", + "type": "text" + }, + { + "bbox": [ + 218, + 568, + 228, + 577 + ], + "score": 0.84, + "content": "r _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 563, + 273, + 581 + ], + "score": 1.0, + "content": ", each such", + "type": "text" + }, + { + "bbox": [ + 274, + 564, + 295, + 578 + ], + "score": 0.92, + "content": "Q ^ { ( r ^ { \\prime } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 563, + 317, + 581 + ], + "score": 1.0, + "content": "is in", + "type": "text" + }, + { + "bbox": [ + 317, + 565, + 368, + 578 + ], + "score": 0.93, + "content": "\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { D } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 563, + 431, + 581 + ], + "score": 1.0, + "content": ". It follows that", + "type": "text" + }, + { + "bbox": [ + 431, + 563, + 505, + 578 + ], + "score": 0.93, + "content": "P _ { D } Q _ { a } ^ { ( r ^ { \\prime } ) } ( \\bar { X } ) , a =", + "type": "inline_equation" + } + ], + "index": 30 + }, + { + "bbox": [ + 107, + 577, + 494, + 591 + ], + "spans": [ + { + "bbox": [ + 107, + 579, + 141, + 590 + ], + "score": 0.89, + "content": "1 , \\ldots , n", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 577, + 181, + 591 + ], + "score": 1.0, + "content": ", and thus", + "type": "text" + }, + { + "bbox": [ + 181, + 577, + 217, + 591 + ], + "score": 0.93, + "content": "Q ^ { ( \\vec { r } ) } ( \\bar { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 577, + 246, + 591 + ], + "score": 1.0, + "content": ", are in", + "type": "text" + }, + { + "bbox": [ + 246, + 578, + 298, + 591 + ], + "score": 0.93, + "content": "\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { D } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 577, + 494, + 591 + ], + "score": 1.0, + "content": "as well. This concludes the proof of Theorem 2.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 537, + 506, + 591 + ] + }, + { + "type": "title", + "bbox": [ + 106, + 622, + 311, + 636 + ], + "lines": [ + { + "bbox": [ + 105, + 622, + 312, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 312, + 637 + ], + "score": 1.0, + "content": "E ALTERNATIVE TFN ARCHITECTURE", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 105, + 646, + 505, + 670 + ], + "lines": [ + { + "bbox": [ + 105, + 646, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 105, + 646, + 505, + 659 + ], + "score": 1.0, + "content": "In this appendix we show that replacing the standard TFN convolutional layer with the layer defined", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 658, + 169, + 670 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 169, + 670 + ], + "score": 1.0, + "content": "in equation 15:", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 646, + 505, + 670 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 145, + 673, + 464, + 707 + ], + "lines": [ + { + "bbox": [ + 145, + 673, + 464, + 707 + ], + "spans": [ + { + "bbox": [ + 145, + 673, + 464, + 707 + ], + "score": 0.93, + "content": "\\tilde { V } _ { a } ( X , V ) = U ( l _ { f } , l _ { i } ) \\left( \\theta _ { 1 } \\sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \\otimes V _ { b } + \\theta _ { 2 } \\sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \\otimes V _ { a } \\right) ,", + "type": "interline_equation", + "image_path": "eef53c7de4865ff017c01000e5c67857f897f3177ae29f9e8392e839781cbb2c.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 145, + 673, + 464, + 684.3333333333334 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 145, + 684.3333333333334, + 464, + 695.6666666666667 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 145, + 695.6666666666667, + 464, + 707.0000000000001 + ], + "spans": [], + "index": 37 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 164, + 722 + ], + "score": 1.0, + "content": "we can obtain", + "type": "text" + }, + { + "bbox": [ + 164, + 710, + 173, + 720 + ], + "score": 0.84, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 709, + 277, + 722 + ], + "score": 1.0, + "content": "-spanning networks using", + "type": "text" + }, + { + "bbox": [ + 277, + 712, + 291, + 720 + ], + "score": 0.85, + "content": "2 D", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "consecutive convolutions with 1-filters (that is, filters", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 116, + 733 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 117, + 721, + 133, + 732 + ], + "score": 0.91, + "content": "W _ { l _ { 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 720, + 164, + 733 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 164, + 720, + 213, + 732 + ], + "score": 0.93, + "content": "l _ { 1 } \\stackrel { \\cdot } { = } [ 0 , 1 ] ^ { \\frac { \\cdot } { T } }", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "). Our discussion here is somewhat informal, meant to provide the general", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 106, + 709, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 95 + ], + "score": 1.0, + "content": "ideas without delving into the details as we have done for the standard TFN architecture in the proof", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "of Theorem 2. In the end of our discussion we will explain what is necessary to make this argument", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 103, + 192, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 192, + 118 + ], + "score": 1.0, + "content": "completely rigorous.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 104, + 122, + 504, + 147 + ], + "lines": [ + { + "bbox": [ + 106, + 120, + 504, + 136 + ], + "spans": [ + { + "bbox": [ + 106, + 120, + 425, + 136 + ], + "score": 1.0, + "content": "We will only need two fixed filters for our argument here: The first is the 1-filter", + "type": "text" + }, + { + "bbox": [ + 425, + 121, + 504, + 135 + ], + "score": 0.92, + "content": "F _ { I d } = ( F ^ { ( 0 ) } , F ^ { ( 1 ) } )", + "type": "inline_equation" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 133, + 371, + 147 + ], + "spans": [ + { + "bbox": [ + 105, + 133, + 180, + 147 + ], + "score": 1.0, + "content": "defined by setting", + "type": "text" + }, + { + "bbox": [ + 180, + 133, + 241, + 147 + ], + "score": 0.94, + "content": "R ^ { ( 0 ) } ( \\| x \\| ) = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 133, + 259, + 147 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 259, + 133, + 331, + 147 + ], + "score": 0.93, + "content": "R ^ { ( 1 ) } ( \\| x \\| ) = \\| x \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 133, + 371, + 147 + ], + "score": 1.0, + "content": "to obtain", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "interline_equation", + "bbox": [ + 234, + 151, + 376, + 166 + ], + "lines": [ + { + "bbox": [ + 234, + 151, + 376, + 166 + ], + "spans": [ + { + "bbox": [ + 234, + 151, + 376, + 166 + ], + "score": 0.91, + "content": "F _ { I d } ( x ) = \\| x \\| Y ^ { 1 } ( \\hat { x } ) = \\| x \\| \\hat { x } = x .", + "type": "interline_equation", + "image_path": "9a1921dd046cebbb5ce79bcb65fe1fb560017efd5db92d99b0b00a6c1fa4eac1.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 234, + 151, + 376, + 166 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 171, + 461, + 185 + ], + "lines": [ + { + "bbox": [ + 104, + 169, + 461, + 187 + ], + "spans": [ + { + "bbox": [ + 104, + 169, + 200, + 187 + ], + "score": 1.0, + "content": "The second is the filter", + "type": "text" + }, + { + "bbox": [ + 200, + 173, + 212, + 183 + ], + "score": 0.88, + "content": "F _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 169, + 286, + 187 + ], + "score": 1.0, + "content": "defined by setting", + "type": "text" + }, + { + "bbox": [ + 287, + 171, + 348, + 185 + ], + "score": 0.93, + "content": "R ^ { ( 0 ) } ( \\| x \\| ) = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 169, + 366, + 187 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 366, + 171, + 427, + 185 + ], + "score": 0.92, + "content": "R ^ { ( 1 ) } ( \\| x \\| ) = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 169, + 461, + 187 + ], + "score": 1.0, + "content": ", so that", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "interline_equation", + "bbox": [ + 281, + 189, + 329, + 203 + ], + "lines": [ + { + "bbox": [ + 281, + 189, + 329, + 203 + ], + "spans": [ + { + "bbox": [ + 281, + 189, + 329, + 203 + ], + "score": 0.89, + "content": "F _ { 1 } ( x ) = 1 .", + "type": "interline_equation", + "image_path": "5875576ebd591118816095a2c2dc95015639c6eac18d036fa8898751f9709c77.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 281, + 189, + 329, + 203 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 213, + 505, + 258 + ], + "lines": [ + { + "bbox": [ + 106, + 214, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 214, + 505, + 226 + ], + "score": 1.0, + "content": "We prove our claim by showing that a pair of convolutions with 1-filters can construct any convolu-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 225, + 505, + 237 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 270, + 237 + ], + "score": 1.0, + "content": "tional layer defined in equation 9 for the", + "type": "text" + }, + { + "bbox": [ + 270, + 226, + 280, + 235 + ], + "score": 0.84, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 225, + 505, + 237 + ], + "score": 1.0, + "content": "-spanning architecture using tensor representations. The", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 234, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 256, + 249 + ], + "score": 1.0, + "content": "claim then follows from the fact that", + "type": "text" + }, + { + "bbox": [ + 257, + 236, + 266, + 246 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 234, + 505, + 249 + ], + "score": 1.0, + "content": "convolutions of the latter architecture suffice for achieving", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 107, + 246, + 252, + 259 + ], + "spans": [ + { + "bbox": [ + 107, + 247, + 116, + 257 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 246, + 252, + 259 + ], + "score": 1.0, + "content": "-spanning, as shown in Lemma 3.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 104, + 263, + 469, + 276 + ], + "lines": [ + { + "bbox": [ + 105, + 262, + 471, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 262, + 471, + 277 + ], + "score": 1.0, + "content": "Convolutions for tensor representations, defined in equation 9, are composed of two terms:", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "interline_equation", + "bbox": [ + 174, + 280, + 437, + 313 + ], + "lines": [ + { + "bbox": [ + 174, + 280, + 437, + 313 + ], + "spans": [ + { + "bbox": [ + 174, + 280, + 437, + 313 + ], + "score": 0.91, + "content": "\\tilde { V } _ { a } ^ { \\mathrm { t e n s o r } , 1 } ( \\bar { X } , V ) = \\bar { x } _ { a } \\otimes V _ { a } \\mathrm { a n d } \\tilde { V } _ { a } ^ { \\mathrm { t e n s o r } , 2 } ( \\bar { X } , V ) = \\sum _ { b = 1 } ^ { n } \\bar { x } _ { b } \\otimes V _ { b } .", + "type": "interline_equation", + "image_path": "888294d450b7cdaa02d29f68368082ecd50c514cf58ba38d5813b40204ef78e0.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 174, + 280, + 437, + 291.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 174, + 291.0, + 437, + 302.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 174, + 302.0, + 437, + 313.0 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 324, + 506, + 372 + ], + "lines": [ + { + "bbox": [ + 105, + 323, + 506, + 339 + ], + "spans": [ + { + "bbox": [ + 105, + 323, + 204, + 339 + ], + "score": 1.0, + "content": "To obtain the first term", + "type": "text" + }, + { + "bbox": [ + 205, + 324, + 241, + 338 + ], + "score": 0.89, + "content": "\\tilde { V } _ { a } ^ { \\mathrm { t e n s o r , 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 323, + 276, + 339 + ], + "score": 1.0, + "content": ", we set", + "type": "text" + }, + { + "bbox": [ + 276, + 326, + 398, + 338 + ], + "score": 0.9, + "content": "\\theta _ { 1 } = 0 , \\theta _ { 2 } = 1 / n , F = F _ { I d }", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 323, + 506, + 339 + ], + "score": 1.0, + "content": "in equation 15 we obtain", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 335, + 506, + 352 + ], + "spans": [ + { + "bbox": [ + 104, + 335, + 268, + 352 + ], + "score": 1.0, + "content": "(the decomposition into irreducibles of)", + "type": "text" + }, + { + "bbox": [ + 269, + 338, + 382, + 351 + ], + "score": 0.9, + "content": "\\tilde { V } _ { a } ^ { \\mathrm { t e n s o r , 1 } } ( \\bar { X } , V ) = \\bar { x } _ { a } \\otimes V _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 335, + 506, + 352 + ], + "score": 1.0, + "content": ". Thus this term can in fact be", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 350, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 350, + 505, + 362 + ], + "score": 1.0, + "content": "expressed by a single convolution. We can leave this outcome unchanged by a second convolution,", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 360, + 292, + 373 + ], + "spans": [ + { + "bbox": [ + 106, + 361, + 180, + 373 + ], + "score": 1.0, + "content": "defined by setting", + "type": "text" + }, + { + "bbox": [ + 180, + 360, + 288, + 372 + ], + "score": 0.93, + "content": "\\bar { \\theta } _ { 1 } = 0 , \\theta _ { 2 } = 1 / n , F = F _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 361, + 292, + 373 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 105, + 378, + 505, + 402 + ], + "lines": [ + { + "bbox": [ + 104, + 375, + 506, + 393 + ], + "spans": [ + { + "bbox": [ + 104, + 375, + 213, + 393 + ], + "score": 1.0, + "content": "To obtain the second term", + "type": "text" + }, + { + "bbox": [ + 214, + 377, + 250, + 391 + ], + "score": 0.87, + "content": "\\tilde { V } _ { a } ^ { \\mathrm { t e n s o r } , 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 251, + 375, + 389, + 393 + ], + "score": 1.0, + "content": ", we apply a first convolution with", + "type": "text" + }, + { + "bbox": [ + 390, + 379, + 501, + 391 + ], + "score": 0.92, + "content": "\\theta _ { 1 } = - 1 , F = F _ { I d } , \\theta _ { 2 } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 375, + 506, + 393 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 389, + 146, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 146, + 402 + ], + "score": 1.0, + "content": "to obtain", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5 + }, + { + "type": "interline_equation", + "bbox": [ + 150, + 404, + 459, + 437 + ], + "lines": [ + { + "bbox": [ + 150, + 404, + 459, + 437 + ], + "spans": [ + { + "bbox": [ + 150, + 404, + 459, + 437 + ], + "score": 0.93, + "content": "\\sum _ { b = 1 } ^ { n } ( x _ { b } - x _ { a } ) \\otimes V _ { b } = \\sum _ { b = 1 } ^ { n } ( { \\bar { x } } _ { b } - { \\bar { x } } _ { a } ) \\otimes V _ { b } = { \\tilde { V } } _ { a } ^ { { \\mathrm { t e n s o r } } , 2 } ( V , { \\bar { X } } ) - { \\bar { x } } _ { a } \\otimes \\sum _ { b = 1 } ^ { n } V _ { b }", + "type": "interline_equation", + "image_path": "1c0c4b0d07b2326bc145517fc85e20b6caa46009e6b670e27b6aa2c2db475bc1.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 150, + 404, + 459, + 415.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 150, + 415.0, + 459, + 426.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 150, + 426.0, + 459, + 437.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 443, + 504, + 469 + ], + "lines": [ + { + "bbox": [ + 105, + 442, + 506, + 457 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 354, + 457 + ], + "score": 1.0, + "content": "By applying an additional averaging filter, defined by setting", + "type": "text" + }, + { + "bbox": [ + 354, + 442, + 459, + 456 + ], + "score": 0.93, + "content": "\\textstyle \\theta _ { 1 } = { \\frac { 1 } { n } } , F = F _ { 1 } , \\theta _ { 2 } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 442, + 506, + 457 + ], + "score": 1.0, + "content": ", we obtain", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 107, + 451, + 327, + 473 + ], + "spans": [ + { + "bbox": [ + 107, + 456, + 171, + 469 + ], + "score": 0.89, + "content": "\\tilde { V } _ { a } ^ { \\mathrm { t e n s o r } , 2 } ( V , \\bar { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 451, + 327, + 473 + ], + "score": 1.0, + "content": ". This concludes our ‘informal proof’.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 106, + 473, + 505, + 529 + ], + "lines": [ + { + "bbox": [ + 105, + 473, + 504, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 383, + 487 + ], + "score": 1.0, + "content": "Our discussion here has been somewhat inaccurate, since in practice", + "type": "text" + }, + { + "bbox": [ + 384, + 474, + 504, + 486 + ], + "score": 0.93, + "content": "F _ { I d } ( x ) = ( 0 , x ) \\in W _ { 0 } \\oplus W _ { 1 }", + "type": "inline_equation" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 484, + 505, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 124, + 498 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 485, + 241, + 497 + ], + "score": 0.91, + "content": "F _ { 1 } ( x ) = ( 1 , 0 ) \\in W _ { 0 } \\oplus W _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 484, + 505, + 498 + ], + "score": 1.0, + "content": ". Moreover, in our proof we have glossed over the multiplication", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 496, + 504, + 507 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 504, + 507 + ], + "score": 1.0, + "content": "by the unitary matrix used to obtain decomposition into irreducible representations. However the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 506, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 286, + 519 + ], + "score": 1.0, + "content": "ideas discussed here can be used to show that", + "type": "text" + }, + { + "bbox": [ + 286, + 507, + 301, + 517 + ], + "score": 0.81, + "content": "2 D", + "type": "inline_equation" + }, + { + "bbox": [ + 301, + 506, + 505, + 519 + ], + "score": 1.0, + "content": "convolutions with 1-filters can satisfy the sufficient", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 518, + 493, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 160, + 530 + ], + "score": 1.0, + "content": "condition for", + "type": "text" + }, + { + "bbox": [ + 161, + 518, + 171, + 528 + ], + "score": 0.84, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 518, + 493, + 530 + ], + "score": 1.0, + "content": "-spanning defined in Lemma 4. See our treatment of Theorem 2 for more details.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29 + }, + { + "type": "title", + "bbox": [ + 107, + 545, + 343, + 558 + ], + "lines": [ + { + "bbox": [ + 105, + 545, + 343, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 343, + 559 + ], + "score": 1.0, + "content": "F COMPARISON WITH ORIGINAL TFN PAPER", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 570, + 504, + 592 + ], + "lines": [ + { + "bbox": [ + 105, + 570, + 504, + 582 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 504, + 582 + ], + "score": 1.0, + "content": "In this Appendix we discuss three superficial differences between the presentation of the TFN archi-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 581, + 338, + 594 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 338, + 594 + ], + "score": 1.0, + "content": "tecture in Thomas et al. (2018) and our presentation here:", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 130, + 601, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 129, + 601, + 504, + 615 + ], + "spans": [ + { + "bbox": [ + 129, + 601, + 504, + 615 + ], + "score": 1.0, + "content": "1. We define convolutional layers between features residing in direct sums of irreducible rep-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 141, + 613, + 505, + 625 + ], + "spans": [ + { + "bbox": [ + 141, + 613, + 505, + 625 + ], + "score": 1.0, + "content": "resentations, while (Thomas et al., 2018) focuses on features which inhabit a single irre-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 141, + 624, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 141, + 624, + 505, + 637 + ], + "score": 1.0, + "content": "ducible representation. This difference is non-essential, as direct sums of irreducible rep-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 141, + 635, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 141, + 635, + 505, + 648 + ], + "score": 1.0, + "content": "resentations can be represented as multiple channels where each feature inhabits a single", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 142, + 646, + 249, + 658 + ], + "spans": [ + { + "bbox": [ + 142, + 646, + 249, + 658 + ], + "score": 1.0, + "content": "irreducible representation.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 130, + 660, + 505, + 674 + ], + "spans": [ + { + "bbox": [ + 130, + 660, + 180, + 674 + ], + "score": 1.0, + "content": "2. The term", + "type": "text" + }, + { + "bbox": [ + 181, + 660, + 201, + 672 + ], + "score": 0.91, + "content": "\\theta _ { 0 } V _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 202, + 660, + 505, + 674 + ], + "score": 1.0, + "content": "in equation 14 appears in (Fuchs et al., 2020), but does not appear explicitly", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 141, + 671, + 506, + 685 + ], + "spans": [ + { + "bbox": [ + 141, + 671, + 506, + 685 + ], + "score": 1.0, + "content": "in (Thomas et al., 2018). However it can be obtained by concatenation of the input of a", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 141, + 683, + 451, + 696 + ], + "spans": [ + { + "bbox": [ + 141, + 683, + 451, + 696 + ], + "score": 1.0, + "content": "self-interaction layer to the output, and then applying a self-interaction layer.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 128, + 697, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 128, + 697, + 256, + 712 + ], + "score": 1.0, + "content": "3. We take the scalar functions", + "type": "text" + }, + { + "bbox": [ + 257, + 698, + 275, + 709 + ], + "score": 0.9, + "content": "R ^ { ( \\ell ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 697, + 506, + 712 + ], + "score": 1.0, + "content": "to be polynomials, while (Thomas et al., 2018) take them", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 141, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 141, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "to be fully connected networks composed with radial basis functions. Using polynomial", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 141, + 720, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 141, + 720, + 506, + 734 + ], + "score": 1.0, + "content": "scalar bases is convenient for our presentation here since it enables exact expression of", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 40 + } + ], + "page_idx": 18, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 301, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 95 + ], + "score": 1.0, + "content": "ideas without delving into the details as we have done for the standard TFN architecture in the proof", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "of Theorem 2. In the end of our discussion we will explain what is necessary to make this argument", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 103, + 192, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 192, + 118 + ], + "score": 1.0, + "content": "completely rigorous.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 82, + 506, + 118 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 122, + 504, + 147 + ], + "lines": [ + { + "bbox": [ + 106, + 120, + 504, + 136 + ], + "spans": [ + { + "bbox": [ + 106, + 120, + 425, + 136 + ], + "score": 1.0, + "content": "We will only need two fixed filters for our argument here: The first is the 1-filter", + "type": "text" + }, + { + "bbox": [ + 425, + 121, + 504, + 135 + ], + "score": 0.92, + "content": "F _ { I d } = ( F ^ { ( 0 ) } , F ^ { ( 1 ) } )", + "type": "inline_equation" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 133, + 371, + 147 + ], + "spans": [ + { + "bbox": [ + 105, + 133, + 180, + 147 + ], + "score": 1.0, + "content": "defined by setting", + "type": "text" + }, + { + "bbox": [ + 180, + 133, + 241, + 147 + ], + "score": 0.94, + "content": "R ^ { ( 0 ) } ( \\| x \\| ) = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 133, + 259, + 147 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 259, + 133, + 331, + 147 + ], + "score": 0.93, + "content": "R ^ { ( 1 ) } ( \\| x \\| ) = \\| x \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 133, + 371, + 147 + ], + "score": 1.0, + "content": "to obtain", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 105, + 120, + 504, + 147 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 234, + 151, + 376, + 166 + ], + "lines": [ + { + "bbox": [ + 234, + 151, + 376, + 166 + ], + "spans": [ + { + "bbox": [ + 234, + 151, + 376, + 166 + ], + "score": 0.91, + "content": "F _ { I d } ( x ) = \\| x \\| Y ^ { 1 } ( \\hat { x } ) = \\| x \\| \\hat { x } = x .", + "type": "interline_equation", + "image_path": "9a1921dd046cebbb5ce79bcb65fe1fb560017efd5db92d99b0b00a6c1fa4eac1.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 234, + 151, + 376, + 166 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 171, + 461, + 185 + ], + "lines": [ + { + "bbox": [ + 104, + 169, + 461, + 187 + ], + "spans": [ + { + "bbox": [ + 104, + 169, + 200, + 187 + ], + "score": 1.0, + "content": "The second is the filter", + "type": "text" + }, + { + "bbox": [ + 200, + 173, + 212, + 183 + ], + "score": 0.88, + "content": "F _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 169, + 286, + 187 + ], + "score": 1.0, + "content": "defined by setting", + "type": "text" + }, + { + "bbox": [ + 287, + 171, + 348, + 185 + ], + "score": 0.93, + "content": "R ^ { ( 0 ) } ( \\| x \\| ) = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 169, + 366, + 187 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 366, + 171, + 427, + 185 + ], + "score": 0.92, + "content": "R ^ { ( 1 ) } ( \\| x \\| ) = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 169, + 461, + 187 + ], + "score": 1.0, + "content": ", so that", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6, + "bbox_fs": [ + 104, + 169, + 461, + 187 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 281, + 189, + 329, + 203 + ], + "lines": [ + { + "bbox": [ + 281, + 189, + 329, + 203 + ], + "spans": [ + { + "bbox": [ + 281, + 189, + 329, + 203 + ], + "score": 0.89, + "content": "F _ { 1 } ( x ) = 1 .", + "type": "interline_equation", + "image_path": "5875576ebd591118816095a2c2dc95015639c6eac18d036fa8898751f9709c77.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 281, + 189, + 329, + 203 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 213, + 505, + 258 + ], + "lines": [ + { + "bbox": [ + 106, + 214, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 106, + 214, + 505, + 226 + ], + "score": 1.0, + "content": "We prove our claim by showing that a pair of convolutions with 1-filters can construct any convolu-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 225, + 505, + 237 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 270, + 237 + ], + "score": 1.0, + "content": "tional layer defined in equation 9 for the", + "type": "text" + }, + { + "bbox": [ + 270, + 226, + 280, + 235 + ], + "score": 0.84, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 225, + 505, + 237 + ], + "score": 1.0, + "content": "-spanning architecture using tensor representations. The", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 234, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 256, + 249 + ], + "score": 1.0, + "content": "claim then follows from the fact that", + "type": "text" + }, + { + "bbox": [ + 257, + 236, + 266, + 246 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 234, + 505, + 249 + ], + "score": 1.0, + "content": "convolutions of the latter architecture suffice for achieving", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 107, + 246, + 252, + 259 + ], + "spans": [ + { + "bbox": [ + 107, + 247, + 116, + 257 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 246, + 252, + 259 + ], + "score": 1.0, + "content": "-spanning, as shown in Lemma 3.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 214, + 505, + 259 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 263, + 469, + 276 + ], + "lines": [ + { + "bbox": [ + 105, + 262, + 471, + 277 + ], + "spans": [ + { + "bbox": [ + 105, + 262, + 471, + 277 + ], + "score": 1.0, + "content": "Convolutions for tensor representations, defined in equation 9, are composed of two terms:", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12, + "bbox_fs": [ + 105, + 262, + 471, + 277 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 174, + 280, + 437, + 313 + ], + "lines": [ + { + "bbox": [ + 174, + 280, + 437, + 313 + ], + "spans": [ + { + "bbox": [ + 174, + 280, + 437, + 313 + ], + "score": 0.91, + "content": "\\tilde { V } _ { a } ^ { \\mathrm { t e n s o r } , 1 } ( \\bar { X } , V ) = \\bar { x } _ { a } \\otimes V _ { a } \\mathrm { a n d } \\tilde { V } _ { a } ^ { \\mathrm { t e n s o r } , 2 } ( \\bar { X } , V ) = \\sum _ { b = 1 } ^ { n } \\bar { x } _ { b } \\otimes V _ { b } .", + "type": "interline_equation", + "image_path": "888294d450b7cdaa02d29f68368082ecd50c514cf58ba38d5813b40204ef78e0.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 174, + 280, + 437, + 291.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 174, + 291.0, + 437, + 302.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 174, + 302.0, + 437, + 313.0 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 324, + 506, + 372 + ], + "lines": [ + { + "bbox": [ + 105, + 323, + 506, + 339 + ], + "spans": [ + { + "bbox": [ + 105, + 323, + 204, + 339 + ], + "score": 1.0, + "content": "To obtain the first term", + "type": "text" + }, + { + "bbox": [ + 205, + 324, + 241, + 338 + ], + "score": 0.89, + "content": "\\tilde { V } _ { a } ^ { \\mathrm { t e n s o r , 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 323, + 276, + 339 + ], + "score": 1.0, + "content": ", we set", + "type": "text" + }, + { + "bbox": [ + 276, + 326, + 398, + 338 + ], + "score": 0.9, + "content": "\\theta _ { 1 } = 0 , \\theta _ { 2 } = 1 / n , F = F _ { I d }", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 323, + 506, + 339 + ], + "score": 1.0, + "content": "in equation 15 we obtain", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 335, + 506, + 352 + ], + "spans": [ + { + "bbox": [ + 104, + 335, + 268, + 352 + ], + "score": 1.0, + "content": "(the decomposition into irreducibles of)", + "type": "text" + }, + { + "bbox": [ + 269, + 338, + 382, + 351 + ], + "score": 0.9, + "content": "\\tilde { V } _ { a } ^ { \\mathrm { t e n s o r , 1 } } ( \\bar { X } , V ) = \\bar { x } _ { a } \\otimes V _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 335, + 506, + 352 + ], + "score": 1.0, + "content": ". Thus this term can in fact be", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 350, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 106, + 350, + 505, + 362 + ], + "score": 1.0, + "content": "expressed by a single convolution. We can leave this outcome unchanged by a second convolution,", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 360, + 292, + 373 + ], + "spans": [ + { + "bbox": [ + 106, + 361, + 180, + 373 + ], + "score": 1.0, + "content": "defined by setting", + "type": "text" + }, + { + "bbox": [ + 180, + 360, + 288, + 372 + ], + "score": 0.93, + "content": "\\bar { \\theta } _ { 1 } = 0 , \\theta _ { 2 } = 1 / n , F = F _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 361, + 292, + 373 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17.5, + "bbox_fs": [ + 104, + 323, + 506, + 373 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 378, + 505, + 402 + ], + "lines": [ + { + "bbox": [ + 104, + 375, + 506, + 393 + ], + "spans": [ + { + "bbox": [ + 104, + 375, + 213, + 393 + ], + "score": 1.0, + "content": "To obtain the second term", + "type": "text" + }, + { + "bbox": [ + 214, + 377, + 250, + 391 + ], + "score": 0.87, + "content": "\\tilde { V } _ { a } ^ { \\mathrm { t e n s o r } , 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 251, + 375, + 389, + 393 + ], + "score": 1.0, + "content": ", we apply a first convolution with", + "type": "text" + }, + { + "bbox": [ + 390, + 379, + 501, + 391 + ], + "score": 0.92, + "content": "\\theta _ { 1 } = - 1 , F = F _ { I d } , \\theta _ { 2 } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 375, + 506, + 393 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 389, + 146, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 146, + 402 + ], + "score": 1.0, + "content": "to obtain", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5, + "bbox_fs": [ + 104, + 375, + 506, + 402 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 150, + 404, + 459, + 437 + ], + "lines": [ + { + "bbox": [ + 150, + 404, + 459, + 437 + ], + "spans": [ + { + "bbox": [ + 150, + 404, + 459, + 437 + ], + "score": 0.93, + "content": "\\sum _ { b = 1 } ^ { n } ( x _ { b } - x _ { a } ) \\otimes V _ { b } = \\sum _ { b = 1 } ^ { n } ( { \\bar { x } } _ { b } - { \\bar { x } } _ { a } ) \\otimes V _ { b } = { \\tilde { V } } _ { a } ^ { { \\mathrm { t e n s o r } } , 2 } ( V , { \\bar { X } } ) - { \\bar { x } } _ { a } \\otimes \\sum _ { b = 1 } ^ { n } V _ { b }", + "type": "interline_equation", + "image_path": "1c0c4b0d07b2326bc145517fc85e20b6caa46009e6b670e27b6aa2c2db475bc1.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 150, + 404, + 459, + 415.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 150, + 415.0, + 459, + 426.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 150, + 426.0, + 459, + 437.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 443, + 504, + 469 + ], + "lines": [ + { + "bbox": [ + 105, + 442, + 506, + 457 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 354, + 457 + ], + "score": 1.0, + "content": "By applying an additional averaging filter, defined by setting", + "type": "text" + }, + { + "bbox": [ + 354, + 442, + 459, + 456 + ], + "score": 0.93, + "content": "\\textstyle \\theta _ { 1 } = { \\frac { 1 } { n } } , F = F _ { 1 } , \\theta _ { 2 } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 442, + 506, + 457 + ], + "score": 1.0, + "content": ", we obtain", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 107, + 451, + 327, + 473 + ], + "spans": [ + { + "bbox": [ + 107, + 456, + 171, + 469 + ], + "score": 0.89, + "content": "\\tilde { V } _ { a } ^ { \\mathrm { t e n s o r } , 2 } ( V , \\bar { X } )", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 451, + 327, + 473 + ], + "score": 1.0, + "content": ". This concludes our ‘informal proof’.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 105, + 442, + 506, + 473 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 473, + 505, + 529 + ], + "lines": [ + { + "bbox": [ + 105, + 473, + 504, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 383, + 487 + ], + "score": 1.0, + "content": "Our discussion here has been somewhat inaccurate, since in practice", + "type": "text" + }, + { + "bbox": [ + 384, + 474, + 504, + 486 + ], + "score": 0.93, + "content": "F _ { I d } ( x ) = ( 0 , x ) \\in W _ { 0 } \\oplus W _ { 1 }", + "type": "inline_equation" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 484, + 505, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 124, + 498 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 485, + 241, + 497 + ], + "score": 0.91, + "content": "F _ { 1 } ( x ) = ( 1 , 0 ) \\in W _ { 0 } \\oplus W _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 484, + 505, + 498 + ], + "score": 1.0, + "content": ". Moreover, in our proof we have glossed over the multiplication", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 496, + 504, + 507 + ], + "spans": [ + { + "bbox": [ + 106, + 496, + 504, + 507 + ], + "score": 1.0, + "content": "by the unitary matrix used to obtain decomposition into irreducible representations. However the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 506, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 286, + 519 + ], + "score": 1.0, + "content": "ideas discussed here can be used to show that", + "type": "text" + }, + { + "bbox": [ + 286, + 507, + 301, + 517 + ], + "score": 0.81, + "content": "2 D", + "type": "inline_equation" + }, + { + "bbox": [ + 301, + 506, + 505, + 519 + ], + "score": 1.0, + "content": "convolutions with 1-filters can satisfy the sufficient", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 518, + 493, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 160, + 530 + ], + "score": 1.0, + "content": "condition for", + "type": "text" + }, + { + "bbox": [ + 161, + 518, + 171, + 528 + ], + "score": 0.84, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 518, + 493, + 530 + ], + "score": 1.0, + "content": "-spanning defined in Lemma 4. See our treatment of Theorem 2 for more details.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 473, + 505, + 530 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 545, + 343, + 558 + ], + "lines": [ + { + "bbox": [ + 105, + 545, + 343, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 343, + 559 + ], + "score": 1.0, + "content": "F COMPARISON WITH ORIGINAL TFN PAPER", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 570, + 504, + 592 + ], + "lines": [ + { + "bbox": [ + 105, + 570, + 504, + 582 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 504, + 582 + ], + "score": 1.0, + "content": "In this Appendix we discuss three superficial differences between the presentation of the TFN archi-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 581, + 338, + 594 + ], + "spans": [ + { + "bbox": [ + 105, + 581, + 338, + 594 + ], + "score": 1.0, + "content": "tecture in Thomas et al. (2018) and our presentation here:", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 570, + 504, + 594 + ] + }, + { + "type": "text", + "bbox": [ + 130, + 601, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 129, + 601, + 504, + 615 + ], + "spans": [ + { + "bbox": [ + 129, + 601, + 504, + 615 + ], + "score": 1.0, + "content": "1. We define convolutional layers between features residing in direct sums of irreducible rep-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 141, + 613, + 505, + 625 + ], + "spans": [ + { + "bbox": [ + 141, + 613, + 505, + 625 + ], + "score": 1.0, + "content": "resentations, while (Thomas et al., 2018) focuses on features which inhabit a single irre-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 141, + 624, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 141, + 624, + 505, + 637 + ], + "score": 1.0, + "content": "ducible representation. This difference is non-essential, as direct sums of irreducible rep-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 141, + 635, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 141, + 635, + 505, + 648 + ], + "score": 1.0, + "content": "resentations can be represented as multiple channels where each feature inhabits a single", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 142, + 646, + 249, + 658 + ], + "spans": [ + { + "bbox": [ + 142, + 646, + 249, + 658 + ], + "score": 1.0, + "content": "irreducible representation.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 130, + 660, + 505, + 674 + ], + "spans": [ + { + "bbox": [ + 130, + 660, + 180, + 674 + ], + "score": 1.0, + "content": "2. The term", + "type": "text" + }, + { + "bbox": [ + 181, + 660, + 201, + 672 + ], + "score": 0.91, + "content": "\\theta _ { 0 } V _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 202, + 660, + 505, + 674 + ], + "score": 1.0, + "content": "in equation 14 appears in (Fuchs et al., 2020), but does not appear explicitly", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 141, + 671, + 506, + 685 + ], + "spans": [ + { + "bbox": [ + 141, + 671, + 506, + 685 + ], + "score": 1.0, + "content": "in (Thomas et al., 2018). However it can be obtained by concatenation of the input of a", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 141, + 683, + 451, + 696 + ], + "spans": [ + { + "bbox": [ + 141, + 683, + 451, + 696 + ], + "score": 1.0, + "content": "self-interaction layer to the output, and then applying a self-interaction layer.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 128, + 697, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 128, + 697, + 256, + 712 + ], + "score": 1.0, + "content": "3. We take the scalar functions", + "type": "text" + }, + { + "bbox": [ + 257, + 698, + 275, + 709 + ], + "score": 0.9, + "content": "R ^ { ( \\ell ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 697, + 506, + 712 + ], + "score": 1.0, + "content": "to be polynomials, while (Thomas et al., 2018) take them", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 141, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 141, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "to be fully connected networks composed with radial basis functions. Using polynomial", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 141, + 720, + 506, + 734 + ], + "spans": [ + { + "bbox": [ + 141, + 720, + 506, + 734 + ], + "score": 1.0, + "content": "scalar bases is convenient for our presentation here since it enables exact expression of", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 141, + 83, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 141, + 83, + 506, + 95 + ], + "score": 1.0, + "content": "equivariant polynomials. Replacing polynomial bases with fully connected networks, we", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 141, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 141, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "obtain approximation of equivariant polynomials instead of exact expression. It can be", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 141, + 105, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 141, + 105, + 199, + 117 + ], + "score": 1.0, + "content": "shown that if", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 199, + 106, + 206, + 116 + ], + "score": 0.81, + "content": "p", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 206, + 105, + 225, + 117 + ], + "score": 1.0, + "content": "is a", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 225, + 105, + 234, + 114 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 235, + 105, + 506, + 117 + ], + "score": 1.0, + "content": "-equivariant polynomial which can be expressed by some network", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 142, + 115, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 142, + 115, + 166, + 128 + ], + "score": 0.91, + "content": "\\mathcal { F } _ { C , D }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 167, + 115, + 433, + 128 + ], + "score": 1.0, + "content": "defined with filters coming from a polynomial scalar basis, then", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 433, + 117, + 440, + 127 + ], + "score": 0.75, + "content": "p", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 441, + 115, + 506, + 128 + ], + "score": 1.0, + "content": "can be approx-", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 141, + 126, + 506, + 139 + ], + "spans": [ + { + "bbox": [ + 141, + 126, + 246, + 139 + ], + "score": 1.0, + "content": "imated on a compact set", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 246, + 127, + 256, + 137 + ], + "score": 0.79, + "content": "K", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 257, + 126, + 336, + 139 + ], + "score": 1.0, + "content": ", up to an arbitrary", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 337, + 129, + 343, + 136 + ], + "score": 0.65, + "content": "\\epsilon", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 343, + 126, + 506, + 139 + ], + "score": 1.0, + "content": "error, by a similar network with scalar", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 141, + 137, + 414, + 150 + ], + "spans": [ + { + "bbox": [ + 141, + 137, + 414, + 150 + ], + "score": 1.0, + "content": "functions coming from a sufficiently large fully connected network.", + "type": "text", + "cross_page": true + } + ], + "index": 5 + } + ], + "index": 40, + "bbox_fs": [ + 128, + 601, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 142, + 82, + 505, + 149 + ], + "lines": [ + { + "bbox": [ + 141, + 83, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 141, + 83, + 506, + 95 + ], + "score": 1.0, + "content": "equivariant polynomials. Replacing polynomial bases with fully connected networks, we", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 141, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 141, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "obtain approximation of equivariant polynomials instead of exact expression. It can be", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 141, + 105, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 141, + 105, + 199, + 117 + ], + "score": 1.0, + "content": "shown that if", + "type": "text" + }, + { + "bbox": [ + 199, + 106, + 206, + 116 + ], + "score": 0.81, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 105, + 225, + 117 + ], + "score": 1.0, + "content": "is a", + "type": "text" + }, + { + "bbox": [ + 225, + 105, + 234, + 114 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 105, + 506, + 117 + ], + "score": 1.0, + "content": "-equivariant polynomial which can be expressed by some network", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 142, + 115, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 142, + 115, + 166, + 128 + ], + "score": 0.91, + "content": "\\mathcal { F } _ { C , D }", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 115, + 433, + 128 + ], + "score": 1.0, + "content": "defined with filters coming from a polynomial scalar basis, then", + "type": "text" + }, + { + "bbox": [ + 433, + 117, + 440, + 127 + ], + "score": 0.75, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 441, + 115, + 506, + 128 + ], + "score": 1.0, + "content": "can be approx-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 141, + 126, + 506, + 139 + ], + "spans": [ + { + "bbox": [ + 141, + 126, + 246, + 139 + ], + "score": 1.0, + "content": "imated on a compact set", + "type": "text" + }, + { + "bbox": [ + 246, + 127, + 256, + 137 + ], + "score": 0.79, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 126, + 336, + 139 + ], + "score": 1.0, + "content": ", up to an arbitrary", + "type": "text" + }, + { + "bbox": [ + 337, + 129, + 343, + 136 + ], + "score": 0.65, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 126, + 506, + 139 + ], + "score": 1.0, + "content": "error, by a similar network with scalar", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 141, + 137, + 414, + 150 + ], + "spans": [ + { + "bbox": [ + 141, + 137, + 414, + 150 + ], + "score": 1.0, + "content": "functions coming from a sufficiently large fully connected network.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 107, + 164, + 253, + 177 + ], + "lines": [ + { + "bbox": [ + 105, + 163, + 254, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 163, + 254, + 180 + ], + "score": 1.0, + "content": "G TENSOR UNIVERSALITY", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 189, + 504, + 216 + ], + "lines": [ + { + "bbox": [ + 105, + 187, + 506, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 187, + 333, + 204 + ], + "score": 1.0, + "content": "In this section we show how to construct the complete set", + "type": "text" + }, + { + "bbox": [ + 334, + 190, + 357, + 202 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { \\mathrm { p o o l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 187, + 393, + 204 + ], + "score": 1.0, + "content": "of linear", + "type": "text" + }, + { + "bbox": [ + 393, + 189, + 420, + 201 + ], + "score": 0.25, + "content": "\\mathrm { S O ( 3 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 187, + 506, + 204 + ], + "score": 1.0, + "content": "invariant functionals", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 199, + 407, + 220 + ], + "spans": [ + { + "bbox": [ + 104, + 199, + 128, + 220 + ], + "score": 1.0, + "content": "from", + "type": "text" + }, + { + "bbox": [ + 129, + 201, + 207, + 216 + ], + "score": 0.93, + "content": "W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } } = \\bigoplus _ { T = 0 } ^ { D } \\mathcal { T } _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 199, + 218, + 220 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 218, + 203, + 226, + 213 + ], + "score": 0.8, + "content": "\\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 199, + 340, + 220 + ], + "score": 1.0, + "content": ". Since each such functional", + "type": "text" + }, + { + "bbox": [ + 340, + 204, + 348, + 213 + ], + "score": 0.84, + "content": "\\Lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 199, + 407, + 220 + ], + "score": 1.0, + "content": "is of the form", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "interline_equation", + "bbox": [ + 239, + 219, + 371, + 255 + ], + "lines": [ + { + "bbox": [ + 239, + 219, + 371, + 255 + ], + "spans": [ + { + "bbox": [ + 239, + 219, + 371, + 255 + ], + "score": 0.94, + "content": "\\Lambda ( w _ { 0 } , . . . , w _ { D } ) = \\sum _ { T = 0 } ^ { D } \\Lambda _ { T } ( w _ { T } ) ,", + "type": "interline_equation", + "image_path": "b57cd45c52bd31dbfd6a60a04df3da149300563b442ea80051d826ec75367796.jpg" + } + ] + } + ], + "index": 9.5, + "virtual_lines": [ + { + "bbox": [ + 239, + 219, + 371, + 237.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 239, + 237.0, + 371, + 255.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 259, + 504, + 281 + ], + "lines": [ + { + "bbox": [ + 106, + 257, + 504, + 271 + ], + "spans": [ + { + "bbox": [ + 106, + 257, + 153, + 271 + ], + "score": 1.0, + "content": "where each", + "type": "text" + }, + { + "bbox": [ + 154, + 260, + 168, + 270 + ], + "score": 0.88, + "content": "\\Lambda _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 257, + 178, + 271 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 178, + 259, + 205, + 271 + ], + "score": 0.34, + "content": "\\mathrm { S O ( 3 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 257, + 504, + 271 + ], + "score": 1.0, + "content": "-invariant, it is sufficient to characterize all linear SO(3)-invariant function-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 270, + 176, + 281 + ], + "spans": [ + { + "bbox": [ + 106, + 270, + 120, + 281 + ], + "score": 1.0, + "content": "als", + "type": "text" + }, + { + "bbox": [ + 120, + 270, + 172, + 281 + ], + "score": 0.91, + "content": "\\Lambda : \\mathcal { T } _ { D } \\to \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 270, + 176, + 281 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 107, + 287, + 230, + 298 + ], + "lines": [ + { + "bbox": [ + 105, + 286, + 230, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 286, + 230, + 299 + ], + "score": 1.0, + "content": "It will be convenient to denote", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "interline_equation", + "bbox": [ + 234, + 300, + 374, + 316 + ], + "lines": [ + { + "bbox": [ + 234, + 300, + 374, + 316 + ], + "spans": [ + { + "bbox": [ + 234, + 300, + 374, + 316 + ], + "score": 0.89, + "content": "W = \\mathbb { R } ^ { 3 } \\operatorname { a n d } W ^ { \\otimes D } \\cong \\mathbb { R } ^ { 3 ^ { D } } = \\mathcal { T } _ { D } .", + "type": "interline_equation", + "image_path": "45327efe75350802e18c9364be7f442c296b8749e8ed5baaa3d7bb1befff1543.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 234, + 300, + 374, + 316 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 321, + 505, + 354 + ], + "lines": [ + { + "bbox": [ + 106, + 321, + 505, + 333 + ], + "spans": [ + { + "bbox": [ + 106, + 321, + 489, + 333 + ], + "score": 1.0, + "content": "We achieve our characterization using the bijective correspondence between linear functional", + "type": "text" + }, + { + "bbox": [ + 489, + 322, + 505, + 332 + ], + "score": 0.71, + "content": "\\Lambda :", + "type": "inline_equation" + } + ], + "index": 15 + }, + { + "bbox": [ + 107, + 331, + 505, + 347 + ], + "spans": [ + { + "bbox": [ + 107, + 333, + 155, + 344 + ], + "score": 0.91, + "content": "W ^ { \\otimes D } \\to \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 331, + 262, + 347 + ], + "score": 1.0, + "content": "and multi-linear functions", + "type": "text" + }, + { + "bbox": [ + 262, + 333, + 319, + 344 + ], + "score": 0.92, + "content": "\\tilde { \\Lambda } : W ^ { D } \\to \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 331, + 365, + 347 + ], + "score": 1.0, + "content": ": each such", + "type": "text" + }, + { + "bbox": [ + 366, + 334, + 374, + 344 + ], + "score": 0.81, + "content": "\\Lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 331, + 472, + 347 + ], + "score": 1.0, + "content": "corresponds to a unique", + "type": "text" + }, + { + "bbox": [ + 473, + 332, + 480, + 344 + ], + "score": 0.84, + "content": "\\hat { \\Lambda }", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 331, + 505, + 347 + ], + "score": 1.0, + "content": ", such", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 344, + 126, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 126, + 357 + ], + "score": 1.0, + "content": "that", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "interline_equation", + "bbox": [ + 181, + 353, + 429, + 369 + ], + "lines": [ + { + "bbox": [ + 181, + 353, + 429, + 369 + ], + "spans": [ + { + "bbox": [ + 181, + 353, + 429, + 369 + ], + "score": 0.87, + "content": "\\tilde { \\Lambda } ( e _ { i _ { 1 } } , \\ldots , e _ { i _ { D } } ) = \\Lambda ( e _ { i _ { 1 } } \\otimes \\ldots \\otimes e _ { i _ { D } } ) , \\forall ( i _ { 1 } , \\ldots , i _ { D } ) \\in [ 3 ] ^ { D } ,", + "type": "interline_equation", + "image_path": "df8f8bbd5775a47d4532415c4acfb54c22bdcbfff704dc096711765598b481ea.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 181, + 353, + 429, + 369 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 370, + 506, + 402 + ], + "lines": [ + { + "bbox": [ + 105, + 369, + 506, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 133, + 382 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 372, + 171, + 381 + ], + "score": 0.9, + "content": "e _ { 1 } , e _ { 2 } , e _ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 369, + 326, + 382 + ], + "score": 1.0, + "content": "denote the standard basis elements of", + "type": "text" + }, + { + "bbox": [ + 326, + 370, + 338, + 380 + ], + "score": 0.87, + "content": "\\mathbb { R } ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 369, + 506, + 382 + ], + "score": 1.0, + "content": ". We define a spanning set of equivariant", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 379, + 505, + 393 + ], + "spans": [ + { + "bbox": [ + 105, + 379, + 190, + 393 + ], + "score": 1.0, + "content": "linear functionals on", + "type": "text" + }, + { + "bbox": [ + 190, + 380, + 214, + 391 + ], + "score": 0.91, + "content": "W ^ { \\otimes D }", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 379, + 482, + 393 + ], + "score": 1.0, + "content": "via a corresponding characterization for multi-linear functionals on", + "type": "text" + }, + { + "bbox": [ + 483, + 381, + 501, + 391 + ], + "score": 0.89, + "content": "W ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 379, + 505, + 393 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 392, + 173, + 405 + ], + "spans": [ + { + "bbox": [ + 106, + 392, + 173, + 405 + ], + "score": 1.0, + "content": "Specifically, set", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + }, + { + "type": "interline_equation", + "bbox": [ + 197, + 402, + 412, + 416 + ], + "lines": [ + { + "bbox": [ + 197, + 402, + 412, + 416 + ], + "spans": [ + { + "bbox": [ + 197, + 402, + 412, + 416 + ], + "score": 0.61, + "content": "K _ { D } = \\{ k \\in \\mathbb { N } _ { + } | D - 3 k { \\mathrm { i s ~ e v e n ~ a n d ~ n o n } } { \\mathrm { - n e g a t i v e . ~ } } \\}", + "type": "interline_equation", + "image_path": "e2231805e6f80db582c8bc67d23c8cd5d73149e2fa25c05289d7491ad933fe3e.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 197, + 402, + 412, + 416 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 417, + 301, + 429 + ], + "lines": [ + { + "bbox": [ + 106, + 417, + 301, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 417, + 122, + 430 + ], + "score": 1.0, + "content": "For", + "type": "text" + }, + { + "bbox": [ + 123, + 418, + 157, + 429 + ], + "score": 0.91, + "content": "k \\in K _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 417, + 301, + 430 + ], + "score": 1.0, + "content": "we define a multi-linear functional:", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "interline_equation", + "bbox": [ + 113, + 433, + 497, + 462 + ], + "lines": [ + { + "bbox": [ + 113, + 433, + 497, + 462 + ], + "spans": [ + { + "bbox": [ + 113, + 433, + 497, + 462 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\tilde { \\Lambda } _ { k } ( w _ { 1 } , \\dots , w _ { D } ) = \\operatorname* { d e t } ( w _ { 1 } , w _ { 2 } , w _ { 3 } ) \\times \\dots \\times \\operatorname* { d e t } ( w _ { 3 k - 2 } , w _ { 3 k - 1 } , w _ { 3 k } ) \\times \\langle w _ { 3 k + 1 } , w _ { 3 k + 2 } \\rangle \\times \\dots } \\\\ { \\times \\langle w _ { D - 1 } , w _ { D } \\rangle , \\qquad ( 2 \\tilde { \\lambda } _ { 1 } - \\tilde { \\lambda } _ { 1 } ) \\times \\dots , } \\end{array}", + "type": "interline_equation", + "image_path": "b96e41e5ec77b65c2f7c30bdd31e0d9db744478990c4a7bf5915fcc7fd0a2048.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 113, + 433, + 497, + 442.6666666666667 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 113, + 442.6666666666667, + 497, + 452.33333333333337 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 113, + 452.33333333333337, + 497, + 462.00000000000006 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 465, + 258, + 478 + ], + "lines": [ + { + "bbox": [ + 106, + 464, + 258, + 479 + ], + "spans": [ + { + "bbox": [ + 106, + 464, + 138, + 479 + ], + "score": 1.0, + "content": "and for", + "type": "text" + }, + { + "bbox": [ + 138, + 465, + 216, + 478 + ], + "score": 0.92, + "content": "( k , \\sigma ) \\in K _ { D } \\times S _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 464, + 258, + 479 + ], + "score": 1.0, + "content": "we define", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "interline_equation", + "bbox": [ + 215, + 482, + 396, + 498 + ], + "lines": [ + { + "bbox": [ + 215, + 482, + 396, + 498 + ], + "spans": [ + { + "bbox": [ + 215, + 482, + 396, + 498 + ], + "score": 0.91, + "content": "\\tilde { \\Lambda } _ { k , \\sigma } ( w _ { 1 } , \\ldots , w _ { D } ) = \\tilde { \\Lambda } _ { k } ( w _ { \\sigma ( 1 ) } , \\ldots , w _ { \\sigma ( D ) } )", + "type": "interline_equation", + "image_path": "ed66adc05cff02565a6b59ffff05877253169c4dcf9eadec0440a5192dfa0043.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 215, + 482, + 396, + 498 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 501, + 504, + 537 + ], + "lines": [ + { + "bbox": [ + 106, + 501, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 106, + 501, + 348, + 515 + ], + "score": 1.0, + "content": "Proposition 1. The space of linear invariant functions from", + "type": "text" + }, + { + "bbox": [ + 348, + 502, + 362, + 513 + ], + "score": 0.87, + "content": "\\mathcal { T } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 501, + 373, + 515 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 374, + 502, + 382, + 511 + ], + "score": 0.76, + "content": "\\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 501, + 506, + 515 + ], + "score": 1.0, + "content": "is spanned by the set of linear", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 512, + 504, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 192, + 528 + ], + "score": 1.0, + "content": "invariant functionals", + "type": "text" + }, + { + "bbox": [ + 192, + 514, + 329, + 526 + ], + "score": 0.91, + "content": "\\lambda _ { D } = \\{ \\Lambda _ { k , \\sigma } | \\left( k , \\sigma \\right) \\in K _ { D } \\times S _ { D } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 512, + 484, + 528 + ], + "score": 1.0, + "content": "induced by the multi-linear functional", + "type": "text" + }, + { + "bbox": [ + 484, + 513, + 504, + 526 + ], + "score": 0.91, + "content": "\\tilde { \\Lambda } _ { k , \\sigma }", + "type": "inline_equation" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 525, + 275, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 262, + 537 + ], + "score": 1.0, + "content": "described in equation 25 and equation", + "type": "text" + }, + { + "bbox": [ + 263, + 526, + 275, + 535 + ], + "score": 0.32, + "content": "2 6", + "type": "inline_equation" + } + ], + "index": 31 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 106, + 545, + 505, + 592 + ], + "lines": [ + { + "bbox": [ + 106, + 544, + 505, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 544, + 505, + 558 + ], + "score": 1.0, + "content": "We note that (i) equation 24 provides a (cumbersome) way to compute all linear invariant functionals", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 107, + 556, + 506, + 570 + ], + "spans": [ + { + "bbox": [ + 107, + 558, + 127, + 570 + ], + "score": 0.9, + "content": "\\Lambda _ { k , \\sigma }", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 556, + 300, + 570 + ], + "score": 1.0, + "content": "explicitly by evaluating the corresponding", + "type": "text" + }, + { + "bbox": [ + 300, + 556, + 320, + 570 + ], + "score": 0.92, + "content": "\\tilde { \\Lambda } _ { k , \\sigma }", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 556, + 349, + 570 + ], + "score": 1.0, + "content": "on the", + "type": "text" + }, + { + "bbox": [ + 349, + 557, + 363, + 568 + ], + "score": 0.88, + "content": "3 ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 556, + 506, + 570 + ], + "score": 1.0, + "content": "elements of the standard basis and", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 569, + 506, + 582 + ], + "spans": [ + { + "bbox": [ + 105, + 569, + 148, + 582 + ], + "score": 1.0, + "content": "(ii) the set", + "type": "text" + }, + { + "bbox": [ + 149, + 569, + 163, + 580 + ], + "score": 0.89, + "content": "\\lambda _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 569, + 416, + 582 + ], + "score": 1.0, + "content": "is spanning, but is not linearly independent. For example, since", + "type": "text" + }, + { + "bbox": [ + 416, + 569, + 501, + 581 + ], + "score": 0.9, + "content": "\\langle w _ { 1 } , w _ { 2 } \\rangle = \\langle w _ { 2 } , w _ { 1 } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 569, + 506, + 582 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 578, + 471, + 593 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 156, + 593 + ], + "score": 1.0, + "content": "the space of", + "type": "text" + }, + { + "bbox": [ + 157, + 580, + 183, + 592 + ], + "score": 0.35, + "content": "\\mathrm { S O ( 3 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 578, + 281, + 593 + ], + "score": 1.0, + "content": "invariant functionals on", + "type": "text" + }, + { + "bbox": [ + 282, + 579, + 327, + 591 + ], + "score": 0.93, + "content": "{ \\mathcal { T } } _ { 2 } = { \\overline { { W } } } ^ { \\otimes 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 578, + 431, + 593 + ], + "score": 1.0, + "content": "is one dimensional while", + "type": "text" + }, + { + "bbox": [ + 431, + 581, + 466, + 592 + ], + "score": 0.92, + "content": "| \\lambda _ { 2 } | = 2", + "type": "inline_equation" + }, + { + "bbox": [ + 466, + 578, + 471, + 593 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 106, + 603, + 505, + 639 + ], + "lines": [ + { + "bbox": [ + 106, + 603, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 603, + 191, + 617 + ], + "score": 1.0, + "content": "Proof of Proposition", + "type": "text" + }, + { + "bbox": [ + 191, + 604, + 196, + 613 + ], + "score": 0.55, + "content": "^ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 603, + 505, + 617 + ], + "score": 1.0, + "content": ". We first show that the bijective correspondence between linear functional", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 107, + 613, + 506, + 629 + ], + "spans": [ + { + "bbox": [ + 107, + 615, + 175, + 626 + ], + "score": 0.9, + "content": "\\Lambda : W ^ { \\otimes D } \\to \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 613, + 284, + 629 + ], + "score": 1.0, + "content": "and multi-linear functions", + "type": "text" + }, + { + "bbox": [ + 285, + 614, + 346, + 626 + ], + "score": 0.91, + "content": "\\tilde { \\Lambda } : \\dot { W } ^ { D } \\to \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 613, + 506, + 629 + ], + "score": 1.0, + "content": ", extends to a bijective correspondence", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 625, + 504, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 430, + 640 + ], + "score": 1.0, + "content": "between SO(3)-invariant linear/multi-linear functionals. The action of SO(3) on", + "type": "text" + }, + { + "bbox": [ + 430, + 626, + 449, + 637 + ], + "score": 0.89, + "content": "W ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 625, + 504, + 640 + ], + "score": 1.0, + "content": "is defined by", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37 + }, + { + "type": "interline_equation", + "bbox": [ + 222, + 642, + 388, + 656 + ], + "lines": [ + { + "bbox": [ + 222, + 642, + 388, + 656 + ], + "spans": [ + { + "bbox": [ + 222, + 642, + 388, + 656 + ], + "score": 0.91, + "content": "\\tilde { \\rho } ( R ) ( w _ { 1 } , \\ldots , w _ { D } ) = ( R w _ { 1 } , \\ldots , R w _ { D } ) .", + "type": "interline_equation", + "image_path": "19eb1123b02d75c7439a3d1622ba1c49afde3b10a10160e4d16e44a73ee0dcef.jpg" + } + ] + } + ], + "index": 39, + "virtual_lines": [ + { + "bbox": [ + 222, + 642, + 388, + 656 + ], + "spans": [], + "index": 39 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 659, + 370, + 673 + ], + "lines": [ + { + "bbox": [ + 104, + 656, + 370, + 677 + ], + "spans": [ + { + "bbox": [ + 104, + 656, + 151, + 677 + ], + "score": 1.0, + "content": "The action", + "type": "text" + }, + { + "bbox": [ + 152, + 660, + 207, + 673 + ], + "score": 0.94, + "content": "\\rho ( R ) = R ^ { \\otimes D }", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 656, + 261, + 677 + ], + "score": 1.0, + "content": "of SO(3) on", + "type": "text" + }, + { + "bbox": [ + 261, + 660, + 285, + 671 + ], + "score": 0.9, + "content": "W ^ { \\otimes D }", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 656, + 370, + 677 + ], + "score": 1.0, + "content": "is such that the map", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40 + }, + { + "type": "interline_equation", + "bbox": [ + 236, + 677, + 374, + 691 + ], + "lines": [ + { + "bbox": [ + 236, + 677, + 374, + 691 + ], + "spans": [ + { + "bbox": [ + 236, + 677, + 374, + 691 + ], + "score": 0.9, + "content": "( w _ { 1 } , \\dots , w _ { D } ) \\mapsto w _ { 1 } \\otimes w _ { 2 } \\dots w _ { D }", + "type": "interline_equation", + "image_path": "8cfd515d2d6196c80a511c7922c80f1e2366a1b77bbb16f659ac4cef32c23b68.jpg" + } + ] + } + ], + "index": 41, + "virtual_lines": [ + { + "bbox": [ + 236, + 677, + 374, + 691 + ], + "spans": [], + "index": 41 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 696, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 695, + 505, + 709 + ], + "spans": [ + { + "bbox": [ + 105, + 695, + 269, + 709 + ], + "score": 1.0, + "content": "is SO(3)- equivariant. It follows that if", + "type": "text" + }, + { + "bbox": [ + 270, + 695, + 277, + 707 + ], + "score": 0.87, + "content": "\\tilde { \\Lambda }", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 695, + 297, + 709 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 297, + 697, + 305, + 707 + ], + "score": 0.82, + "content": "\\Lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 695, + 437, + 709 + ], + "score": 1.0, + "content": "satisfy equation 24, then for all", + "type": "text" + }, + { + "bbox": [ + 437, + 697, + 486, + 709 + ], + "score": 0.85, + "content": "R \\in \\mathrm { S O ( 3 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 695, + 505, + 709 + ], + "score": 1.0, + "content": ", the", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 708, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 238, + 722 + ], + "score": 1.0, + "content": "same equation holds for the pair", + "type": "text" + }, + { + "bbox": [ + 238, + 708, + 275, + 722 + ], + "score": 0.93, + "content": "\\tilde { \\Lambda } \\circ \\tilde { \\rho } ( R )", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 708, + 294, + 722 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 294, + 709, + 332, + 722 + ], + "score": 0.92, + "content": "\\Lambda \\circ \\rho ( R )", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 708, + 443, + 722 + ], + "score": 1.0, + "content": ". Thus SO(3)-invariance of", + "type": "text" + }, + { + "bbox": [ + 443, + 708, + 451, + 720 + ], + "score": 0.86, + "content": "\\tilde { \\Lambda }", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 708, + 505, + 722 + ], + "score": 1.0, + "content": "is equivalent", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 720, + 213, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 117, + 732 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 117, + 721, + 144, + 732 + ], + "score": 0.37, + "content": "\\mathrm { S O ( 3 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 720, + 200, + 732 + ], + "score": 1.0, + "content": "-invariance of", + "type": "text" + }, + { + "bbox": [ + 201, + 721, + 208, + 730 + ], + "score": 0.82, + "content": "\\Lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 720, + 213, + 732 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43 + } + ], + "page_idx": 19, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 142, + 82, + 505, + 149 + ], + "lines": [], + "index": 2.5, + "bbox_fs": [ + 141, + 83, + 506, + 150 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 107, + 164, + 253, + 177 + ], + "lines": [ + { + "bbox": [ + 105, + 163, + 254, + 180 + ], + "spans": [ + { + "bbox": [ + 105, + 163, + 254, + 180 + ], + "score": 1.0, + "content": "G TENSOR UNIVERSALITY", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 189, + 504, + 216 + ], + "lines": [ + { + "bbox": [ + 105, + 187, + 506, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 187, + 333, + 204 + ], + "score": 1.0, + "content": "In this section we show how to construct the complete set", + "type": "text" + }, + { + "bbox": [ + 334, + 190, + 357, + 202 + ], + "score": 0.92, + "content": "\\mathcal { F } _ { \\mathrm { p o o l } }", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 187, + 393, + 204 + ], + "score": 1.0, + "content": "of linear", + "type": "text" + }, + { + "bbox": [ + 393, + 189, + 420, + 201 + ], + "score": 0.25, + "content": "\\mathrm { S O ( 3 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 187, + 506, + 204 + ], + "score": 1.0, + "content": "invariant functionals", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 199, + 407, + 220 + ], + "spans": [ + { + "bbox": [ + 104, + 199, + 128, + 220 + ], + "score": 1.0, + "content": "from", + "type": "text" + }, + { + "bbox": [ + 129, + 201, + 207, + 216 + ], + "score": 0.93, + "content": "W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } } = \\bigoplus _ { T = 0 } ^ { D } \\mathcal { T } _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 199, + 218, + 220 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 218, + 203, + 226, + 213 + ], + "score": 0.8, + "content": "\\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 199, + 340, + 220 + ], + "score": 1.0, + "content": ". Since each such functional", + "type": "text" + }, + { + "bbox": [ + 340, + 204, + 348, + 213 + ], + "score": 0.84, + "content": "\\Lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 199, + 407, + 220 + ], + "score": 1.0, + "content": "is of the form", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 104, + 187, + 506, + 220 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 239, + 219, + 371, + 255 + ], + "lines": [ + { + "bbox": [ + 239, + 219, + 371, + 255 + ], + "spans": [ + { + "bbox": [ + 239, + 219, + 371, + 255 + ], + "score": 0.94, + "content": "\\Lambda ( w _ { 0 } , . . . , w _ { D } ) = \\sum _ { T = 0 } ^ { D } \\Lambda _ { T } ( w _ { T } ) ,", + "type": "interline_equation", + "image_path": "b57cd45c52bd31dbfd6a60a04df3da149300563b442ea80051d826ec75367796.jpg" + } + ] + } + ], + "index": 9.5, + "virtual_lines": [ + { + "bbox": [ + 239, + 219, + 371, + 237.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 239, + 237.0, + 371, + 255.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 259, + 504, + 281 + ], + "lines": [ + { + "bbox": [ + 106, + 257, + 504, + 271 + ], + "spans": [ + { + "bbox": [ + 106, + 257, + 153, + 271 + ], + "score": 1.0, + "content": "where each", + "type": "text" + }, + { + "bbox": [ + 154, + 260, + 168, + 270 + ], + "score": 0.88, + "content": "\\Lambda _ { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 257, + 178, + 271 + ], + "score": 1.0, + "content": "is", + "type": "text" + }, + { + "bbox": [ + 178, + 259, + 205, + 271 + ], + "score": 0.34, + "content": "\\mathrm { S O ( 3 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 257, + 504, + 271 + ], + "score": 1.0, + "content": "-invariant, it is sufficient to characterize all linear SO(3)-invariant function-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 270, + 176, + 281 + ], + "spans": [ + { + "bbox": [ + 106, + 270, + 120, + 281 + ], + "score": 1.0, + "content": "als", + "type": "text" + }, + { + "bbox": [ + 120, + 270, + 172, + 281 + ], + "score": 0.91, + "content": "\\Lambda : \\mathcal { T } _ { D } \\to \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 270, + 176, + 281 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5, + "bbox_fs": [ + 106, + 257, + 504, + 281 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 287, + 230, + 298 + ], + "lines": [ + { + "bbox": [ + 105, + 286, + 230, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 286, + 230, + 299 + ], + "score": 1.0, + "content": "It will be convenient to denote", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13, + "bbox_fs": [ + 105, + 286, + 230, + 299 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 234, + 300, + 374, + 316 + ], + "lines": [ + { + "bbox": [ + 234, + 300, + 374, + 316 + ], + "spans": [ + { + "bbox": [ + 234, + 300, + 374, + 316 + ], + "score": 0.89, + "content": "W = \\mathbb { R } ^ { 3 } \\operatorname { a n d } W ^ { \\otimes D } \\cong \\mathbb { R } ^ { 3 ^ { D } } = \\mathcal { T } _ { D } .", + "type": "interline_equation", + "image_path": "45327efe75350802e18c9364be7f442c296b8749e8ed5baaa3d7bb1befff1543.jpg" + } + ] + } + ], + "index": 14, + "virtual_lines": [ + { + "bbox": [ + 234, + 300, + 374, + 316 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 321, + 505, + 354 + ], + "lines": [ + { + "bbox": [ + 106, + 321, + 505, + 333 + ], + "spans": [ + { + "bbox": [ + 106, + 321, + 489, + 333 + ], + "score": 1.0, + "content": "We achieve our characterization using the bijective correspondence between linear functional", + "type": "text" + }, + { + "bbox": [ + 489, + 322, + 505, + 332 + ], + "score": 0.71, + "content": "\\Lambda :", + "type": "inline_equation" + } + ], + "index": 15 + }, + { + "bbox": [ + 107, + 331, + 505, + 347 + ], + "spans": [ + { + "bbox": [ + 107, + 333, + 155, + 344 + ], + "score": 0.91, + "content": "W ^ { \\otimes D } \\to \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 331, + 262, + 347 + ], + "score": 1.0, + "content": "and multi-linear functions", + "type": "text" + }, + { + "bbox": [ + 262, + 333, + 319, + 344 + ], + "score": 0.92, + "content": "\\tilde { \\Lambda } : W ^ { D } \\to \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 331, + 365, + 347 + ], + "score": 1.0, + "content": ": each such", + "type": "text" + }, + { + "bbox": [ + 366, + 334, + 374, + 344 + ], + "score": 0.81, + "content": "\\Lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 331, + 472, + 347 + ], + "score": 1.0, + "content": "corresponds to a unique", + "type": "text" + }, + { + "bbox": [ + 473, + 332, + 480, + 344 + ], + "score": 0.84, + "content": "\\hat { \\Lambda }", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 331, + 505, + 347 + ], + "score": 1.0, + "content": ", such", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 344, + 126, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 126, + 357 + ], + "score": 1.0, + "content": "that", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 321, + 505, + 357 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 181, + 353, + 429, + 369 + ], + "lines": [ + { + "bbox": [ + 181, + 353, + 429, + 369 + ], + "spans": [ + { + "bbox": [ + 181, + 353, + 429, + 369 + ], + "score": 0.87, + "content": "\\tilde { \\Lambda } ( e _ { i _ { 1 } } , \\ldots , e _ { i _ { D } } ) = \\Lambda ( e _ { i _ { 1 } } \\otimes \\ldots \\otimes e _ { i _ { D } } ) , \\forall ( i _ { 1 } , \\ldots , i _ { D } ) \\in [ 3 ] ^ { D } ,", + "type": "interline_equation", + "image_path": "df8f8bbd5775a47d4532415c4acfb54c22bdcbfff704dc096711765598b481ea.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 181, + 353, + 429, + 369 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 370, + 506, + 402 + ], + "lines": [ + { + "bbox": [ + 105, + 369, + 506, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 133, + 382 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 372, + 171, + 381 + ], + "score": 0.9, + "content": "e _ { 1 } , e _ { 2 } , e _ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 369, + 326, + 382 + ], + "score": 1.0, + "content": "denote the standard basis elements of", + "type": "text" + }, + { + "bbox": [ + 326, + 370, + 338, + 380 + ], + "score": 0.87, + "content": "\\mathbb { R } ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 369, + 506, + 382 + ], + "score": 1.0, + "content": ". We define a spanning set of equivariant", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 379, + 505, + 393 + ], + "spans": [ + { + "bbox": [ + 105, + 379, + 190, + 393 + ], + "score": 1.0, + "content": "linear functionals on", + "type": "text" + }, + { + "bbox": [ + 190, + 380, + 214, + 391 + ], + "score": 0.91, + "content": "W ^ { \\otimes D }", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 379, + 482, + 393 + ], + "score": 1.0, + "content": "via a corresponding characterization for multi-linear functionals on", + "type": "text" + }, + { + "bbox": [ + 483, + 381, + 501, + 391 + ], + "score": 0.89, + "content": "W ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 379, + 505, + 393 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 392, + 173, + 405 + ], + "spans": [ + { + "bbox": [ + 106, + 392, + 173, + 405 + ], + "score": 1.0, + "content": "Specifically, set", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 369, + 506, + 405 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 197, + 402, + 412, + 416 + ], + "lines": [ + { + "bbox": [ + 197, + 402, + 412, + 416 + ], + "spans": [ + { + "bbox": [ + 197, + 402, + 412, + 416 + ], + "score": 0.61, + "content": "K _ { D } = \\{ k \\in \\mathbb { N } _ { + } | D - 3 k { \\mathrm { i s ~ e v e n ~ a n d ~ n o n } } { \\mathrm { - n e g a t i v e . ~ } } \\}", + "type": "interline_equation", + "image_path": "e2231805e6f80db582c8bc67d23c8cd5d73149e2fa25c05289d7491ad933fe3e.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 197, + 402, + 412, + 416 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 417, + 301, + 429 + ], + "lines": [ + { + "bbox": [ + 106, + 417, + 301, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 417, + 122, + 430 + ], + "score": 1.0, + "content": "For", + "type": "text" + }, + { + "bbox": [ + 123, + 418, + 157, + 429 + ], + "score": 0.91, + "content": "k \\in K _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 417, + 301, + 430 + ], + "score": 1.0, + "content": "we define a multi-linear functional:", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23, + "bbox_fs": [ + 106, + 417, + 301, + 430 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 113, + 433, + 497, + 462 + ], + "lines": [ + { + "bbox": [ + 113, + 433, + 497, + 462 + ], + "spans": [ + { + "bbox": [ + 113, + 433, + 497, + 462 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\tilde { \\Lambda } _ { k } ( w _ { 1 } , \\dots , w _ { D } ) = \\operatorname* { d e t } ( w _ { 1 } , w _ { 2 } , w _ { 3 } ) \\times \\dots \\times \\operatorname* { d e t } ( w _ { 3 k - 2 } , w _ { 3 k - 1 } , w _ { 3 k } ) \\times \\langle w _ { 3 k + 1 } , w _ { 3 k + 2 } \\rangle \\times \\dots } \\\\ { \\times \\langle w _ { D - 1 } , w _ { D } \\rangle , \\qquad ( 2 \\tilde { \\lambda } _ { 1 } - \\tilde { \\lambda } _ { 1 } ) \\times \\dots , } \\end{array}", + "type": "interline_equation", + "image_path": "b96e41e5ec77b65c2f7c30bdd31e0d9db744478990c4a7bf5915fcc7fd0a2048.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 113, + 433, + 497, + 442.6666666666667 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 113, + 442.6666666666667, + 497, + 452.33333333333337 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 113, + 452.33333333333337, + 497, + 462.00000000000006 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 465, + 258, + 478 + ], + "lines": [ + { + "bbox": [ + 106, + 464, + 258, + 479 + ], + "spans": [ + { + "bbox": [ + 106, + 464, + 138, + 479 + ], + "score": 1.0, + "content": "and for", + "type": "text" + }, + { + "bbox": [ + 138, + 465, + 216, + 478 + ], + "score": 0.92, + "content": "( k , \\sigma ) \\in K _ { D } \\times S _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 464, + 258, + 479 + ], + "score": 1.0, + "content": "we define", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27, + "bbox_fs": [ + 106, + 464, + 258, + 479 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 215, + 482, + 396, + 498 + ], + "lines": [ + { + "bbox": [ + 215, + 482, + 396, + 498 + ], + "spans": [ + { + "bbox": [ + 215, + 482, + 396, + 498 + ], + "score": 0.91, + "content": "\\tilde { \\Lambda } _ { k , \\sigma } ( w _ { 1 } , \\ldots , w _ { D } ) = \\tilde { \\Lambda } _ { k } ( w _ { \\sigma ( 1 ) } , \\ldots , w _ { \\sigma ( D ) } )", + "type": "interline_equation", + "image_path": "ed66adc05cff02565a6b59ffff05877253169c4dcf9eadec0440a5192dfa0043.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 215, + 482, + 396, + 498 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 501, + 504, + 537 + ], + "lines": [ + { + "bbox": [ + 106, + 501, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 106, + 501, + 348, + 515 + ], + "score": 1.0, + "content": "Proposition 1. The space of linear invariant functions from", + "type": "text" + }, + { + "bbox": [ + 348, + 502, + 362, + 513 + ], + "score": 0.87, + "content": "\\mathcal { T } _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 501, + 373, + 515 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 374, + 502, + 382, + 511 + ], + "score": 0.76, + "content": "\\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 501, + 506, + 515 + ], + "score": 1.0, + "content": "is spanned by the set of linear", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 512, + 504, + 528 + ], + "spans": [ + { + "bbox": [ + 105, + 512, + 192, + 528 + ], + "score": 1.0, + "content": "invariant functionals", + "type": "text" + }, + { + "bbox": [ + 192, + 514, + 329, + 526 + ], + "score": 0.91, + "content": "\\lambda _ { D } = \\{ \\Lambda _ { k , \\sigma } | \\left( k , \\sigma \\right) \\in K _ { D } \\times S _ { D } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 512, + 484, + 528 + ], + "score": 1.0, + "content": "induced by the multi-linear functional", + "type": "text" + }, + { + "bbox": [ + 484, + 513, + 504, + 526 + ], + "score": 0.91, + "content": "\\tilde { \\Lambda } _ { k , \\sigma }", + "type": "inline_equation" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 525, + 275, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 262, + 537 + ], + "score": 1.0, + "content": "described in equation 25 and equation", + "type": "text" + }, + { + "bbox": [ + 263, + 526, + 275, + 535 + ], + "score": 0.32, + "content": "2 6", + "type": "inline_equation" + } + ], + "index": 31 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 501, + 506, + 537 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 545, + 505, + 592 + ], + "lines": [ + { + "bbox": [ + 106, + 544, + 505, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 544, + 505, + 558 + ], + "score": 1.0, + "content": "We note that (i) equation 24 provides a (cumbersome) way to compute all linear invariant functionals", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 107, + 556, + 506, + 570 + ], + "spans": [ + { + "bbox": [ + 107, + 558, + 127, + 570 + ], + "score": 0.9, + "content": "\\Lambda _ { k , \\sigma }", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 556, + 300, + 570 + ], + "score": 1.0, + "content": "explicitly by evaluating the corresponding", + "type": "text" + }, + { + "bbox": [ + 300, + 556, + 320, + 570 + ], + "score": 0.92, + "content": "\\tilde { \\Lambda } _ { k , \\sigma }", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 556, + 349, + 570 + ], + "score": 1.0, + "content": "on the", + "type": "text" + }, + { + "bbox": [ + 349, + 557, + 363, + 568 + ], + "score": 0.88, + "content": "3 ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 556, + 506, + 570 + ], + "score": 1.0, + "content": "elements of the standard basis and", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 569, + 506, + 582 + ], + "spans": [ + { + "bbox": [ + 105, + 569, + 148, + 582 + ], + "score": 1.0, + "content": "(ii) the set", + "type": "text" + }, + { + "bbox": [ + 149, + 569, + 163, + 580 + ], + "score": 0.89, + "content": "\\lambda _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 569, + 416, + 582 + ], + "score": 1.0, + "content": "is spanning, but is not linearly independent. For example, since", + "type": "text" + }, + { + "bbox": [ + 416, + 569, + 501, + 581 + ], + "score": 0.9, + "content": "\\langle w _ { 1 } , w _ { 2 } \\rangle = \\langle w _ { 2 } , w _ { 1 } \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 501, + 569, + 506, + 582 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 578, + 471, + 593 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 156, + 593 + ], + "score": 1.0, + "content": "the space of", + "type": "text" + }, + { + "bbox": [ + 157, + 580, + 183, + 592 + ], + "score": 0.35, + "content": "\\mathrm { S O ( 3 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 578, + 281, + 593 + ], + "score": 1.0, + "content": "invariant functionals on", + "type": "text" + }, + { + "bbox": [ + 282, + 579, + 327, + 591 + ], + "score": 0.93, + "content": "{ \\mathcal { T } } _ { 2 } = { \\overline { { W } } } ^ { \\otimes 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 578, + 431, + 593 + ], + "score": 1.0, + "content": "is one dimensional while", + "type": "text" + }, + { + "bbox": [ + 431, + 581, + 466, + 592 + ], + "score": 0.92, + "content": "| \\lambda _ { 2 } | = 2", + "type": "inline_equation" + }, + { + "bbox": [ + 466, + 578, + 471, + 593 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 544, + 506, + 593 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 603, + 505, + 639 + ], + "lines": [ + { + "bbox": [ + 106, + 603, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 603, + 191, + 617 + ], + "score": 1.0, + "content": "Proof of Proposition", + "type": "text" + }, + { + "bbox": [ + 191, + 604, + 196, + 613 + ], + "score": 0.55, + "content": "^ { l }", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 603, + 505, + 617 + ], + "score": 1.0, + "content": ". We first show that the bijective correspondence between linear functional", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 107, + 613, + 506, + 629 + ], + "spans": [ + { + "bbox": [ + 107, + 615, + 175, + 626 + ], + "score": 0.9, + "content": "\\Lambda : W ^ { \\otimes D } \\to \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 613, + 284, + 629 + ], + "score": 1.0, + "content": "and multi-linear functions", + "type": "text" + }, + { + "bbox": [ + 285, + 614, + 346, + 626 + ], + "score": 0.91, + "content": "\\tilde { \\Lambda } : \\dot { W } ^ { D } \\to \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 613, + 506, + 629 + ], + "score": 1.0, + "content": ", extends to a bijective correspondence", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 625, + 504, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 430, + 640 + ], + "score": 1.0, + "content": "between SO(3)-invariant linear/multi-linear functionals. The action of SO(3) on", + "type": "text" + }, + { + "bbox": [ + 430, + 626, + 449, + 637 + ], + "score": 0.89, + "content": "W ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 625, + 504, + 640 + ], + "score": 1.0, + "content": "is defined by", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37, + "bbox_fs": [ + 105, + 603, + 506, + 640 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 222, + 642, + 388, + 656 + ], + "lines": [ + { + "bbox": [ + 222, + 642, + 388, + 656 + ], + "spans": [ + { + "bbox": [ + 222, + 642, + 388, + 656 + ], + "score": 0.91, + "content": "\\tilde { \\rho } ( R ) ( w _ { 1 } , \\ldots , w _ { D } ) = ( R w _ { 1 } , \\ldots , R w _ { D } ) .", + "type": "interline_equation", + "image_path": "19eb1123b02d75c7439a3d1622ba1c49afde3b10a10160e4d16e44a73ee0dcef.jpg" + } + ] + } + ], + "index": 39, + "virtual_lines": [ + { + "bbox": [ + 222, + 642, + 388, + 656 + ], + "spans": [], + "index": 39 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 659, + 370, + 673 + ], + "lines": [ + { + "bbox": [ + 104, + 656, + 370, + 677 + ], + "spans": [ + { + "bbox": [ + 104, + 656, + 151, + 677 + ], + "score": 1.0, + "content": "The action", + "type": "text" + }, + { + "bbox": [ + 152, + 660, + 207, + 673 + ], + "score": 0.94, + "content": "\\rho ( R ) = R ^ { \\otimes D }", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 656, + 261, + 677 + ], + "score": 1.0, + "content": "of SO(3) on", + "type": "text" + }, + { + "bbox": [ + 261, + 660, + 285, + 671 + ], + "score": 0.9, + "content": "W ^ { \\otimes D }", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 656, + 370, + 677 + ], + "score": 1.0, + "content": "is such that the map", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40, + "bbox_fs": [ + 104, + 656, + 370, + 677 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 236, + 677, + 374, + 691 + ], + "lines": [ + { + "bbox": [ + 236, + 677, + 374, + 691 + ], + "spans": [ + { + "bbox": [ + 236, + 677, + 374, + 691 + ], + "score": 0.9, + "content": "( w _ { 1 } , \\dots , w _ { D } ) \\mapsto w _ { 1 } \\otimes w _ { 2 } \\dots w _ { D }", + "type": "interline_equation", + "image_path": "8cfd515d2d6196c80a511c7922c80f1e2366a1b77bbb16f659ac4cef32c23b68.jpg" + } + ] + } + ], + "index": 41, + "virtual_lines": [ + { + "bbox": [ + 236, + 677, + 374, + 691 + ], + "spans": [], + "index": 41 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 696, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 695, + 505, + 709 + ], + "spans": [ + { + "bbox": [ + 105, + 695, + 269, + 709 + ], + "score": 1.0, + "content": "is SO(3)- equivariant. It follows that if", + "type": "text" + }, + { + "bbox": [ + 270, + 695, + 277, + 707 + ], + "score": 0.87, + "content": "\\tilde { \\Lambda }", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 695, + 297, + 709 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 297, + 697, + 305, + 707 + ], + "score": 0.82, + "content": "\\Lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 695, + 437, + 709 + ], + "score": 1.0, + "content": "satisfy equation 24, then for all", + "type": "text" + }, + { + "bbox": [ + 437, + 697, + 486, + 709 + ], + "score": 0.85, + "content": "R \\in \\mathrm { S O ( 3 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 695, + 505, + 709 + ], + "score": 1.0, + "content": ", the", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 708, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 238, + 722 + ], + "score": 1.0, + "content": "same equation holds for the pair", + "type": "text" + }, + { + "bbox": [ + 238, + 708, + 275, + 722 + ], + "score": 0.93, + "content": "\\tilde { \\Lambda } \\circ \\tilde { \\rho } ( R )", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 708, + 294, + 722 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 294, + 709, + 332, + 722 + ], + "score": 0.92, + "content": "\\Lambda \\circ \\rho ( R )", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 708, + 443, + 722 + ], + "score": 1.0, + "content": ". Thus SO(3)-invariance of", + "type": "text" + }, + { + "bbox": [ + 443, + 708, + 451, + 720 + ], + "score": 0.86, + "content": "\\tilde { \\Lambda }", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 708, + 505, + 722 + ], + "score": 1.0, + "content": "is equivalent", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 720, + 213, + 732 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 117, + 732 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 117, + 721, + 144, + 732 + ], + "score": 0.37, + "content": "\\mathrm { S O ( 3 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 720, + 200, + 732 + ], + "score": 1.0, + "content": "-invariance of", + "type": "text" + }, + { + "bbox": [ + 201, + 721, + 208, + 730 + ], + "score": 0.82, + "content": "\\Lambda", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 720, + 213, + 732 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43, + "bbox_fs": [ + 105, + 695, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 215, + 95 + ], + "score": 1.0, + "content": "Multi-linear functionals on", + "type": "text" + }, + { + "bbox": [ + 215, + 82, + 234, + 93 + ], + "score": 0.9, + "content": "W ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 81, + 281, + 95 + ], + "score": 1.0, + "content": "invariant to", + "type": "text" + }, + { + "bbox": [ + 282, + 83, + 288, + 94 + ], + "score": 0.85, + "content": "\\tilde { \\rho }", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 81, + 448, + 95 + ], + "score": 1.0, + "content": "are a subset of the set of polynomials on", + "type": "text" + }, + { + "bbox": [ + 448, + 82, + 466, + 93 + ], + "score": 0.9, + "content": "W ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 467, + 81, + 506, + 95 + ], + "score": 1.0, + "content": "invariant", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 104, + 93, + 505, + 107 + ], + "spans": [ + { + "bbox": [ + 104, + 93, + 116, + 107 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 117, + 94, + 123, + 105 + ], + "score": 0.77, + "content": "\\tilde { \\rho }", + "type": "inline_equation" + }, + { + "bbox": [ + 123, + 93, + 505, + 107 + ], + "score": 1.0, + "content": ". It is known (see (Kraft & Procesi, 2000), page 114), that all such polynomials are algebraically", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 105, + 247, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 247, + 117 + ], + "score": 1.0, + "content": "generated by functions of the form", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "interline_equation", + "bbox": [ + 175, + 121, + 432, + 135 + ], + "lines": [ + { + "bbox": [ + 175, + 121, + 432, + 135 + ], + "spans": [ + { + "bbox": [ + 175, + 121, + 432, + 135 + ], + "score": 0.83, + "content": "\\begin{array} { r } { \\operatorname* { d e t } ( w _ { i _ { 1 } } , w _ { i _ { 2 } } , w _ { i _ { 3 } } ) \\mathrm { ~ a n d ~ } \\langle w _ { j _ { 1 } } , w _ { j _ { 2 } } \\rangle , \\mathrm { ~ w h e r e ~ } i _ { 1 } , i _ { 2 } , i _ { 3 } , j _ { 1 } , j _ { 2 } \\in [ D ] . } \\end{array}", + "type": "interline_equation", + "image_path": "5c1bc250cffdd612aa0182d31b2a2ccdf1ef9897cc9acb334763c60639da03d5.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 175, + 121, + 432, + 135 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 140, + 505, + 162 + ], + "lines": [ + { + "bbox": [ + 105, + 139, + 506, + 154 + ], + "spans": [ + { + "bbox": [ + 105, + 139, + 162, + 154 + ], + "score": 1.0, + "content": "Equivalently,", + "type": "text" + }, + { + "bbox": [ + 162, + 140, + 189, + 153 + ], + "score": 0.3, + "content": "\\mathrm { S O ( 3 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 139, + 506, + 154 + ], + "score": 1.0, + "content": "-invariant polynomials are spanned by linear combinations of polynomials of", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 150, + 144, + 164 + ], + "spans": [ + { + "bbox": [ + 105, + 150, + 144, + 164 + ], + "score": 1.0, + "content": "the form", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "interline_equation", + "bbox": [ + 172, + 162, + 439, + 175 + ], + "lines": [ + { + "bbox": [ + 172, + 162, + 439, + 175 + ], + "spans": [ + { + "bbox": [ + 172, + 162, + 439, + 175 + ], + "score": 0.86, + "content": "\\operatorname * { d e t } \\bigl ( w _ { i _ { 1 } } , w _ { i _ { 2 } } , w _ { i _ { 3 } } \\bigr ) \\operatorname * { d e t } \\bigl ( w _ { i _ { 4 } } , w _ { i _ { 5 } } , w _ { i _ { 6 } } \\bigr ) \\ldots \\langle w _ { j _ { 1 } } , w _ { j _ { 2 } } \\rangle \\langle w _ { j _ { 3 } } , w _ { j _ { 4 } } \\rangle \\ldots .", + "type": "interline_equation", + "image_path": "aeb93f1dc0eb0f9e7c84a28c079f6e15a61d4012660eb399737422057ff3481f.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 172, + 162, + 439, + 175 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 177, + 505, + 211 + ], + "lines": [ + { + "bbox": [ + 106, + 176, + 505, + 190 + ], + "spans": [ + { + "bbox": [ + 106, + 176, + 505, + 190 + ], + "score": 1.0, + "content": "When considering the subset of multi-linear invariant polynomials, we see that they must be spanned", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 188, + 505, + 202 + ], + "spans": [ + { + "bbox": [ + 105, + 188, + 290, + 202 + ], + "score": 1.0, + "content": "by polynomials as in equation 27, where each", + "type": "text" + }, + { + "bbox": [ + 291, + 190, + 340, + 200 + ], + "score": 0.9, + "content": "w _ { 1 } , \\ldots , w _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 188, + 505, + 202 + ], + "score": 1.0, + "content": "appears exactly once in each polynomial", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 199, + 391, + 212 + ], + "spans": [ + { + "bbox": [ + 105, + 199, + 372, + 212 + ], + "score": 1.0, + "content": "in the spanning set. These precisely correspond to the functions in", + "type": "text" + }, + { + "bbox": [ + 373, + 200, + 387, + 210 + ], + "score": 0.9, + "content": "\\lambda _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 199, + 391, + 212 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8 + }, + { + "type": "title", + "bbox": [ + 107, + 250, + 203, + 263 + ], + "lines": [ + { + "bbox": [ + 105, + 250, + 205, + 265 + ], + "spans": [ + { + "bbox": [ + 105, + 250, + 205, + 265 + ], + "score": 1.0, + "content": "H EXPERIMENTS", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "table", + "bbox": [ + 153, + 318, + 458, + 409 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 287, + 504, + 318 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 288, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 505, + 299 + ], + "score": 1.0, + "content": "Table 1: Results obtained on the QM9 dataset Ramakrishnan et al. (2014) for different design choices", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 298, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 104, + 298, + 324, + 311 + ], + "score": 1.0, + "content": "in the TFN architecture. Results are reported for the", + "type": "text" + }, + { + "bbox": [ + 324, + 300, + 350, + 310 + ], + "score": 0.64, + "content": "\\epsilon _ { h o m o }", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 298, + 505, + 311 + ], + "score": 1.0, + "content": "target variable, and are multiplied by", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 308, + 128, + 321 + ], + "spans": [ + { + "bbox": [ + 106, + 309, + 122, + 320 + ], + "score": 0.84, + "content": "1 0 ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 122, + 308, + 128, + 321 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12 + }, + { + "type": "table_body", + "bbox": [ + 153, + 318, + 458, + 409 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 153, + 318, + 458, + 409 + ], + "spans": [ + { + "bbox": [ + 153, + 318, + 458, + 409 + ], + "score": 0.98, + "html": "
Model variant Mean l1error
Order 0irreps 111.2 ± 0.3
Order 0-1 ireps 82.1 ± 0.4
Order 0-2 irreps 62.1 ± 1.5
Order 0-3 irreps 53.2 ± 1.9
Order 0-4 irreps 51.2 ± 0.4
Order O-3 irreps without non-linearity 53.1 ± 0.8
Order O-3 irreps + self-interaction only in the final layer 55.2 ± 0.3
", + "type": "table", + "image_path": "30712253f5955d3ca6f795324f935ac71bda0d63fa1692f647560e32ee1bf29d.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 153, + 318, + 458, + 348.3333333333333 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 153, + 348.3333333333333, + 458, + 378.66666666666663 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 153, + 378.66666666666663, + 458, + 408.99999999999994 + ], + "spans": [], + "index": 16 + } + ] + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 106, + 422, + 503, + 445 + ], + "lines": [ + { + "bbox": [ + 106, + 421, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 106, + 421, + 505, + 436 + ], + "score": 1.0, + "content": "This section provides an experimental evaluation of different design choices of the TFN architecture,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 433, + 390, + 446 + ], + "spans": [ + { + "bbox": [ + 106, + 433, + 390, + 446 + ], + "score": 1.0, + "content": "inspired by our theoretical analysis. We study the following questions:", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 129, + 455, + 505, + 584 + ], + "lines": [ + { + "bbox": [ + 130, + 455, + 504, + 468 + ], + "spans": [ + { + "bbox": [ + 130, + 455, + 504, + 468 + ], + "score": 1.0, + "content": "1. The importance of non-linear activation. Our proof shows that using non-linear acti-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 466, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 141, + 466, + 505, + 479 + ], + "score": 1.0, + "content": "vation functions is not necessary for proving universality. Here, we empirically test the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 477, + 289, + 491 + ], + "spans": [ + { + "bbox": [ + 141, + 477, + 289, + 491 + ], + "score": 1.0, + "content": "possibility of removing these layers.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 129, + 492, + 505, + 504 + ], + "spans": [ + { + "bbox": [ + 129, + 492, + 505, + 504 + ], + "score": 1.0, + "content": "2. The importance of high-dimensional irreducible representations. Our theoretical anal-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 141, + 502, + 505, + 517 + ], + "spans": [ + { + "bbox": [ + 141, + 502, + 505, + 517 + ], + "score": 1.0, + "content": "ysis shows that in order to represent/approximate high degree polynomials, high-order rep-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 513, + 505, + 527 + ], + "spans": [ + { + "bbox": [ + 141, + 513, + 505, + 527 + ], + "score": 1.0, + "content": "resentations should be used. Here, we check whether using high-order representations has", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 141, + 526, + 215, + 536 + ], + "spans": [ + { + "bbox": [ + 141, + 526, + 215, + 536 + ], + "score": 1.0, + "content": "practical benefits.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 130, + 540, + 506, + 553 + ], + "spans": [ + { + "bbox": [ + 130, + 540, + 506, + 553 + ], + "score": 1.0, + "content": "3. The effect of self-interaction layers. Our proof suggests that it is enough to use self", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 141, + 551, + 506, + 564 + ], + "spans": [ + { + "bbox": [ + 141, + 551, + 506, + 564 + ], + "score": 1.0, + "content": "interaction linear layers at the end of the model. We empirically compare this approach", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 141, + 562, + 506, + 575 + ], + "spans": [ + { + "bbox": [ + 141, + 562, + 506, + 575 + ], + "score": 1.0, + "content": "with the more common approach of using self-interaction layers after each convolutional", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 140, + 572, + 168, + 587 + ], + "spans": [ + { + "bbox": [ + 140, + 572, + 168, + 587 + ], + "score": 1.0, + "content": "layer.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 108, + 596, + 503, + 630 + ], + "lines": [ + { + "bbox": [ + 106, + 596, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 106, + 596, + 505, + 609 + ], + "score": 1.0, + "content": "Dataset. We use the QM9 (Ramakrishnan et al., 2014) dataset for our experiments. The dataset", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 606, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 105, + 606, + 505, + 621 + ], + "score": 1.0, + "content": "contains 134K molecules, with node 3D positions, 5 categorical node features and 4 categorical edge", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 618, + 379, + 632 + ], + "spans": [ + { + "bbox": [ + 106, + 618, + 379, + 632 + ], + "score": 1.0, + "content": "features. The task is a molecule property prediction regression task.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 107, + 641, + 505, + 675 + ], + "lines": [ + { + "bbox": [ + 105, + 640, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 505, + 655 + ], + "score": 1.0, + "content": "Framework. We used pytorch (Paszke et al., 2017)as the deep learning framework and the Deep", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 652, + 505, + 666 + ], + "spans": [ + { + "bbox": [ + 105, + 652, + 505, + 666 + ], + "score": 1.0, + "content": "Graph Library (DGL) (Wang et al., 2019a) as the graph learning framework. All experiments ran on", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 664, + 205, + 675 + ], + "spans": [ + { + "bbox": [ + 106, + 664, + 205, + 675 + ], + "score": 1.0, + "content": "NVIDIA GV100 GPUs.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "Experimental setup. We use the the TFN implementation from Fuchs et al. (2020). We trained", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 698, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 276, + 713 + ], + "score": 1.0, + "content": "each model variant for 50 epochs on the", + "type": "text" + }, + { + "bbox": [ + 277, + 700, + 303, + 710 + ], + "score": 0.83, + "content": "\\epsilon _ { h o m o }", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 698, + 404, + 713 + ], + "score": 1.0, + "content": "target variable using an", + "type": "text" + }, + { + "bbox": [ + 404, + 699, + 414, + 710 + ], + "score": 0.87, + "content": "\\ell _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 698, + 506, + 713 + ], + "score": 1.0, + "content": "loss function and the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 254, + 722 + ], + "score": 1.0, + "content": "ADAM optimizer with learning rate", + "type": "text" + }, + { + "bbox": [ + 254, + 709, + 276, + 720 + ], + "score": 0.9, + "content": "1 0 ^ { - 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "and report results on the test set on the final epoch aver-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 720, + 476, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 476, + 733 + ], + "score": 1.0, + "content": "aged over two runs. We used the default parameters and data splits from Fuchs et al. (2020).", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37.5 + } + ], + "page_idx": 20, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 14 + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 495, + 216, + 504, + 227 + ], + "lines": [ + { + "bbox": [ + 496, + 218, + 504, + 226 + ], + "spans": [ + { + "bbox": [ + 496, + 218, + 504, + 226 + ], + "score": 0.994, + "content": "□", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 81, + 504, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 215, + 95 + ], + "score": 1.0, + "content": "Multi-linear functionals on", + "type": "text" + }, + { + "bbox": [ + 215, + 82, + 234, + 93 + ], + "score": 0.9, + "content": "W ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 81, + 281, + 95 + ], + "score": 1.0, + "content": "invariant to", + "type": "text" + }, + { + "bbox": [ + 282, + 83, + 288, + 94 + ], + "score": 0.85, + "content": "\\tilde { \\rho }", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 81, + 448, + 95 + ], + "score": 1.0, + "content": "are a subset of the set of polynomials on", + "type": "text" + }, + { + "bbox": [ + 448, + 82, + 466, + 93 + ], + "score": 0.9, + "content": "W ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 467, + 81, + 506, + 95 + ], + "score": 1.0, + "content": "invariant", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 104, + 93, + 505, + 107 + ], + "spans": [ + { + "bbox": [ + 104, + 93, + 116, + 107 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 117, + 94, + 123, + 105 + ], + "score": 0.77, + "content": "\\tilde { \\rho }", + "type": "inline_equation" + }, + { + "bbox": [ + 123, + 93, + 505, + 107 + ], + "score": 1.0, + "content": ". It is known (see (Kraft & Procesi, 2000), page 114), that all such polynomials are algebraically", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 105, + 247, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 247, + 117 + ], + "score": 1.0, + "content": "generated by functions of the form", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 104, + 81, + 506, + 117 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 175, + 121, + 432, + 135 + ], + "lines": [ + { + "bbox": [ + 175, + 121, + 432, + 135 + ], + "spans": [ + { + "bbox": [ + 175, + 121, + 432, + 135 + ], + "score": 0.83, + "content": "\\begin{array} { r } { \\operatorname* { d e t } ( w _ { i _ { 1 } } , w _ { i _ { 2 } } , w _ { i _ { 3 } } ) \\mathrm { ~ a n d ~ } \\langle w _ { j _ { 1 } } , w _ { j _ { 2 } } \\rangle , \\mathrm { ~ w h e r e ~ } i _ { 1 } , i _ { 2 } , i _ { 3 } , j _ { 1 } , j _ { 2 } \\in [ D ] . } \\end{array}", + "type": "interline_equation", + "image_path": "5c1bc250cffdd612aa0182d31b2a2ccdf1ef9897cc9acb334763c60639da03d5.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 175, + 121, + 432, + 135 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 140, + 505, + 162 + ], + "lines": [ + { + "bbox": [ + 105, + 139, + 506, + 154 + ], + "spans": [ + { + "bbox": [ + 105, + 139, + 162, + 154 + ], + "score": 1.0, + "content": "Equivalently,", + "type": "text" + }, + { + "bbox": [ + 162, + 140, + 189, + 153 + ], + "score": 0.3, + "content": "\\mathrm { S O ( 3 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 139, + 506, + 154 + ], + "score": 1.0, + "content": "-invariant polynomials are spanned by linear combinations of polynomials of", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 150, + 144, + 164 + ], + "spans": [ + { + "bbox": [ + 105, + 150, + 144, + 164 + ], + "score": 1.0, + "content": "the form", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 105, + 139, + 506, + 164 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 172, + 162, + 439, + 175 + ], + "lines": [ + { + "bbox": [ + 172, + 162, + 439, + 175 + ], + "spans": [ + { + "bbox": [ + 172, + 162, + 439, + 175 + ], + "score": 0.86, + "content": "\\operatorname * { d e t } \\bigl ( w _ { i _ { 1 } } , w _ { i _ { 2 } } , w _ { i _ { 3 } } \\bigr ) \\operatorname * { d e t } \\bigl ( w _ { i _ { 4 } } , w _ { i _ { 5 } } , w _ { i _ { 6 } } \\bigr ) \\ldots \\langle w _ { j _ { 1 } } , w _ { j _ { 2 } } \\rangle \\langle w _ { j _ { 3 } } , w _ { j _ { 4 } } \\rangle \\ldots .", + "type": "interline_equation", + "image_path": "aeb93f1dc0eb0f9e7c84a28c079f6e15a61d4012660eb399737422057ff3481f.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 172, + 162, + 439, + 175 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 177, + 505, + 211 + ], + "lines": [ + { + "bbox": [ + 106, + 176, + 505, + 190 + ], + "spans": [ + { + "bbox": [ + 106, + 176, + 505, + 190 + ], + "score": 1.0, + "content": "When considering the subset of multi-linear invariant polynomials, we see that they must be spanned", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 188, + 505, + 202 + ], + "spans": [ + { + "bbox": [ + 105, + 188, + 290, + 202 + ], + "score": 1.0, + "content": "by polynomials as in equation 27, where each", + "type": "text" + }, + { + "bbox": [ + 291, + 190, + 340, + 200 + ], + "score": 0.9, + "content": "w _ { 1 } , \\ldots , w _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 188, + 505, + 202 + ], + "score": 1.0, + "content": "appears exactly once in each polynomial", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 199, + 391, + 212 + ], + "spans": [ + { + "bbox": [ + 105, + 199, + 372, + 212 + ], + "score": 1.0, + "content": "in the spanning set. These precisely correspond to the functions in", + "type": "text" + }, + { + "bbox": [ + 373, + 200, + 387, + 210 + ], + "score": 0.9, + "content": "\\lambda _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 199, + 391, + 212 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8, + "bbox_fs": [ + 105, + 176, + 505, + 212 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 250, + 203, + 263 + ], + "lines": [ + { + "bbox": [ + 105, + 250, + 205, + 265 + ], + "spans": [ + { + "bbox": [ + 105, + 250, + 205, + 265 + ], + "score": 1.0, + "content": "H EXPERIMENTS", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "table", + "bbox": [ + 153, + 318, + 458, + 409 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 287, + 504, + 318 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 288, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 505, + 299 + ], + "score": 1.0, + "content": "Table 1: Results obtained on the QM9 dataset Ramakrishnan et al. (2014) for different design choices", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 298, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 104, + 298, + 324, + 311 + ], + "score": 1.0, + "content": "in the TFN architecture. Results are reported for the", + "type": "text" + }, + { + "bbox": [ + 324, + 300, + 350, + 310 + ], + "score": 0.64, + "content": "\\epsilon _ { h o m o }", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 298, + 505, + 311 + ], + "score": 1.0, + "content": "target variable, and are multiplied by", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 308, + 128, + 321 + ], + "spans": [ + { + "bbox": [ + 106, + 309, + 122, + 320 + ], + "score": 0.84, + "content": "1 0 ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 122, + 308, + 128, + 321 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12 + }, + { + "type": "table_body", + "bbox": [ + 153, + 318, + 458, + 409 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 153, + 318, + 458, + 409 + ], + "spans": [ + { + "bbox": [ + 153, + 318, + 458, + 409 + ], + "score": 0.98, + "html": "
Model variant Mean l1error
Order 0irreps 111.2 ± 0.3
Order 0-1 ireps 82.1 ± 0.4
Order 0-2 irreps 62.1 ± 1.5
Order 0-3 irreps 53.2 ± 1.9
Order 0-4 irreps 51.2 ± 0.4
Order O-3 irreps without non-linearity 53.1 ± 0.8
Order O-3 irreps + self-interaction only in the final layer 55.2 ± 0.3
", + "type": "table", + "image_path": "30712253f5955d3ca6f795324f935ac71bda0d63fa1692f647560e32ee1bf29d.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 153, + 318, + 458, + 348.3333333333333 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 153, + 348.3333333333333, + 458, + 378.66666666666663 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 153, + 378.66666666666663, + 458, + 408.99999999999994 + ], + "spans": [], + "index": 16 + } + ] + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 106, + 422, + 503, + 445 + ], + "lines": [ + { + "bbox": [ + 106, + 421, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 106, + 421, + 505, + 436 + ], + "score": 1.0, + "content": "This section provides an experimental evaluation of different design choices of the TFN architecture,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 433, + 390, + 446 + ], + "spans": [ + { + "bbox": [ + 106, + 433, + 390, + 446 + ], + "score": 1.0, + "content": "inspired by our theoretical analysis. We study the following questions:", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 106, + 421, + 505, + 446 + ] + }, + { + "type": "list", + "bbox": [ + 129, + 455, + 505, + 584 + ], + "lines": [ + { + "bbox": [ + 130, + 455, + 504, + 468 + ], + "spans": [ + { + "bbox": [ + 130, + 455, + 504, + 468 + ], + "score": 1.0, + "content": "1. The importance of non-linear activation. Our proof shows that using non-linear acti-", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 466, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 141, + 466, + 505, + 479 + ], + "score": 1.0, + "content": "vation functions is not necessary for proving universality. Here, we empirically test the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 477, + 289, + 491 + ], + "spans": [ + { + "bbox": [ + 141, + 477, + 289, + 491 + ], + "score": 1.0, + "content": "possibility of removing these layers.", + "type": "text" + } + ], + "index": 21, + "is_list_end_line": true + }, + { + "bbox": [ + 129, + 492, + 505, + 504 + ], + "spans": [ + { + "bbox": [ + 129, + 492, + 505, + 504 + ], + "score": 1.0, + "content": "2. The importance of high-dimensional irreducible representations. Our theoretical anal-", + "type": "text" + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 502, + 505, + 517 + ], + "spans": [ + { + "bbox": [ + 141, + 502, + 505, + 517 + ], + "score": 1.0, + "content": "ysis shows that in order to represent/approximate high degree polynomials, high-order rep-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 513, + 505, + 527 + ], + "spans": [ + { + "bbox": [ + 141, + 513, + 505, + 527 + ], + "score": 1.0, + "content": "resentations should be used. Here, we check whether using high-order representations has", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 141, + 526, + 215, + 536 + ], + "spans": [ + { + "bbox": [ + 141, + 526, + 215, + 536 + ], + "score": 1.0, + "content": "practical benefits.", + "type": "text" + } + ], + "index": 25, + "is_list_end_line": true + }, + { + "bbox": [ + 130, + 540, + 506, + 553 + ], + "spans": [ + { + "bbox": [ + 130, + 540, + 506, + 553 + ], + "score": 1.0, + "content": "3. The effect of self-interaction layers. Our proof suggests that it is enough to use self", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 551, + 506, + 564 + ], + "spans": [ + { + "bbox": [ + 141, + 551, + 506, + 564 + ], + "score": 1.0, + "content": "interaction linear layers at the end of the model. We empirically compare this approach", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 141, + 562, + 506, + 575 + ], + "spans": [ + { + "bbox": [ + 141, + 562, + 506, + 575 + ], + "score": 1.0, + "content": "with the more common approach of using self-interaction layers after each convolutional", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 140, + 572, + 168, + 587 + ], + "spans": [ + { + "bbox": [ + 140, + 572, + 168, + 587 + ], + "score": 1.0, + "content": "layer.", + "type": "text" + } + ], + "index": 29, + "is_list_end_line": true + } + ], + "index": 24, + "bbox_fs": [ + 129, + 455, + 506, + 587 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 596, + 503, + 630 + ], + "lines": [ + { + "bbox": [ + 106, + 596, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 106, + 596, + 505, + 609 + ], + "score": 1.0, + "content": "Dataset. We use the QM9 (Ramakrishnan et al., 2014) dataset for our experiments. The dataset", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 606, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 105, + 606, + 505, + 621 + ], + "score": 1.0, + "content": "contains 134K molecules, with node 3D positions, 5 categorical node features and 4 categorical edge", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 618, + 379, + 632 + ], + "spans": [ + { + "bbox": [ + 106, + 618, + 379, + 632 + ], + "score": 1.0, + "content": "features. The task is a molecule property prediction regression task.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 596, + 505, + 632 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 641, + 505, + 675 + ], + "lines": [ + { + "bbox": [ + 105, + 640, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 505, + 655 + ], + "score": 1.0, + "content": "Framework. We used pytorch (Paszke et al., 2017)as the deep learning framework and the Deep", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 652, + 505, + 666 + ], + "spans": [ + { + "bbox": [ + 105, + 652, + 505, + 666 + ], + "score": 1.0, + "content": "Graph Library (DGL) (Wang et al., 2019a) as the graph learning framework. All experiments ran on", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 664, + 205, + 675 + ], + "spans": [ + { + "bbox": [ + 106, + 664, + 205, + 675 + ], + "score": 1.0, + "content": "NVIDIA GV100 GPUs.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 640, + 505, + 675 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "Experimental setup. We use the the TFN implementation from Fuchs et al. (2020). We trained", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 698, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 276, + 713 + ], + "score": 1.0, + "content": "each model variant for 50 epochs on the", + "type": "text" + }, + { + "bbox": [ + 277, + 700, + 303, + 710 + ], + "score": 0.83, + "content": "\\epsilon _ { h o m o }", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 698, + 404, + 713 + ], + "score": 1.0, + "content": "target variable using an", + "type": "text" + }, + { + "bbox": [ + 404, + 699, + 414, + 710 + ], + "score": 0.87, + "content": "\\ell _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 698, + 506, + 713 + ], + "score": 1.0, + "content": "loss function and the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 254, + 722 + ], + "score": 1.0, + "content": "ADAM optimizer with learning rate", + "type": "text" + }, + { + "bbox": [ + 254, + 709, + 276, + 720 + ], + "score": 0.9, + "content": "1 0 ^ { - 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "and report results on the test set on the final epoch aver-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 720, + 476, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 476, + 733 + ], + "score": 1.0, + "content": "aged over two runs. We used the default parameters and data splits from Fuchs et al. (2020).", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 687, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 138 + ], + "lines": [ + { + "bbox": [ + 106, + 83, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 83, + 505, + 95 + ], + "score": 1.0, + "content": "Architecture. The architecture consists of 4 TFN convolutional layers, each followed by a linear", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 504, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 504, + 106 + ], + "score": 1.0, + "content": "self-interaction layer. We used 16 copies of each irreducible representation used. We used norm-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "score": 1.0, + "content": "based non-linearity as in the original TFN paper (Thomas et al., 2018). These convolutional layers", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 399, + 128 + ], + "score": 1.0, + "content": "are followed by a max-pooling layer and two fully connected layers with", + "type": "text" + }, + { + "bbox": [ + 399, + 115, + 416, + 126 + ], + "score": 0.76, + "content": "1 6 d", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 115, + 506, + 128 + ], + "score": 1.0, + "content": "features in the hidden", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 126, + 398, + 140 + ], + "spans": [ + { + "bbox": [ + 106, + 126, + 158, + 140 + ], + "score": 1.0, + "content": "layer, where", + "type": "text" + }, + { + "bbox": [ + 158, + 127, + 164, + 136 + ], + "score": 0.78, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 126, + 398, + 140 + ], + "score": 1.0, + "content": "is the maximal degree of irreducible representations used.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 107, + 149, + 296, + 348 + ], + "lines": [ + { + "bbox": [ + 106, + 149, + 297, + 163 + ], + "spans": [ + { + "bbox": [ + 106, + 149, + 297, + 163 + ], + "score": 1.0, + "content": "Results. Table 1 and Figure 1 present the re-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 160, + 297, + 173 + ], + "spans": [ + { + "bbox": [ + 106, + 160, + 297, + 173 + ], + "score": 1.0, + "content": "sults. The main conclusions are: (1) The ex-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 171, + 298, + 185 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 298, + 185 + ], + "score": 1.0, + "content": "periments show that, at least for this task, using", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 183, + 297, + 195 + ], + "spans": [ + { + "bbox": [ + 106, + 183, + 297, + 195 + ], + "score": 1.0, + "content": "non-linear activations does not improve perfor-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 194, + 297, + 206 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 297, + 206 + ], + "score": 1.0, + "content": "mance. This result fits our theoretical analysis", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 204, + 297, + 216 + ], + "spans": [ + { + "bbox": [ + 106, + 204, + 297, + 216 + ], + "score": 1.0, + "content": "which shows that these layers are not needed", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 215, + 298, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 298, + 228 + ], + "score": 1.0, + "content": "for universality. (2) Figure 1 presents a plot", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 226, + 298, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 226, + 298, + 239 + ], + "score": 1.0, + "content": "of error vs representation degrees used. The", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 237, + 297, + 250 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 297, + 250 + ], + "score": 1.0, + "content": "plot clearly shows that using high-dimensional", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 249, + 297, + 261 + ], + "spans": [ + { + "bbox": [ + 106, + 249, + 297, + 261 + ], + "score": 1.0, + "content": "representations (up to order 3) improves per-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 258, + 298, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 258, + 298, + 272 + ], + "score": 1.0, + "content": "formance, which also fits our analysis. Using", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 271, + 297, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 297, + 282 + ], + "score": 1.0, + "content": "representation orders higher than 3 is signifi-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 282, + 297, + 293 + ], + "spans": [ + { + "bbox": [ + 106, + 282, + 297, + 293 + ], + "score": 1.0, + "content": "cantly more time consuming, and was found to", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 292, + 297, + 304 + ], + "spans": [ + { + "bbox": [ + 106, + 292, + 297, + 304 + ], + "score": 1.0, + "content": "have little effect on the results (as in in Fuchs", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 303, + 297, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 297, + 316 + ], + "score": 1.0, + "content": "et al. (2020)), though we believe this to be", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 314, + 297, + 327 + ], + "spans": [ + { + "bbox": [ + 105, + 314, + 297, + 327 + ], + "score": 1.0, + "content": "application-dependent. (3) Using self interac-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 325, + 297, + 337 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 297, + 337 + ], + "score": 1.0, + "content": "tion layers only at the end of the model is shown", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 337, + 294, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 337, + 294, + 348 + ], + "score": 1.0, + "content": "to have marginal negative effect on the results.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 13.5 + }, + { + "type": "image", + "bbox": [ + 308, + 155, + 481, + 285 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 308, + 155, + 481, + 285 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 308, + 155, + 481, + 285 + ], + "spans": [ + { + "bbox": [ + 308, + 155, + 481, + 285 + ], + "score": 0.968, + "type": "image", + "image_path": "0a3168963674044c22256599e104a6aa2cefc267903dcecc69ad907cfab638f0.jpg" + } + ] + } + ], + "index": 27.5, + "virtual_lines": [ + { + "bbox": [ + 308, + 155, + 481, + 168.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 308, + 168.0, + 481, + 181.0 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 308, + 181.0, + 481, + 194.0 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 308, + 194.0, + 481, + 207.0 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 308, + 207.0, + 481, + 220.0 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 308, + 220.0, + 481, + 233.0 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 308, + 233.0, + 481, + 246.0 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 308, + 246.0, + 481, + 259.0 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 308, + 259.0, + 481, + 272.0 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 308, + 272.0, + 481, + 285.0 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 304, + 300, + 505, + 355 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 304, + 300, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 304, + 300, + 351, + 312 + ], + "score": 1.0, + "content": "Figure 1:", + "type": "text" + }, + { + "bbox": [ + 351, + 300, + 361, + 311 + ], + "score": 0.83, + "content": "\\ell _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 300, + 505, + 312 + ], + "score": 1.0, + "content": "error versus maximal irreducible", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 303, + 311, + 505, + 323 + ], + "spans": [ + { + "bbox": [ + 303, + 311, + 505, + 323 + ], + "score": 1.0, + "content": "representation used. It is clear that the error is", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 304, + 323, + 504, + 334 + ], + "spans": [ + { + "bbox": [ + 304, + 323, + 504, + 334 + ], + "score": 1.0, + "content": "reduced as higher order representation are used.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 304, + 333, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 304, + 333, + 505, + 344 + ], + "score": 1.0, + "content": "Semi-transparent color represents standard devia-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 303, + 344, + 326, + 356 + ], + "spans": [ + { + "bbox": [ + 303, + 344, + 326, + 356 + ], + "score": 1.0, + "content": "tion.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35 + } + ], + "index": 31.25 + } + ], + "page_idx": 21, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 292, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2021", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "22", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 138 + ], + "lines": [ + { + "bbox": [ + 106, + 83, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 83, + 505, + 95 + ], + "score": 1.0, + "content": "Architecture. The architecture consists of 4 TFN convolutional layers, each followed by a linear", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 504, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 504, + 106 + ], + "score": 1.0, + "content": "self-interaction layer. We used 16 copies of each irreducible representation used. We used norm-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "score": 1.0, + "content": "based non-linearity as in the original TFN paper (Thomas et al., 2018). These convolutional layers", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 399, + 128 + ], + "score": 1.0, + "content": "are followed by a max-pooling layer and two fully connected layers with", + "type": "text" + }, + { + "bbox": [ + 399, + 115, + 416, + 126 + ], + "score": 0.76, + "content": "1 6 d", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 115, + 506, + 128 + ], + "score": 1.0, + "content": "features in the hidden", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 126, + 398, + 140 + ], + "spans": [ + { + "bbox": [ + 106, + 126, + 158, + 140 + ], + "score": 1.0, + "content": "layer, where", + "type": "text" + }, + { + "bbox": [ + 158, + 127, + 164, + 136 + ], + "score": 0.78, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 126, + 398, + 140 + ], + "score": 1.0, + "content": "is the maximal degree of irreducible representations used.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 83, + 506, + 140 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 149, + 296, + 348 + ], + "lines": [ + { + "bbox": [ + 106, + 149, + 297, + 163 + ], + "spans": [ + { + "bbox": [ + 106, + 149, + 297, + 163 + ], + "score": 1.0, + "content": "Results. Table 1 and Figure 1 present the re-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 160, + 297, + 173 + ], + "spans": [ + { + "bbox": [ + 106, + 160, + 297, + 173 + ], + "score": 1.0, + "content": "sults. The main conclusions are: (1) The ex-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 171, + 298, + 185 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 298, + 185 + ], + "score": 1.0, + "content": "periments show that, at least for this task, using", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 183, + 297, + 195 + ], + "spans": [ + { + "bbox": [ + 106, + 183, + 297, + 195 + ], + "score": 1.0, + "content": "non-linear activations does not improve perfor-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 194, + 297, + 206 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 297, + 206 + ], + "score": 1.0, + "content": "mance. This result fits our theoretical analysis", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 204, + 297, + 216 + ], + "spans": [ + { + "bbox": [ + 106, + 204, + 297, + 216 + ], + "score": 1.0, + "content": "which shows that these layers are not needed", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 215, + 298, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 298, + 228 + ], + "score": 1.0, + "content": "for universality. (2) Figure 1 presents a plot", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 226, + 298, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 226, + 298, + 239 + ], + "score": 1.0, + "content": "of error vs representation degrees used. The", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 237, + 297, + 250 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 297, + 250 + ], + "score": 1.0, + "content": "plot clearly shows that using high-dimensional", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 249, + 297, + 261 + ], + "spans": [ + { + "bbox": [ + 106, + 249, + 297, + 261 + ], + "score": 1.0, + "content": "representations (up to order 3) improves per-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 258, + 298, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 258, + 298, + 272 + ], + "score": 1.0, + "content": "formance, which also fits our analysis. Using", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 271, + 297, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 297, + 282 + ], + "score": 1.0, + "content": "representation orders higher than 3 is signifi-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 282, + 297, + 293 + ], + "spans": [ + { + "bbox": [ + 106, + 282, + 297, + 293 + ], + "score": 1.0, + "content": "cantly more time consuming, and was found to", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 292, + 297, + 304 + ], + "spans": [ + { + "bbox": [ + 106, + 292, + 297, + 304 + ], + "score": 1.0, + "content": "have little effect on the results (as in in Fuchs", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 303, + 297, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 297, + 316 + ], + "score": 1.0, + "content": "et al. (2020)), though we believe this to be", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 314, + 297, + 327 + ], + "spans": [ + { + "bbox": [ + 105, + 314, + 297, + 327 + ], + "score": 1.0, + "content": "application-dependent. (3) Using self interac-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 325, + 297, + 337 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 297, + 337 + ], + "score": 1.0, + "content": "tion layers only at the end of the model is shown", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 337, + 294, + 348 + ], + "spans": [ + { + "bbox": [ + 105, + 337, + 294, + 348 + ], + "score": 1.0, + "content": "to have marginal negative effect on the results.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 13.5, + "bbox_fs": [ + 105, + 149, + 298, + 348 + ] + }, + { + "type": "image", + "bbox": [ + 308, + 155, + 481, + 285 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 308, + 155, + 481, + 285 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 308, + 155, + 481, + 285 + ], + "spans": [ + { + "bbox": [ + 308, + 155, + 481, + 285 + ], + "score": 0.968, + "type": "image", + "image_path": "0a3168963674044c22256599e104a6aa2cefc267903dcecc69ad907cfab638f0.jpg" + } + ] + } + ], + "index": 27.5, + "virtual_lines": [ + { + "bbox": [ + 308, + 155, + 481, + 168.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 308, + 168.0, + 481, + 181.0 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 308, + 181.0, + 481, + 194.0 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 308, + 194.0, + 481, + 207.0 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 308, + 207.0, + 481, + 220.0 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 308, + 220.0, + 481, + 233.0 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 308, + 233.0, + 481, + 246.0 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 308, + 246.0, + 481, + 259.0 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 308, + 259.0, + 481, + 272.0 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 308, + 272.0, + 481, + 285.0 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 304, + 300, + 505, + 355 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 304, + 300, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 304, + 300, + 351, + 312 + ], + "score": 1.0, + "content": "Figure 1:", + "type": "text" + }, + { + "bbox": [ + 351, + 300, + 361, + 311 + ], + "score": 0.83, + "content": "\\ell _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 300, + 505, + 312 + ], + "score": 1.0, + "content": "error versus maximal irreducible", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 303, + 311, + 505, + 323 + ], + "spans": [ + { + "bbox": [ + 303, + 311, + 505, + 323 + ], + "score": 1.0, + "content": "representation used. It is clear that the error is", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 304, + 323, + 504, + 334 + ], + "spans": [ + { + "bbox": [ + 304, + 323, + 504, + 334 + ], + "score": 1.0, + "content": "reduced as higher order representation are used.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 304, + 333, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 304, + 333, + 505, + 344 + ], + "score": 1.0, + "content": "Semi-transparent color represents standard devia-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 303, + 344, + 326, + 356 + ], + "spans": [ + { + "bbox": [ + 303, + 344, + 326, + 356 + ], + "score": 1.0, + "content": "tion.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35 + } + ], + "index": 31.25 + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/train/6NFBvWlRXaG/6NFBvWlRXaG_model.json b/parse/train/6NFBvWlRXaG/6NFBvWlRXaG_model.json new file mode 100644 index 0000000000000000000000000000000000000000..eab46fa3482ce10ce3b71fac582aa84bdc6fa53a --- /dev/null +++ b/parse/train/6NFBvWlRXaG/6NFBvWlRXaG_model.json @@ -0,0 +1,45213 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 398, + 617, + 1302, + 617, + 1302, + 922, + 398, + 922 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 299, + 1780, + 1403, + 1780, + 1403, + 1966, + 299, + 1966 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 299, + 1262, + 936, + 1262, + 936, + 1505, + 299, + 1505 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 297, + 1048, + 1404, + 1048, + 1404, + 1260, + 297, + 1260 + ], + "score": 0.967 + }, + { + "category_id": 1, + "poly": [ + 298, + 1551, + 1405, + 1551, + 1405, + 1765, + 298, + 1765 + ], + "score": 0.965 + }, + { + "category_id": 0, + "poly": [ + 300, + 219, + 1397, + 219, + 1397, + 323, + 300, + 323 + ], + "score": 0.962 + }, + { + "category_id": 3, + "poly": [ + 994, + 1302, + 1330, + 1302, + 1330, + 1538, + 994, + 1538 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 312, + 377, + 619, + 377, + 619, + 468, + 312, + 468 + ], + "score": 0.944 + }, + { + "category_id": 1, + "poly": [ + 870, + 377, + 1160, + 377, + 1160, + 468, + 870, + 468 + ], + "score": 0.942 + }, + { + "category_id": 1, + "poly": [ + 298, + 1972, + 1401, + 1972, + 1401, + 2034, + 298, + 2034 + ], + "score": 0.925 + }, + { + "category_id": 0, + "poly": [ + 302, + 978, + 580, + 978, + 580, + 1013, + 302, + 1013 + ], + "score": 0.898 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 813, + 75, + 813, + 105, + 299, + 105 + ], + "score": 0.879 + }, + { + "category_id": 0, + "poly": [ + 773, + 550, + 927, + 550, + 927, + 583, + 773, + 583 + ], + "score": 0.873 + }, + { + "category_id": 1, + "poly": [ + 298, + 1521, + 935, + 1521, + 935, + 1551, + 298, + 1551 + ], + "score": 0.789 + }, + { + "category_id": 2, + "poly": [ + 842, + 2088, + 857, + 2088, + 857, + 2112, + 842, + 2112 + ], + "score": 0.629 + }, + { + "category_id": 13, + "poly": [ + 1039, + 1643, + 1112, + 1643, + 1112, + 1672, + 1039, + 1672 + ], + "score": 0.56, + "latex": "2 0 1 9 \\mathrm { c }" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 211.0, + 1406.0, + 211.0, + 1406.0, + 279.0, + 293.0, + 279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 272.0, + 846.0, + 272.0, + 846.0, + 328.0, + 293.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 1308.0, + 1220.0, + 1308.0, + 1220.0, + 1340.0, + 1035.0, + 1340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1070.0, + 1342.0, + 1255.0, + 1342.0, + 1255.0, + 1376.0, + 1070.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 1391.0, + 1097.0, + 1391.0, + 1097.0, + 1420.0, + 1050.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1108.0, + 1395.0, + 1297.0, + 1395.0, + 1297.0, + 1430.0, + 1108.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 973.0, + 587.0, + 973.0, + 587.0, + 1024.0, + 291.0, + 1024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 546.0, + 934.0, + 546.0, + 934.0, + 589.0, + 768.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 841.0, + 2088.0, + 858.0, + 2088.0, + 858.0, + 2116.0, + 841.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 619.0, + 1303.0, + 619.0, + 1303.0, + 651.0, + 394.0, + 651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 649.0, + 1305.0, + 649.0, + 1305.0, + 681.0, + 391.0, + 681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 680.0, + 1306.0, + 680.0, + 1306.0, + 711.0, + 391.0, + 711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 707.0, + 1306.0, + 707.0, + 1306.0, + 745.0, + 393.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 741.0, + 1306.0, + 741.0, + 1306.0, + 774.0, + 394.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 771.0, + 1305.0, + 771.0, + 1305.0, + 802.0, + 395.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 799.0, + 1308.0, + 799.0, + 1308.0, + 838.0, + 393.0, + 838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 833.0, + 1304.0, + 833.0, + 1304.0, + 861.0, + 395.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 861.0, + 1305.0, + 861.0, + 1305.0, + 894.0, + 394.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 891.0, + 1251.0, + 891.0, + 1251.0, + 926.0, + 394.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1781.0, + 1404.0, + 1781.0, + 1404.0, + 1813.0, + 298.0, + 1813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1809.0, + 1405.0, + 1809.0, + 1405.0, + 1846.0, + 293.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1840.0, + 1407.0, + 1840.0, + 1407.0, + 1877.0, + 292.0, + 1877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1872.0, + 1404.0, + 1872.0, + 1404.0, + 1904.0, + 295.0, + 1904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1899.0, + 1407.0, + 1899.0, + 1407.0, + 1939.0, + 292.0, + 1939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1937.0, + 374.0, + 1937.0, + 374.0, + 1968.0, + 291.0, + 1968.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1260.0, + 937.0, + 1260.0, + 937.0, + 1293.0, + 297.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1291.0, + 938.0, + 1291.0, + 938.0, + 1324.0, + 296.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1320.0, + 937.0, + 1320.0, + 937.0, + 1355.0, + 295.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1351.0, + 938.0, + 1351.0, + 938.0, + 1385.0, + 295.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1382.0, + 937.0, + 1382.0, + 937.0, + 1415.0, + 294.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1411.0, + 937.0, + 1411.0, + 937.0, + 1447.0, + 293.0, + 1447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1442.0, + 937.0, + 1442.0, + 937.0, + 1476.0, + 296.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1473.0, + 870.0, + 1473.0, + 870.0, + 1506.0, + 296.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1046.0, + 1405.0, + 1046.0, + 1405.0, + 1083.0, + 294.0, + 1083.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1076.0, + 1405.0, + 1076.0, + 1405.0, + 1111.0, + 292.0, + 1111.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1106.0, + 1405.0, + 1106.0, + 1405.0, + 1144.0, + 292.0, + 1144.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1138.0, + 1405.0, + 1138.0, + 1405.0, + 1171.0, + 292.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1166.0, + 1409.0, + 1166.0, + 1409.0, + 1204.0, + 291.0, + 1204.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1199.0, + 1407.0, + 1199.0, + 1407.0, + 1233.0, + 294.0, + 1233.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1230.0, + 1405.0, + 1230.0, + 1405.0, + 1264.0, + 294.0, + 1264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1549.0, + 1405.0, + 1549.0, + 1405.0, + 1587.0, + 295.0, + 1587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1581.0, + 1405.0, + 1581.0, + 1405.0, + 1615.0, + 293.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1611.0, + 1405.0, + 1611.0, + 1405.0, + 1646.0, + 293.0, + 1646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1641.0, + 1038.0, + 1641.0, + 1038.0, + 1679.0, + 292.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1113.0, + 1641.0, + 1405.0, + 1641.0, + 1405.0, + 1679.0, + 1113.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1673.0, + 1405.0, + 1673.0, + 1405.0, + 1707.0, + 295.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1701.0, + 1408.0, + 1701.0, + 1408.0, + 1740.0, + 292.0, + 1740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1737.0, + 636.0, + 1737.0, + 636.0, + 1766.0, + 294.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 374.0, + 462.0, + 374.0, + 462.0, + 412.0, + 310.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 403.0, + 504.0, + 403.0, + 504.0, + 443.0, + 310.0, + 443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 439.0, + 619.0, + 439.0, + 619.0, + 471.0, + 313.0, + 471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 375.0, + 1053.0, + 375.0, + 1053.0, + 412.0, + 869.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 406.0, + 1085.0, + 406.0, + 1085.0, + 437.0, + 870.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 438.0, + 1162.0, + 438.0, + 1162.0, + 469.0, + 870.0, + 469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1969.0, + 1405.0, + 1969.0, + 1405.0, + 2011.0, + 294.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2003.0, + 1149.0, + 2003.0, + 1149.0, + 2038.0, + 294.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1516.0, + 937.0, + 1516.0, + 937.0, + 1558.0, + 293.0, + 1558.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1028, + 1404, + 1028, + 1404, + 1395, + 298, + 1395 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 297, + 645, + 1405, + 645, + 1405, + 1012, + 297, + 1012 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 1682, + 1404, + 1682, + 1404, + 1838, + 298, + 1838 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 299, + 229, + 1402, + 229, + 1402, + 384, + 299, + 384 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 299, + 398, + 1402, + 398, + 1402, + 521, + 299, + 521 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 299, + 1518, + 1402, + 1518, + 1402, + 1581, + 299, + 1581 + ], + "score": 0.954 + }, + { + "category_id": 1, + "poly": [ + 295, + 1972, + 1403, + 1972, + 1403, + 2035, + 295, + 2035 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 293, + 1852, + 1400, + 1852, + 1400, + 1916, + 293, + 1916 + ], + "score": 0.945 + }, + { + "category_id": 8, + "poly": [ + 692, + 1920, + 1003, + 1920, + 1003, + 1958, + 692, + 1958 + ], + "score": 0.941 + }, + { + "category_id": 0, + "poly": [ + 301, + 572, + 593, + 572, + 593, + 608, + 301, + 608 + ], + "score": 0.907 + }, + { + "category_id": 0, + "poly": [ + 300, + 1448, + 994, + 1448, + 994, + 1480, + 300, + 1480 + ], + "score": 0.905 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 813, + 76, + 813, + 104, + 300, + 104 + ], + "score": 0.894 + }, + { + "category_id": 0, + "poly": [ + 300, + 1624, + 645, + 1624, + 645, + 1656, + 300, + 1656 + ], + "score": 0.894 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.739 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.157 + }, + { + "category_id": 13, + "poly": [ + 934, + 1716, + 1058, + 1716, + 1058, + 1748, + 934, + 1748 + ], + "score": 0.92, + "latex": "g _ { 1 } , g _ { 2 } \\in G" + }, + { + "category_id": 13, + "poly": [ + 1176, + 1746, + 1228, + 1746, + 1228, + 1779, + 1176, + 1779 + ], + "score": 0.92, + "latex": "\\rho ( g )" + }, + { + "category_id": 13, + "poly": [ + 797, + 1272, + 875, + 1272, + 875, + 1305, + 797, + 1305 + ], + "score": 0.92, + "latex": "S E ( 2 )" + }, + { + "category_id": 13, + "poly": [ + 1122, + 1684, + 1311, + 1684, + 1311, + 1718, + 1122, + 1718 + ], + "score": 0.91, + "latex": "\\rho ( g ) : W \\to W" + }, + { + "category_id": 14, + "poly": [ + 692, + 1916, + 1007, + 1916, + 1007, + 1957, + 692, + 1957 + ], + "score": 0.91, + "latex": "X = ( x _ { 1 } , \\ldots , x _ { n } ) \\in \\mathbb { R } ^ { 3 \\times n } ." + }, + { + "category_id": 13, + "poly": [ + 578, + 1715, + 855, + 1715, + 855, + 1749, + 578, + 1749 + ], + "score": 0.91, + "latex": "\\rho ( g _ { 1 } ) \\circ \\rho ( g _ { 2 } ) = \\rho ( g _ { 1 } g _ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 405, + 1777, + 480, + 1777, + 480, + 1809, + 405, + 1809 + ], + "score": 0.9, + "latex": "g \\in G" + }, + { + "category_id": 13, + "poly": [ + 384, + 1716, + 459, + 1716, + 459, + 1748, + 384, + 1748 + ], + "score": 0.9, + "latex": "g \\in G" + }, + { + "category_id": 13, + "poly": [ + 1077, + 1120, + 1112, + 1120, + 1112, + 1150, + 1077, + 1150 + ], + "score": 0.89, + "latex": "S _ { n }" + }, + { + "category_id": 13, + "poly": [ + 1365, + 1852, + 1400, + 1852, + 1400, + 1881, + 1365, + 1881 + ], + "score": 0.87, + "latex": "\\mathbb { R } ^ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 1122, + 1747, + 1148, + 1747, + 1148, + 1774, + 1122, + 1774 + ], + "score": 0.82, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 856, + 1752, + 875, + 1752, + 875, + 1777, + 856, + 1777 + ], + "score": 0.82, + "latex": "\\rho" + }, + { + "category_id": 13, + "poly": [ + 1376, + 1716, + 1401, + 1716, + 1401, + 1743, + 1376, + 1743 + ], + "score": 0.81, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 492, + 1808, + 518, + 1808, + 518, + 1834, + 492, + 1834 + ], + "score": 0.8, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 540, + 1686, + 565, + 1686, + 565, + 1712, + 540, + 1712 + ], + "score": 0.8, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 1260, + 1777, + 1293, + 1777, + 1293, + 1804, + 1260, + 1804 + ], + "score": 0.8, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 580, + 1332, + 658, + 1332, + 658, + 1366, + 580, + 1366 + ], + "score": 0.77, + "latex": "S O ( 3 )" + }, + { + "category_id": 13, + "poly": [ + 822, + 1686, + 855, + 1686, + 855, + 1712, + 822, + 1712 + ], + "score": 0.74, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 721, + 1748, + 753, + 1748, + 753, + 1774, + 721, + 1774 + ], + "score": 0.72, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 971, + 1090, + 1043, + 1090, + 1043, + 1119, + 971, + 1119 + ], + "score": 0.44, + "latex": "2 0 1 7 \\mathrm { a }" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 571.0, + 599.0, + 571.0, + 599.0, + 614.0, + 292.0, + 614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1442.0, + 999.0, + 1442.0, + 999.0, + 1487.0, + 291.0, + 1487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 72.0, + 816.0, + 72.0, + 816.0, + 110.0, + 294.0, + 110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1624.0, + 648.0, + 1624.0, + 648.0, + 1660.0, + 295.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1028.0, + 1405.0, + 1028.0, + 1405.0, + 1066.0, + 296.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1057.0, + 1404.0, + 1057.0, + 1404.0, + 1096.0, + 293.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1089.0, + 970.0, + 1089.0, + 970.0, + 1125.0, + 292.0, + 1125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 1089.0, + 1406.0, + 1089.0, + 1406.0, + 1125.0, + 1044.0, + 1125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1120.0, + 1076.0, + 1120.0, + 1076.0, + 1154.0, + 293.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1113.0, + 1120.0, + 1404.0, + 1120.0, + 1404.0, + 1154.0, + 1113.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1147.0, + 1405.0, + 1147.0, + 1405.0, + 1187.0, + 292.0, + 1187.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1180.0, + 1402.0, + 1180.0, + 1402.0, + 1216.0, + 293.0, + 1216.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1210.0, + 1405.0, + 1210.0, + 1405.0, + 1249.0, + 291.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1239.0, + 1405.0, + 1239.0, + 1405.0, + 1276.0, + 293.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1271.0, + 796.0, + 1271.0, + 796.0, + 1308.0, + 293.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 876.0, + 1271.0, + 1403.0, + 1271.0, + 1403.0, + 1308.0, + 876.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1301.0, + 1406.0, + 1301.0, + 1406.0, + 1339.0, + 292.0, + 1339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1332.0, + 579.0, + 1332.0, + 579.0, + 1370.0, + 292.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 1332.0, + 1404.0, + 1332.0, + 1404.0, + 1370.0, + 659.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1364.0, + 1404.0, + 1364.0, + 1404.0, + 1398.0, + 294.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 645.0, + 1405.0, + 645.0, + 1405.0, + 683.0, + 294.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 675.0, + 1406.0, + 675.0, + 1406.0, + 714.0, + 291.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 708.0, + 1405.0, + 708.0, + 1405.0, + 742.0, + 294.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 736.0, + 1403.0, + 736.0, + 1403.0, + 771.0, + 294.0, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 766.0, + 1405.0, + 766.0, + 1405.0, + 804.0, + 292.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 799.0, + 1405.0, + 799.0, + 1405.0, + 833.0, + 294.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 830.0, + 1403.0, + 830.0, + 1403.0, + 864.0, + 294.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 860.0, + 1405.0, + 860.0, + 1405.0, + 894.0, + 294.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 891.0, + 1403.0, + 891.0, + 1403.0, + 925.0, + 294.0, + 925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 920.0, + 1405.0, + 920.0, + 1405.0, + 954.0, + 295.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 951.0, + 1405.0, + 951.0, + 1405.0, + 985.0, + 295.0, + 985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 982.0, + 1140.0, + 982.0, + 1140.0, + 1016.0, + 294.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1680.0, + 539.0, + 1680.0, + 539.0, + 1720.0, + 293.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 566.0, + 1680.0, + 821.0, + 1680.0, + 821.0, + 1720.0, + 566.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 856.0, + 1680.0, + 1121.0, + 1680.0, + 1121.0, + 1720.0, + 856.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1312.0, + 1680.0, + 1407.0, + 1680.0, + 1407.0, + 1720.0, + 1312.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1712.0, + 383.0, + 1712.0, + 383.0, + 1752.0, + 293.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 460.0, + 1712.0, + 577.0, + 1712.0, + 577.0, + 1752.0, + 460.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 856.0, + 1712.0, + 933.0, + 1712.0, + 933.0, + 1752.0, + 856.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 1712.0, + 1375.0, + 1712.0, + 1375.0, + 1752.0, + 1059.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1712.0, + 1405.0, + 1712.0, + 1405.0, + 1752.0, + 1402.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1737.0, + 720.0, + 1737.0, + 720.0, + 1788.0, + 290.0, + 1788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 754.0, + 1737.0, + 855.0, + 1737.0, + 855.0, + 1788.0, + 754.0, + 1788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 876.0, + 1737.0, + 1121.0, + 1737.0, + 1121.0, + 1788.0, + 876.0, + 1788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 1737.0, + 1175.0, + 1737.0, + 1175.0, + 1788.0, + 1149.0, + 1788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1229.0, + 1737.0, + 1406.0, + 1737.0, + 1406.0, + 1788.0, + 1229.0, + 1788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1773.0, + 404.0, + 1773.0, + 404.0, + 1815.0, + 293.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 481.0, + 1773.0, + 1259.0, + 1773.0, + 1259.0, + 1815.0, + 481.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1294.0, + 1773.0, + 1409.0, + 1773.0, + 1409.0, + 1815.0, + 1294.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1805.0, + 491.0, + 1805.0, + 491.0, + 1841.0, + 293.0, + 1841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 519.0, + 1805.0, + 535.0, + 1805.0, + 535.0, + 1841.0, + 519.0, + 1841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 229.0, + 1406.0, + 229.0, + 1406.0, + 267.0, + 297.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 260.0, + 1403.0, + 260.0, + 1403.0, + 296.0, + 295.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 288.0, + 1406.0, + 288.0, + 1406.0, + 328.0, + 293.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 320.0, + 1406.0, + 320.0, + 1406.0, + 357.0, + 295.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 352.0, + 857.0, + 352.0, + 857.0, + 388.0, + 294.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 399.0, + 1403.0, + 399.0, + 1403.0, + 435.0, + 294.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 430.0, + 1404.0, + 430.0, + 1404.0, + 463.0, + 295.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 458.0, + 1406.0, + 458.0, + 1406.0, + 496.0, + 293.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 492.0, + 453.0, + 492.0, + 453.0, + 522.0, + 297.0, + 522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1516.0, + 1404.0, + 1516.0, + 1404.0, + 1555.0, + 293.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1551.0, + 803.0, + 1551.0, + 803.0, + 1583.0, + 297.0, + 1583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1970.0, + 1404.0, + 1970.0, + 1404.0, + 2006.0, + 294.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2004.0, + 1404.0, + 2004.0, + 1404.0, + 2035.0, + 295.0, + 2035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1851.0, + 1364.0, + 1851.0, + 1364.0, + 1887.0, + 292.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1884.0, + 540.0, + 1884.0, + 540.0, + 1920.0, + 295.0, + 1920.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 727, + 1405, + 727, + 1405, + 1004, + 297, + 1004 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 1222, + 1404, + 1222, + 1404, + 1438, + 297, + 1438 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 528, + 1404, + 528, + 1404, + 712, + 297, + 712 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1609, + 826, + 1609, + 826, + 1946, + 298, + 1946 + ], + "score": 0.969 + }, + { + "category_id": 1, + "poly": [ + 298, + 355, + 1404, + 355, + 1404, + 446, + 298, + 446 + ], + "score": 0.962 + }, + { + "category_id": 1, + "poly": [ + 296, + 1453, + 1404, + 1453, + 1404, + 1608, + 296, + 1608 + ], + "score": 0.96 + }, + { + "category_id": 8, + "poly": [ + 418, + 1955, + 1278, + 1955, + 1278, + 2042, + 418, + 2042 + ], + "score": 0.944 + }, + { + "category_id": 1, + "poly": [ + 297, + 1016, + 1399, + 1016, + 1399, + 1087, + 297, + 1087 + ], + "score": 0.937 + }, + { + "category_id": 8, + "poly": [ + 644, + 266, + 1053, + 266, + 1053, + 306, + 644, + 306 + ], + "score": 0.923 + }, + { + "category_id": 1, + "poly": [ + 295, + 227, + 1052, + 227, + 1052, + 262, + 295, + 262 + ], + "score": 0.921 + }, + { + "category_id": 8, + "poly": [ + 584, + 446, + 1117, + 446, + 1117, + 481, + 584, + 481 + ], + "score": 0.921 + }, + { + "category_id": 1, + "poly": [ + 301, + 309, + 1201, + 309, + 1201, + 342, + 301, + 342 + ], + "score": 0.919 + }, + { + "category_id": 1, + "poly": [ + 299, + 1098, + 1105, + 1098, + 1105, + 1132, + 299, + 1132 + ], + "score": 0.919 + }, + { + "category_id": 2, + "poly": [ + 298, + 74, + 814, + 74, + 814, + 105, + 298, + 105 + ], + "score": 0.913 + }, + { + "category_id": 0, + "poly": [ + 303, + 1166, + 757, + 1166, + 757, + 1197, + 303, + 1197 + ], + "score": 0.903 + }, + { + "category_id": 1, + "poly": [ + 302, + 482, + 1399, + 482, + 1399, + 515, + 302, + 515 + ], + "score": 0.897 + }, + { + "category_id": 9, + "poly": [ + 1366, + 273, + 1400, + 273, + 1400, + 303, + 1366, + 303 + ], + "score": 0.893 + }, + { + "category_id": 9, + "poly": [ + 1364, + 1982, + 1401, + 1982, + 1401, + 2012, + 1364, + 2012 + ], + "score": 0.862 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.657 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.454 + }, + { + "category_id": 3, + "poly": [ + 887, + 1614, + 1305, + 1614, + 1305, + 1899, + 887, + 1899 + ], + "score": 0.259 + }, + { + "category_id": 3, + "poly": [ + 888, + 1615, + 1303, + 1615, + 1303, + 1900, + 888, + 1900 + ], + "score": 0.249 + }, + { + "category_id": 13, + "poly": [ + 491, + 943, + 539, + 943, + 539, + 974, + 491, + 974 + ], + "score": 0.94, + "latex": "W _ { T } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 637, + 228, + 902, + 228, + 902, + 264, + 637, + 264 + ], + "score": 0.93, + "latex": "G = \\mathbb { R } ^ { 3 } \\rtimes \\mathrm { S O ( 3 ) } \\times S _ { n }" + }, + { + "category_id": 13, + "poly": [ + 1148, + 1017, + 1354, + 1017, + 1354, + 1052, + 1148, + 1052 + ], + "score": 0.93, + "latex": "[ D ] = \\{ 1 , \\dots , D \\}" + }, + { + "category_id": 13, + "poly": [ + 866, + 1018, + 1052, + 1018, + 1052, + 1056, + 866, + 1056 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\mathbb { N } _ { + } ^ { * } = \\bigcup _ { r \\in \\mathbb { N } } \\mathbb { N } _ { + } ^ { r } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 643, + 1018, + 815, + 1018, + 815, + 1052, + 643, + 1052 + ], + "score": 0.93, + "latex": " { \\mathbb { N } } _ { + } = { \\mathbb { N } } \\cup \\lbrace 0 \\rbrace" + }, + { + "category_id": 13, + "poly": [ + 297, + 1575, + 571, + 1575, + 571, + 1610, + 297, + 1610 + ], + "score": 0.92, + "latex": "\\begin{array} { c c l } { W _ { \\mathrm { f e a t } } ^ { n } } & { = } & { \\bigoplus _ { i = 1 } ^ { n } W _ { \\mathrm { f e a t } } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 417, + 1949, + 1280, + 1949, + 1280, + 2044, + 417, + 2044 + ], + "score": 0.92, + "latex": "\\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } , \\mathcal { F } _ { \\mathrm { p o o l } } ) = \\{ f | f ( X ) = \\sum _ { c = 1 } ^ { C } \\hat { \\Lambda } _ { c } ( g _ { c } ( X ) ) , \\ : ( \\Lambda _ { c } , g _ { c } ) \\in \\mathcal { F } _ { \\mathrm { p o o l } } \\times \\mathcal { F } _ { \\mathrm { f e a t } } \\} ." + }, + { + "category_id": 13, + "poly": [ + 843, + 386, + 908, + 386, + 908, + 420, + 843, + 420 + ], + "score": 0.92, + "latex": "\\rho _ { 2 } ( g )" + }, + { + "category_id": 13, + "poly": [ + 539, + 911, + 645, + 911, + 645, + 942, + 539, + 942 + ], + "score": 0.92, + "latex": "W _ { T } = \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 1016, + 879, + 1133, + 879, + 1133, + 911, + 1016, + 911 + ], + "score": 0.92, + "latex": "W _ { T } = \\mathbb { R } ^ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 531, + 1789, + 742, + 1789, + 742, + 1828, + 531, + 1828 + ], + "score": 0.92, + "latex": "\\hat { \\Lambda } : W _ { \\mathrm { f e a t } } ^ { n } \\to W _ { T } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 1189, + 387, + 1372, + 387, + 1372, + 419, + 1189, + 419 + ], + "score": 0.92, + "latex": "f : W _ { 1 } \\to W _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 920, + 482, + 985, + 482, + 985, + 515, + 920, + 515 + ], + "score": 0.92, + "latex": "\\rho _ { 2 } ( g )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1758, + 451, + 1758, + 451, + 1789, + 298, + 1789 + ], + "score": 0.91, + "latex": "W _ { \\mathrm { f e a t } } \\to W _ { T }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1051, + 516, + 1051, + 516, + 1087, + 297, + 1087 + ], + "score": 0.91, + "latex": "[ D ] _ { 0 } = \\{ 0 , \\dots , D \\}" + }, + { + "category_id": 13, + "poly": [ + 297, + 386, + 362, + 386, + 362, + 421, + 297, + 421 + ], + "score": 0.91, + "latex": "\\bar { \\rho _ { 1 } ( g ) }" + }, + { + "category_id": 13, + "poly": [ + 372, + 308, + 451, + 308, + 451, + 338, + 372, + 338 + ], + "score": 0.91, + "latex": "t \\in \\mathbb { R } ^ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 685, + 1757, + 822, + 1757, + 822, + 1791, + 685, + 1791 + ], + "score": 0.91, + "latex": "\\mathrm { S O } ( 3 ) \\times S _ { n }" + }, + { + "category_id": 13, + "poly": [ + 749, + 1636, + 816, + 1636, + 816, + 1670, + 749, + 1670 + ], + "score": 0.91, + "latex": "\\mathcal { F } _ { \\mathrm { p o o l } }" + }, + { + "category_id": 13, + "poly": [ + 1270, + 730, + 1318, + 730, + 1318, + 762, + 1270, + 762 + ], + "score": 0.9, + "latex": "W _ { T } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 710, + 972, + 756, + 972, + 756, + 1002, + 710, + 1002 + ], + "score": 0.9, + "latex": "W _ { T }" + }, + { + "category_id": 13, + "poly": [ + 677, + 1575, + 746, + 1575, + 746, + 1607, + 677, + 1607 + ], + "score": 0.9, + "latex": "W _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 1321, + 482, + 1393, + 482, + 1393, + 514, + 1321, + 514 + ], + "score": 0.9, + "latex": "g \\in G" + }, + { + "category_id": 13, + "poly": [ + 525, + 1886, + 567, + 1886, + 567, + 1916, + 525, + 1916 + ], + "score": 0.9, + "latex": "\\mathcal { F } _ { C }" + }, + { + "category_id": 13, + "poly": [ + 1086, + 790, + 1134, + 790, + 1134, + 823, + 1086, + 823 + ], + "score": 0.9, + "latex": "W _ { T } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 1172, + 357, + 1214, + 357, + 1214, + 387, + 1172, + 387 + ], + "score": 0.9, + "latex": "W _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 465, + 387, + 506, + 387, + 506, + 418, + 465, + 418 + ], + "score": 0.9, + "latex": "W _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 962, + 1543, + 1030, + 1543, + 1030, + 1573, + 962, + 1573 + ], + "score": 0.89, + "latex": "\\mathbb { R } ^ { 3 \\times n }" + }, + { + "category_id": 13, + "poly": [ + 1035, + 448, + 1108, + 448, + 1108, + 480, + 1035, + 480 + ], + "score": 0.89, + "latex": "g \\in G" + }, + { + "category_id": 13, + "poly": [ + 353, + 1697, + 422, + 1697, + 422, + 1727, + 353, + 1727 + ], + "score": 0.89, + "latex": "W _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1728, + 344, + 1728, + 344, + 1758, + 298, + 1758 + ], + "score": 0.89, + "latex": "W _ { T }" + }, + { + "category_id": 13, + "poly": [ + 941, + 228, + 1007, + 228, + 1007, + 259, + 941, + 259 + ], + "score": 0.89, + "latex": "\\mathbb { R } ^ { 3 \\times n }" + }, + { + "category_id": 13, + "poly": [ + 298, + 759, + 344, + 759, + 344, + 789, + 298, + 789 + ], + "score": 0.89, + "latex": "W _ { T }" + }, + { + "category_id": 13, + "poly": [ + 1308, + 1286, + 1393, + 1286, + 1393, + 1314, + 1308, + 1314 + ], + "score": 0.89, + "latex": "n \\gg 3" + }, + { + "category_id": 13, + "poly": [ + 835, + 1515, + 897, + 1515, + 897, + 1545, + 835, + 1545 + ], + "score": 0.88, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } }" + }, + { + "category_id": 14, + "poly": [ + 644, + 267, + 1052, + 267, + 1052, + 306, + 644, + 306 + ], + "score": 0.86, + "latex": "\\rho _ { G } ( t , R , P ) ( X ) = R ( X - t 1 _ { n } ^ { T } ) P ^ { T } ," + }, + { + "category_id": 13, + "poly": [ + 372, + 794, + 408, + 794, + 408, + 822, + 372, + 822 + ], + "score": 0.85, + "latex": "\\rho _ { G }" + }, + { + "category_id": 13, + "poly": [ + 903, + 311, + 928, + 311, + 928, + 337, + 903, + 337 + ], + "score": 0.84, + "latex": "P" + }, + { + "category_id": 13, + "poly": [ + 690, + 311, + 715, + 311, + 715, + 337, + 690, + 337 + ], + "score": 0.84, + "latex": "R" + }, + { + "category_id": 13, + "poly": [ + 298, + 1917, + 322, + 1917, + 322, + 1943, + 298, + 1943 + ], + "score": 0.83, + "latex": "C" + }, + { + "category_id": 13, + "poly": [ + 420, + 1545, + 478, + 1545, + 478, + 1575, + 420, + 1575 + ], + "score": 0.83, + "latex": "f _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 1053, + 358, + 1079, + 358, + 1079, + 384, + 1053, + 384 + ], + "score": 0.83, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 340, + 1856, + 365, + 1856, + 365, + 1883, + 340, + 1883 + ], + "score": 0.81, + "latex": "C" + }, + { + "category_id": 13, + "poly": [ + 1024, + 790, + 1049, + 790, + 1049, + 818, + 1024, + 818 + ], + "score": 0.81, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 581, + 851, + 606, + 851, + 606, + 879, + 581, + 879 + ], + "score": 0.81, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 1243, + 1515, + 1268, + 1515, + 1268, + 1543, + 1243, + 1543 + ], + "score": 0.78, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 376, + 1825, + 399, + 1825, + 399, + 1851, + 376, + 1851 + ], + "score": 0.77, + "latex": "\\Lambda" + }, + { + "category_id": 13, + "poly": [ + 581, + 446, + 985, + 446, + 985, + 482, + 581, + 482 + ], + "score": 0.59, + "latex": "f ( \\rho _ { 1 } ( g ) w ) = \\rho _ { 2 } ( g ) f ( w ) , \\forall w \\in W _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 785, + 1728, + 808, + 1728, + 808, + 1755, + 785, + 1755 + ], + "score": 0.53, + "latex": "\\Lambda" + }, + { + "category_id": 14, + "poly": [ + 581, + 445, + 985, + 445, + 985, + 482, + 581, + 482 + ], + "score": 0.51, + "latex": "f ( \\rho _ { 1 } ( g ) w ) = \\rho _ { 2 } ( g ) f ( w ) , \\forall w \\in W _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 473, + 1727, + 548, + 1727, + 548, + 1761, + 473, + 1761 + ], + "score": 0.26, + "latex": "\\mathrm { S O ( 3 ) }" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 815.0, + 72.0, + 815.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1165.0, + 762.0, + 1165.0, + 762.0, + 1201.0, + 296.0, + 1201.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 1619.0, + 957.0, + 1619.0, + 957.0, + 1657.0, + 891.0, + 1657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 1610.0, + 1127.0, + 1610.0, + 1127.0, + 1644.0, + 1054.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1201.0, + 1618.0, + 1282.0, + 1618.0, + 1282.0, + 1656.0, + 1201.0, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1015.0, + 1642.0, + 1164.0, + 1642.0, + 1164.0, + 1672.0, + 1015.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 960.0, + 1701.0, + 1014.0, + 1701.0, + 1014.0, + 1738.0, + 960.0, + 1738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1164.0, + 1709.0, + 1185.0, + 1709.0, + 1185.0, + 1732.0, + 1164.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 1858.0, + 1124.0, + 1858.0, + 1124.0, + 1907.0, + 1056.0, + 1907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1223.0, + 1863.0, + 1271.0, + 1863.0, + 1271.0, + 1901.0, + 1223.0, + 1901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 1757.5, + 1244.0, + 1757.5, + 1244.0, + 1765.5, + 1236.0, + 1765.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 1862.0, + 960.0, + 1862.0, + 960.0, + 1893.0, + 896.0, + 1893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 1619.0, + 956.0, + 1619.0, + 956.0, + 1657.0, + 890.0, + 1657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1610.0, + 1128.0, + 1610.0, + 1128.0, + 1644.0, + 1053.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1201.0, + 1618.0, + 1282.0, + 1618.0, + 1282.0, + 1656.0, + 1201.0, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1015.0, + 1642.0, + 1165.0, + 1642.0, + 1165.0, + 1672.0, + 1015.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1162.0, + 1707.0, + 1185.0, + 1707.0, + 1185.0, + 1733.0, + 1162.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 1858.0, + 1125.0, + 1858.0, + 1125.0, + 1908.0, + 1056.0, + 1908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1223.0, + 1863.0, + 1272.0, + 1863.0, + 1272.0, + 1902.0, + 1223.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 967.0, + 1706.0, + 1007.0, + 1706.0, + 1007.0, + 1738.5, + 967.0, + 1738.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 1862.0, + 960.0, + 1862.0, + 960.0, + 1893.0, + 896.0, + 1893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 725.0, + 1269.0, + 725.0, + 1269.0, + 766.0, + 292.0, + 766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1319.0, + 725.0, + 1407.0, + 725.0, + 1407.0, + 766.0, + 1319.0, + 766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 759.0, + 1403.0, + 759.0, + 1403.0, + 792.0, + 345.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 787.0, + 371.0, + 787.0, + 371.0, + 828.0, + 292.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 409.0, + 787.0, + 1023.0, + 787.0, + 1023.0, + 828.0, + 409.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 787.0, + 1085.0, + 787.0, + 1085.0, + 828.0, + 1050.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1135.0, + 787.0, + 1406.0, + 787.0, + 1406.0, + 828.0, + 1135.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 818.0, + 1405.0, + 818.0, + 1405.0, + 855.0, + 292.0, + 855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 851.0, + 580.0, + 851.0, + 580.0, + 884.0, + 294.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 607.0, + 851.0, + 1405.0, + 851.0, + 1405.0, + 884.0, + 607.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 879.0, + 1015.0, + 879.0, + 1015.0, + 915.0, + 295.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1134.0, + 879.0, + 1403.0, + 879.0, + 1403.0, + 915.0, + 1134.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 910.0, + 538.0, + 910.0, + 538.0, + 947.0, + 295.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 646.0, + 910.0, + 1405.0, + 910.0, + 1405.0, + 947.0, + 646.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 939.0, + 490.0, + 939.0, + 490.0, + 979.0, + 292.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 939.0, + 1405.0, + 939.0, + 1405.0, + 979.0, + 540.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 973.0, + 709.0, + 973.0, + 709.0, + 1006.0, + 296.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 973.0, + 765.0, + 973.0, + 765.0, + 1006.0, + 757.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1220.0, + 1406.0, + 1220.0, + 1406.0, + 1261.0, + 294.0, + 1261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1253.0, + 1404.0, + 1253.0, + 1404.0, + 1288.0, + 294.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1284.0, + 1307.0, + 1284.0, + 1307.0, + 1319.0, + 295.0, + 1319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 1284.0, + 1405.0, + 1284.0, + 1405.0, + 1319.0, + 1394.0, + 1319.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1315.0, + 1404.0, + 1315.0, + 1404.0, + 1348.0, + 293.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1344.0, + 1405.0, + 1344.0, + 1405.0, + 1381.0, + 292.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1374.0, + 1405.0, + 1374.0, + 1405.0, + 1413.0, + 294.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1407.0, + 671.0, + 1407.0, + 671.0, + 1441.0, + 293.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 530.0, + 1404.0, + 530.0, + 1404.0, + 562.0, + 296.0, + 562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 559.0, + 1406.0, + 559.0, + 1406.0, + 595.0, + 294.0, + 595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 588.0, + 1406.0, + 588.0, + 1406.0, + 625.0, + 292.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 623.0, + 1405.0, + 623.0, + 1405.0, + 654.0, + 296.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 649.0, + 1407.0, + 649.0, + 1407.0, + 686.0, + 291.0, + 686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 685.0, + 455.0, + 685.0, + 455.0, + 714.0, + 293.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1604.0, + 825.0, + 1604.0, + 825.0, + 1639.0, + 295.0, + 1639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1630.0, + 748.0, + 1630.0, + 748.0, + 1674.0, + 293.0, + 1674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 817.0, + 1630.0, + 829.0, + 1630.0, + 829.0, + 1674.0, + 817.0, + 1674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1664.0, + 826.0, + 1664.0, + 826.0, + 1700.0, + 295.0, + 1700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1695.0, + 352.0, + 1695.0, + 352.0, + 1731.0, + 294.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 1695.0, + 826.0, + 1695.0, + 826.0, + 1731.0, + 423.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1726.0, + 472.0, + 1726.0, + 472.0, + 1760.0, + 345.0, + 1760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 549.0, + 1726.0, + 784.0, + 1726.0, + 784.0, + 1760.0, + 549.0, + 1760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 809.0, + 1726.0, + 826.0, + 1726.0, + 826.0, + 1760.0, + 809.0, + 1760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 1756.0, + 684.0, + 1756.0, + 684.0, + 1793.0, + 452.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1790.0, + 530.0, + 1790.0, + 530.0, + 1834.0, + 292.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 1790.0, + 830.0, + 1790.0, + 830.0, + 1834.0, + 743.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1821.0, + 375.0, + 1821.0, + 375.0, + 1858.0, + 294.0, + 1858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 1821.0, + 827.0, + 1821.0, + 827.0, + 1858.0, + 400.0, + 1858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1853.0, + 339.0, + 1853.0, + 339.0, + 1889.0, + 293.0, + 1889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 366.0, + 1853.0, + 828.0, + 1853.0, + 828.0, + 1889.0, + 366.0, + 1889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1880.0, + 524.0, + 1880.0, + 524.0, + 1921.0, + 293.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 1880.0, + 828.0, + 1880.0, + 828.0, + 1921.0, + 568.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1914.0, + 795.0, + 1914.0, + 795.0, + 1949.0, + 323.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 356.0, + 1052.0, + 356.0, + 1052.0, + 390.0, + 294.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1080.0, + 356.0, + 1171.0, + 356.0, + 1171.0, + 390.0, + 1080.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 356.0, + 1402.0, + 356.0, + 1402.0, + 390.0, + 1215.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 386.0, + 296.0, + 386.0, + 296.0, + 423.0, + 293.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 386.0, + 464.0, + 386.0, + 464.0, + 423.0, + 363.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 507.0, + 386.0, + 842.0, + 386.0, + 842.0, + 423.0, + 507.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 909.0, + 386.0, + 1188.0, + 386.0, + 1188.0, + 423.0, + 909.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1373.0, + 386.0, + 1405.0, + 386.0, + 1405.0, + 423.0, + 1373.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 419.0, + 459.0, + 419.0, + 459.0, + 448.0, + 292.0, + 448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1448.0, + 1406.0, + 1448.0, + 1406.0, + 1487.0, + 294.0, + 1487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1480.0, + 1404.0, + 1480.0, + 1404.0, + 1519.0, + 291.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1511.0, + 834.0, + 1511.0, + 834.0, + 1552.0, + 290.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 898.0, + 1511.0, + 1242.0, + 1511.0, + 1242.0, + 1552.0, + 898.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1269.0, + 1511.0, + 1407.0, + 1511.0, + 1407.0, + 1552.0, + 1269.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1537.0, + 419.0, + 1537.0, + 419.0, + 1581.0, + 289.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 1537.0, + 961.0, + 1537.0, + 961.0, + 1581.0, + 479.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 1537.0, + 1409.0, + 1537.0, + 1409.0, + 1581.0, + 1031.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 1561.0, + 296.0, + 1561.0, + 296.0, + 1623.0, + 286.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 1561.0, + 676.0, + 1561.0, + 676.0, + 1623.0, + 572.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 1561.0, + 1411.0, + 1561.0, + 1411.0, + 1623.0, + 747.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1010.0, + 642.0, + 1010.0, + 642.0, + 1058.0, + 291.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 816.0, + 1010.0, + 865.0, + 1010.0, + 865.0, + 1058.0, + 816.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1010.0, + 1147.0, + 1010.0, + 1147.0, + 1058.0, + 1053.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1355.0, + 1010.0, + 1406.0, + 1010.0, + 1406.0, + 1058.0, + 1355.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1050.0, + 527.0, + 1050.0, + 527.0, + 1089.0, + 517.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 224.0, + 636.0, + 224.0, + 636.0, + 268.0, + 294.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 903.0, + 224.0, + 940.0, + 224.0, + 940.0, + 268.0, + 903.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1008.0, + 224.0, + 1054.0, + 224.0, + 1054.0, + 268.0, + 1008.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 305.0, + 371.0, + 305.0, + 371.0, + 345.0, + 296.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 305.0, + 689.0, + 305.0, + 689.0, + 345.0, + 452.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 305.0, + 902.0, + 305.0, + 902.0, + 345.0, + 716.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 929.0, + 305.0, + 1201.0, + 305.0, + 1201.0, + 345.0, + 929.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1095.0, + 1106.0, + 1095.0, + 1106.0, + 1137.0, + 294.0, + 1137.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 476.0, + 919.0, + 476.0, + 919.0, + 519.0, + 296.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 476.0, + 1320.0, + 476.0, + 1320.0, + 519.0, + 986.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 476.0, + 1405.0, + 476.0, + 1405.0, + 519.0, + 1394.0, + 519.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 423, + 1404, + 423, + 1404, + 763, + 298, + 763 + ], + "score": 0.985 + }, + { + "category_id": 1, + "poly": [ + 298, + 1172, + 1404, + 1172, + 1404, + 1297, + 298, + 1297 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 298, + 1308, + 1404, + 1308, + 1404, + 1403, + 298, + 1403 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 297, + 1502, + 1404, + 1502, + 1404, + 1598, + 297, + 1598 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 297, + 771, + 1405, + 771, + 1405, + 867, + 297, + 867 + ], + "score": 0.967 + }, + { + "category_id": 8, + "poly": [ + 705, + 890, + 990, + 890, + 990, + 979, + 705, + 979 + ], + "score": 0.96 + }, + { + "category_id": 8, + "poly": [ + 706, + 1744, + 990, + 1744, + 990, + 1836, + 706, + 1836 + ], + "score": 0.959 + }, + { + "category_id": 1, + "poly": [ + 297, + 228, + 1400, + 228, + 1400, + 325, + 297, + 325 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 299, + 1096, + 1400, + 1096, + 1400, + 1160, + 299, + 1160 + ], + "score": 0.952 + }, + { + "category_id": 1, + "poly": [ + 295, + 1658, + 1402, + 1658, + 1402, + 1722, + 295, + 1722 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 294, + 1002, + 1404, + 1002, + 1404, + 1070, + 294, + 1070 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 294, + 1852, + 1404, + 1852, + 1404, + 1918, + 294, + 1918 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 298, + 1429, + 1402, + 1429, + 1402, + 1492, + 298, + 1492 + ], + "score": 0.944 + }, + { + "category_id": 1, + "poly": [ + 294, + 1930, + 1400, + 1930, + 1400, + 1997, + 294, + 1997 + ], + "score": 0.94 + }, + { + "category_id": 1, + "poly": [ + 295, + 333, + 1400, + 333, + 1400, + 399, + 295, + 399 + ], + "score": 0.935 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 813, + 75, + 813, + 105, + 299, + 105 + ], + "score": 0.907 + }, + { + "category_id": 9, + "poly": [ + 1366, + 918, + 1400, + 918, + 1400, + 949, + 1366, + 949 + ], + "score": 0.893 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1774, + 1400, + 1774, + 1400, + 1803, + 1366, + 1803 + ], + "score": 0.89 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2111, + 841, + 2111 + ], + "score": 0.783 + }, + { + "category_id": 14, + "poly": [ + 706, + 886, + 990, + 886, + 990, + 981, + 706, + 981 + ], + "score": 0.95, + "latex": "p ( X ) = \\sum _ { k = 1 } ^ { K } \\hat { \\Lambda } _ { k } ( f _ { k } ( X ) ) ," + }, + { + "category_id": 14, + "poly": [ + 706, + 1740, + 991, + 1740, + 991, + 1837, + 706, + 1837 + ], + "score": 0.95, + "latex": "p ( X ) = \\sum _ { k = 1 } ^ { K } \\hat { \\Lambda } _ { k } ( f _ { k } ( X ) ) ," + }, + { + "category_id": 13, + "poly": [ + 912, + 1001, + 1117, + 1001, + 1117, + 1041, + 912, + 1041 + ], + "score": 0.94, + "latex": "{ \\hat { \\boldsymbol { \\Lambda } } } _ { k } : { \\cal W } _ { \\mathrm { f e a t } } ^ { n } \\mathbb { R } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 750, + 1506, + 817, + 1506, + 817, + 1538, + 750, + 1538 + ], + "score": 0.93, + "latex": "\\mathcal { F } _ { \\mathrm { p o o l } }" + }, + { + "category_id": 13, + "poly": [ + 524, + 1536, + 665, + 1536, + 665, + 1568, + 524, + 1568 + ], + "score": 0.92, + "latex": "C \\geq C ( D )" + }, + { + "category_id": 13, + "poly": [ + 408, + 290, + 586, + 290, + 586, + 326, + 408, + 326 + ], + "score": 0.92, + "latex": "\\mathcal { C } _ { G } \\mathrm { ( } \\mathbb { R } ^ { 3 \\times n } , W _ { T } ^ { n } \\mathrm { ) }" + }, + { + "category_id": 13, + "poly": [ + 934, + 1660, + 1160, + 1660, + 1160, + 1692, + 934, + 1692 + ], + "score": 0.92, + "latex": "f _ { 1 } , \\dots , f _ { K } \\in { \\mathcal { F } } _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 1167, + 291, + 1349, + 291, + 1349, + 326, + 1167, + 326 + ], + "score": 0.92, + "latex": "\\mathcal { P } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , \\bar { W } _ { T } ^ { n } )" + }, + { + "category_id": 13, + "poly": [ + 490, + 701, + 550, + 701, + 550, + 731, + 490, + 731 + ], + "score": 0.92, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 477, + 487, + 672, + 487, + 672, + 520, + 477, + 520 + ], + "score": 0.91, + "latex": "\\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } , \\mathcal { F } _ { \\mathrm { p o o l } } )" + }, + { + "category_id": 13, + "poly": [ + 372, + 1007, + 564, + 1007, + 564, + 1038, + 372, + 1038 + ], + "score": 0.91, + "latex": "\\Lambda _ { k } : W _ { \\mathrm { f e a t } } \\to \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 892, + 1535, + 1085, + 1535, + 1085, + 1570, + 892, + 1570 + ], + "score": 0.91, + "latex": "\\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } , \\mathcal { F } _ { \\mathrm { p o o l } } )" + }, + { + "category_id": 13, + "poly": [ + 966, + 609, + 1033, + 609, + 1033, + 643, + 966, + 643 + ], + "score": 0.91, + "latex": "W _ { \\mathrm { f e a t } } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 326, + 1886, + 374, + 1886, + 374, + 1919, + 326, + 1919 + ], + "score": 0.91, + "latex": "W _ { T } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 625, + 1176, + 692, + 1176, + 692, + 1208, + 625, + 1208 + ], + "score": 0.91, + "latex": "\\mathcal { F } _ { \\mathrm { p o o l } }" + }, + { + "category_id": 13, + "poly": [ + 1216, + 804, + 1399, + 804, + 1399, + 836, + 1216, + 836 + ], + "score": 0.9, + "latex": "p : \\mathbb { R } ^ { 3 \\times n } \\mathbb { R } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 1305, + 1504, + 1404, + 1504, + 1404, + 1538, + 1305, + 1538 + ], + "score": 0.9, + "latex": "C ( D ) \\in" + }, + { + "category_id": 13, + "poly": [ + 966, + 1310, + 1032, + 1310, + 1032, + 1343, + 966, + 1343 + ], + "score": 0.9, + "latex": "\\mathcal { F } _ { \\mathrm { p o o l } }" + }, + { + "category_id": 13, + "poly": [ + 1063, + 700, + 1129, + 700, + 1129, + 734, + 1063, + 734 + ], + "score": 0.9, + "latex": "W _ { \\mathrm { f e a t } } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 666, + 774, + 769, + 774, + 769, + 806, + 666, + 806 + ], + "score": 0.9, + "latex": "D \\in { \\mathbb { N } } _ { + }" + }, + { + "category_id": 13, + "poly": [ + 471, + 1506, + 531, + 1506, + 531, + 1536, + 471, + 1536 + ], + "score": 0.9, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 622, + 1341, + 690, + 1341, + 690, + 1372, + 622, + 1372 + ], + "score": 0.9, + "latex": "W _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 1093, + 1934, + 1128, + 1934, + 1128, + 1963, + 1093, + 1963 + ], + "score": 0.9, + "latex": "\\Lambda _ { k }" + }, + { + "category_id": 13, + "poly": [ + 680, + 806, + 896, + 806, + 896, + 837, + 680, + 837 + ], + "score": 0.89, + "latex": "f _ { 1 } , \\dots , f _ { K } \\in \\mathcal { F } _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 1053, + 771, + 1249, + 771, + 1249, + 807, + 1053, + 807 + ], + "score": 0.89, + "latex": "\\mathcal { C } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , W _ { \\mathrm { f e a t } } ^ { n } )" + }, + { + "category_id": 13, + "poly": [ + 1114, + 1207, + 1150, + 1207, + 1150, + 1236, + 1114, + 1236 + ], + "score": 0.89, + "latex": "\\Lambda _ { k }" + }, + { + "category_id": 13, + "poly": [ + 821, + 775, + 880, + 775, + 880, + 805, + 821, + 805 + ], + "score": 0.89, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 745, + 1341, + 791, + 1341, + 791, + 1372, + 745, + 1372 + ], + "score": 0.89, + "latex": "W _ { T }" + }, + { + "category_id": 13, + "poly": [ + 645, + 1038, + 679, + 1038, + 679, + 1067, + 645, + 1067 + ], + "score": 0.88, + "latex": "\\Lambda _ { k }" + }, + { + "category_id": 13, + "poly": [ + 921, + 333, + 1099, + 333, + 1099, + 367, + 921, + 367 + ], + "score": 0.88, + "latex": "\\mathcal { C } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , W _ { T } ^ { n } )" + }, + { + "category_id": 13, + "poly": [ + 298, + 805, + 358, + 805, + 358, + 836, + 298, + 836 + ], + "score": 0.88, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 972, + 1856, + 1009, + 1856, + 1009, + 1882, + 972, + 1882 + ], + "score": 0.87, + "latex": "\\mathbb { R } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 673, + 1129, + 699, + 1129, + 699, + 1155, + 673, + 1155 + ], + "score": 0.83, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 1113, + 1237, + 1138, + 1237, + 1138, + 1263, + 1113, + 1263 + ], + "score": 0.82, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 740, + 1099, + 767, + 1099, + 767, + 1125, + 740, + 1125 + ], + "score": 0.82, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 1244, + 263, + 1268, + 263, + 1268, + 289, + 1244, + 289 + ], + "score": 0.82, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 862, + 293, + 887, + 293, + 887, + 319, + 862, + 319 + ], + "score": 0.82, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 888, + 427, + 913, + 427, + 913, + 454, + 888, + 454 + ], + "score": 0.81, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 715, + 458, + 740, + 458, + 740, + 485, + 715, + 485 + ], + "score": 0.81, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 520, + 1661, + 547, + 1661, + 547, + 1687, + 520, + 1687 + ], + "score": 0.81, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 619, + 336, + 644, + 336, + 644, + 364, + 619, + 364 + ], + "score": 0.8, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 777, + 1862, + 794, + 1862, + 794, + 1887, + 777, + 1887 + ], + "score": 0.8, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 593, + 367, + 618, + 367, + 618, + 393, + 593, + 393 + ], + "score": 0.8, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 548, + 1567, + 603, + 1567, + 603, + 1596, + 548, + 1596 + ], + "score": 0.8, + "latex": "\\le D" + }, + { + "category_id": 13, + "poly": [ + 888, + 1267, + 913, + 1267, + 913, + 1293, + 888, + 1293 + ], + "score": 0.79, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 1137, + 458, + 1165, + 458, + 1165, + 484, + 1137, + 484 + ], + "score": 0.79, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 388, + 805, + 415, + 805, + 415, + 832, + 388, + 832 + ], + "score": 0.79, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 601, + 1241, + 620, + 1241, + 620, + 1267, + 601, + 1267 + ], + "score": 0.79, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 298, + 732, + 324, + 732, + 324, + 758, + 298, + 758 + ], + "score": 0.78, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 929, + 366, + 1110, + 366, + 1110, + 400, + 929, + 400 + ], + "score": 0.78, + "latex": "\\mathcal { P } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , \\bar { W } _ { T } ^ { n } )" + }, + { + "category_id": 13, + "poly": [ + 560, + 1506, + 586, + 1506, + 586, + 1532, + 560, + 1532 + ], + "score": 0.78, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 617, + 1944, + 633, + 1944, + 633, + 1965, + 617, + 1965 + ], + "score": 0.78, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 1239, + 1537, + 1264, + 1537, + 1264, + 1563, + 1239, + 1563 + ], + "score": 0.75, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 406, + 836, + 433, + 836, + 433, + 862, + 406, + 862 + ], + "score": 0.74, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 455, + 775, + 481, + 775, + 481, + 802, + 455, + 802 + ], + "score": 0.71, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 297, + 1536, + 323, + 1536, + 323, + 1564, + 297, + 1564 + ], + "score": 0.55, + "latex": "\\mathbb { N }" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 862.0, + 2086.0, + 862.0, + 2118.0, + 839.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 425.0, + 887.0, + 425.0, + 887.0, + 461.0, + 296.0, + 461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 425.0, + 1404.0, + 425.0, + 1404.0, + 461.0, + 914.0, + 461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 457.0, + 714.0, + 457.0, + 714.0, + 492.0, + 294.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 457.0, + 1136.0, + 457.0, + 1136.0, + 492.0, + 741.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1166.0, + 457.0, + 1404.0, + 457.0, + 1404.0, + 492.0, + 1166.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 484.0, + 476.0, + 484.0, + 476.0, + 523.0, + 292.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 673.0, + 484.0, + 1407.0, + 484.0, + 1407.0, + 523.0, + 673.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 518.0, + 1406.0, + 518.0, + 1406.0, + 553.0, + 294.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 547.0, + 1405.0, + 547.0, + 1405.0, + 584.0, + 293.0, + 584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 579.0, + 1405.0, + 579.0, + 1405.0, + 614.0, + 294.0, + 614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 599.0, + 965.0, + 599.0, + 965.0, + 654.0, + 288.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 599.0, + 1410.0, + 599.0, + 1410.0, + 654.0, + 1034.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 641.0, + 1405.0, + 641.0, + 1405.0, + 672.0, + 296.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 669.0, + 1406.0, + 669.0, + 1406.0, + 703.0, + 293.0, + 703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 697.0, + 489.0, + 697.0, + 489.0, + 735.0, + 291.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 551.0, + 697.0, + 1062.0, + 697.0, + 1062.0, + 735.0, + 551.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 697.0, + 1135.0, + 697.0, + 1135.0, + 735.0, + 1130.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 700.0, + 1404.0, + 700.0, + 1404.0, + 732.0, + 1147.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 725.0, + 297.0, + 725.0, + 297.0, + 772.0, + 294.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 725.0, + 446.0, + 725.0, + 446.0, + 772.0, + 325.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1172.0, + 624.0, + 1172.0, + 624.0, + 1212.0, + 293.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 1172.0, + 1405.0, + 1172.0, + 1405.0, + 1212.0, + 693.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1204.0, + 1113.0, + 1204.0, + 1113.0, + 1240.0, + 295.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1151.0, + 1204.0, + 1404.0, + 1204.0, + 1404.0, + 1240.0, + 1151.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1236.0, + 600.0, + 1236.0, + 600.0, + 1272.0, + 293.0, + 1272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 621.0, + 1236.0, + 1112.0, + 1236.0, + 1112.0, + 1272.0, + 621.0, + 1272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1139.0, + 1236.0, + 1405.0, + 1236.0, + 1405.0, + 1272.0, + 1139.0, + 1272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1266.0, + 887.0, + 1266.0, + 887.0, + 1299.0, + 294.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 1266.0, + 1136.0, + 1266.0, + 1136.0, + 1299.0, + 914.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1308.0, + 965.0, + 1308.0, + 965.0, + 1346.0, + 294.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 1308.0, + 1405.0, + 1308.0, + 1405.0, + 1346.0, + 1033.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1340.0, + 621.0, + 1340.0, + 621.0, + 1377.0, + 293.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 1340.0, + 744.0, + 1340.0, + 744.0, + 1377.0, + 691.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 792.0, + 1340.0, + 1406.0, + 1340.0, + 1406.0, + 1377.0, + 792.0, + 1377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1368.0, + 997.0, + 1368.0, + 997.0, + 1409.0, + 293.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1499.0, + 470.0, + 1499.0, + 470.0, + 1543.0, + 292.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1499.0, + 559.0, + 1499.0, + 559.0, + 1543.0, + 532.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 1499.0, + 749.0, + 1499.0, + 749.0, + 1543.0, + 587.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 818.0, + 1499.0, + 1304.0, + 1499.0, + 1304.0, + 1543.0, + 818.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1530.0, + 296.0, + 1530.0, + 296.0, + 1575.0, + 291.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1530.0, + 523.0, + 1530.0, + 523.0, + 1575.0, + 324.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 666.0, + 1530.0, + 891.0, + 1530.0, + 891.0, + 1575.0, + 666.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 1530.0, + 1238.0, + 1530.0, + 1238.0, + 1575.0, + 1086.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1265.0, + 1530.0, + 1407.0, + 1530.0, + 1407.0, + 1575.0, + 1265.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1566.0, + 547.0, + 1566.0, + 547.0, + 1599.0, + 292.0, + 1599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 604.0, + 1566.0, + 613.0, + 1566.0, + 613.0, + 1599.0, + 604.0, + 1599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 755.0, + 454.0, + 755.0, + 454.0, + 822.0, + 286.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 755.0, + 665.0, + 755.0, + 665.0, + 822.0, + 482.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 770.0, + 755.0, + 820.0, + 755.0, + 820.0, + 822.0, + 770.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 755.0, + 1052.0, + 755.0, + 1052.0, + 822.0, + 881.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1250.0, + 755.0, + 1412.0, + 755.0, + 1412.0, + 822.0, + 1250.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 801.0, + 297.0, + 801.0, + 297.0, + 840.0, + 292.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 801.0, + 387.0, + 801.0, + 387.0, + 840.0, + 359.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 801.0, + 679.0, + 801.0, + 679.0, + 840.0, + 416.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 801.0, + 1215.0, + 801.0, + 1215.0, + 840.0, + 897.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1400.0, + 801.0, + 1406.0, + 801.0, + 1406.0, + 840.0, + 1400.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 832.0, + 405.0, + 832.0, + 405.0, + 870.0, + 295.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 832.0, + 1379.0, + 832.0, + 1379.0, + 870.0, + 434.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 226.0, + 1405.0, + 226.0, + 1405.0, + 267.0, + 295.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 262.0, + 1243.0, + 262.0, + 1243.0, + 296.0, + 295.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1269.0, + 262.0, + 1405.0, + 262.0, + 1405.0, + 296.0, + 1269.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 283.0, + 407.0, + 283.0, + 407.0, + 332.0, + 291.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 283.0, + 861.0, + 283.0, + 861.0, + 332.0, + 587.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 888.0, + 283.0, + 1166.0, + 283.0, + 1166.0, + 332.0, + 888.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1350.0, + 283.0, + 1362.0, + 283.0, + 1362.0, + 332.0, + 1350.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1096.0, + 739.0, + 1096.0, + 739.0, + 1132.0, + 295.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 1096.0, + 1406.0, + 1096.0, + 1406.0, + 1132.0, + 768.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1125.0, + 672.0, + 1125.0, + 672.0, + 1161.0, + 294.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 1125.0, + 1204.0, + 1125.0, + 1204.0, + 1161.0, + 700.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1657.0, + 519.0, + 1657.0, + 519.0, + 1695.0, + 294.0, + 1695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 1657.0, + 933.0, + 1657.0, + 933.0, + 1695.0, + 548.0, + 1695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1161.0, + 1657.0, + 1405.0, + 1657.0, + 1405.0, + 1695.0, + 1161.0, + 1695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1689.0, + 1291.0, + 1689.0, + 1291.0, + 1725.0, + 297.0, + 1725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 994.0, + 371.0, + 994.0, + 371.0, + 1056.0, + 288.0, + 1056.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 565.0, + 994.0, + 911.0, + 994.0, + 911.0, + 1056.0, + 565.0, + 1056.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1118.0, + 994.0, + 1407.0, + 994.0, + 1407.0, + 1056.0, + 1118.0, + 1056.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1036.0, + 644.0, + 1036.0, + 644.0, + 1073.0, + 296.0, + 1073.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 1036.0, + 689.0, + 1036.0, + 689.0, + 1073.0, + 680.0, + 1073.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1850.0, + 776.0, + 1850.0, + 776.0, + 1890.0, + 294.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 1850.0, + 971.0, + 1850.0, + 971.0, + 1890.0, + 795.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1010.0, + 1850.0, + 1405.0, + 1850.0, + 1405.0, + 1890.0, + 1010.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1884.0, + 325.0, + 1884.0, + 325.0, + 1921.0, + 294.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 1884.0, + 470.0, + 1884.0, + 470.0, + 1921.0, + 375.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1425.0, + 1407.0, + 1425.0, + 1407.0, + 1468.0, + 293.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1458.0, + 404.0, + 1458.0, + 404.0, + 1497.0, + 294.0, + 1497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1931.0, + 616.0, + 1931.0, + 616.0, + 1967.0, + 294.0, + 1967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 634.0, + 1931.0, + 1092.0, + 1931.0, + 1092.0, + 1967.0, + 634.0, + 1967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 1931.0, + 1404.0, + 1931.0, + 1404.0, + 1967.0, + 1129.0, + 1967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1961.0, + 829.0, + 1961.0, + 829.0, + 1997.0, + 296.0, + 1997.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1381.0, + 1970.0, + 1398.0, + 1970.0, + 1398.0, + 1987.0, + 1381.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 326.0, + 618.0, + 326.0, + 618.0, + 374.0, + 292.0, + 374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 645.0, + 326.0, + 920.0, + 326.0, + 920.0, + 374.0, + 645.0, + 374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 326.0, + 1406.0, + 326.0, + 1406.0, + 374.0, + 1100.0, + 374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 360.0, + 592.0, + 360.0, + 592.0, + 405.0, + 289.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 619.0, + 360.0, + 928.0, + 360.0, + 928.0, + 405.0, + 619.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1111.0, + 360.0, + 1125.0, + 360.0, + 1125.0, + 405.0, + 1111.0, + 405.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 629, + 1405, + 629, + 1405, + 878, + 296, + 878 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 890, + 1405, + 890, + 1405, + 1076, + 297, + 1076 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 1832, + 1404, + 1832, + 1404, + 1941, + 298, + 1941 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 295, + 1187, + 1401, + 1187, + 1401, + 1311, + 295, + 1311 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 297, + 1356, + 938, + 1356, + 938, + 1515, + 297, + 1515 + ], + "score": 0.967 + }, + { + "category_id": 8, + "poly": [ + 737, + 469, + 960, + 469, + 960, + 541, + 737, + 541 + ], + "score": 0.951 + }, + { + "category_id": 8, + "poly": [ + 751, + 1732, + 945, + 1732, + 945, + 1820, + 751, + 1820 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 298, + 229, + 934, + 229, + 934, + 293, + 298, + 293 + ], + "score": 0.948 + }, + { + "category_id": 8, + "poly": [ + 519, + 1953, + 1176, + 1953, + 1176, + 2042, + 519, + 2042 + ], + "score": 0.947 + }, + { + "category_id": 1, + "poly": [ + 296, + 1653, + 1407, + 1653, + 1407, + 1717, + 296, + 1717 + ], + "score": 0.945 + }, + { + "category_id": 3, + "poly": [ + 970, + 1350, + 1349, + 1350, + 1349, + 1509, + 970, + 1509 + ], + "score": 0.944 + }, + { + "category_id": 8, + "poly": [ + 649, + 1561, + 1048, + 1561, + 1048, + 1598, + 649, + 1598 + ], + "score": 0.929 + }, + { + "category_id": 0, + "poly": [ + 298, + 1121, + 1052, + 1121, + 1052, + 1154, + 298, + 1154 + ], + "score": 0.922 + }, + { + "category_id": 1, + "poly": [ + 297, + 1605, + 535, + 1605, + 535, + 1639, + 297, + 1639 + ], + "score": 0.915 + }, + { + "category_id": 1, + "poly": [ + 295, + 298, + 935, + 298, + 935, + 423, + 295, + 423 + ], + "score": 0.913 + }, + { + "category_id": 0, + "poly": [ + 300, + 574, + 832, + 574, + 832, + 605, + 300, + 605 + ], + "score": 0.907 + }, + { + "category_id": 1, + "poly": [ + 296, + 1515, + 1304, + 1515, + 1304, + 1549, + 296, + 1549 + ], + "score": 0.903 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 813, + 75, + 813, + 105, + 299, + 105 + ], + "score": 0.901 + }, + { + "category_id": 1, + "poly": [ + 287, + 424, + 1322, + 424, + 1322, + 456, + 287, + 456 + ], + "score": 0.882 + }, + { + "category_id": 9, + "poly": [ + 1365, + 1979, + 1401, + 1979, + 1401, + 2009, + 1365, + 2009 + ], + "score": 0.87 + }, + { + "category_id": 9, + "poly": [ + 1365, + 1563, + 1400, + 1563, + 1400, + 1593, + 1365, + 1593 + ], + "score": 0.869 + }, + { + "category_id": 3, + "poly": [ + 979, + 190, + 1339, + 190, + 1339, + 367, + 979, + 367 + ], + "score": 0.786 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 859, + 2088, + 859, + 2112, + 840, + 2112 + ], + "score": 0.777 + }, + { + "category_id": 3, + "poly": [ + 979, + 191, + 1339, + 191, + 1339, + 367, + 979, + 367 + ], + "score": 0.432 + }, + { + "category_id": 14, + "poly": [ + 736, + 467, + 962, + 467, + 962, + 542, + 736, + 542 + ], + "score": 0.94, + "latex": "{ \\mathcal { F } } = \\bigcup _ { D \\in \\mathbb { N } } { \\mathcal { F } } _ { C ( D ) , D } ." + }, + { + "category_id": 13, + "poly": [ + 846, + 1833, + 1039, + 1833, + 1039, + 1867, + 846, + 1867 + ], + "score": 0.94, + "latex": "\\vec { r } = \\left( r _ { 1 } , \\ldots , r _ { K } \\right)" + }, + { + "category_id": 14, + "poly": [ + 751, + 1727, + 949, + 1727, + 949, + 1822, + 751, + 1822 + ], + "score": 0.94, + "latex": "W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } } = \\bigoplus _ { T = 0 } ^ { D } \\mathcal { T } _ { T } ." + }, + { + "category_id": 14, + "poly": [ + 520, + 1948, + 1180, + 1948, + 1180, + 2044, + 520, + 2044 + ], + "score": 0.93, + "latex": "Q _ { j } ^ { ( \\vec { r } ) } ( X ) = \\sum _ { i _ { 2 } , \\ldots , i _ { K } = 1 } ^ { n } x _ { j } ^ { \\otimes r _ { 1 } } \\otimes x _ { i _ { 2 } } ^ { \\otimes r _ { 2 } } \\otimes x _ { i _ { 3 } } ^ { \\otimes r _ { 3 } } \\otimes \\ldots \\otimes x _ { i _ { K } } ^ { \\otimes r _ { K } } ." + }, + { + "category_id": 13, + "poly": [ + 780, + 1869, + 1004, + 1869, + 1004, + 1905, + 780, + 1905 + ], + "score": 0.93, + "latex": "Q ^ { ( \\vec { r } ) } : \\mathbb { R } ^ { 3 \\times n } \\mathcal { T } _ { T } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 500, + 1870, + 615, + 1870, + 615, + 1905, + 500, + 1905 + ], + "score": 0.93, + "latex": "T = \\| r \\| _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 359, + 1905, + 440, + 1905, + 440, + 1940, + 359, + 1940 + ], + "score": 0.92, + "latex": "j \\in [ n ]" + }, + { + "category_id": 13, + "poly": [ + 675, + 1388, + 789, + 1388, + 789, + 1425, + 675, + 1425 + ], + "score": 0.92, + "latex": "\\mathcal { T } _ { k } = \\mathbb { R } ^ { 3 ^ { k } }" + }, + { + "category_id": 13, + "poly": [ + 707, + 1515, + 881, + 1515, + 881, + 1549, + 707, + 1549 + ], + "score": 0.92, + "latex": "\\bar { i } _ { 1 } , \\ldots , i _ { k } \\in [ n ]" + }, + { + "category_id": 13, + "poly": [ + 1024, + 1865, + 1230, + 1865, + 1230, + 1910, + 1024, + 1910 + ], + "score": 0.92, + "latex": "Q ^ { ( \\vec { r } ) } = ( Q _ { j } ^ { ( \\vec { r } ) } ) _ { j = 1 } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 298, + 422, + 476, + 422, + 476, + 457, + 298, + 457 + ], + "score": 0.92, + "latex": "\\mathcal { C } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , W _ { T } ^ { n } )" + }, + { + "category_id": 13, + "poly": [ + 494, + 1393, + 588, + 1393, + 588, + 1427, + 494, + 1427 + ], + "score": 0.91, + "latex": "k \\in \\mathbb { N } _ { + }" + }, + { + "category_id": 13, + "poly": [ + 561, + 300, + 717, + 300, + 717, + 333, + 561, + 333 + ], + "score": 0.91, + "latex": "{ \\cal C } , { \\cal D } \\in \\mathbb { N } _ { + }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1516, + 419, + 1516, + 419, + 1547, + 297, + 1547 + ], + "score": 0.91, + "latex": "k = 1 , 2 , 3" + }, + { + "category_id": 13, + "poly": [ + 779, + 301, + 845, + 301, + 845, + 335, + 779, + 335 + ], + "score": 0.91, + "latex": "\\mathcal { F } _ { C , D }" + }, + { + "category_id": 13, + "poly": [ + 819, + 1455, + 934, + 1455, + 934, + 1487, + 819, + 1487 + ], + "score": 0.91, + "latex": "\\rho _ { k } ( R ) : =" + }, + { + "category_id": 14, + "poly": [ + 646, + 1560, + 1051, + 1560, + 1051, + 1596, + 646, + 1596 + ], + "score": 0.91, + "latex": "( x _ { 1 } , \\ldots , x _ { n } ) \\mapsto x _ { i _ { 1 } } \\otimes x _ { i _ { 2 } } \\ldots \\otimes x _ { i _ { k } }" + }, + { + "category_id": 13, + "poly": [ + 1048, + 1514, + 1115, + 1514, + 1115, + 1543, + 1048, + 1543 + ], + "score": 0.91, + "latex": "\\mathbb { R } ^ { 3 \\times n }" + }, + { + "category_id": 13, + "poly": [ + 835, + 1685, + 902, + 1685, + 902, + 1716, + 835, + 1716 + ], + "score": 0.9, + "latex": "W _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1484, + 352, + 1484, + 352, + 1514, + 298, + 1514 + ], + "score": 0.89, + "latex": "\\bar { R } ^ { \\otimes k }" + }, + { + "category_id": 13, + "poly": [ + 1147, + 1516, + 1179, + 1516, + 1179, + 1546, + 1147, + 1546 + ], + "score": 0.88, + "latex": "\\mathcal { T } _ { k }" + }, + { + "category_id": 13, + "poly": [ + 335, + 1425, + 367, + 1425, + 367, + 1455, + 335, + 1455 + ], + "score": 0.87, + "latex": "\\mathcal { T } _ { k }" + }, + { + "category_id": 13, + "poly": [ + 1180, + 1839, + 1208, + 1839, + 1208, + 1864, + 1180, + 1864 + ], + "score": 0.83, + "latex": "r _ { k }" + }, + { + "category_id": 13, + "poly": [ + 346, + 1835, + 373, + 1835, + 373, + 1861, + 346, + 1861 + ], + "score": 0.81, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 722, + 785, + 748, + 785, + 748, + 811, + 722, + 811 + ], + "score": 0.81, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 298, + 1686, + 324, + 1686, + 324, + 1712, + 298, + 1712 + ], + "score": 0.8, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 1020, + 633, + 1045, + 633, + 1045, + 659, + 1020, + 659 + ], + "score": 0.8, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 716, + 1456, + 736, + 1456, + 736, + 1482, + 716, + 1482 + ], + "score": 0.79, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 638, + 394, + 663, + 394, + 663, + 420, + 638, + 420 + ], + "score": 0.79, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 414, + 363, + 440, + 363, + 440, + 389, + 414, + 389 + ], + "score": 0.75, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 325, + 1654, + 351, + 1654, + 351, + 1681, + 325, + 1681 + ], + "score": 0.74, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 1264, + 1219, + 1291, + 1219, + 1291, + 1247, + 1264, + 1247 + ], + "score": 0.73, + "latex": "D" + }, + { + "category_id": 15, + "poly": [ + 1104.0, + 1353.0, + 1146.0, + 1353.0, + 1146.0, + 1382.0, + 1104.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1230.0, + 1349.0, + 1274.0, + 1349.0, + 1274.0, + 1379.0, + 1230.0, + 1379.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1027.0, + 1401.0, + 1044.0, + 1401.0, + 1044.0, + 1420.0, + 1027.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1164.0, + 1397.0, + 1190.0, + 1397.0, + 1190.0, + 1437.0, + 1164.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1320.0, + 1410.0, + 1352.0, + 1410.0, + 1352.0, + 1460.0, + 1320.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 1479.0, + 1040.0, + 1479.0, + 1040.0, + 1510.0, + 963.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1306.75, + 1368.5, + 1312.75, + 1368.5, + 1312.75, + 1389.5, + 1306.75, + 1389.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1070.0, + 1477.0, + 1156.0, + 1477.0, + 1156.0, + 1511.5, + 1070.0, + 1511.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1234.0, + 1476.0, + 1323.0, + 1476.0, + 1323.0, + 1513.5, + 1234.0, + 1513.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1118.0, + 1054.0, + 1118.0, + 1054.0, + 1161.0, + 293.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 573.0, + 837.0, + 573.0, + 837.0, + 609.0, + 294.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 255.0, + 1326.0, + 255.0, + 1326.0, + 312.0, + 1002.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 861.0, + 2086.0, + 861.0, + 2118.0, + 839.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1000.0, + 254.0, + 1325.0, + 254.0, + 1325.0, + 316.0, + 1000.0, + 316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 629.0, + 1019.0, + 629.0, + 1019.0, + 667.0, + 291.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 629.0, + 1406.0, + 629.0, + 1406.0, + 667.0, + 1046.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 664.0, + 1403.0, + 664.0, + 1403.0, + 694.0, + 295.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 692.0, + 1406.0, + 692.0, + 1406.0, + 729.0, + 293.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 723.0, + 1405.0, + 723.0, + 1405.0, + 757.0, + 295.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 752.0, + 1407.0, + 752.0, + 1407.0, + 790.0, + 294.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 783.0, + 721.0, + 783.0, + 721.0, + 820.0, + 293.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 749.0, + 783.0, + 1405.0, + 783.0, + 1405.0, + 820.0, + 749.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 814.0, + 1403.0, + 814.0, + 1403.0, + 850.0, + 293.0, + 850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 847.0, + 405.0, + 847.0, + 405.0, + 882.0, + 293.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 889.0, + 1405.0, + 889.0, + 1405.0, + 926.0, + 292.0, + 926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 924.0, + 1405.0, + 924.0, + 1405.0, + 956.0, + 296.0, + 956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 951.0, + 1406.0, + 951.0, + 1406.0, + 988.0, + 294.0, + 988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 982.0, + 1405.0, + 982.0, + 1405.0, + 1017.0, + 295.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1011.0, + 1406.0, + 1011.0, + 1406.0, + 1049.0, + 294.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1042.0, + 1187.0, + 1042.0, + 1187.0, + 1079.0, + 294.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1829.0, + 345.0, + 1829.0, + 345.0, + 1870.0, + 294.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1829.0, + 845.0, + 1829.0, + 845.0, + 1870.0, + 374.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 1829.0, + 1179.0, + 1829.0, + 1179.0, + 1870.0, + 1040.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1209.0, + 1829.0, + 1405.0, + 1829.0, + 1405.0, + 1870.0, + 1209.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 280.0, + 1857.0, + 499.0, + 1857.0, + 499.0, + 1924.0, + 280.0, + 1924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 1857.0, + 779.0, + 1857.0, + 779.0, + 1924.0, + 616.0, + 1924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1005.0, + 1857.0, + 1023.0, + 1857.0, + 1023.0, + 1924.0, + 1005.0, + 1924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1899.0, + 358.0, + 1899.0, + 358.0, + 1946.0, + 293.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 441.0, + 1899.0, + 482.0, + 1899.0, + 482.0, + 1946.0, + 441.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1164.25, + 1859.5, + 1409.25, + 1859.5, + 1409.25, + 1913.5, + 1164.25, + 1913.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1185.0, + 1405.0, + 1185.0, + 1405.0, + 1225.0, + 293.0, + 1225.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1216.0, + 1263.0, + 1216.0, + 1263.0, + 1255.0, + 292.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1292.0, + 1216.0, + 1406.0, + 1216.0, + 1406.0, + 1255.0, + 1292.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1249.0, + 1405.0, + 1249.0, + 1405.0, + 1284.0, + 293.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1281.0, + 1405.0, + 1281.0, + 1405.0, + 1313.0, + 295.0, + 1313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1355.0, + 936.0, + 1355.0, + 936.0, + 1393.0, + 295.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1388.0, + 493.0, + 1388.0, + 493.0, + 1429.0, + 292.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 589.0, + 1388.0, + 674.0, + 1388.0, + 674.0, + 1429.0, + 589.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 790.0, + 1388.0, + 939.0, + 1388.0, + 939.0, + 1429.0, + 790.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1422.0, + 334.0, + 1422.0, + 334.0, + 1460.0, + 293.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 1422.0, + 936.0, + 1422.0, + 936.0, + 1460.0, + 368.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1456.0, + 715.0, + 1456.0, + 715.0, + 1487.0, + 296.0, + 1487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 737.0, + 1456.0, + 818.0, + 1456.0, + 818.0, + 1487.0, + 737.0, + 1487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1481.0, + 297.0, + 1481.0, + 297.0, + 1520.0, + 294.0, + 1520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1481.0, + 938.0, + 1481.0, + 938.0, + 1520.0, + 353.0, + 1520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 227.0, + 937.0, + 227.0, + 937.0, + 265.0, + 296.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 261.0, + 789.0, + 261.0, + 789.0, + 295.0, + 295.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1652.0, + 324.0, + 1652.0, + 324.0, + 1688.0, + 295.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1652.0, + 1405.0, + 1652.0, + 1405.0, + 1688.0, + 352.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1682.0, + 834.0, + 1682.0, + 834.0, + 1721.0, + 325.0, + 1721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 903.0, + 1682.0, + 1089.0, + 1682.0, + 1089.0, + 1721.0, + 903.0, + 1721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1601.0, + 540.0, + 1601.0, + 540.0, + 1644.0, + 293.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 296.0, + 560.0, + 296.0, + 560.0, + 335.0, + 295.0, + 335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 296.0, + 778.0, + 296.0, + 778.0, + 335.0, + 718.0, + 335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 296.0, + 938.0, + 296.0, + 938.0, + 335.0, + 846.0, + 335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 332.0, + 937.0, + 332.0, + 937.0, + 367.0, + 293.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 362.0, + 413.0, + 362.0, + 413.0, + 396.0, + 295.0, + 396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 441.0, + 362.0, + 937.0, + 362.0, + 937.0, + 396.0, + 441.0, + 396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 391.0, + 637.0, + 391.0, + 637.0, + 427.0, + 294.0, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 664.0, + 391.0, + 938.0, + 391.0, + 938.0, + 427.0, + 664.0, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1509.0, + 296.0, + 1509.0, + 296.0, + 1555.0, + 290.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 420.0, + 1509.0, + 706.0, + 1509.0, + 706.0, + 1555.0, + 420.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 882.0, + 1509.0, + 1047.0, + 1509.0, + 1047.0, + 1555.0, + 882.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1116.0, + 1509.0, + 1146.0, + 1509.0, + 1146.0, + 1555.0, + 1116.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 1509.0, + 1308.0, + 1509.0, + 1308.0, + 1555.0, + 1180.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 412.0, + 297.0, + 412.0, + 297.0, + 463.0, + 293.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 412.0, + 1326.0, + 412.0, + 1326.0, + 463.0, + 477.0, + 463.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1368, + 1404, + 1368, + 1404, + 1494, + 296, + 1494 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1910, + 1404, + 1910, + 1404, + 2036, + 298, + 2036 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 297, + 1026, + 1406, + 1026, + 1406, + 1152, + 297, + 1152 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 296, + 225, + 1405, + 225, + 1405, + 395, + 296, + 395 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 299, + 920, + 1404, + 920, + 1404, + 993, + 299, + 993 + ], + "score": 0.959 + }, + { + "category_id": 8, + "poly": [ + 646, + 813, + 1047, + 813, + 1047, + 904, + 646, + 904 + ], + "score": 0.958 + }, + { + "category_id": 1, + "poly": [ + 293, + 720, + 1400, + 720, + 1400, + 793, + 293, + 793 + ], + "score": 0.956 + }, + { + "category_id": 8, + "poly": [ + 587, + 486, + 1112, + 486, + 1112, + 550, + 587, + 550 + ], + "score": 0.954 + }, + { + "category_id": 8, + "poly": [ + 560, + 1272, + 1139, + 1272, + 1139, + 1344, + 560, + 1344 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 295, + 1163, + 1396, + 1163, + 1396, + 1241, + 295, + 1241 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 298, + 567, + 1400, + 567, + 1400, + 634, + 298, + 634 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 298, + 408, + 1398, + 408, + 1398, + 474, + 298, + 474 + ], + "score": 0.946 + }, + { + "category_id": 1, + "poly": [ + 296, + 1562, + 1406, + 1562, + 1406, + 1628, + 296, + 1628 + ], + "score": 0.946 + }, + { + "category_id": 8, + "poly": [ + 627, + 1510, + 1072, + 1510, + 1072, + 1549, + 627, + 1549 + ], + "score": 0.941 + }, + { + "category_id": 8, + "poly": [ + 390, + 1685, + 1306, + 1685, + 1306, + 1748, + 390, + 1748 + ], + "score": 0.933 + }, + { + "category_id": 1, + "poly": [ + 297, + 1840, + 1036, + 1840, + 1036, + 1875, + 297, + 1875 + ], + "score": 0.932 + }, + { + "category_id": 1, + "poly": [ + 299, + 1761, + 591, + 1761, + 591, + 1795, + 299, + 1795 + ], + "score": 0.926 + }, + { + "category_id": 1, + "poly": [ + 294, + 1635, + 832, + 1635, + 832, + 1669, + 294, + 1669 + ], + "score": 0.922 + }, + { + "category_id": 2, + "poly": [ + 299, + 74, + 814, + 74, + 814, + 105, + 299, + 105 + ], + "score": 0.917 + }, + { + "category_id": 1, + "poly": [ + 295, + 641, + 952, + 641, + 952, + 677, + 295, + 677 + ], + "score": 0.911 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1513, + 1400, + 1513, + 1400, + 1544, + 1352, + 1544 + ], + "score": 0.9 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1701, + 1400, + 1701, + 1400, + 1733, + 1352, + 1733 + ], + "score": 0.896 + }, + { + "category_id": 9, + "poly": [ + 1366, + 839, + 1400, + 839, + 1400, + 869, + 1366, + 869 + ], + "score": 0.888 + }, + { + "category_id": 9, + "poly": [ + 1365, + 1284, + 1400, + 1284, + 1400, + 1314, + 1365, + 1314 + ], + "score": 0.88 + }, + { + "category_id": 2, + "poly": [ + 1375, + 1842, + 1402, + 1842, + 1402, + 1872, + 1375, + 1872 + ], + "score": 0.816 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 859, + 2088, + 859, + 2112, + 840, + 2112 + ], + "score": 0.809 + }, + { + "category_id": 9, + "poly": [ + 1366, + 503, + 1400, + 503, + 1400, + 534, + 1366, + 534 + ], + "score": 0.595 + }, + { + "category_id": 9, + "poly": [ + 1366, + 503, + 1400, + 503, + 1400, + 534, + 1366, + 534 + ], + "score": 0.385 + }, + { + "category_id": 14, + "poly": [ + 647, + 809, + 1048, + 809, + 1048, + 904, + 647, + 904 + ], + "score": 0.95, + "latex": "p _ { \\vec { \\alpha } } ^ { j } ( X ) = \\sum _ { i _ { 2 } , \\ldots , i _ { K } = 1 } ^ { n } x _ { j } ^ { \\alpha _ { 1 } } x _ { i _ { 2 } } ^ { \\alpha _ { 2 } } \\cdot \\cdot \\cdot x _ { i _ { k } } ^ { \\alpha _ { k } }" + }, + { + "category_id": 13, + "poly": [ + 1039, + 754, + 1199, + 754, + 1199, + 796, + 1039, + 796 + ], + "score": 0.95, + "latex": "p _ { \\vec { \\alpha } } = ( p _ { \\vec { \\alpha } } ^ { j } ) _ { j = 1 } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 1126, + 1164, + 1275, + 1164, + 1275, + 1203, + 1126, + 1203 + ], + "score": 0.95, + "latex": "\\mathbb { R } ^ { 3 \\times n } \\oplus \\mathcal { T } _ { k + 1 } ^ { n }" + }, + { + "category_id": 14, + "poly": [ + 587, + 483, + 1111, + 483, + 1111, + 550, + 587, + 550 + ], + "score": 0.94, + "latex": "\\mathcal { Q } _ { D } = \\{ \\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac { 1 } { n } X 1 _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) | \\| \\vec { r } \\| _ { 1 } \\leq D \\} ," + }, + { + "category_id": 13, + "poly": [ + 691, + 920, + 797, + 920, + 797, + 958, + 691, + 958 + ], + "score": 0.93, + "latex": "\\alpha _ { k } \\in \\mathbb { N } _ { + } ^ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 664, + 1370, + 968, + 1370, + 968, + 1404, + 664, + 1404 + ], + "score": 0.93, + "latex": "( X , V ) \\mapsto f ( X , V | \\theta _ { 1 } , \\theta _ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 963, + 565, + 1190, + 565, + 1190, + 607, + 963, + 607 + ], + "score": 0.93, + "latex": "W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } } = \\oplus _ { T = 0 } ^ { D } \\mathcal { T } _ { T }" + }, + { + "category_id": 13, + "poly": [ + 966, + 1164, + 1094, + 1164, + 1094, + 1201, + 966, + 1201 + ], + "score": 0.92, + "latex": "\\mathbb { R } ^ { 3 \\times n } \\oplus \\mathcal { T } _ { k } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 455, + 224, + 509, + 224, + 509, + 269, + 455, + 269 + ], + "score": 0.92, + "latex": "Q _ { j } ^ { ( \\vec { r } ) }" + }, + { + "category_id": 13, + "poly": [ + 568, + 1202, + 879, + 1202, + 879, + 1240, + 568, + 1240 + ], + "score": 0.92, + "latex": "f _ { j } ( X , V ) = ( x _ { j } , \\tilde { V } _ { j } ( X , V ) )" + }, + { + "category_id": 13, + "poly": [ + 1014, + 408, + 1067, + 408, + 1067, + 444, + 1014, + 444 + ], + "score": 0.92, + "latex": "Q ^ { ( \\vec { r } ) }" + }, + { + "category_id": 13, + "poly": [ + 377, + 1204, + 458, + 1204, + 458, + 1239, + 377, + 1239 + ], + "score": 0.92, + "latex": "j \\in [ n ]" + }, + { + "category_id": 13, + "poly": [ + 510, + 1939, + 577, + 1939, + 577, + 1976, + 510, + 1976 + ], + "score": 0.92, + "latex": "W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } }" + }, + { + "category_id": 13, + "poly": [ + 655, + 1166, + 823, + 1166, + 823, + 1200, + 655, + 1200 + ], + "score": 0.92, + "latex": "f ( X , V | \\theta _ { 1 } , \\theta _ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 965, + 1564, + 1031, + 1564, + 1031, + 1595, + 965, + 1595 + ], + "score": 0.92, + "latex": "\\mathcal { F } _ { m i n }" + }, + { + "category_id": 13, + "poly": [ + 545, + 957, + 598, + 957, + 598, + 992, + 545, + 992 + ], + "score": 0.92, + "latex": "Q ^ { ( \\vec { r } ) }" + }, + { + "category_id": 13, + "poly": [ + 678, + 1973, + 783, + 1973, + 783, + 2003, + 678, + 2003 + ], + "score": 0.92, + "latex": "W _ { T } = \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 373, + 922, + 581, + 922, + 581, + 956, + 373, + 956 + ], + "score": 0.92, + "latex": "\\vec { \\alpha } = ( \\alpha _ { 1 } , \\ldots , \\alpha _ { K } )" + }, + { + "category_id": 13, + "poly": [ + 542, + 643, + 639, + 643, + 639, + 676, + 542, + 676 + ], + "score": 0.91, + "latex": "D \\in { \\mathbb { N } } _ { + }" + }, + { + "category_id": 13, + "poly": [ + 436, + 1434, + 517, + 1434, + 517, + 1465, + 436, + 1465 + ], + "score": 0.91, + "latex": "( X , V )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1402, + 419, + 1402, + 419, + 1433, + 298, + 1433 + ], + "score": 0.91, + "latex": "\\theta _ { 1 } , \\theta _ { 2 } \\in \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 360, + 267, + 413, + 267, + 413, + 303, + 360, + 303 + ], + "score": 0.91, + "latex": "Q ^ { ( \\vec { r } ) }" + }, + { + "category_id": 13, + "poly": [ + 297, + 2003, + 363, + 2003, + 363, + 2037, + 297, + 2037 + ], + "score": 0.91, + "latex": "\\mathcal { F } _ { \\mathrm { p o o l } }" + }, + { + "category_id": 13, + "poly": [ + 442, + 270, + 579, + 270, + 579, + 303, + 442, + 303 + ], + "score": 0.91, + "latex": "\\mathrm { S O } ( 3 ) \\times S _ { n }" + }, + { + "category_id": 14, + "poly": [ + 624, + 1510, + 1073, + 1510, + 1073, + 1548, + 624, + 1548 + ], + "score": 0.91, + "latex": "\\operatorname { e x t } ( X ) = ( X , 1 _ { n } ) \\operatorname { a n d } \\pi _ { V } ( X , V ) = V ." + }, + { + "category_id": 13, + "poly": [ + 617, + 2004, + 677, + 2004, + 677, + 2035, + 617, + 2035 + ], + "score": 0.9, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 1205, + 721, + 1400, + 721, + 1400, + 755, + 1205, + 755 + ], + "score": 0.9, + "latex": "p : \\mathbb { R } ^ { 3 \\times n } \\mathbb { R } ^ { n }" + }, + { + "category_id": 14, + "poly": [ + 388, + 1682, + 1309, + 1682, + 1309, + 1749, + 388, + 1749 + ], + "score": 0.9, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } } = \\big \\{ \\iota \\circ \\pi _ { V } \\circ f ^ { 1 } \\circ f ^ { 2 } \\circ \\ldots \\circ f ^ { T } \\circ \\mathrm { e x t } ( X - \\frac { 1 } { n } X \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) | f ^ { j } \\in \\mathcal { F } _ { m i n } , T \\leq D \\big \\} ." + }, + { + "category_id": 14, + "poly": [ + 560, + 1272, + 1140, + 1272, + 1140, + 1347, + 560, + 1347 + ], + "score": 0.9, + "latex": "\\tilde { V } _ { j } ( X , V | \\theta _ { 1 } , \\theta _ { 2 } ) = \\theta _ { 1 } \\left( x _ { j } \\otimes V _ { j } \\right) + \\theta _ { 2 } \\sum _ { i } \\left( x _ { i } \\otimes V _ { i } \\right)" + }, + { + "category_id": 13, + "poly": [ + 1007, + 1029, + 1052, + 1029, + 1052, + 1059, + 1007, + 1059 + ], + "score": 0.9, + "latex": "\\mathcal { Q } _ { D }" + }, + { + "category_id": 13, + "poly": [ + 1303, + 1059, + 1347, + 1059, + 1347, + 1090, + 1303, + 1090 + ], + "score": 0.89, + "latex": "\\mathcal { Q } _ { D }" + }, + { + "category_id": 13, + "poly": [ + 615, + 1638, + 658, + 1638, + 658, + 1667, + 615, + 1667 + ], + "score": 0.89, + "latex": "\\mathcal { Q } _ { D }" + }, + { + "category_id": 13, + "poly": [ + 413, + 759, + 447, + 759, + 447, + 789, + 413, + 789 + ], + "score": 0.89, + "latex": "S _ { n }" + }, + { + "category_id": 13, + "poly": [ + 866, + 572, + 902, + 572, + 902, + 602, + 866, + 602 + ], + "score": 0.89, + "latex": "\\mathcal { T } _ { T }" + }, + { + "category_id": 13, + "poly": [ + 886, + 1944, + 931, + 1944, + 931, + 1973, + 886, + 1973 + ], + "score": 0.89, + "latex": "W _ { T }" + }, + { + "category_id": 13, + "poly": [ + 466, + 1403, + 531, + 1403, + 531, + 1433, + 466, + 1433 + ], + "score": 0.88, + "latex": "\\mathcal { F } _ { m i n }" + }, + { + "category_id": 13, + "poly": [ + 731, + 644, + 774, + 644, + 774, + 675, + 731, + 675 + ], + "score": 0.88, + "latex": "\\mathcal { Q } _ { D }" + }, + { + "category_id": 13, + "poly": [ + 359, + 1763, + 419, + 1763, + 419, + 1794, + 359, + 1794 + ], + "score": 0.88, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 654, + 603, + 680, + 603, + 680, + 628, + 654, + 628 + ], + "score": 0.85, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 1014, + 1434, + 1042, + 1434, + 1042, + 1460, + 1014, + 1460 + ], + "score": 0.83, + "latex": "X" + }, + { + "category_id": 13, + "poly": [ + 1086, + 1030, + 1113, + 1030, + 1113, + 1056, + 1086, + 1056 + ], + "score": 0.82, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 1001, + 1845, + 1027, + 1845, + 1027, + 1870, + 1001, + 1870 + ], + "score": 0.82, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 416, + 1465, + 441, + 1465, + 441, + 1490, + 416, + 1490 + ], + "score": 0.82, + "latex": "V" + }, + { + "category_id": 13, + "poly": [ + 478, + 1565, + 504, + 1565, + 504, + 1591, + 478, + 1591 + ], + "score": 0.81, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 298, + 1060, + 324, + 1060, + 324, + 1086, + 298, + 1086 + ], + "score": 0.81, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 805, + 645, + 831, + 645, + 831, + 671, + 805, + 671 + ], + "score": 0.8, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 947, + 967, + 964, + 967, + 964, + 992, + 947, + 992 + ], + "score": 0.79, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 448, + 1764, + 474, + 1764, + 474, + 1791, + 448, + 1791 + ], + "score": 0.78, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 375, + 578, + 388, + 578, + 388, + 599, + 375, + 599 + ], + "score": 0.73, + "latex": "\\iota" + }, + { + "category_id": 13, + "poly": [ + 1091, + 1944, + 1117, + 1944, + 1117, + 1971, + 1091, + 1971 + ], + "score": 0.38, + "latex": "\\mathbf { G }" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 815.0, + 72.0, + 815.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1380.0, + 1848.0, + 1401.0, + 1848.0, + 1401.0, + 1870.0, + 1380.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2086.0, + 862.0, + 2086.0, + 862.0, + 2116.0, + 840.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1367.0, + 663.0, + 1367.0, + 663.0, + 1407.0, + 294.0, + 1407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 969.0, + 1367.0, + 1405.0, + 1367.0, + 1405.0, + 1407.0, + 969.0, + 1407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1400.0, + 297.0, + 1400.0, + 297.0, + 1437.0, + 294.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 420.0, + 1400.0, + 465.0, + 1400.0, + 465.0, + 1437.0, + 420.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 532.0, + 1400.0, + 1406.0, + 1400.0, + 1406.0, + 1437.0, + 532.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1433.0, + 435.0, + 1433.0, + 435.0, + 1466.0, + 294.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 1433.0, + 1013.0, + 1433.0, + 1013.0, + 1466.0, + 518.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 1433.0, + 1404.0, + 1433.0, + 1404.0, + 1466.0, + 1043.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1463.0, + 415.0, + 1463.0, + 415.0, + 1495.0, + 293.0, + 1495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 442.0, + 1463.0, + 978.0, + 1463.0, + 978.0, + 1495.0, + 442.0, + 1495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1908.0, + 1404.0, + 1908.0, + 1404.0, + 1946.0, + 292.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 1931.0, + 509.0, + 1931.0, + 509.0, + 1989.0, + 286.0, + 1989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 578.0, + 1931.0, + 885.0, + 1931.0, + 885.0, + 1989.0, + 578.0, + 1989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 932.0, + 1931.0, + 1090.0, + 1931.0, + 1090.0, + 1989.0, + 932.0, + 1989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1118.0, + 1931.0, + 1414.0, + 1931.0, + 1414.0, + 1989.0, + 1118.0, + 1989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1972.0, + 677.0, + 1972.0, + 677.0, + 2007.0, + 293.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1972.0, + 1404.0, + 1972.0, + 1404.0, + 2007.0, + 784.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2001.0, + 296.0, + 2001.0, + 296.0, + 2041.0, + 293.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 2001.0, + 616.0, + 2001.0, + 616.0, + 2041.0, + 364.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 2001.0, + 1406.0, + 2001.0, + 1406.0, + 2041.0, + 678.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1025.0, + 1006.0, + 1025.0, + 1006.0, + 1062.0, + 294.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1025.0, + 1085.0, + 1025.0, + 1085.0, + 1062.0, + 1053.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1114.0, + 1025.0, + 1404.0, + 1025.0, + 1404.0, + 1062.0, + 1114.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1057.0, + 1302.0, + 1057.0, + 1302.0, + 1094.0, + 325.0, + 1094.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1348.0, + 1057.0, + 1406.0, + 1057.0, + 1406.0, + 1094.0, + 1348.0, + 1094.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1090.0, + 1404.0, + 1090.0, + 1404.0, + 1123.0, + 295.0, + 1123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1124.0, + 392.0, + 1124.0, + 392.0, + 1151.0, + 295.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 223.0, + 454.0, + 223.0, + 454.0, + 269.0, + 294.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 223.0, + 1407.0, + 223.0, + 1407.0, + 269.0, + 510.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 264.0, + 359.0, + 264.0, + 359.0, + 308.0, + 294.0, + 308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 414.0, + 264.0, + 441.0, + 264.0, + 441.0, + 308.0, + 414.0, + 308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.0, + 264.0, + 1407.0, + 264.0, + 1407.0, + 308.0, + 580.0, + 308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 298.0, + 1407.0, + 298.0, + 1407.0, + 337.0, + 293.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 330.0, + 1406.0, + 330.0, + 1406.0, + 368.0, + 294.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 362.0, + 703.0, + 362.0, + 703.0, + 394.0, + 294.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 920.0, + 372.0, + 920.0, + 372.0, + 958.0, + 297.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 582.0, + 920.0, + 690.0, + 920.0, + 690.0, + 958.0, + 582.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 798.0, + 920.0, + 1404.0, + 920.0, + 1404.0, + 958.0, + 798.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 951.0, + 544.0, + 951.0, + 544.0, + 1000.0, + 292.0, + 1000.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 599.0, + 951.0, + 946.0, + 951.0, + 946.0, + 1000.0, + 599.0, + 1000.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 951.0, + 977.0, + 951.0, + 977.0, + 1000.0, + 965.0, + 1000.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1376.0, + 962.0, + 1404.0, + 962.0, + 1404.0, + 989.0, + 1376.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 719.0, + 1204.0, + 719.0, + 1204.0, + 759.0, + 291.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 719.0, + 1406.0, + 719.0, + 1406.0, + 759.0, + 1401.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 752.0, + 412.0, + 752.0, + 412.0, + 798.0, + 292.0, + 798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 752.0, + 1038.0, + 752.0, + 1038.0, + 798.0, + 448.0, + 798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 752.0, + 1331.0, + 752.0, + 1331.0, + 798.0, + 1200.0, + 798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1157.0, + 654.0, + 1157.0, + 654.0, + 1210.0, + 293.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 824.0, + 1157.0, + 965.0, + 1157.0, + 965.0, + 1210.0, + 824.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 1157.0, + 1125.0, + 1157.0, + 1125.0, + 1210.0, + 1095.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1276.0, + 1157.0, + 1406.0, + 1157.0, + 1406.0, + 1210.0, + 1276.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1201.0, + 376.0, + 1201.0, + 376.0, + 1244.0, + 294.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 459.0, + 1201.0, + 567.0, + 1201.0, + 567.0, + 1244.0, + 459.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 880.0, + 1201.0, + 967.0, + 1201.0, + 967.0, + 1244.0, + 880.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 561.0, + 374.0, + 561.0, + 374.0, + 615.0, + 292.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 561.0, + 865.0, + 561.0, + 865.0, + 615.0, + 389.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 903.0, + 561.0, + 962.0, + 561.0, + 962.0, + 615.0, + 903.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 561.0, + 1409.0, + 561.0, + 1409.0, + 615.0, + 1191.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 598.0, + 653.0, + 598.0, + 653.0, + 638.0, + 293.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 598.0, + 800.0, + 598.0, + 800.0, + 638.0, + 681.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 405.0, + 1013.0, + 405.0, + 1013.0, + 449.0, + 293.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 405.0, + 1404.0, + 405.0, + 1404.0, + 449.0, + 1068.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 442.0, + 687.0, + 442.0, + 687.0, + 475.0, + 297.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1560.0, + 477.0, + 1560.0, + 477.0, + 1600.0, + 294.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 1560.0, + 964.0, + 1560.0, + 964.0, + 1600.0, + 505.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 1560.0, + 1404.0, + 1560.0, + 1404.0, + 1600.0, + 1032.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1589.0, + 372.0, + 1589.0, + 372.0, + 1635.0, + 291.0, + 1635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1840.0, + 1000.0, + 1840.0, + 1000.0, + 1880.0, + 294.0, + 1880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 1840.0, + 1037.0, + 1840.0, + 1037.0, + 1880.0, + 1028.0, + 1880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1756.0, + 358.0, + 1756.0, + 358.0, + 1802.0, + 294.0, + 1802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 420.0, + 1756.0, + 447.0, + 1756.0, + 447.0, + 1802.0, + 420.0, + 1802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 1756.0, + 595.0, + 1756.0, + 595.0, + 1802.0, + 475.0, + 1802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1633.0, + 614.0, + 1633.0, + 614.0, + 1673.0, + 296.0, + 1673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 1633.0, + 833.0, + 1633.0, + 833.0, + 1673.0, + 659.0, + 1673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 634.0, + 541.0, + 634.0, + 541.0, + 686.0, + 293.0, + 686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 640.0, + 634.0, + 730.0, + 634.0, + 730.0, + 686.0, + 640.0, + 686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 775.0, + 634.0, + 804.0, + 634.0, + 804.0, + 686.0, + 775.0, + 686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 634.0, + 953.0, + 634.0, + 953.0, + 686.0, + 832.0, + 686.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 881, + 1405, + 881, + 1405, + 1069, + 296, + 1069 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 1160, + 1405, + 1160, + 1405, + 1354, + 297, + 1354 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 296, + 1848, + 1405, + 1848, + 1405, + 2035, + 296, + 2035 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 296, + 1525, + 1404, + 1525, + 1404, + 1640, + 296, + 1640 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 296, + 1367, + 1404, + 1367, + 1404, + 1499, + 296, + 1499 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 299, + 307, + 1402, + 307, + 1402, + 401, + 299, + 401 + ], + "score": 0.962 + }, + { + "category_id": 8, + "poly": [ + 608, + 457, + 1089, + 457, + 1089, + 546, + 608, + 546 + ], + "score": 0.959 + }, + { + "category_id": 1, + "poly": [ + 290, + 228, + 1402, + 228, + 1402, + 295, + 290, + 295 + ], + "score": 0.955 + }, + { + "category_id": 8, + "poly": [ + 733, + 1657, + 967, + 1657, + 967, + 1748, + 733, + 1748 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 297, + 561, + 1403, + 561, + 1403, + 630, + 297, + 630 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 296, + 718, + 996, + 718, + 996, + 751, + 296, + 751 + ], + "score": 0.931 + }, + { + "category_id": 1, + "poly": [ + 296, + 1759, + 941, + 1759, + 941, + 1808, + 296, + 1808 + ], + "score": 0.923 + }, + { + "category_id": 1, + "poly": [ + 299, + 651, + 1034, + 651, + 1034, + 686, + 299, + 686 + ], + "score": 0.921 + }, + { + "category_id": 0, + "poly": [ + 301, + 816, + 1132, + 816, + 1132, + 851, + 301, + 851 + ], + "score": 0.92 + }, + { + "category_id": 1, + "poly": [ + 296, + 408, + 873, + 408, + 873, + 442, + 296, + 442 + ], + "score": 0.916 + }, + { + "category_id": 2, + "poly": [ + 299, + 74, + 813, + 74, + 813, + 105, + 299, + 105 + ], + "score": 0.916 + }, + { + "category_id": 0, + "poly": [ + 298, + 1104, + 880, + 1104, + 880, + 1137, + 298, + 1137 + ], + "score": 0.915 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1685, + 1400, + 1685, + 1400, + 1717, + 1352, + 1717 + ], + "score": 0.905 + }, + { + "category_id": 2, + "poly": [ + 1374, + 720, + 1403, + 720, + 1403, + 749, + 1374, + 749 + ], + "score": 0.829 + }, + { + "category_id": 2, + "poly": [ + 841, + 2087, + 858, + 2087, + 858, + 2111, + 841, + 2111 + ], + "score": 0.702 + }, + { + "category_id": 2, + "poly": [ + 841, + 2087, + 859, + 2087, + 859, + 2111, + 841, + 2111 + ], + "score": 0.226 + }, + { + "category_id": 14, + "poly": [ + 732, + 1652, + 968, + 1652, + 968, + 1749, + 732, + 1749 + ], + "score": 0.94, + "latex": "\\Lambda _ { j } ( V ) = \\sum _ { i = 1 } ^ { K _ { 1 } } M _ { i j } V _ { i }" + }, + { + "category_id": 14, + "poly": [ + 608, + 452, + 1088, + 452, + 1088, + 549, + 608, + 549 + ], + "score": 0.94, + "latex": "\\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac { 1 } { n } X 1 _ { n } 1 _ { n } ^ { T } ) = \\sum _ { k = 1 } ^ { K } \\hat { A } _ { k } f _ { k } ( X ) ," + }, + { + "category_id": 13, + "poly": [ + 606, + 1596, + 726, + 1596, + 726, + 1642, + 606, + 1642 + ], + "score": 0.94, + "latex": "\\bar { \\ell } _ { i } ^ { ( 1 ) } \\neq \\ell _ { j } ^ { ( 2 ) }" + }, + { + "category_id": 13, + "poly": [ + 910, + 1254, + 1006, + 1254, + 1006, + 1290, + 910, + 1290 + ], + "score": 0.94, + "latex": "D ^ { ( \\ell ) } ( R )" + }, + { + "category_id": 13, + "poly": [ + 372, + 1761, + 516, + 1761, + 516, + 1801, + 372, + 1801 + ], + "score": 0.93, + "latex": "V = ( V _ { i } ) _ { i = 1 } ^ { K _ { 1 } }" + }, + { + "category_id": 13, + "poly": [ + 1193, + 1458, + 1393, + 1458, + 1393, + 1501, + 1193, + 1501 + ], + "score": 0.93, + "latex": "W _ { l } = \\bigoplus _ { k = 1 } ^ { K } W _ { \\ell _ { k } }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1318, + 448, + 1318, + 448, + 1355, + 297, + 1355 + ], + "score": 0.93, + "latex": "D ^ { ( 0 ) } ( R ) = 1" + }, + { + "category_id": 13, + "poly": [ + 498, + 1318, + 656, + 1318, + 656, + 1355, + 498, + 1355 + ], + "score": 0.93, + "latex": "D ^ { ( 1 ) } ( R ) = R" + }, + { + "category_id": 13, + "poly": [ + 409, + 1285, + 567, + 1285, + 567, + 1318, + 409, + 1318 + ], + "score": 0.92, + "latex": "W _ { \\ell } : = \\overset { \\bullet } { \\mathbb { R } } ^ { 2 \\ell + 1 }" + }, + { + "category_id": 13, + "poly": [ + 774, + 1767, + 933, + 1767, + 933, + 1799, + 774, + 1799 + ], + "score": 0.92, + "latex": "i = 1 , \\ldots , K _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1027, + 1193, + 1137, + 1193, + 1137, + 1223, + 1027, + 1223 + ], + "score": 0.92, + "latex": "h w \\in W _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 730, + 1193, + 838, + 1193, + 838, + 1223, + 730, + 1223 + ], + "score": 0.92, + "latex": "W _ { 0 } \\subset W" + }, + { + "category_id": 13, + "poly": [ + 820, + 562, + 1105, + 562, + 1105, + 602, + 820, + 602 + ], + "score": 0.92, + "latex": "{ \\hat { A } } _ { k } : W _ { \\mathrm { f e a t } } ^ { n } \\to ( W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } } ) ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 570, + 1765, + 691, + 1765, + 691, + 1810, + 570, + 1810 + ], + "score": 0.92, + "latex": "V _ { i } \\in W _ { \\ell _ { i } ^ { ( 1 ) } }" + }, + { + "category_id": 13, + "poly": [ + 757, + 262, + 823, + 262, + 823, + 296, + 757, + 296 + ], + "score": 0.92, + "latex": "\\mathcal { F } _ { \\mathrm { p o o l } }" + }, + { + "category_id": 13, + "poly": [ + 1263, + 1288, + 1359, + 1288, + 1359, + 1319, + 1263, + 1319 + ], + "score": 0.92, + "latex": "\\ell = 0 , 1" + }, + { + "category_id": 13, + "poly": [ + 771, + 1528, + 1017, + 1528, + 1017, + 1569, + 771, + 1569 + ], + "score": 0.91, + "latex": "l ^ { ( 2 ) } = ( \\ell _ { 1 } ^ { ( 2 ) } , \\dots , \\ell _ { K _ { 2 } } ^ { ( 2 ) } )" + }, + { + "category_id": 13, + "poly": [ + 389, + 1604, + 489, + 1604, + 489, + 1639, + 389, + 1639 + ], + "score": 0.91, + "latex": "M _ { i j } = 0" + }, + { + "category_id": 13, + "poly": [ + 473, + 1528, + 718, + 1528, + 718, + 1571, + 473, + 1571 + ], + "score": 0.91, + "latex": "l ^ { ( 1 ) } = ( \\ell _ { 1 } ^ { ( 1 ) } , \\dots , \\ell _ { K _ { 1 } } ^ { ( 1 ) } )" + }, + { + "category_id": 13, + "poly": [ + 879, + 1461, + 1142, + 1461, + 1142, + 1500, + 879, + 1500 + ], + "score": 0.91, + "latex": "\\boldsymbol { l } = ( \\ell _ { 1 } , \\dots , \\ell _ { K } ) \\in \\mathbb { N } _ { + } ^ { K }" + }, + { + "category_id": 13, + "poly": [ + 1212, + 1193, + 1393, + 1193, + 1393, + 1224, + 1212, + 1224 + ], + "score": 0.91, + "latex": "h \\in H , w \\in W _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 1151, + 1535, + 1374, + 1535, + 1374, + 1568, + 1151, + 1568 + ], + "score": 0.9, + "latex": "\\boldsymbol { \\Lambda } = \\left( \\Lambda _ { 1 } , \\ldots , \\Lambda _ { K _ { 2 } } \\right)" + }, + { + "category_id": 13, + "poly": [ + 843, + 1569, + 903, + 1569, + 903, + 1599, + 843, + 1599 + ], + "score": 0.89, + "latex": "W _ { l ^ { ( 2 ) } }" + }, + { + "category_id": 13, + "poly": [ + 1214, + 1569, + 1321, + 1569, + 1321, + 1597, + 1214, + 1597 + ], + "score": 0.89, + "latex": "K _ { 1 } \\times K _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 558, + 655, + 595, + 655, + 595, + 684, + 558, + 684 + ], + "score": 0.89, + "latex": "A _ { k }" + }, + { + "category_id": 13, + "poly": [ + 711, + 339, + 757, + 339, + 757, + 369, + 711, + 369 + ], + "score": 0.89, + "latex": "W _ { T }" + }, + { + "category_id": 13, + "poly": [ + 726, + 1568, + 787, + 1568, + 787, + 1599, + 726, + 1599 + ], + "score": 0.89, + "latex": "W _ { l ^ { ( 1 ) } }" + }, + { + "category_id": 13, + "poly": [ + 627, + 410, + 671, + 410, + 671, + 440, + 627, + 440 + ], + "score": 0.88, + "latex": "\\mathcal { Q } _ { D }" + }, + { + "category_id": 13, + "poly": [ + 459, + 599, + 495, + 599, + 495, + 628, + 459, + 628 + ], + "score": 0.87, + "latex": "A _ { k }" + }, + { + "category_id": 13, + "poly": [ + 517, + 563, + 763, + 563, + 763, + 599, + 517, + 599 + ], + "score": 0.87, + "latex": "A _ { k } : W _ { \\mathrm { f e a t } } \\to W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } }" + }, + { + "category_id": 13, + "poly": [ + 1362, + 1429, + 1400, + 1429, + 1400, + 1460, + 1362, + 1460 + ], + "score": 0.85, + "latex": "W _ { l }" + }, + { + "category_id": 13, + "poly": [ + 800, + 1106, + 879, + 1106, + 879, + 1139, + 800, + 1139 + ], + "score": 0.84, + "latex": "S O ( 3 )" + }, + { + "category_id": 13, + "poly": [ + 866, + 1194, + 894, + 1194, + 894, + 1221, + 866, + 1221 + ], + "score": 0.83, + "latex": "H" + }, + { + "category_id": 13, + "poly": [ + 379, + 370, + 406, + 370, + 406, + 396, + 379, + 396 + ], + "score": 0.82, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 867, + 340, + 893, + 340, + 893, + 366, + 867, + 366 + ], + "score": 0.82, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 562, + 599, + 622, + 599, + 622, + 628, + 562, + 628 + ], + "score": 0.81, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 1099, + 1164, + 1127, + 1164, + 1127, + 1190, + 1099, + 1190 + ], + "score": 0.81, + "latex": "H" + }, + { + "category_id": 13, + "poly": [ + 785, + 946, + 812, + 946, + 812, + 972, + 785, + 972 + ], + "score": 0.78, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 298, + 1604, + 331, + 1604, + 331, + 1633, + 298, + 1633 + ], + "score": 0.72, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 841, + 1163, + 874, + 1163, + 874, + 1190, + 841, + 1190 + ], + "score": 0.71, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 485, + 1224, + 518, + 1224, + 518, + 1252, + 485, + 1252 + ], + "score": 0.7, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 374, + 567, + 502, + 567, + 502, + 598, + 374, + 598 + ], + "score": 0.67, + "latex": "f _ { k } \\in \\mathcal { F } _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 652, + 599, + 678, + 599, + 678, + 624, + 652, + 624 + ], + "score": 0.67, + "latex": "D" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 811.0, + 1137.0, + 811.0, + 1137.0, + 862.0, + 290.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 815.0, + 72.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1103.0, + 799.0, + 1103.0, + 799.0, + 1141.0, + 293.0, + 1141.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 724.0, + 1403.0, + 724.0, + 1403.0, + 750.0, + 1378.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2118.0, + 839.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 884.0, + 1405.0, + 884.0, + 1405.0, + 916.0, + 295.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 913.0, + 1405.0, + 913.0, + 1405.0, + 949.0, + 294.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 943.0, + 784.0, + 943.0, + 784.0, + 980.0, + 293.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 813.0, + 943.0, + 1406.0, + 943.0, + 1406.0, + 980.0, + 813.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 975.0, + 1405.0, + 975.0, + 1405.0, + 1008.0, + 293.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1006.0, + 1405.0, + 1006.0, + 1405.0, + 1038.0, + 294.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1035.0, + 412.0, + 1035.0, + 412.0, + 1069.0, + 294.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1161.0, + 840.0, + 1161.0, + 840.0, + 1198.0, + 294.0, + 1198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1161.0, + 1098.0, + 1161.0, + 1098.0, + 1198.0, + 875.0, + 1198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 1161.0, + 1406.0, + 1161.0, + 1406.0, + 1198.0, + 1128.0, + 1198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1190.0, + 729.0, + 1190.0, + 729.0, + 1228.0, + 292.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 1190.0, + 865.0, + 1190.0, + 865.0, + 1228.0, + 839.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1190.0, + 1026.0, + 1190.0, + 1026.0, + 1228.0, + 895.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1138.0, + 1190.0, + 1211.0, + 1190.0, + 1211.0, + 1228.0, + 1138.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 1190.0, + 1405.0, + 1190.0, + 1405.0, + 1228.0, + 1394.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1222.0, + 484.0, + 1222.0, + 484.0, + 1258.0, + 294.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 519.0, + 1222.0, + 1405.0, + 1222.0, + 1405.0, + 1258.0, + 519.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1253.0, + 909.0, + 1253.0, + 909.0, + 1294.0, + 292.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1007.0, + 1253.0, + 1406.0, + 1253.0, + 1406.0, + 1294.0, + 1007.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1278.0, + 408.0, + 1278.0, + 408.0, + 1327.0, + 291.0, + 1327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 1278.0, + 1262.0, + 1278.0, + 1262.0, + 1327.0, + 568.0, + 1327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1360.0, + 1278.0, + 1410.0, + 1278.0, + 1410.0, + 1327.0, + 1360.0, + 1327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1315.0, + 497.0, + 1315.0, + 497.0, + 1355.0, + 449.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 657.0, + 1315.0, + 666.0, + 1315.0, + 666.0, + 1355.0, + 657.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1848.0, + 1405.0, + 1848.0, + 1405.0, + 1885.0, + 294.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1879.0, + 1406.0, + 1879.0, + 1406.0, + 1916.0, + 293.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1911.0, + 1403.0, + 1911.0, + 1403.0, + 1947.0, + 295.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1941.0, + 1403.0, + 1941.0, + 1403.0, + 1976.0, + 295.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1974.0, + 1406.0, + 1974.0, + 1406.0, + 2006.0, + 296.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2000.0, + 436.0, + 2000.0, + 436.0, + 2038.0, + 294.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1374.0, + 2003.0, + 1406.0, + 2003.0, + 1406.0, + 2034.0, + 1374.0, + 2034.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 1517.0, + 472.0, + 1517.0, + 472.0, + 1581.0, + 284.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 1517.0, + 770.0, + 1517.0, + 770.0, + 1581.0, + 719.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1018.0, + 1517.0, + 1150.0, + 1517.0, + 1150.0, + 1581.0, + 1018.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1375.0, + 1517.0, + 1414.0, + 1517.0, + 1414.0, + 1581.0, + 1375.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1565.0, + 725.0, + 1565.0, + 725.0, + 1603.0, + 294.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 1565.0, + 842.0, + 1565.0, + 842.0, + 1603.0, + 788.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 1565.0, + 1213.0, + 1565.0, + 1213.0, + 1603.0, + 904.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 1565.0, + 1406.0, + 1565.0, + 1406.0, + 1603.0, + 1322.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1584.0, + 297.0, + 1584.0, + 297.0, + 1651.0, + 288.0, + 1651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 1584.0, + 388.0, + 1584.0, + 388.0, + 1651.0, + 332.0, + 1651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1584.0, + 605.0, + 1584.0, + 605.0, + 1651.0, + 490.0, + 1651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 727.0, + 1584.0, + 856.0, + 1584.0, + 856.0, + 1651.0, + 727.0, + 1651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1368.0, + 1404.0, + 1368.0, + 1404.0, + 1402.0, + 294.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1399.0, + 1406.0, + 1399.0, + 1406.0, + 1432.0, + 294.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1429.0, + 1361.0, + 1429.0, + 1361.0, + 1463.0, + 294.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 1450.0, + 878.0, + 1450.0, + 878.0, + 1514.0, + 284.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1143.0, + 1450.0, + 1192.0, + 1450.0, + 1192.0, + 1514.0, + 1143.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 1450.0, + 1412.0, + 1450.0, + 1412.0, + 1514.0, + 1394.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 307.0, + 1404.0, + 307.0, + 1404.0, + 343.0, + 293.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 339.0, + 710.0, + 339.0, + 710.0, + 373.0, + 295.0, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 339.0, + 866.0, + 339.0, + 866.0, + 373.0, + 758.0, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 339.0, + 1404.0, + 339.0, + 1404.0, + 373.0, + 894.0, + 373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 370.0, + 378.0, + 370.0, + 378.0, + 404.0, + 295.0, + 404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 370.0, + 1235.0, + 370.0, + 1235.0, + 404.0, + 407.0, + 404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 228.0, + 1405.0, + 228.0, + 1405.0, + 268.0, + 294.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 260.0, + 756.0, + 260.0, + 756.0, + 297.0, + 296.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 824.0, + 260.0, + 1126.0, + 260.0, + 1126.0, + 297.0, + 824.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 551.0, + 373.0, + 551.0, + 373.0, + 617.0, + 287.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 551.0, + 516.0, + 551.0, + 516.0, + 617.0, + 503.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 551.0, + 819.0, + 551.0, + 819.0, + 617.0, + 764.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1106.0, + 551.0, + 1413.0, + 551.0, + 1413.0, + 617.0, + 1106.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 593.0, + 458.0, + 593.0, + 458.0, + 634.0, + 295.0, + 634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 593.0, + 561.0, + 593.0, + 561.0, + 634.0, + 496.0, + 634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 623.0, + 593.0, + 651.0, + 593.0, + 651.0, + 634.0, + 623.0, + 634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 679.0, + 593.0, + 797.0, + 593.0, + 797.0, + 634.0, + 679.0, + 634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 716.0, + 999.0, + 716.0, + 999.0, + 756.0, + 294.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1755.0, + 371.0, + 1755.0, + 371.0, + 1808.0, + 290.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 1755.0, + 569.0, + 1755.0, + 569.0, + 1808.0, + 517.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 1755.0, + 773.0, + 1755.0, + 773.0, + 1808.0, + 692.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 1755.0, + 949.0, + 1755.0, + 949.0, + 1808.0, + 934.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 651.0, + 557.0, + 651.0, + 557.0, + 691.0, + 296.0, + 691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 651.0, + 1035.0, + 651.0, + 1035.0, + 691.0, + 596.0, + 691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 406.0, + 626.0, + 406.0, + 626.0, + 446.0, + 294.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 406.0, + 874.0, + 406.0, + 874.0, + 446.0, + 672.0, + 446.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 229, + 1404, + 229, + 1404, + 507, + 297, + 507 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 736, + 1404, + 736, + 1404, + 892, + 298, + 892 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 299, + 520, + 1404, + 520, + 1404, + 644, + 299, + 644 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 297, + 1091, + 1403, + 1091, + 1403, + 1222, + 297, + 1222 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 298, + 1583, + 1404, + 1583, + 1404, + 1717, + 298, + 1717 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 298, + 1285, + 1404, + 1285, + 1404, + 1381, + 298, + 1381 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 298, + 905, + 1404, + 905, + 1404, + 999, + 298, + 999 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 297, + 1733, + 1404, + 1733, + 1404, + 1829, + 297, + 1829 + ], + "score": 0.97 + }, + { + "category_id": 8, + "poly": [ + 529, + 1482, + 1167, + 1482, + 1167, + 1572, + 529, + 1572 + ], + "score": 0.954 + }, + { + "category_id": 8, + "poly": [ + 600, + 1234, + 1102, + 1234, + 1102, + 1276, + 600, + 1276 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 298, + 1393, + 1401, + 1393, + 1401, + 1469, + 298, + 1469 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 300, + 1012, + 1401, + 1012, + 1401, + 1078, + 300, + 1078 + ], + "score": 0.943 + }, + { + "category_id": 1, + "poly": [ + 298, + 1843, + 1401, + 1843, + 1401, + 1906, + 298, + 1906 + ], + "score": 0.941 + }, + { + "category_id": 1, + "poly": [ + 298, + 1919, + 1402, + 1919, + 1402, + 1984, + 298, + 1984 + ], + "score": 0.935 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1511, + 1400, + 1511, + 1400, + 1541, + 1352, + 1541 + ], + "score": 0.904 + }, + { + "category_id": 2, + "poly": [ + 326, + 2004, + 1240, + 2004, + 1240, + 2034, + 326, + 2034 + ], + "score": 0.896 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1239, + 1399, + 1239, + 1399, + 1270, + 1352, + 1270 + ], + "score": 0.893 + }, + { + "category_id": 2, + "poly": [ + 300, + 75, + 813, + 75, + 813, + 104, + 300, + 104 + ], + "score": 0.89 + }, + { + "category_id": 0, + "poly": [ + 300, + 680, + 681, + 680, + 681, + 711, + 300, + 711 + ], + "score": 0.888 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 858, + 2088, + 858, + 2112, + 840, + 2112 + ], + "score": 0.808 + }, + { + "category_id": 14, + "poly": [ + 528, + 1480, + 1170, + 1480, + 1170, + 1574, + 528, + 1574 + ], + "score": 0.94, + "latex": "\\tilde { V } _ { a } ( X , V ) = U ( l _ { f } , l _ { i } ) \\left( \\theta _ { 0 } V _ { a } + \\sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \\otimes V _ { b } \\right) ." + }, + { + "category_id": 13, + "poly": [ + 831, + 1183, + 1197, + 1183, + 1197, + 1223, + 831, + 1223 + ], + "score": 0.93, + "latex": "F ( \\boldsymbol { x } ) = \\left[ F ^ { ( 0 ) } ( x ) , \\ldots , F ^ { ( D ) } ( x ) \\right]" + }, + { + "category_id": 13, + "poly": [ + 719, + 443, + 858, + 443, + 858, + 477, + 719, + 477 + ], + "score": 0.93, + "latex": "l = l ( l _ { 1 } , l _ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1142, + 1951, + 1247, + 1951, + 1247, + 1985, + 1142, + 1985 + ], + "score": 0.93, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )" + }, + { + "category_id": 13, + "poly": [ + 382, + 1429, + 615, + 1429, + 615, + 1472, + 382, + 1472 + ], + "score": 0.93, + "latex": "\\tilde { V } = ( \\tilde { V } _ { a } ) _ { a = 1 } ^ { n } \\in W _ { l _ { 0 } } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 391, + 383, + 522, + 383, + 522, + 416, + 391, + 416 + ], + "score": 0.93, + "latex": "W _ { l _ { 1 } } \\otimes W _ { l _ { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 702, + 1434, + 849, + 1434, + 849, + 1469, + 702, + 1469 + ], + "score": 0.92, + "latex": "\\boldsymbol { l _ { o } } = \\boldsymbol { l } ( \\boldsymbol { l _ { f } } , \\boldsymbol { l _ { i } } )" + }, + { + "category_id": 13, + "poly": [ + 376, + 1287, + 517, + 1287, + 517, + 1321, + 376, + 1321 + ], + "score": 0.92, + "latex": "\\hat { x } = x / \\| x \\|" + }, + { + "category_id": 13, + "poly": [ + 298, + 1185, + 458, + 1185, + 458, + 1221, + 298, + 1221 + ], + "score": 0.92, + "latex": "[ 0 , 1 , \\ldots , D ] ^ { T }" + }, + { + "category_id": 13, + "poly": [ + 1164, + 322, + 1212, + 322, + 1212, + 355, + 1164, + 355 + ], + "score": 0.92, + "latex": "W _ { \\ell _ { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 864, + 1653, + 950, + 1653, + 950, + 1684, + 864, + 1684 + ], + "score": 0.92, + "latex": "\\theta _ { 0 } \\in \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 790, + 799, + 956, + 799, + 956, + 835, + 790, + 835 + ], + "score": 0.92, + "latex": "W _ { l _ { T } } ^ { n } , l _ { T } \\in \\mathbb { N } _ { + } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 706, + 1395, + 815, + 1395, + 815, + 1433, + 706, + 1433 + ], + "score": 0.91, + "latex": "V \\in W _ { l _ { i } } ^ { n }" + }, + { + "category_id": 14, + "poly": [ + 597, + 1232, + 1102, + 1232, + 1102, + 1274, + 597, + 1274 + ], + "score": 0.91, + "latex": "F _ { m } ^ { \\ell } ( x ) = R ^ { ( \\ell ) } ( \\| x \\| ) Y _ { m } ^ { \\ell } ( \\hat { x } ) , m = - \\ell , \\ldots , \\ell" + }, + { + "category_id": 13, + "poly": [ + 553, + 1288, + 637, + 1288, + 637, + 1319, + 553, + 1319 + ], + "score": 0.91, + "latex": "x \\neq 0" + }, + { + "category_id": 13, + "poly": [ + 1075, + 1767, + 1249, + 1767, + 1249, + 1798, + 1075, + 1798 + ], + "score": 0.91, + "latex": "\\Lambda : W _ { l } \\stackrel { \\cdot } { } W _ { l _ { T } }" + }, + { + "category_id": 13, + "poly": [ + 1147, + 383, + 1207, + 383, + 1207, + 414, + 1147, + 414 + ], + "score": 0.91, + "latex": "l _ { 1 } , l _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1272, + 414, + 1399, + 414, + 1399, + 447, + 1272, + 447 + ], + "score": 0.91, + "latex": "W _ { l _ { 1 } } \\otimes W _ { l _ { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 1031, + 446, + 1133, + 446, + 1133, + 477, + 1031, + 477 + ], + "score": 0.91, + "latex": "U ( l _ { 1 } , l _ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 705, + 1953, + 783, + 1953, + 783, + 1985, + 705, + 1985 + ], + "score": 0.91, + "latex": "( C , D )" + }, + { + "category_id": 13, + "poly": [ + 911, + 1285, + 951, + 1285, + 951, + 1321, + 911, + 1321 + ], + "score": 0.91, + "latex": "Y _ { m } ^ { \\ell }" + }, + { + "category_id": 13, + "poly": [ + 1065, + 323, + 1114, + 323, + 1114, + 354, + 1065, + 354 + ], + "score": 0.91, + "latex": "W _ { \\ell _ { 1 } }" + }, + { + "category_id": 13, + "poly": [ + 1060, + 1152, + 1245, + 1152, + 1245, + 1186, + 1060, + 1186 + ], + "score": 0.91, + "latex": "F : \\bar { \\mathbb { R } ^ { 3 } } W _ { l _ { D } }" + }, + { + "category_id": 13, + "poly": [ + 694, + 1288, + 779, + 1288, + 779, + 1316, + 694, + 1316 + ], + "score": 0.91, + "latex": "\\hat { x } ~ = ~ 0" + }, + { + "category_id": 13, + "poly": [ + 1299, + 1284, + 1350, + 1284, + 1350, + 1316, + 1299, + 1316 + ], + "score": 0.9, + "latex": "R ^ { ( \\ell ) }" + }, + { + "category_id": 13, + "poly": [ + 1096, + 1731, + 1278, + 1731, + 1278, + 1767, + 1096, + 1767 + ], + "score": 0.9, + "latex": "\\hat { \\Lambda } : W _ { l } ^ { n } \\to W _ { l , r } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 1256, + 1583, + 1402, + 1583, + 1402, + 1619, + 1256, + 1619 + ], + "score": 0.89, + "latex": "f ( X , V ) \\quad = \\quad" + }, + { + "category_id": 13, + "poly": [ + 523, + 1685, + 565, + 1685, + 565, + 1715, + 523, + 1715 + ], + "score": 0.89, + "latex": "\\mathcal { F } _ { D }" + }, + { + "category_id": 13, + "poly": [ + 1030, + 414, + 1132, + 414, + 1132, + 445, + 1030, + 445 + ], + "score": 0.89, + "latex": "U ( l _ { 1 } , l _ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 299, + 1615, + 458, + 1615, + 458, + 1653, + 299, + 1653 + ], + "score": 0.89, + "latex": "( X , { \\tilde { V } } ( X , V ) )" + }, + { + "category_id": 13, + "poly": [ + 354, + 444, + 392, + 444, + 392, + 474, + 354, + 474 + ], + "score": 0.88, + "latex": "W _ { l }" + }, + { + "category_id": 13, + "poly": [ + 1335, + 1154, + 1404, + 1154, + 1404, + 1185, + 1335, + 1185 + ], + "score": 0.88, + "latex": "\\boldsymbol { l } _ { D } =" + }, + { + "category_id": 13, + "poly": [ + 527, + 475, + 555, + 475, + 555, + 505, + 527, + 505 + ], + "score": 0.88, + "latex": "\\ell _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1653, + 512, + 1653, + 512, + 1685, + 298, + 1685 + ], + "score": 0.87, + "latex": "R ^ { ( \\ell ) } , \\ell = 0 , \\dots , D" + }, + { + "category_id": 13, + "poly": [ + 538, + 1321, + 593, + 1321, + 593, + 1349, + 538, + 1349 + ], + "score": 0.87, + "latex": "\\leq D" + }, + { + "category_id": 13, + "poly": [ + 605, + 475, + 633, + 475, + 633, + 505, + 605, + 505 + ], + "score": 0.86, + "latex": "\\ell _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 466, + 1687, + 484, + 1687, + 484, + 1716, + 466, + 1716 + ], + "score": 0.85, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 952, + 1397, + 977, + 1397, + 977, + 1423, + 952, + 1423 + ], + "score": 0.84, + "latex": "F" + }, + { + "category_id": 13, + "poly": [ + 925, + 1955, + 951, + 1955, + 951, + 1979, + 925, + 1979 + ], + "score": 0.83, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 808, + 553, + 835, + 553, + 835, + 579, + 808, + 579 + ], + "score": 0.8, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 990, + 1620, + 1017, + 1620, + 1017, + 1647, + 990, + 1647 + ], + "score": 0.8, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 833, + 1797, + 867, + 1797, + 867, + 1824, + 833, + 1824 + ], + "score": 0.78, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 521, + 1188, + 536, + 1188, + 536, + 1215, + 521, + 1215 + ], + "score": 0.77, + "latex": "\\ell" + }, + { + "category_id": 13, + "poly": [ + 298, + 800, + 322, + 800, + 322, + 827, + 298, + 827 + ], + "score": 0.73, + "latex": "G" + }, + { + "category_id": 15, + "poly": [ + 333.0, + 2001.0, + 1239.0, + 2001.0, + 1239.0, + 2038.0, + 333.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 815.0, + 72.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 677.0, + 684.0, + 677.0, + 684.0, + 715.0, + 293.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 2085.0, + 859.0, + 2085.0, + 859.0, + 2116.0, + 836.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 228.0, + 1406.0, + 228.0, + 1406.0, + 266.0, + 295.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 260.0, + 1405.0, + 260.0, + 1405.0, + 296.0, + 295.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 291.0, + 1406.0, + 291.0, + 1406.0, + 327.0, + 294.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 321.0, + 1064.0, + 321.0, + 1064.0, + 357.0, + 295.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1115.0, + 321.0, + 1163.0, + 321.0, + 1163.0, + 357.0, + 1115.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 321.0, + 1405.0, + 321.0, + 1405.0, + 357.0, + 1213.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 351.0, + 1405.0, + 351.0, + 1405.0, + 389.0, + 292.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 382.0, + 390.0, + 382.0, + 390.0, + 419.0, + 294.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 382.0, + 1146.0, + 382.0, + 1146.0, + 419.0, + 523.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1208.0, + 382.0, + 1405.0, + 382.0, + 1405.0, + 419.0, + 1208.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 407.0, + 1029.0, + 407.0, + 1029.0, + 453.0, + 292.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1133.0, + 407.0, + 1271.0, + 407.0, + 1271.0, + 453.0, + 1133.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1400.0, + 407.0, + 1404.0, + 407.0, + 1404.0, + 453.0, + 1400.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 438.0, + 353.0, + 438.0, + 353.0, + 481.0, + 291.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 438.0, + 718.0, + 438.0, + 718.0, + 481.0, + 393.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 859.0, + 438.0, + 1030.0, + 438.0, + 1030.0, + 481.0, + 859.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1134.0, + 438.0, + 1405.0, + 438.0, + 1405.0, + 481.0, + 1134.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 473.0, + 526.0, + 473.0, + 526.0, + 508.0, + 294.0, + 508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 473.0, + 604.0, + 473.0, + 604.0, + 508.0, + 556.0, + 508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 634.0, + 473.0, + 765.0, + 473.0, + 765.0, + 508.0, + 634.0, + 508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 738.0, + 1402.0, + 738.0, + 1402.0, + 771.0, + 297.0, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 769.0, + 1404.0, + 769.0, + 1404.0, + 802.0, + 296.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 791.0, + 297.0, + 791.0, + 297.0, + 841.0, + 291.0, + 841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 791.0, + 789.0, + 791.0, + 789.0, + 841.0, + 323.0, + 841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 791.0, + 1409.0, + 791.0, + 1409.0, + 841.0, + 957.0, + 841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 823.0, + 1404.0, + 823.0, + 1404.0, + 869.0, + 291.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 860.0, + 409.0, + 860.0, + 409.0, + 891.0, + 292.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 522.0, + 1403.0, + 522.0, + 1403.0, + 555.0, + 295.0, + 555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 549.0, + 807.0, + 549.0, + 807.0, + 589.0, + 293.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 549.0, + 1405.0, + 549.0, + 1405.0, + 589.0, + 836.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 583.0, + 1405.0, + 583.0, + 1405.0, + 618.0, + 292.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 614.0, + 569.0, + 614.0, + 569.0, + 646.0, + 295.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1089.0, + 1407.0, + 1089.0, + 1407.0, + 1128.0, + 294.0, + 1128.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1123.0, + 1405.0, + 1123.0, + 1405.0, + 1156.0, + 293.0, + 1156.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1147.0, + 1059.0, + 1147.0, + 1059.0, + 1191.0, + 291.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 1147.0, + 1334.0, + 1147.0, + 1334.0, + 1191.0, + 1246.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1405.0, + 1147.0, + 1408.0, + 1147.0, + 1408.0, + 1191.0, + 1405.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1181.0, + 297.0, + 1181.0, + 297.0, + 1228.0, + 291.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 459.0, + 1181.0, + 520.0, + 1181.0, + 520.0, + 1228.0, + 459.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 537.0, + 1181.0, + 830.0, + 1181.0, + 830.0, + 1228.0, + 537.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 1181.0, + 1389.0, + 1181.0, + 1389.0, + 1228.0, + 1198.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1581.0, + 1255.0, + 1581.0, + 1255.0, + 1620.0, + 292.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 459.0, + 1618.0, + 989.0, + 1618.0, + 989.0, + 1652.0, + 459.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1018.0, + 1618.0, + 1405.0, + 1618.0, + 1405.0, + 1652.0, + 1018.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1646.0, + 297.0, + 1646.0, + 297.0, + 1691.0, + 293.0, + 1691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 1646.0, + 863.0, + 1646.0, + 863.0, + 1691.0, + 513.0, + 1691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 951.0, + 1646.0, + 1408.0, + 1646.0, + 1408.0, + 1691.0, + 951.0, + 1691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1680.0, + 465.0, + 1680.0, + 465.0, + 1720.0, + 294.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 485.0, + 1680.0, + 522.0, + 1680.0, + 522.0, + 1720.0, + 485.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 566.0, + 1680.0, + 577.0, + 1680.0, + 577.0, + 1720.0, + 566.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1280.0, + 375.0, + 1280.0, + 375.0, + 1326.0, + 292.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 1280.0, + 552.0, + 1280.0, + 552.0, + 1326.0, + 518.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 638.0, + 1280.0, + 693.0, + 1280.0, + 693.0, + 1326.0, + 638.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 1280.0, + 910.0, + 1280.0, + 910.0, + 1326.0, + 780.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 952.0, + 1280.0, + 1298.0, + 1280.0, + 1298.0, + 1326.0, + 952.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1351.0, + 1280.0, + 1408.0, + 1280.0, + 1408.0, + 1326.0, + 1351.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1317.0, + 537.0, + 1317.0, + 537.0, + 1353.0, + 292.0, + 1353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 594.0, + 1317.0, + 1405.0, + 1317.0, + 1405.0, + 1353.0, + 594.0, + 1353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1349.0, + 1284.0, + 1349.0, + 1284.0, + 1383.0, + 296.0, + 1383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 906.0, + 1404.0, + 906.0, + 1404.0, + 940.0, + 296.0, + 940.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 937.0, + 1405.0, + 937.0, + 1405.0, + 971.0, + 293.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 968.0, + 1370.0, + 968.0, + 1370.0, + 1002.0, + 294.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1730.0, + 1095.0, + 1730.0, + 1095.0, + 1775.0, + 291.0, + 1775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 1730.0, + 1407.0, + 1730.0, + 1407.0, + 1775.0, + 1279.0, + 1775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1764.0, + 1074.0, + 1764.0, + 1074.0, + 1802.0, + 294.0, + 1802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1250.0, + 1764.0, + 1405.0, + 1764.0, + 1405.0, + 1802.0, + 1250.0, + 1802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1794.0, + 832.0, + 1794.0, + 832.0, + 1831.0, + 295.0, + 1831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 868.0, + 1794.0, + 1404.0, + 1794.0, + 1404.0, + 1831.0, + 868.0, + 1831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1390.0, + 705.0, + 1390.0, + 705.0, + 1432.0, + 294.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 816.0, + 1390.0, + 951.0, + 1390.0, + 951.0, + 1432.0, + 816.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 1390.0, + 1403.0, + 1390.0, + 1403.0, + 1432.0, + 978.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1427.0, + 381.0, + 1427.0, + 381.0, + 1475.0, + 293.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 1427.0, + 701.0, + 1427.0, + 701.0, + 1475.0, + 616.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 1427.0, + 1063.0, + 1427.0, + 1063.0, + 1475.0, + 850.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1015.0, + 1402.0, + 1015.0, + 1402.0, + 1048.0, + 298.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1042.0, + 1402.0, + 1042.0, + 1402.0, + 1081.0, + 294.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1843.0, + 1405.0, + 1843.0, + 1405.0, + 1878.0, + 296.0, + 1878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1875.0, + 1269.0, + 1875.0, + 1269.0, + 1906.0, + 296.0, + 1906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1919.0, + 1404.0, + 1919.0, + 1404.0, + 1955.0, + 294.0, + 1955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1950.0, + 704.0, + 1950.0, + 704.0, + 1986.0, + 293.0, + 1986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1950.0, + 924.0, + 1950.0, + 924.0, + 1986.0, + 784.0, + 1986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 952.0, + 1950.0, + 1141.0, + 1950.0, + 1141.0, + 1986.0, + 952.0, + 1986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1248.0, + 1950.0, + 1406.0, + 1950.0, + 1406.0, + 1986.0, + 1248.0, + 1986.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1555, + 1403, + 1555, + 1403, + 1802, + 298, + 1802 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 297, + 1385, + 1404, + 1385, + 1404, + 1542, + 297, + 1542 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 297, + 319, + 1405, + 319, + 1405, + 492, + 297, + 492 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 297, + 1815, + 1403, + 1815, + 1403, + 1940, + 297, + 1940 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 832, + 1404, + 832, + 1404, + 959, + 298, + 959 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 299, + 989, + 1405, + 989, + 1405, + 1085, + 299, + 1085 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 303, + 706, + 1404, + 706, + 1404, + 800, + 303, + 800 + ], + "score": 0.966 + }, + { + "category_id": 1, + "poly": [ + 295, + 1211, + 1407, + 1211, + 1407, + 1276, + 295, + 1276 + ], + "score": 0.953 + }, + { + "category_id": 8, + "poly": [ + 404, + 1095, + 1288, + 1095, + 1288, + 1187, + 404, + 1187 + ], + "score": 0.952 + }, + { + "category_id": 1, + "poly": [ + 296, + 1971, + 1400, + 1971, + 1400, + 2035, + 296, + 2035 + ], + "score": 0.943 + }, + { + "category_id": 8, + "poly": [ + 515, + 271, + 1180, + 271, + 1180, + 313, + 515, + 313 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 301, + 229, + 829, + 229, + 829, + 262, + 301, + 262 + ], + "score": 0.921 + }, + { + "category_id": 0, + "poly": [ + 299, + 1318, + 545, + 1318, + 545, + 1354, + 299, + 1354 + ], + "score": 0.918 + }, + { + "category_id": 2, + "poly": [ + 299, + 74, + 813, + 74, + 813, + 106, + 299, + 106 + ], + "score": 0.915 + }, + { + "category_id": 1, + "poly": [ + 299, + 497, + 719, + 497, + 719, + 532, + 299, + 532 + ], + "score": 0.898 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1124, + 1400, + 1124, + 1400, + 1154, + 1352, + 1154 + ], + "score": 0.885 + }, + { + "category_id": 1, + "poly": [ + 353, + 554, + 1403, + 554, + 1403, + 687, + 353, + 687 + ], + "score": 0.865 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 859, + 2088, + 859, + 2111, + 840, + 2111 + ], + "score": 0.795 + }, + { + "category_id": 13, + "poly": [ + 527, + 497, + 711, + 497, + 711, + 534, + 527, + 534 + ], + "score": 0.93, + "latex": "n \\in \\mathbb { N } , l _ { T } \\in \\mathbb { N } _ { + } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 786, + 389, + 891, + 389, + 891, + 423, + 786, + 423 + ], + "score": 0.93, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )" + }, + { + "category_id": 14, + "poly": [ + 406, + 1094, + 1294, + 1094, + 1294, + 1188, + 406, + 1188 + ], + "score": 0.93, + "latex": "\\tilde { V } _ { a } ( X , V ) = U ( l _ { f } , l _ { i } ) \\left( \\theta _ { 1 } \\sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \\otimes V _ { b } + \\theta _ { 2 } \\sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \\otimes V _ { a } \\right) ." + }, + { + "category_id": 13, + "poly": [ + 527, + 926, + 631, + 926, + 631, + 958, + 527, + 958 + ], + "score": 0.93, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )" + }, + { + "category_id": 13, + "poly": [ + 431, + 392, + 504, + 392, + 504, + 430, + 431, + 430 + ], + "score": 0.93, + "latex": "W _ { l ( D ) } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 788, + 1213, + 893, + 1213, + 893, + 1247, + 788, + 1247 + ], + "score": 0.93, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )" + }, + { + "category_id": 13, + "poly": [ + 1319, + 322, + 1393, + 322, + 1393, + 362, + 1319, + 362 + ], + "score": 0.93, + "latex": "W _ { l ( D ) } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 1154, + 321, + 1260, + 321, + 1260, + 356, + 1154, + 356 + ], + "score": 0.93, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )" + }, + { + "category_id": 13, + "poly": [ + 910, + 558, + 1101, + 558, + 1101, + 594, + 910, + 594 + ], + "score": 0.93, + "latex": "p : \\mathbb { R } ^ { 3 \\times n } W _ { T } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 823, + 865, + 1014, + 865, + 1014, + 896, + 823, + 896 + ], + "score": 0.93, + "latex": "\\ell = 0 , 1 , \\ldots , D" + }, + { + "category_id": 13, + "poly": [ + 709, + 1053, + 828, + 1053, + 828, + 1085, + 709, + 1085 + ], + "score": 0.92, + "latex": "\\theta _ { 1 } , \\theta _ { 2 } \\in \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 584, + 738, + 690, + 738, + 690, + 772, + 584, + 772 + ], + "score": 0.92, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )" + }, + { + "category_id": 13, + "poly": [ + 946, + 391, + 1057, + 391, + 1057, + 423, + 946, + 423 + ], + "score": 0.92, + "latex": "\\mathcal { F } _ { \\mathrm { p o o l } } ( D )" + }, + { + "category_id": 13, + "poly": [ + 922, + 357, + 1034, + 357, + 1034, + 391, + 922, + 391 + ], + "score": 0.92, + "latex": "\\mathcal { F } _ { \\mathrm { p o o l } } ( D )" + }, + { + "category_id": 13, + "poly": [ + 542, + 390, + 594, + 390, + 594, + 427, + 542, + 427 + ], + "score": 0.92, + "latex": "W _ { l _ { T } } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 1291, + 556, + 1392, + 556, + 1392, + 600, + 1291, + 600 + ], + "score": 0.92, + "latex": "\\mathcal { F } _ { C ( D ) , D } ^ { \\mathrm { T F N } }" + }, + { + "category_id": 13, + "poly": [ + 408, + 358, + 466, + 358, + 466, + 391, + 408, + 391 + ], + "score": 0.91, + "latex": "\\boldsymbol { l } ( D )" + }, + { + "category_id": 13, + "poly": [ + 870, + 709, + 937, + 709, + 937, + 740, + 870, + 740 + ], + "score": 0.91, + "latex": "\\mathcal { F } _ { \\mathrm { p o o l } }" + }, + { + "category_id": 13, + "poly": [ + 977, + 426, + 1370, + 426, + 1370, + 465, + 977, + 465 + ], + "score": 0.91, + "latex": "\\mathcal { F } _ { C , D } ^ { \\mathrm { T F N } } = \\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } ( D ) , \\mathcal { F } _ { \\mathrm { p o o l } } ( D ) )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1906, + 517, + 1906, + 517, + 1938, + 298, + 1938 + ], + "score": 0.91, + "latex": "G = \\mathbb { R } ^ { d } \\rtimes \\breve { H } \\times \\bar { S } _ { n }" + }, + { + "category_id": 13, + "poly": [ + 567, + 644, + 749, + 644, + 749, + 687, + 567, + 687 + ], + "score": 0.91, + "latex": "\\cup _ { D \\in \\mathbb { N } _ { + } } \\mathcal { F } _ { C ( D ) , D } ^ { \\mathrm { T F N } }" + }, + { + "category_id": 13, + "poly": [ + 441, + 560, + 539, + 560, + 539, + 592, + 441, + 592 + ], + "score": 0.91, + "latex": "D \\in { \\mathbb { N } } _ { + }" + }, + { + "category_id": 13, + "poly": [ + 631, + 1906, + 696, + 1906, + 696, + 1935, + 631, + 1935 + ], + "score": 0.9, + "latex": "\\mathbb { R } ^ { d \\times n }" + }, + { + "category_id": 13, + "poly": [ + 715, + 927, + 775, + 927, + 775, + 956, + 715, + 956 + ], + "score": 0.9, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } }" + }, + { + "category_id": 14, + "poly": [ + 518, + 271, + 1178, + 271, + 1178, + 312, + 518, + 312 + ], + "score": 0.89, + "latex": "{ \\mathcal { F } } _ { \\mathrm { f e a t } } ( D ) = \\{ \\pi _ { V } \\circ f ^ { 2 D } \\circ \\ldots f ^ { 2 } \\circ f ^ { 1 } \\circ \\operatorname { e x t } ( X ) | f ^ { j } \\in { \\mathcal { F } } _ { D } \\} ," + }, + { + "category_id": 13, + "poly": [ + 1356, + 896, + 1400, + 896, + 1400, + 926, + 1356, + 926 + ], + "score": 0.87, + "latex": "\\mathcal { Q } _ { D }" + }, + { + "category_id": 13, + "poly": [ + 717, + 894, + 751, + 894, + 751, + 923, + 717, + 923 + ], + "score": 0.86, + "latex": "\\mathbb { R } ^ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 534, + 232, + 574, + 232, + 574, + 258, + 534, + 258 + ], + "score": 0.86, + "latex": "2 D" + }, + { + "category_id": 13, + "poly": [ + 465, + 327, + 503, + 327, + 503, + 352, + 465, + 352 + ], + "score": 0.86, + "latex": "\\pi _ { V }" + }, + { + "category_id": 13, + "poly": [ + 394, + 1848, + 418, + 1848, + 418, + 1875, + 394, + 1875 + ], + "score": 0.83, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 1243, + 993, + 1268, + 993, + 1268, + 1020, + 1243, + 1020 + ], + "score": 0.83, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 926, + 1215, + 952, + 1215, + 952, + 1241, + 926, + 1241 + ], + "score": 0.83, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 805, + 927, + 831, + 927, + 831, + 953, + 805, + 953 + ], + "score": 0.82, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 1182, + 1389, + 1206, + 1389, + 1206, + 1415, + 1182, + 1415 + ], + "score": 0.82, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 724, + 740, + 750, + 740, + 750, + 766, + 724, + 766 + ], + "score": 0.81, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 523, + 431, + 547, + 431, + 547, + 458, + 523, + 458 + ], + "score": 0.81, + "latex": "C" + }, + { + "category_id": 13, + "poly": [ + 298, + 770, + 324, + 770, + 324, + 796, + 298, + 796 + ], + "score": 0.81, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 1080, + 1480, + 1104, + 1480, + 1104, + 1507, + 1080, + 1507 + ], + "score": 0.81, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 651, + 896, + 678, + 896, + 678, + 923, + 651, + 923 + ], + "score": 0.81, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 551, + 1024, + 578, + 1024, + 578, + 1050, + 551, + 1050 + ], + "score": 0.8, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 691, + 359, + 718, + 359, + 718, + 386, + 691, + 386 + ], + "score": 0.8, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 783, + 1909, + 811, + 1909, + 811, + 1935, + 783, + 1935 + ], + "score": 0.79, + "latex": "H" + }, + { + "category_id": 13, + "poly": [ + 599, + 618, + 623, + 618, + 623, + 644, + 599, + 644 + ], + "score": 0.79, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 615, + 561, + 640, + 561, + 640, + 588, + 615, + 588 + ], + "score": 0.78, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 1210, + 561, + 1237, + 561, + 1237, + 587, + 1210, + 587 + ], + "score": 0.78, + "latex": "D" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1311.0, + 550.0, + 1311.0, + 550.0, + 1363.0, + 292.0, + 1363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 815.0, + 72.0, + 815.0, + 109.0, + 295.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2087.0, + 861.0, + 2087.0, + 861.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1557.0, + 1404.0, + 1557.0, + 1404.0, + 1591.0, + 296.0, + 1591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1586.0, + 1403.0, + 1586.0, + 1403.0, + 1622.0, + 293.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1617.0, + 1404.0, + 1617.0, + 1404.0, + 1652.0, + 293.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1648.0, + 1405.0, + 1648.0, + 1405.0, + 1682.0, + 293.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1680.0, + 1405.0, + 1680.0, + 1405.0, + 1713.0, + 293.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1707.0, + 1405.0, + 1707.0, + 1405.0, + 1742.0, + 293.0, + 1742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1738.0, + 1407.0, + 1738.0, + 1407.0, + 1774.0, + 293.0, + 1774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1769.0, + 1371.0, + 1769.0, + 1371.0, + 1805.0, + 294.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1386.0, + 1181.0, + 1386.0, + 1181.0, + 1423.0, + 294.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 1386.0, + 1405.0, + 1386.0, + 1405.0, + 1423.0, + 1207.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1414.0, + 1402.0, + 1414.0, + 1402.0, + 1453.0, + 295.0, + 1453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1449.0, + 1404.0, + 1449.0, + 1404.0, + 1482.0, + 296.0, + 1482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1477.0, + 1079.0, + 1477.0, + 1079.0, + 1514.0, + 295.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1105.0, + 1477.0, + 1404.0, + 1477.0, + 1404.0, + 1514.0, + 1105.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1509.0, + 996.0, + 1509.0, + 996.0, + 1545.0, + 295.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 353.0, + 407.0, + 353.0, + 407.0, + 396.0, + 292.0, + 396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 467.0, + 353.0, + 690.0, + 353.0, + 690.0, + 396.0, + 467.0, + 396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 353.0, + 921.0, + 353.0, + 921.0, + 396.0, + 719.0, + 396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 353.0, + 1406.0, + 353.0, + 1406.0, + 396.0, + 1035.0, + 396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 379.0, + 430.0, + 379.0, + 430.0, + 438.0, + 287.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 379.0, + 541.0, + 379.0, + 541.0, + 438.0, + 505.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 379.0, + 785.0, + 379.0, + 785.0, + 438.0, + 595.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 379.0, + 945.0, + 379.0, + 945.0, + 438.0, + 892.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 379.0, + 1412.0, + 379.0, + 1412.0, + 438.0, + 1058.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 282.0, + 410.0, + 522.0, + 410.0, + 522.0, + 483.0, + 282.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 410.0, + 976.0, + 410.0, + 976.0, + 483.0, + 548.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1371.0, + 410.0, + 1405.0, + 410.0, + 1405.0, + 483.0, + 1371.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 459.0, + 1020.0, + 459.0, + 1020.0, + 493.0, + 294.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.25, + 311.0, + 1409.25, + 311.0, + 1409.25, + 364.5, + 290.25, + 364.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1814.0, + 1402.0, + 1814.0, + 1402.0, + 1851.0, + 294.0, + 1851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1846.0, + 393.0, + 1846.0, + 393.0, + 1882.0, + 293.0, + 1882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 419.0, + 1846.0, + 1405.0, + 1846.0, + 1405.0, + 1882.0, + 419.0, + 1882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1877.0, + 1404.0, + 1877.0, + 1404.0, + 1914.0, + 293.0, + 1914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1902.0, + 297.0, + 1902.0, + 297.0, + 1947.0, + 292.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 1902.0, + 630.0, + 1902.0, + 630.0, + 1947.0, + 518.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1902.0, + 782.0, + 1902.0, + 782.0, + 1947.0, + 697.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 1902.0, + 1251.0, + 1902.0, + 1251.0, + 1947.0, + 812.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 834.0, + 1402.0, + 834.0, + 1402.0, + 867.0, + 296.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 865.0, + 822.0, + 865.0, + 822.0, + 898.0, + 294.0, + 898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1015.0, + 865.0, + 1405.0, + 865.0, + 1405.0, + 898.0, + 1015.0, + 898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 892.0, + 650.0, + 892.0, + 650.0, + 932.0, + 291.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 679.0, + 892.0, + 716.0, + 892.0, + 716.0, + 932.0, + 679.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 752.0, + 892.0, + 1355.0, + 892.0, + 1355.0, + 932.0, + 752.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 892.0, + 1406.0, + 892.0, + 1406.0, + 932.0, + 1401.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 925.0, + 526.0, + 925.0, + 526.0, + 962.0, + 293.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 632.0, + 925.0, + 714.0, + 925.0, + 714.0, + 962.0, + 632.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 776.0, + 925.0, + 804.0, + 925.0, + 804.0, + 962.0, + 776.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 925.0, + 1201.0, + 925.0, + 1201.0, + 962.0, + 832.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1376.0, + 929.0, + 1402.0, + 929.0, + 1402.0, + 953.0, + 1376.0, + 953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 990.0, + 1242.0, + 990.0, + 1242.0, + 1025.0, + 295.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1269.0, + 990.0, + 1405.0, + 990.0, + 1405.0, + 1025.0, + 1269.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1021.0, + 550.0, + 1021.0, + 550.0, + 1058.0, + 293.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 579.0, + 1021.0, + 1406.0, + 1021.0, + 1406.0, + 1058.0, + 579.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1052.0, + 708.0, + 1052.0, + 708.0, + 1087.0, + 293.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 1052.0, + 1405.0, + 1052.0, + 1405.0, + 1087.0, + 829.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 706.0, + 869.0, + 706.0, + 869.0, + 743.0, + 297.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 706.0, + 1405.0, + 706.0, + 1405.0, + 743.0, + 938.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 738.0, + 583.0, + 738.0, + 583.0, + 772.0, + 297.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 738.0, + 723.0, + 738.0, + 723.0, + 772.0, + 691.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 738.0, + 1404.0, + 738.0, + 1404.0, + 772.0, + 751.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 769.0, + 676.0, + 769.0, + 676.0, + 803.0, + 325.0, + 803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1210.0, + 787.0, + 1210.0, + 787.0, + 1249.0, + 294.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1210.0, + 925.0, + 1210.0, + 925.0, + 1249.0, + 894.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 953.0, + 1210.0, + 1406.0, + 1210.0, + 1406.0, + 1249.0, + 953.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1242.0, + 1062.0, + 1242.0, + 1062.0, + 1277.0, + 294.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1971.0, + 1405.0, + 1971.0, + 1405.0, + 2006.0, + 294.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2002.0, + 1400.0, + 2002.0, + 1400.0, + 2038.0, + 295.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 229.0, + 533.0, + 229.0, + 533.0, + 265.0, + 296.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 229.0, + 832.0, + 229.0, + 832.0, + 265.0, + 575.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 491.0, + 526.0, + 491.0, + 526.0, + 539.0, + 294.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 712.0, + 491.0, + 722.0, + 491.0, + 722.0, + 539.0, + 712.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 536.0, + 440.0, + 536.0, + 440.0, + 619.0, + 347.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 536.0, + 614.0, + 536.0, + 614.0, + 619.0, + 540.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 536.0, + 909.0, + 536.0, + 909.0, + 619.0, + 641.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1102.0, + 536.0, + 1209.0, + 536.0, + 1209.0, + 619.0, + 1102.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 536.0, + 1290.0, + 536.0, + 1290.0, + 619.0, + 1238.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1393.0, + 536.0, + 1415.0, + 536.0, + 1415.0, + 619.0, + 1393.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 612.0, + 566.0, + 612.0, + 566.0, + 703.0, + 357.0, + 703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 750.0, + 612.0, + 1404.0, + 612.0, + 1404.0, + 703.0, + 750.0, + 703.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 300, + 75, + 813, + 75, + 813, + 105, + 300, + 105 + ], + "score": 0.889 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2112, + 836, + 2112 + ], + "score": 0.822 + }, + { + "category_id": 0, + "poly": [ + 299, + 227, + 489, + 227, + 489, + 262, + 299, + 262 + ], + "score": 0.808 + }, + { + "category_id": 1, + "poly": [ + 295, + 279, + 1400, + 279, + 1400, + 343, + 295, + 343 + ], + "score": 0.703 + }, + { + "category_id": 1, + "poly": [ + 297, + 1124, + 1399, + 1124, + 1399, + 1189, + 297, + 1189 + ], + "score": 0.675 + }, + { + "category_id": 1, + "poly": [ + 292, + 1490, + 1401, + 1490, + 1401, + 1556, + 292, + 1556 + ], + "score": 0.675 + }, + { + "category_id": 1, + "poly": [ + 299, + 1574, + 1402, + 1574, + 1402, + 1668, + 299, + 1668 + ], + "score": 0.671 + }, + { + "category_id": 1, + "poly": [ + 296, + 1209, + 1400, + 1209, + 1400, + 1273, + 296, + 1273 + ], + "score": 0.67 + }, + { + "category_id": 1, + "poly": [ + 293, + 1406, + 1402, + 1406, + 1402, + 1472, + 293, + 1472 + ], + "score": 0.637 + }, + { + "category_id": 1, + "poly": [ + 297, + 1291, + 1402, + 1291, + 1402, + 1386, + 297, + 1386 + ], + "score": 0.628 + }, + { + "category_id": 1, + "poly": [ + 296, + 1688, + 1402, + 1688, + 1402, + 1753, + 296, + 1753 + ], + "score": 0.621 + }, + { + "category_id": 1, + "poly": [ + 296, + 1009, + 1400, + 1009, + 1400, + 1105, + 296, + 1105 + ], + "score": 0.611 + }, + { + "category_id": 1, + "poly": [ + 295, + 925, + 1395, + 925, + 1395, + 992, + 295, + 992 + ], + "score": 0.607 + }, + { + "category_id": 1, + "poly": [ + 297, + 842, + 1398, + 842, + 1398, + 907, + 297, + 907 + ], + "score": 0.602 + }, + { + "category_id": 1, + "poly": [ + 297, + 362, + 1399, + 362, + 1399, + 456, + 297, + 456 + ], + "score": 0.599 + }, + { + "category_id": 1, + "poly": [ + 294, + 560, + 1401, + 560, + 1401, + 625, + 294, + 625 + ], + "score": 0.595 + }, + { + "category_id": 1, + "poly": [ + 292, + 476, + 1402, + 476, + 1402, + 541, + 292, + 541 + ], + "score": 0.589 + }, + { + "category_id": 1, + "poly": [ + 295, + 758, + 1400, + 758, + 1400, + 823, + 295, + 823 + ], + "score": 0.585 + }, + { + "category_id": 1, + "poly": [ + 301, + 643, + 1401, + 643, + 1401, + 740, + 301, + 740 + ], + "score": 0.574 + }, + { + "category_id": 1, + "poly": [ + 296, + 1971, + 1397, + 1971, + 1397, + 2035, + 296, + 2035 + ], + "score": 0.56 + }, + { + "category_id": 1, + "poly": [ + 295, + 1773, + 1401, + 1773, + 1401, + 1837, + 295, + 1837 + ], + "score": 0.518 + }, + { + "category_id": 1, + "poly": [ + 301, + 1856, + 1398, + 1856, + 1398, + 1953, + 301, + 1953 + ], + "score": 0.503 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 230.0, + 490.0, + 230.0, + 490.0, + 263.0, + 297.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 276.0, + 1402.0, + 276.0, + 1402.0, + 315.0, + 294.0, + 315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 310.0, + 948.0, + 310.0, + 948.0, + 344.0, + 323.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1127.0, + 1401.0, + 1127.0, + 1401.0, + 1159.0, + 296.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1154.0, + 1259.0, + 1154.0, + 1259.0, + 1191.0, + 323.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1487.0, + 1406.0, + 1487.0, + 1406.0, + 1530.0, + 292.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1518.0, + 398.0, + 1518.0, + 398.0, + 1557.0, + 319.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1574.0, + 1403.0, + 1574.0, + 1403.0, + 1613.0, + 293.0, + 1613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1603.0, + 1404.0, + 1603.0, + 1404.0, + 1642.0, + 320.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1633.0, + 398.0, + 1633.0, + 398.0, + 1670.0, + 319.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1204.0, + 1405.0, + 1204.0, + 1405.0, + 1249.0, + 292.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1242.0, + 981.0, + 1242.0, + 981.0, + 1272.0, + 322.0, + 1272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1409.0, + 1403.0, + 1409.0, + 1403.0, + 1442.0, + 296.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1439.0, + 1288.0, + 1439.0, + 1288.0, + 1473.0, + 322.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1290.0, + 1404.0, + 1290.0, + 1404.0, + 1328.0, + 295.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1323.0, + 1406.0, + 1323.0, + 1406.0, + 1359.0, + 321.0, + 1359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1350.0, + 491.0, + 1350.0, + 491.0, + 1389.0, + 323.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1687.0, + 1404.0, + 1687.0, + 1404.0, + 1727.0, + 295.0, + 1727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1721.0, + 585.0, + 1721.0, + 585.0, + 1750.0, + 322.0, + 1750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1007.0, + 1405.0, + 1007.0, + 1405.0, + 1048.0, + 295.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1042.0, + 1405.0, + 1042.0, + 1405.0, + 1077.0, + 324.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1072.0, + 537.0, + 1072.0, + 537.0, + 1107.0, + 324.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 927.0, + 1399.0, + 927.0, + 1399.0, + 964.0, + 294.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 959.0, + 1235.0, + 959.0, + 1235.0, + 992.0, + 322.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 841.0, + 1402.0, + 841.0, + 1402.0, + 880.0, + 295.0, + 880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 875.0, + 612.0, + 875.0, + 612.0, + 905.0, + 323.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 363.0, + 1404.0, + 363.0, + 1404.0, + 397.0, + 297.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 392.0, + 1405.0, + 392.0, + 1405.0, + 431.0, + 321.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 425.0, + 610.0, + 425.0, + 610.0, + 454.0, + 322.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 558.0, + 1404.0, + 558.0, + 1404.0, + 599.0, + 293.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 591.0, + 603.0, + 591.0, + 603.0, + 626.0, + 322.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 472.0, + 1405.0, + 472.0, + 1405.0, + 515.0, + 292.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 506.0, + 611.0, + 506.0, + 611.0, + 542.0, + 321.0, + 542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 761.0, + 1402.0, + 761.0, + 1402.0, + 793.0, + 295.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 789.0, + 1249.0, + 789.0, + 1249.0, + 824.0, + 322.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 640.0, + 1406.0, + 640.0, + 1406.0, + 683.0, + 294.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 677.0, + 1404.0, + 677.0, + 1404.0, + 711.0, + 323.0, + 711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 706.0, + 850.0, + 706.0, + 850.0, + 739.0, + 322.0, + 739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1971.0, + 1403.0, + 1971.0, + 1403.0, + 2007.0, + 295.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 2002.0, + 1079.0, + 2002.0, + 1079.0, + 2036.0, + 321.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1770.0, + 1405.0, + 1770.0, + 1405.0, + 1811.0, + 293.0, + 1811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1805.0, + 904.0, + 1805.0, + 904.0, + 1837.0, + 323.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1855.0, + 1404.0, + 1855.0, + 1404.0, + 1895.0, + 293.0, + 1895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1885.0, + 1403.0, + 1885.0, + 1403.0, + 1925.0, + 320.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 1919.0, + 936.0, + 1919.0, + 936.0, + 1954.0, + 326.0, + 1954.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 299, + 75, + 814, + 75, + 814, + 105, + 299, + 105 + ], + "score": 0.884 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 863, + 2088, + 863, + 2113, + 836, + 2113 + ], + "score": 0.828 + }, + { + "category_id": 1, + "poly": [ + 295, + 1125, + 1402, + 1125, + 1402, + 1193, + 295, + 1193 + ], + "score": 0.672 + }, + { + "category_id": 1, + "poly": [ + 292, + 1043, + 1405, + 1043, + 1405, + 1108, + 292, + 1108 + ], + "score": 0.642 + }, + { + "category_id": 1, + "poly": [ + 293, + 961, + 1404, + 961, + 1404, + 1025, + 293, + 1025 + ], + "score": 0.619 + }, + { + "category_id": 1, + "poly": [ + 298, + 1208, + 1401, + 1208, + 1401, + 1304, + 298, + 1304 + ], + "score": 0.615 + }, + { + "category_id": 1, + "poly": [ + 296, + 878, + 1400, + 878, + 1400, + 942, + 296, + 942 + ], + "score": 0.613 + }, + { + "category_id": 1, + "poly": [ + 298, + 1548, + 1401, + 1548, + 1401, + 1643, + 298, + 1643 + ], + "score": 0.59 + }, + { + "category_id": 1, + "poly": [ + 297, + 764, + 1401, + 764, + 1401, + 859, + 297, + 859 + ], + "score": 0.589 + }, + { + "category_id": 1, + "poly": [ + 299, + 1434, + 1401, + 1434, + 1401, + 1530, + 299, + 1530 + ], + "score": 0.588 + }, + { + "category_id": 1, + "poly": [ + 295, + 650, + 1402, + 650, + 1402, + 746, + 295, + 746 + ], + "score": 0.572 + }, + { + "category_id": 1, + "poly": [ + 300, + 1321, + 1401, + 1321, + 1401, + 1417, + 300, + 1417 + ], + "score": 0.571 + }, + { + "category_id": 1, + "poly": [ + 292, + 1661, + 1402, + 1661, + 1402, + 1727, + 292, + 1727 + ], + "score": 0.545 + }, + { + "category_id": 1, + "poly": [ + 293, + 229, + 1399, + 229, + 1399, + 293, + 293, + 293 + ], + "score": 0.505 + }, + { + "category_id": 1, + "poly": [ + 298, + 537, + 1403, + 537, + 1403, + 633, + 298, + 633 + ], + "score": 0.503 + }, + { + "category_id": 1, + "poly": [ + 298, + 424, + 1402, + 424, + 1402, + 519, + 298, + 519 + ], + "score": 0.486 + }, + { + "category_id": 1, + "poly": [ + 300, + 1744, + 1400, + 1744, + 1400, + 1839, + 300, + 1839 + ], + "score": 0.47 + }, + { + "category_id": 1, + "poly": [ + 299, + 311, + 1400, + 311, + 1400, + 406, + 299, + 406 + ], + "score": 0.469 + }, + { + "category_id": 1, + "poly": [ + 290, + 1857, + 1402, + 1857, + 1402, + 1922, + 290, + 1922 + ], + "score": 0.467 + }, + { + "category_id": 1, + "poly": [ + 298, + 1941, + 1404, + 1941, + 1404, + 2034, + 298, + 2034 + ], + "score": 0.392 + }, + { + "category_id": 13, + "poly": [ + 742, + 1811, + 760, + 1811, + 760, + 1833, + 742, + 1833 + ], + "score": 0.55, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 1144, + 771, + 1179, + 771, + 1179, + 795, + 1144, + 795 + ], + "score": 0.33, + "latex": "^ { + + }" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2125.0, + 832.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1126.0, + 1404.0, + 1126.0, + 1404.0, + 1163.0, + 295.0, + 1163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 1158.0, + 1384.0, + 1158.0, + 1384.0, + 1194.0, + 327.0, + 1194.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1041.0, + 1406.0, + 1041.0, + 1406.0, + 1081.0, + 294.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1073.0, + 611.0, + 1073.0, + 611.0, + 1108.0, + 320.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 961.0, + 563.0, + 961.0, + 563.0, + 992.0, + 295.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 614.0, + 960.0, + 1173.0, + 960.0, + 1173.0, + 999.0, + 614.0, + 999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 961.0, + 1406.0, + 961.0, + 1406.0, + 998.0, + 1219.0, + 998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 992.0, + 609.0, + 992.0, + 609.0, + 1022.0, + 321.0, + 1022.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1209.0, + 1405.0, + 1209.0, + 1405.0, + 1244.0, + 294.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1240.0, + 1405.0, + 1240.0, + 1405.0, + 1274.0, + 323.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1271.0, + 862.0, + 1271.0, + 862.0, + 1305.0, + 323.0, + 1305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 878.0, + 1404.0, + 878.0, + 1404.0, + 915.0, + 294.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 909.0, + 1327.0, + 909.0, + 1327.0, + 945.0, + 324.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1549.0, + 1405.0, + 1549.0, + 1405.0, + 1583.0, + 296.0, + 1583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1580.0, + 1401.0, + 1580.0, + 1401.0, + 1615.0, + 323.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 1613.0, + 772.0, + 1613.0, + 772.0, + 1643.0, + 327.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 765.0, + 1143.0, + 765.0, + 1143.0, + 799.0, + 297.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 765.0, + 1403.0, + 765.0, + 1403.0, + 799.0, + 1180.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 793.0, + 1405.0, + 793.0, + 1405.0, + 835.0, + 321.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 829.0, + 686.0, + 829.0, + 686.0, + 859.0, + 322.0, + 859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1433.0, + 1405.0, + 1433.0, + 1405.0, + 1472.0, + 294.0, + 1472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1466.0, + 1403.0, + 1466.0, + 1403.0, + 1503.0, + 320.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1499.0, + 476.0, + 1499.0, + 476.0, + 1527.0, + 325.0, + 1527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 648.0, + 1406.0, + 648.0, + 1406.0, + 689.0, + 294.0, + 689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 682.0, + 1403.0, + 682.0, + 1403.0, + 719.0, + 322.0, + 719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 711.0, + 561.0, + 711.0, + 561.0, + 746.0, + 323.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1321.0, + 1405.0, + 1321.0, + 1405.0, + 1359.0, + 294.0, + 1359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1351.0, + 1406.0, + 1351.0, + 1406.0, + 1392.0, + 323.0, + 1392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1385.0, + 913.0, + 1385.0, + 913.0, + 1418.0, + 324.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1660.0, + 1407.0, + 1660.0, + 1407.0, + 1698.0, + 294.0, + 1698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1693.0, + 1340.0, + 1693.0, + 1340.0, + 1728.0, + 323.0, + 1728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 226.0, + 1404.0, + 226.0, + 1404.0, + 267.0, + 293.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 261.0, + 773.0, + 261.0, + 773.0, + 293.0, + 323.0, + 293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 537.0, + 1402.0, + 537.0, + 1402.0, + 574.0, + 294.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 573.0, + 1403.0, + 573.0, + 1403.0, + 603.0, + 325.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 601.0, + 749.0, + 601.0, + 749.0, + 635.0, + 326.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 420.0, + 1406.0, + 420.0, + 1406.0, + 463.0, + 293.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 458.0, + 1404.0, + 458.0, + 1404.0, + 489.0, + 323.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 489.0, + 497.0, + 489.0, + 497.0, + 519.0, + 320.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1743.0, + 1405.0, + 1743.0, + 1405.0, + 1782.0, + 294.0, + 1782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1774.0, + 1401.0, + 1774.0, + 1401.0, + 1811.0, + 320.0, + 1811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 1806.0, + 741.0, + 1806.0, + 741.0, + 1840.0, + 326.0, + 1840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 1806.0, + 938.0, + 1806.0, + 938.0, + 1840.0, + 761.0, + 1840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 310.0, + 1404.0, + 310.0, + 1404.0, + 349.0, + 294.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 345.0, + 1401.0, + 345.0, + 1401.0, + 376.0, + 324.0, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 374.0, + 777.0, + 374.0, + 777.0, + 406.0, + 323.0, + 406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1855.0, + 1405.0, + 1855.0, + 1405.0, + 1896.0, + 293.0, + 1896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1888.0, + 611.0, + 1888.0, + 611.0, + 1922.0, + 322.0, + 1922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1939.0, + 1404.0, + 1939.0, + 1404.0, + 1977.0, + 293.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1968.0, + 1405.0, + 1968.0, + 1405.0, + 2012.0, + 321.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 2001.0, + 535.0, + 2001.0, + 535.0, + 2035.0, + 322.0, + 2035.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1847, + 1404, + 1847, + 1404, + 2035, + 297, + 2035 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 718, + 1406, + 718, + 1406, + 813, + 298, + 813 + ], + "score": 0.973 + }, + { + "category_id": 8, + "poly": [ + 334, + 1339, + 1362, + 1339, + 1362, + 1599, + 334, + 1599 + ], + "score": 0.965 + }, + { + "category_id": 1, + "poly": [ + 298, + 228, + 1402, + 228, + 1402, + 323, + 298, + 323 + ], + "score": 0.96 + }, + { + "category_id": 8, + "poly": [ + 695, + 1092, + 1005, + 1092, + 1005, + 1157, + 695, + 1157 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 296, + 1184, + 1403, + 1184, + 1403, + 1249, + 296, + 1249 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 296, + 950, + 1401, + 950, + 1401, + 1011, + 296, + 1011 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 296, + 825, + 1399, + 825, + 1399, + 892, + 296, + 892 + ], + "score": 0.946 + }, + { + "category_id": 1, + "poly": [ + 296, + 443, + 1403, + 443, + 1403, + 509, + 296, + 509 + ], + "score": 0.946 + }, + { + "category_id": 1, + "poly": [ + 298, + 1684, + 1400, + 1684, + 1400, + 1747, + 298, + 1747 + ], + "score": 0.944 + }, + { + "category_id": 1, + "poly": [ + 300, + 1262, + 1401, + 1262, + 1401, + 1329, + 300, + 1329 + ], + "score": 0.943 + }, + { + "category_id": 1, + "poly": [ + 296, + 1016, + 1401, + 1016, + 1401, + 1082, + 296, + 1082 + ], + "score": 0.94 + }, + { + "category_id": 8, + "poly": [ + 692, + 515, + 1006, + 515, + 1006, + 556, + 692, + 556 + ], + "score": 0.939 + }, + { + "category_id": 8, + "poly": [ + 643, + 901, + 1055, + 901, + 1055, + 943, + 643, + 943 + ], + "score": 0.938 + }, + { + "category_id": 1, + "poly": [ + 295, + 1752, + 1397, + 1752, + 1397, + 1818, + 295, + 1818 + ], + "score": 0.93 + }, + { + "category_id": 0, + "poly": [ + 301, + 593, + 698, + 593, + 698, + 630, + 301, + 630 + ], + "score": 0.918 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1109, + 1400, + 1109, + 1400, + 1140, + 1352, + 1140 + ], + "score": 0.908 + }, + { + "category_id": 0, + "poly": [ + 298, + 662, + 889, + 662, + 889, + 694, + 298, + 694 + ], + "score": 0.908 + }, + { + "category_id": 2, + "poly": [ + 299, + 74, + 814, + 74, + 814, + 105, + 299, + 105 + ], + "score": 0.897 + }, + { + "category_id": 0, + "poly": [ + 300, + 376, + 509, + 376, + 509, + 413, + 300, + 413 + ], + "score": 0.896 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2113, + 836, + 2113 + ], + "score": 0.861 + }, + { + "category_id": 2, + "poly": [ + 1375, + 1619, + 1402, + 1619, + 1402, + 1649, + 1375, + 1649 + ], + "score": 0.802 + }, + { + "category_id": 14, + "poly": [ + 332, + 1337, + 1369, + 1337, + 1369, + 1599, + 332, + 1599 + ], + "score": 0.94, + "latex": "\\begin{array} { l } { f \\left( \\rho _ { G } ( t , R , P ) ( X ) \\right) = f ( R ( X + t \\boldsymbol { 1 } _ { n } ) P ^ { T } ) = h ( R ( X + t \\boldsymbol { 1 } _ { n } ) P ^ { T } - \\displaystyle \\frac { 1 } { n } R ( X + t \\boldsymbol { 1 } ) P ^ { T } \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) } \\\\ { \\displaystyle \\quad \\quad = h ( R ( X - \\frac { 1 } { n } X \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) P ^ { T } ) = h \\left( \\rho _ { \\mathbb { R } ^ { 3 } \\times S _ { n } } ( R , P ) ( X - \\frac { 1 } { n } X \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) \\right) } \\\\ { \\displaystyle \\quad = \\rho _ { W _ { T } \\times S _ { n } } ( R , P ) h \\left( X - \\frac { 1 } { n } X \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } \\right) } \\\\ { \\displaystyle \\quad = \\rho _ { G } ( t , R , P ) f ( X ) . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 694, + 1090, + 1007, + 1090, + 1007, + 1156, + 694, + 1156 + ], + "score": 0.94, + "latex": "f ( X ) = h ( X - \\frac 1 n X 1 _ { n } 1 _ { n } ^ { T } )" + }, + { + "category_id": 13, + "poly": [ + 298, + 474, + 464, + 474, + 464, + 512, + 298, + 512 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { X - \\frac { 1 } { n } X \\mathbb { 1 } _ { n } \\mathbb { 1 } _ { n } ^ { T } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 568, + 1848, + 717, + 1848, + 717, + 1882, + 568, + 1882 + ], + "score": 0.92, + "latex": "K ~ \\subseteq ~ \\mathbb { R } ^ { 3 \\times n }" + }, + { + "category_id": 13, + "poly": [ + 720, + 1292, + 1055, + 1292, + 1055, + 1327, + 720, + 1327 + ], + "score": 0.92, + "latex": "P ^ { T } 1 _ { n } 1 _ { n } ^ { T } = 1 _ { n } 1 _ { n } ^ { T } = 1 _ { n } 1 _ { n } ^ { T } P ^ { T }" + }, + { + "category_id": 13, + "poly": [ + 1153, + 1909, + 1393, + 1909, + 1393, + 1948, + 1153, + 1948 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\dot { X } \\stackrel { \\bullet } { \\mapsto } X - \\frac { 1 } { n } X 1 _ { n } 1 _ { n } ^ { T } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 836, + 477, + 979, + 477, + 979, + 510, + 836, + 510 + ], + "score": 0.92, + "latex": "\\left( \\hat { x } _ { 1 } , \\ldots , \\hat { x } _ { n } \\right)" + }, + { + "category_id": 14, + "poly": [ + 642, + 900, + 1058, + 900, + 1058, + 942, + 642, + 942 + ], + "score": 0.92, + "latex": "\\rho _ { W _ { T } \\times S _ { n } } ( R , P ) ( Y ) = \\rho _ { W _ { T } } ( R ) Y P ^ { T }" + }, + { + "category_id": 13, + "poly": [ + 759, + 1186, + 894, + 1186, + 894, + 1219, + 759, + 1219 + ], + "score": 0.91, + "latex": "\\mathrm { S O } ( 3 ) \\times S _ { n }" + }, + { + "category_id": 13, + "poly": [ + 580, + 1015, + 776, + 1015, + 776, + 1053, + 580, + 1053 + ], + "score": 0.91, + "latex": "f : \\mathbb { R } ^ { 3 \\times n } W _ { T } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1293, + 639, + 1293, + 639, + 1328, + 297, + 1328 + ], + "score": 0.91, + "latex": "( t , R , P ) \\in \\mathbb { R } ^ { d } \\rtimes \\mathrm { S O ( 3 ) } \\times S _ { n }" + }, + { + "category_id": 13, + "poly": [ + 845, + 1049, + 984, + 1049, + 984, + 1084, + 845, + 1084 + ], + "score": 0.91, + "latex": "\\mathrm { S O } ( 3 ) \\times S _ { n }" + }, + { + "category_id": 13, + "poly": [ + 1024, + 1048, + 1089, + 1048, + 1089, + 1077, + 1024, + 1077 + ], + "score": 0.91, + "latex": "\\mathbb { R } ^ { 3 \\times n }" + }, + { + "category_id": 13, + "poly": [ + 887, + 1974, + 1028, + 1974, + 1028, + 2006, + 887, + 2006 + ], + "score": 0.91, + "latex": "\\mathrm { S O } ( 3 ) \\times S _ { n }" + }, + { + "category_id": 13, + "poly": [ + 678, + 858, + 817, + 858, + 817, + 892, + 678, + 892 + ], + "score": 0.91, + "latex": "\\mathrm { S O } ( 3 ) \\times S _ { n }" + }, + { + "category_id": 13, + "poly": [ + 846, + 1219, + 917, + 1219, + 917, + 1249, + 846, + 1249 + ], + "score": 0.9, + "latex": "h = f" + }, + { + "category_id": 13, + "poly": [ + 857, + 861, + 903, + 861, + 903, + 893, + 857, + 893 + ], + "score": 0.9, + "latex": "W _ { T } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 544, + 865, + 641, + 865, + 641, + 892, + 544, + 892 + ], + "score": 0.9, + "latex": "\\rho _ { W _ { T } \\times S _ { n } }" + }, + { + "category_id": 14, + "poly": [ + 691, + 516, + 1009, + 516, + 1009, + 556, + 691, + 556 + ], + "score": 0.9, + "latex": "\\Sigma _ { T } = \\{ \\vec { r } \\in \\mathbb { N } _ { + } ^ { * } | \\| \\vec { r } \\| _ { 1 } = T \\}" + }, + { + "category_id": 13, + "poly": [ + 841, + 1943, + 977, + 1943, + 977, + 1974, + 841, + 1974 + ], + "score": 0.9, + "latex": "\\mathrm { S O } ( 3 ) \\times S _ { n }" + }, + { + "category_id": 13, + "poly": [ + 1347, + 828, + 1393, + 828, + 1393, + 860, + 1347, + 860 + ], + "score": 0.89, + "latex": "W _ { T }" + }, + { + "category_id": 13, + "poly": [ + 646, + 1974, + 684, + 1974, + 684, + 2003, + 646, + 2003 + ], + "score": 0.88, + "latex": "K _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 926, + 981, + 991, + 981, + 991, + 1007, + 926, + 1007 + ], + "score": 0.88, + "latex": "\\mathbb { R } ^ { 3 \\times n }" + }, + { + "category_id": 13, + "poly": [ + 853, + 956, + 1029, + 956, + 1029, + 982, + 853, + 982 + ], + "score": 0.87, + "latex": "\\rho _ { G } ~ = ~ \\rho _ { \\underline { { W } } _ { T } \\times S _ { n } }" + }, + { + "category_id": 13, + "poly": [ + 486, + 834, + 539, + 834, + 539, + 861, + 486, + 861 + ], + "score": 0.87, + "latex": "\\rho _ { W _ { T } }" + }, + { + "category_id": 13, + "poly": [ + 768, + 474, + 796, + 474, + 796, + 504, + 768, + 504 + ], + "score": 0.86, + "latex": "\\bar { X }" + }, + { + "category_id": 13, + "poly": [ + 1308, + 1977, + 1338, + 1977, + 1338, + 2005, + 1308, + 2005 + ], + "score": 0.86, + "latex": "p _ { k }" + }, + { + "category_id": 13, + "poly": [ + 322, + 1218, + 341, + 1218, + 341, + 1249, + 322, + 1249 + ], + "score": 0.85, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 505, + 1188, + 530, + 1188, + 530, + 1214, + 505, + 1214 + ], + "score": 0.85, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 1362, + 1882, + 1400, + 1882, + 1400, + 1911, + 1362, + 1911 + ], + "score": 0.85, + "latex": "K _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 554, + 1265, + 574, + 1265, + 574, + 1295, + 554, + 1295 + ], + "score": 0.84, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 609, + 1687, + 633, + 1687, + 633, + 1713, + 609, + 1713 + ], + "score": 0.84, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 1078, + 1687, + 1102, + 1687, + 1102, + 1713, + 1078, + 1713 + ], + "score": 0.84, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 812, + 1913, + 841, + 1913, + 841, + 1939, + 812, + 1939 + ], + "score": 0.83, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 1053, + 1265, + 1077, + 1265, + 1077, + 1291, + 1053, + 1291 + ], + "score": 0.83, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 1341, + 443, + 1405, + 443, + 1405, + 477, + 1341, + 477 + ], + "score": 0.83, + "latex": "\\bar { X } =" + }, + { + "category_id": 13, + "poly": [ + 370, + 1218, + 395, + 1218, + 395, + 1244, + 370, + 1244 + ], + "score": 0.83, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 702, + 721, + 727, + 721, + 727, + 747, + 702, + 747 + ], + "score": 0.83, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 1337, + 1946, + 1356, + 1946, + 1356, + 1974, + 1337, + 1974 + ], + "score": 0.83, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 592, + 951, + 617, + 951, + 617, + 977, + 592, + 977 + ], + "score": 0.83, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 920, + 1753, + 1097, + 1753, + 1097, + 1785, + 920, + 1785 + ], + "score": 0.82, + "latex": "\\mathcal { C } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , W _ { T } ^ { n } )" + }, + { + "category_id": 13, + "poly": [ + 929, + 1883, + 958, + 1883, + 958, + 1909, + 929, + 1909 + ], + "score": 0.82, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 999, + 1882, + 1023, + 1882, + 1023, + 1910, + 999, + 1910 + ], + "score": 0.82, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 1367, + 1852, + 1391, + 1852, + 1391, + 1878, + 1367, + 1878 + ], + "score": 0.81, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 807, + 1019, + 832, + 1019, + 832, + 1046, + 807, + 1046 + ], + "score": 0.81, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 593, + 1786, + 618, + 1786, + 618, + 1812, + 593, + 1812 + ], + "score": 0.8, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 619, + 1756, + 644, + 1756, + 644, + 1782, + 619, + 1782 + ], + "score": 0.8, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 863, + 982, + 887, + 982, + 887, + 1007, + 863, + 1007 + ], + "score": 0.8, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 928, + 1785, + 1110, + 1785, + 1110, + 1819, + 928, + 1819 + ], + "score": 0.78, + "latex": "\\mathcal { P } _ { G } ( \\mathbb { R } ^ { 3 \\times n } , \\dot { W } _ { T } ^ { n } )" + }, + { + "category_id": 13, + "poly": [ + 368, + 663, + 394, + 663, + 394, + 691, + 368, + 691 + ], + "score": 0.75, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 1382, + 1020, + 1401, + 1020, + 1401, + 1046, + 1382, + 1046 + ], + "score": 0.68, + "latex": "h" + }, + { + "category_id": 13, + "poly": [ + 825, + 827, + 898, + 827, + 898, + 860, + 825, + 860 + ], + "score": 0.29, + "latex": "\\mathrm { S O ( 3 ) }" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 591.0, + 700.0, + 591.0, + 700.0, + 635.0, + 294.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 662.0, + 367.0, + 662.0, + 367.0, + 698.0, + 295.0, + 698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 662.0, + 891.0, + 662.0, + 891.0, + 698.0, + 395.0, + 698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 815.0, + 72.0, + 815.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 371.0, + 514.0, + 371.0, + 514.0, + 421.0, + 293.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 1624.0, + 1402.0, + 1624.0, + 1402.0, + 1651.0, + 1378.0, + 1651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1845.0, + 567.0, + 1845.0, + 567.0, + 1886.0, + 293.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 1845.0, + 1366.0, + 1845.0, + 1366.0, + 1886.0, + 718.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1392.0, + 1845.0, + 1405.0, + 1845.0, + 1405.0, + 1886.0, + 1392.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 283.0, + 1878.0, + 811.0, + 1878.0, + 811.0, + 1978.0, + 283.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 1878.0, + 998.0, + 1878.0, + 998.0, + 1978.0, + 978.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1024.0, + 1878.0, + 1152.0, + 1878.0, + 1152.0, + 1978.0, + 1024.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 1878.0, + 1418.0, + 1878.0, + 1418.0, + 1978.0, + 1401.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1972.0, + 645.0, + 1972.0, + 645.0, + 2008.0, + 294.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 1972.0, + 886.0, + 1972.0, + 886.0, + 2008.0, + 685.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1029.0, + 1972.0, + 1307.0, + 1972.0, + 1307.0, + 2008.0, + 1029.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1339.0, + 1972.0, + 1406.0, + 1972.0, + 1406.0, + 2008.0, + 1339.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2003.0, + 976.0, + 2003.0, + 976.0, + 2039.0, + 295.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 719.0, + 701.0, + 719.0, + 701.0, + 753.0, + 295.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 728.0, + 719.0, + 1406.0, + 719.0, + 1406.0, + 753.0, + 728.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 750.0, + 1404.0, + 750.0, + 1404.0, + 783.0, + 293.0, + 783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 780.0, + 730.0, + 780.0, + 730.0, + 814.0, + 295.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 228.0, + 1404.0, + 228.0, + 1404.0, + 266.0, + 297.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 260.0, + 1404.0, + 260.0, + 1404.0, + 296.0, + 322.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 290.0, + 612.0, + 290.0, + 612.0, + 324.0, + 322.0, + 324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1184.0, + 504.0, + 1184.0, + 504.0, + 1220.0, + 295.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 1184.0, + 758.0, + 1184.0, + 758.0, + 1220.0, + 531.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1184.0, + 1405.0, + 1184.0, + 1405.0, + 1220.0, + 895.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1215.0, + 321.0, + 1215.0, + 321.0, + 1251.0, + 294.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 1215.0, + 369.0, + 1215.0, + 369.0, + 1251.0, + 342.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 1215.0, + 845.0, + 1215.0, + 845.0, + 1251.0, + 396.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1215.0, + 930.0, + 1215.0, + 930.0, + 1251.0, + 918.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 944.0, + 591.0, + 944.0, + 591.0, + 987.0, + 294.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 618.0, + 944.0, + 852.0, + 944.0, + 852.0, + 987.0, + 618.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 944.0, + 1405.0, + 944.0, + 1405.0, + 987.0, + 1030.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 973.0, + 862.0, + 973.0, + 862.0, + 1016.0, + 291.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 888.0, + 973.0, + 925.0, + 973.0, + 925.0, + 1016.0, + 888.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 973.0, + 1271.0, + 973.0, + 1271.0, + 1016.0, + 992.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 822.0, + 485.0, + 822.0, + 485.0, + 865.0, + 291.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 822.0, + 824.0, + 822.0, + 824.0, + 865.0, + 540.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 822.0, + 1346.0, + 822.0, + 1346.0, + 865.0, + 899.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 822.0, + 1404.0, + 822.0, + 1404.0, + 865.0, + 1394.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 856.0, + 543.0, + 856.0, + 543.0, + 896.0, + 294.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 642.0, + 856.0, + 677.0, + 856.0, + 677.0, + 896.0, + 642.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 818.0, + 856.0, + 856.0, + 856.0, + 856.0, + 896.0, + 818.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 856.0, + 943.0, + 856.0, + 943.0, + 896.0, + 904.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 441.0, + 1340.0, + 441.0, + 1340.0, + 479.0, + 294.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 472.0, + 297.0, + 472.0, + 297.0, + 512.0, + 294.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 465.0, + 472.0, + 767.0, + 472.0, + 767.0, + 512.0, + 465.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 797.0, + 472.0, + 835.0, + 472.0, + 835.0, + 512.0, + 797.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 472.0, + 1117.0, + 472.0, + 1117.0, + 512.0, + 980.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1682.0, + 608.0, + 1682.0, + 608.0, + 1722.0, + 296.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 634.0, + 1682.0, + 1077.0, + 1682.0, + 1077.0, + 1722.0, + 634.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 1682.0, + 1404.0, + 1682.0, + 1404.0, + 1722.0, + 1103.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1716.0, + 495.0, + 1716.0, + 495.0, + 1749.0, + 296.0, + 1749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1259.0, + 553.0, + 1259.0, + 553.0, + 1297.0, + 294.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 1259.0, + 1052.0, + 1259.0, + 1052.0, + 1297.0, + 575.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1078.0, + 1259.0, + 1405.0, + 1259.0, + 1405.0, + 1297.0, + 1078.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1279.0, + 296.0, + 1279.0, + 296.0, + 1340.0, + 289.0, + 1340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 640.0, + 1279.0, + 719.0, + 1279.0, + 719.0, + 1340.0, + 640.0, + 1340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 1279.0, + 1067.0, + 1279.0, + 1067.0, + 1340.0, + 1056.0, + 1340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1010.0, + 579.0, + 1010.0, + 579.0, + 1057.0, + 291.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 1010.0, + 806.0, + 1010.0, + 806.0, + 1057.0, + 777.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 833.0, + 1010.0, + 1381.0, + 1010.0, + 1381.0, + 1057.0, + 833.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1010.0, + 1408.0, + 1010.0, + 1408.0, + 1057.0, + 1402.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1043.0, + 844.0, + 1043.0, + 844.0, + 1087.0, + 294.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 1043.0, + 1023.0, + 1043.0, + 1023.0, + 1087.0, + 985.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 1043.0, + 1155.0, + 1043.0, + 1155.0, + 1087.0, + 1090.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1745.0, + 618.0, + 1745.0, + 618.0, + 1793.0, + 291.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 645.0, + 1745.0, + 919.0, + 1745.0, + 919.0, + 1793.0, + 645.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1098.0, + 1745.0, + 1404.0, + 1745.0, + 1404.0, + 1793.0, + 1098.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1779.0, + 592.0, + 1779.0, + 592.0, + 1824.0, + 289.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 619.0, + 1779.0, + 927.0, + 1779.0, + 927.0, + 1824.0, + 619.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1111.0, + 1779.0, + 1124.0, + 1779.0, + 1124.0, + 1824.0, + 1111.0, + 1824.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1443, + 1404, + 1443, + 1404, + 1599, + 297, + 1599 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 1611, + 1402, + 1611, + 1402, + 1785, + 297, + 1785 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 665, + 1405, + 665, + 1405, + 772, + 297, + 772 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 296, + 227, + 1404, + 227, + 1404, + 355, + 296, + 355 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 297, + 1139, + 1404, + 1139, + 1404, + 1236, + 297, + 1236 + ], + "score": 0.972 + }, + { + "category_id": 8, + "poly": [ + 343, + 1796, + 1352, + 1796, + 1352, + 1986, + 343, + 1986 + ], + "score": 0.971 + }, + { + "category_id": 8, + "poly": [ + 308, + 857, + 1394, + 857, + 1394, + 1011, + 308, + 1011 + ], + "score": 0.97 + }, + { + "category_id": 8, + "poly": [ + 705, + 1344, + 990, + 1344, + 990, + 1433, + 705, + 1433 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 294, + 783, + 1401, + 783, + 1401, + 848, + 294, + 848 + ], + "score": 0.954 + }, + { + "category_id": 1, + "poly": [ + 297, + 502, + 1401, + 502, + 1401, + 592, + 297, + 592 + ], + "score": 0.952 + }, + { + "category_id": 1, + "poly": [ + 295, + 384, + 1405, + 384, + 1405, + 448, + 295, + 448 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 297, + 1267, + 1402, + 1267, + 1402, + 1331, + 297, + 1331 + ], + "score": 0.945 + }, + { + "category_id": 8, + "poly": [ + 741, + 456, + 955, + 456, + 955, + 495, + 741, + 495 + ], + "score": 0.936 + }, + { + "category_id": 0, + "poly": [ + 299, + 1083, + 637, + 1083, + 637, + 1115, + 299, + 1115 + ], + "score": 0.923 + }, + { + "category_id": 2, + "poly": [ + 299, + 74, + 813, + 74, + 813, + 105, + 299, + 105 + ], + "score": 0.914 + }, + { + "category_id": 1, + "poly": [ + 326, + 2001, + 1166, + 2001, + 1166, + 2038, + 326, + 2038 + ], + "score": 0.91 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1370, + 1400, + 1370, + 1400, + 1402, + 1351, + 1402 + ], + "score": 0.907 + }, + { + "category_id": 8, + "poly": [ + 620, + 594, + 1076, + 594, + 1076, + 661, + 620, + 661 + ], + "score": 0.875 + }, + { + "category_id": 2, + "poly": [ + 836, + 2087, + 865, + 2087, + 865, + 2113, + 836, + 2113 + ], + "score": 0.872 + }, + { + "category_id": 2, + "poly": [ + 1374, + 1017, + 1403, + 1017, + 1403, + 1046, + 1374, + 1046 + ], + "score": 0.777 + }, + { + "category_id": 8, + "poly": [ + 622, + 594, + 1076, + 594, + 1076, + 661, + 622, + 661 + ], + "score": 0.221 + }, + { + "category_id": 14, + "poly": [ + 706, + 1339, + 991, + 1339, + 991, + 1435, + 706, + 1435 + ], + "score": 0.95, + "latex": "p ( X ) = \\sum _ { k = 1 } ^ { K } \\hat { \\Lambda } _ { k } ( f _ { k } ( X ) ) ," + }, + { + "category_id": 14, + "poly": [ + 623, + 590, + 1076, + 590, + 1076, + 663, + 623, + 663 + ], + "score": 0.94, + "latex": "\\langle p \\rangle ( X ) = \\int _ { G } \\rho _ { 2 } ( g ^ { - 1 } ) p ( \\rho _ { 1 } ( g ) X ) d \\mu ( g ) ." + }, + { + "category_id": 13, + "poly": [ + 732, + 1705, + 894, + 1705, + 894, + 1748, + 732, + 1748 + ], + "score": 0.94, + "latex": "f _ { k } = ( f _ { k } ^ { j } ) _ { j = 1 } ^ { n }" + }, + { + "category_id": 14, + "poly": [ + 345, + 1792, + 1353, + 1792, + 1353, + 1988, + 345, + 1988 + ], + "score": 0.94, + "latex": "\\begin{array} { l } { { \\displaystyle p _ { j } ( X ) = \\int _ { \\mathrm { S O } ( 3 ) } \\rho _ { 2 } ( R ^ { - 1 } ) \\circ p _ { j } ( R X ) d \\nu ( R ) = \\sum _ { k = 1 } ^ { K } \\int _ { \\mathrm { S O } ( 3 ) } \\rho _ { 2 } ( R ^ { - 1 } ) \\circ \\Lambda _ { k } \\circ f _ { j } ^ { k } ( R X ) d \\nu ( R ) } } \\\\ { { \\displaystyle \\qquad = \\sum _ { k = 1 } ^ { K } \\int _ { \\mathrm { S O } ( 3 ) } \\rho _ { 2 } ( R ^ { - 1 } ) \\circ \\Lambda _ { k } \\left( \\rho _ { 1 } ( R ) \\circ f _ { k } ^ { j } ( X ) \\right) d \\nu ( R ) = \\sum _ { k = 1 } ^ { K } \\tilde { \\Lambda } _ { k } \\circ f _ { k } ^ { j } ( X ) , } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 836, + 697, + 1170, + 697, + 1170, + 740, + 836, + 740 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\sum _ { j = 1 } ^ { N } w _ { j } \\rho _ { 2 } ( g _ { j } ^ { - 1 } ) p ( \\rho _ { 1 } ( g _ { j } ) X ) } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 310, + 856, + 1394, + 856, + 1394, + 1013, + 310, + 1013 + ], + "score": 0.93, + "latex": "\\begin{array} { l } { \\displaystyle { | \\langle p _ { k } \\rangle ( X ) - f ( X ) | = \\left| \\int _ { G } \\rho _ { 2 } ( g ^ { - 1 } ) p _ { k } ( \\rho _ { 1 } ( g ) X ) - \\rho _ { 2 } ( g ^ { - 1 } ) f ( \\rho _ { 1 } ( g ) X ) d \\mu ( g ) \\right| } } \\\\ { \\displaystyle { \\phantom { \\rho _ { k } ( \\rho _ { 1 } ) ( } = \\left| \\int _ { G } \\rho _ { 2 } ( g ^ { - 1 } ) \\left[ p _ { k } ( \\rho _ { 1 } ( g ) X ) - f ( \\rho _ { 1 } ( g ) X ) \\right] d \\mu ( g ) \\right| \\leq N \\| f - p _ { k } \\| _ { C ( K _ { 1 } ) } \\to 0 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 655, + 2003, + 937, + 2003, + 937, + 2036, + 655, + 2036 + ], + "score": 0.93, + "latex": "( g , X ) \\mapsto \\rho _ { j } ( g ) X , j = 1 , 2" + }, + { + "category_id": 13, + "poly": [ + 687, + 1506, + 857, + 1506, + 857, + 1542, + 687, + 1542 + ], + "score": 0.93, + "latex": "p _ { i } = ( p _ { i j } ) _ { j \\in [ n ] }" + }, + { + "category_id": 13, + "poly": [ + 359, + 1506, + 571, + 1506, + 571, + 1542, + 359, + 1542 + ], + "score": 0.93, + "latex": "p = ( p _ { i j } ) _ { i \\in [ t ] , j \\in [ n ] }" + }, + { + "category_id": 13, + "poly": [ + 1130, + 533, + 1289, + 533, + 1289, + 564, + 1130, + 564 + ], + "score": 0.93, + "latex": "p : W _ { 1 } \\to W _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 529, + 1710, + 676, + 1710, + 676, + 1748, + 529, + 1748 + ], + "score": 0.93, + "latex": "p = ( p _ { j } ) _ { j = 1 } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 868, + 1269, + 1087, + 1269, + 1087, + 1301, + 868, + 1301 + ], + "score": 0.92, + "latex": "f _ { 1 } , \\dots , f _ { K } \\in \\mathcal { F } _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 855, + 783, + 1015, + 783, + 1015, + 819, + 855, + 819 + ], + "score": 0.92, + "latex": "g \\mapsto \\rho _ { 2 } ( g ^ { - 1 } )" + }, + { + "category_id": 13, + "poly": [ + 525, + 1173, + 665, + 1173, + 665, + 1206, + 525, + 1206 + ], + "score": 0.92, + "latex": "C \\geq C ( D )" + }, + { + "category_id": 13, + "poly": [ + 297, + 815, + 392, + 815, + 392, + 849, + 297, + 849 + ], + "score": 0.92, + "latex": "\\rho _ { 2 } ( g ^ { - 1 } )" + }, + { + "category_id": 13, + "poly": [ + 885, + 417, + 969, + 417, + 969, + 444, + 885, + 444 + ], + "score": 0.92, + "latex": "G \\times K" + }, + { + "category_id": 13, + "poly": [ + 750, + 1143, + 816, + 1143, + 816, + 1175, + 750, + 1175 + ], + "score": 0.91, + "latex": "\\mathcal { F } _ { \\mathrm { p o o l } }" + }, + { + "category_id": 13, + "poly": [ + 599, + 262, + 640, + 262, + 640, + 292, + 599, + 292 + ], + "score": 0.91, + "latex": "W _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 705, + 262, + 816, + 262, + 816, + 293, + 705, + 293 + ], + "score": 0.91, + "latex": "K \\subseteq W _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1030, + 1711, + 1181, + 1711, + 1181, + 1743, + 1030, + 1743 + ], + "score": 0.91, + "latex": "j = 1 , \\dotsc , n" + }, + { + "category_id": 13, + "poly": [ + 456, + 665, + 493, + 665, + 493, + 699, + 456, + 699 + ], + "score": 0.91, + "latex": "\\langle p \\rangle" + }, + { + "category_id": 13, + "poly": [ + 1305, + 1142, + 1404, + 1142, + 1404, + 1175, + 1305, + 1175 + ], + "score": 0.9, + "latex": "C ( D ) \\in" + }, + { + "category_id": 13, + "poly": [ + 892, + 1172, + 1086, + 1172, + 1086, + 1206, + 892, + 1206 + ], + "score": 0.9, + "latex": "\\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } , \\mathcal { F } _ { \\mathrm { p o o l } } )" + }, + { + "category_id": 13, + "poly": [ + 431, + 1298, + 611, + 1298, + 611, + 1331, + 431, + 1331 + ], + "score": 0.9, + "latex": "p : \\mathbb { R } ^ { 3 \\bar { \\times } n } \\bar { \\mathbb { R } } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 1298, + 1442, + 1365, + 1442, + 1365, + 1474, + 1298, + 1474 + ], + "score": 0.9, + "latex": "\\mathbb { R } ^ { 3 \\times n }" + }, + { + "category_id": 13, + "poly": [ + 992, + 740, + 1072, + 740, + 1072, + 772, + 992, + 772 + ], + "score": 0.9, + "latex": "C ( K _ { 1 } )" + }, + { + "category_id": 13, + "poly": [ + 395, + 1611, + 594, + 1611, + 594, + 1647, + 395, + 1647 + ], + "score": 0.9, + "latex": "p : \\mathbb { R } ^ { 3 \\times n } W _ { T } ^ { n }" + }, + { + "category_id": 14, + "poly": [ + 742, + 456, + 956, + 456, + 956, + 494, + 742, + 494 + ], + "score": 0.9, + "latex": "( g , X ) \\mapsto \\rho _ { 1 } ( g ) X ." + }, + { + "category_id": 13, + "poly": [ + 761, + 535, + 798, + 535, + 798, + 564, + 761, + 564 + ], + "score": 0.9, + "latex": "K _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 554, + 1747, + 584, + 1747, + 584, + 1785, + 554, + 1785 + ], + "score": 0.9, + "latex": "f _ { k } ^ { j }" + }, + { + "category_id": 13, + "poly": [ + 298, + 292, + 479, + 292, + 479, + 323, + 298, + 323 + ], + "score": 0.9, + "latex": "f : W _ { 1 } \\to W _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 297, + 323, + 470, + 323, + 470, + 354, + 297, + 354 + ], + "score": 0.89, + "latex": "p _ { k } : W _ { 1 } \\mapsto W _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1070, + 666, + 1106, + 666, + 1106, + 698, + 1070, + 698 + ], + "score": 0.89, + "latex": "\\langle p \\rangle" + }, + { + "category_id": 13, + "poly": [ + 1235, + 387, + 1274, + 387, + 1274, + 416, + 1235, + 416 + ], + "score": 0.89, + "latex": "K _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 438, + 786, + 543, + 786, + 543, + 816, + 438, + 816 + ], + "score": 0.89, + "latex": "X \\ \\in \\ K _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1202, + 1675, + 1248, + 1675, + 1248, + 1706, + 1202, + 1706 + ], + "score": 0.89, + "latex": "W _ { T }" + }, + { + "category_id": 13, + "poly": [ + 842, + 818, + 919, + 818, + 919, + 844, + 842, + 844 + ], + "score": 0.89, + "latex": "N > 0" + }, + { + "category_id": 13, + "poly": [ + 417, + 703, + 456, + 703, + 456, + 734, + 417, + 734 + ], + "score": 0.89, + "latex": "K _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 899, + 1567, + 934, + 1567, + 934, + 1597, + 899, + 1597 + ], + "score": 0.89, + "latex": "\\Lambda _ { k }" + }, + { + "category_id": 13, + "poly": [ + 471, + 1143, + 530, + 1143, + 530, + 1172, + 471, + 1172 + ], + "score": 0.88, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1476, + 452, + 1476, + 452, + 1509, + 298, + 1509 + ], + "score": 0.88, + "latex": "W _ { T } ^ { n } = \\mathbb { R } ^ { t \\times n }" + }, + { + "category_id": 13, + "poly": [ + 1081, + 1675, + 1149, + 1675, + 1149, + 1706, + 1081, + 1706 + ], + "score": 0.88, + "latex": "W _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 1018, + 1566, + 1050, + 1566, + 1050, + 1594, + 1018, + 1594 + ], + "score": 0.88, + "latex": "\\mathbb { R } ^ { t }" + }, + { + "category_id": 13, + "poly": [ + 501, + 262, + 543, + 262, + 543, + 292, + 501, + 292 + ], + "score": 0.88, + "latex": "W _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1064, + 1614, + 1122, + 1614, + 1122, + 1644, + 1064, + 1644 + ], + "score": 0.87, + "latex": "\\le D" + }, + { + "category_id": 13, + "poly": [ + 379, + 1447, + 415, + 1447, + 415, + 1475, + 379, + 1475 + ], + "score": 0.87, + "latex": "\\Lambda _ { k }" + }, + { + "category_id": 13, + "poly": [ + 469, + 1647, + 505, + 1647, + 505, + 1675, + 469, + 1675 + ], + "score": 0.87, + "latex": "\\Lambda _ { k }" + }, + { + "category_id": 13, + "poly": [ + 759, + 508, + 790, + 508, + 790, + 534, + 759, + 534 + ], + "score": 0.86, + "latex": "p _ { k }" + }, + { + "category_id": 13, + "poly": [ + 474, + 1754, + 502, + 1754, + 502, + 1784, + 474, + 1784 + ], + "score": 0.86, + "latex": "p _ { j }" + }, + { + "category_id": 13, + "poly": [ + 808, + 235, + 837, + 235, + 837, + 262, + 808, + 262 + ], + "score": 0.85, + "latex": "\\rho _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 584, + 534, + 604, + 534, + 604, + 564, + 584, + 564 + ], + "score": 0.85, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 1288, + 1678, + 1318, + 1678, + 1318, + 1707, + 1288, + 1707 + ], + "score": 0.84, + "latex": "\\rho _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 965, + 534, + 992, + 534, + 992, + 560, + 965, + 560 + ], + "score": 0.84, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 548, + 1205, + 603, + 1205, + 603, + 1233, + 548, + 1233 + ], + "score": 0.83, + "latex": "\\le D" + }, + { + "category_id": 13, + "poly": [ + 891, + 234, + 921, + 234, + 921, + 263, + 891, + 263 + ], + "score": 0.83, + "latex": "\\rho _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 627, + 668, + 654, + 668, + 654, + 693, + 627, + 693 + ], + "score": 0.83, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 1370, + 1679, + 1400, + 1679, + 1400, + 1707, + 1370, + 1707 + ], + "score": 0.82, + "latex": "\\rho _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 701, + 1447, + 724, + 1447, + 724, + 1473, + 701, + 1473 + ], + "score": 0.82, + "latex": "\\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 298, + 565, + 322, + 565, + 322, + 591, + 298, + 591 + ], + "score": 0.82, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 656, + 1615, + 681, + 1615, + 681, + 1642, + 656, + 1642 + ], + "score": 0.82, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 560, + 1143, + 586, + 1143, + 586, + 1169, + 560, + 1169 + ], + "score": 0.82, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 789, + 1451, + 808, + 1451, + 808, + 1478, + 789, + 1478 + ], + "score": 0.82, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 343, + 390, + 365, + 390, + 365, + 418, + 343, + 418 + ], + "score": 0.81, + "latex": "\\mu" + }, + { + "category_id": 13, + "poly": [ + 462, + 1270, + 489, + 1270, + 489, + 1296, + 462, + 1296 + ], + "score": 0.81, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 670, + 739, + 696, + 739, + 696, + 766, + 670, + 766 + ], + "score": 0.81, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 889, + 293, + 918, + 293, + 918, + 319, + 889, + 319 + ], + "score": 0.8, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 774, + 1544, + 791, + 1544, + 791, + 1569, + 774, + 1569 + ], + "score": 0.8, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 1374, + 704, + 1400, + 704, + 1400, + 730, + 1374, + 730 + ], + "score": 0.79, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 1145, + 387, + 1170, + 387, + 1170, + 413, + 1145, + 413 + ], + "score": 0.78, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 1240, + 1174, + 1264, + 1174, + 1264, + 1200, + 1240, + 1200 + ], + "score": 0.77, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 740, + 1680, + 758, + 1680, + 758, + 1702, + 740, + 1702 + ], + "score": 0.75, + "latex": "\\nu" + }, + { + "category_id": 13, + "poly": [ + 500, + 232, + 524, + 232, + 524, + 259, + 500, + 259 + ], + "score": 0.73, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 297, + 1173, + 323, + 1173, + 323, + 1201, + 297, + 1201 + ], + "score": 0.54, + "latex": "\\mathbb { N }" + }, + { + "category_id": 13, + "poly": [ + 1299, + 231, + 1325, + 231, + 1325, + 259, + 1299, + 259 + ], + "score": 0.48, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 624, + 1676, + 699, + 1676, + 699, + 1707, + 624, + 1707 + ], + "score": 0.31, + "latex": "\\mathrm { S O ( 3 ) }" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1083.0, + 641.0, + 1083.0, + 641.0, + 1119.0, + 295.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 815.0, + 72.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2123.0, + 831.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1377.0, + 1021.0, + 1403.0, + 1021.0, + 1403.0, + 1048.0, + 1377.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1437.0, + 378.0, + 1437.0, + 378.0, + 1482.0, + 292.0, + 1482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 1437.0, + 700.0, + 1437.0, + 700.0, + 1482.0, + 416.0, + 1482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 725.0, + 1437.0, + 788.0, + 1437.0, + 788.0, + 1482.0, + 725.0, + 1482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 809.0, + 1437.0, + 1297.0, + 1437.0, + 1297.0, + 1482.0, + 809.0, + 1482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1366.0, + 1437.0, + 1410.0, + 1437.0, + 1410.0, + 1482.0, + 1366.0, + 1482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1468.0, + 297.0, + 1468.0, + 297.0, + 1513.0, + 292.0, + 1513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 1468.0, + 1409.0, + 1468.0, + 1409.0, + 1513.0, + 453.0, + 1513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1500.0, + 358.0, + 1500.0, + 358.0, + 1548.0, + 290.0, + 1548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 1500.0, + 686.0, + 1500.0, + 686.0, + 1548.0, + 572.0, + 1548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 1500.0, + 1409.0, + 1500.0, + 1409.0, + 1548.0, + 858.0, + 1548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1533.0, + 773.0, + 1533.0, + 773.0, + 1573.0, + 291.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 792.0, + 1533.0, + 1405.0, + 1533.0, + 1405.0, + 1573.0, + 792.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1564.0, + 898.0, + 1564.0, + 898.0, + 1600.0, + 293.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 1564.0, + 1017.0, + 1564.0, + 1017.0, + 1600.0, + 935.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 1564.0, + 1060.0, + 1564.0, + 1060.0, + 1600.0, + 1051.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1606.0, + 394.0, + 1606.0, + 394.0, + 1652.0, + 291.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 1606.0, + 655.0, + 1606.0, + 655.0, + 1652.0, + 595.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 1606.0, + 1063.0, + 1606.0, + 1063.0, + 1652.0, + 682.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 1606.0, + 1408.0, + 1606.0, + 1408.0, + 1652.0, + 1123.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1643.0, + 468.0, + 1643.0, + 468.0, + 1680.0, + 293.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 506.0, + 1643.0, + 1406.0, + 1643.0, + 1406.0, + 1680.0, + 506.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1674.0, + 623.0, + 1674.0, + 623.0, + 1710.0, + 295.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 1674.0, + 739.0, + 1674.0, + 739.0, + 1710.0, + 700.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 1674.0, + 1080.0, + 1674.0, + 1080.0, + 1710.0, + 759.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1150.0, + 1674.0, + 1201.0, + 1674.0, + 1201.0, + 1710.0, + 1150.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1249.0, + 1674.0, + 1287.0, + 1674.0, + 1287.0, + 1710.0, + 1249.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1319.0, + 1674.0, + 1369.0, + 1674.0, + 1369.0, + 1710.0, + 1319.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1707.0, + 528.0, + 1707.0, + 528.0, + 1749.0, + 290.0, + 1749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 1707.0, + 731.0, + 1707.0, + 731.0, + 1749.0, + 677.0, + 1749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1707.0, + 1029.0, + 1707.0, + 1029.0, + 1749.0, + 895.0, + 1749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1182.0, + 1707.0, + 1405.0, + 1707.0, + 1405.0, + 1749.0, + 1182.0, + 1749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1748.0, + 473.0, + 1748.0, + 473.0, + 1786.0, + 293.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1748.0, + 553.0, + 1748.0, + 553.0, + 1786.0, + 503.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 1748.0, + 694.0, + 1748.0, + 694.0, + 1786.0, + 585.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 281.0, + 662.0, + 416.0, + 662.0, + 416.0, + 759.0, + 281.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 662.0, + 626.0, + 662.0, + 626.0, + 759.0, + 494.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 655.0, + 662.0, + 835.0, + 662.0, + 835.0, + 759.0, + 655.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 662.0, + 1373.0, + 662.0, + 1373.0, + 759.0, + 1171.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 662.0, + 1417.0, + 662.0, + 1417.0, + 759.0, + 1401.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 735.0, + 669.0, + 735.0, + 669.0, + 775.0, + 294.0, + 775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 735.0, + 991.0, + 735.0, + 991.0, + 775.0, + 697.0, + 775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 735.0, + 1085.0, + 735.0, + 1085.0, + 775.0, + 1073.0, + 775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 228.0, + 499.0, + 228.0, + 499.0, + 268.0, + 293.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 228.0, + 807.0, + 228.0, + 807.0, + 268.0, + 525.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 228.0, + 890.0, + 228.0, + 890.0, + 268.0, + 838.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 922.0, + 228.0, + 1298.0, + 228.0, + 1298.0, + 268.0, + 922.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1326.0, + 228.0, + 1406.0, + 228.0, + 1406.0, + 268.0, + 1326.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 259.0, + 500.0, + 259.0, + 500.0, + 298.0, + 293.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 544.0, + 259.0, + 598.0, + 259.0, + 598.0, + 298.0, + 544.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 259.0, + 704.0, + 259.0, + 704.0, + 298.0, + 641.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 817.0, + 259.0, + 1405.0, + 259.0, + 1405.0, + 298.0, + 817.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 290.0, + 297.0, + 290.0, + 297.0, + 327.0, + 293.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 480.0, + 290.0, + 888.0, + 290.0, + 888.0, + 327.0, + 480.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 290.0, + 1405.0, + 290.0, + 1405.0, + 327.0, + 919.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 321.0, + 296.0, + 321.0, + 296.0, + 357.0, + 291.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 471.0, + 321.0, + 483.0, + 321.0, + 483.0, + 357.0, + 471.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1136.0, + 470.0, + 1136.0, + 470.0, + 1179.0, + 292.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 1136.0, + 559.0, + 1136.0, + 559.0, + 1179.0, + 531.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 1136.0, + 749.0, + 1136.0, + 749.0, + 1179.0, + 587.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 817.0, + 1136.0, + 1304.0, + 1136.0, + 1304.0, + 1179.0, + 817.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1169.0, + 296.0, + 1169.0, + 296.0, + 1210.0, + 292.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1169.0, + 524.0, + 1169.0, + 524.0, + 1210.0, + 324.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 666.0, + 1169.0, + 891.0, + 1169.0, + 891.0, + 1210.0, + 666.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1169.0, + 1239.0, + 1169.0, + 1239.0, + 1210.0, + 1087.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1265.0, + 1169.0, + 1406.0, + 1169.0, + 1406.0, + 1210.0, + 1265.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1202.0, + 547.0, + 1202.0, + 547.0, + 1237.0, + 292.0, + 1237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 604.0, + 1202.0, + 613.0, + 1202.0, + 613.0, + 1237.0, + 604.0, + 1237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 782.0, + 437.0, + 782.0, + 437.0, + 822.0, + 293.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 544.0, + 782.0, + 854.0, + 782.0, + 854.0, + 822.0, + 544.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1016.0, + 782.0, + 1407.0, + 782.0, + 1407.0, + 822.0, + 1016.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 813.0, + 841.0, + 813.0, + 841.0, + 853.0, + 393.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 813.0, + 1013.0, + 813.0, + 1013.0, + 853.0, + 920.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 499.0, + 758.0, + 499.0, + 758.0, + 539.0, + 293.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 791.0, + 499.0, + 1402.0, + 499.0, + 1402.0, + 539.0, + 791.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 532.0, + 583.0, + 532.0, + 583.0, + 568.0, + 293.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 605.0, + 532.0, + 760.0, + 532.0, + 760.0, + 568.0, + 605.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 799.0, + 532.0, + 964.0, + 532.0, + 964.0, + 568.0, + 799.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 993.0, + 532.0, + 1129.0, + 532.0, + 1129.0, + 568.0, + 993.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1290.0, + 532.0, + 1406.0, + 532.0, + 1406.0, + 568.0, + 1290.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 561.0, + 558.0, + 561.0, + 558.0, + 598.0, + 323.0, + 598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 384.0, + 342.0, + 384.0, + 342.0, + 420.0, + 294.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 366.0, + 384.0, + 1144.0, + 384.0, + 1144.0, + 420.0, + 366.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 384.0, + 1234.0, + 384.0, + 1234.0, + 420.0, + 1171.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1275.0, + 384.0, + 1405.0, + 384.0, + 1405.0, + 420.0, + 1275.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 411.0, + 884.0, + 411.0, + 884.0, + 454.0, + 293.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 970.0, + 411.0, + 1316.0, + 411.0, + 1316.0, + 454.0, + 970.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1267.0, + 461.0, + 1267.0, + 461.0, + 1303.0, + 295.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1267.0, + 867.0, + 1267.0, + 867.0, + 1303.0, + 490.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 1267.0, + 1406.0, + 1267.0, + 1406.0, + 1303.0, + 1088.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 1289.0, + 430.0, + 1289.0, + 430.0, + 1339.0, + 287.0, + 1339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 612.0, + 1289.0, + 1402.0, + 1289.0, + 1402.0, + 1339.0, + 612.0, + 1339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 1998.0, + 654.0, + 1998.0, + 654.0, + 2041.0, + 330.0, + 2041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 1998.0, + 1165.0, + 1998.0, + 1165.0, + 2041.0, + 938.0, + 2041.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 790, + 1405, + 790, + 1405, + 931, + 296, + 931 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 297, + 1905, + 1404, + 1905, + 1404, + 2035, + 297, + 2035 + ], + "score": 0.973 + }, + { + "category_id": 8, + "poly": [ + 646, + 691, + 1048, + 691, + 1048, + 780, + 646, + 780 + ], + "score": 0.96 + }, + { + "category_id": 8, + "poly": [ + 519, + 1070, + 1176, + 1070, + 1176, + 1161, + 519, + 1161 + ], + "score": 0.954 + }, + { + "category_id": 1, + "poly": [ + 293, + 604, + 1401, + 604, + 1401, + 677, + 293, + 677 + ], + "score": 0.951 + }, + { + "category_id": 8, + "poly": [ + 431, + 1405, + 1261, + 1405, + 1261, + 1490, + 431, + 1490 + ], + "score": 0.946 + }, + { + "category_id": 1, + "poly": [ + 299, + 1829, + 1398, + 1829, + 1398, + 1896, + 299, + 1896 + ], + "score": 0.946 + }, + { + "category_id": 1, + "poly": [ + 290, + 1325, + 1403, + 1325, + 1403, + 1396, + 290, + 1396 + ], + "score": 0.944 + }, + { + "category_id": 8, + "poly": [ + 708, + 942, + 989, + 942, + 989, + 987, + 708, + 987 + ], + "score": 0.942 + }, + { + "category_id": 1, + "poly": [ + 298, + 995, + 1071, + 995, + 1071, + 1057, + 298, + 1057 + ], + "score": 0.942 + }, + { + "category_id": 1, + "poly": [ + 301, + 356, + 1128, + 356, + 1128, + 394, + 301, + 394 + ], + "score": 0.934 + }, + { + "category_id": 8, + "poly": [ + 571, + 274, + 1126, + 274, + 1126, + 346, + 571, + 346 + ], + "score": 0.933 + }, + { + "category_id": 1, + "poly": [ + 300, + 1184, + 1394, + 1184, + 1394, + 1222, + 300, + 1222 + ], + "score": 0.929 + }, + { + "category_id": 1, + "poly": [ + 299, + 1497, + 579, + 1497, + 579, + 1533, + 299, + 1533 + ], + "score": 0.922 + }, + { + "category_id": 2, + "poly": [ + 298, + 73, + 814, + 73, + 814, + 106, + 298, + 106 + ], + "score": 0.921 + }, + { + "category_id": 1, + "poly": [ + 299, + 1717, + 591, + 1717, + 591, + 1752, + 299, + 1752 + ], + "score": 0.92 + }, + { + "category_id": 8, + "poly": [ + 390, + 1649, + 1306, + 1649, + 1306, + 1711, + 390, + 1711 + ], + "score": 0.918 + }, + { + "category_id": 1, + "poly": [ + 302, + 226, + 1388, + 226, + 1388, + 265, + 302, + 265 + ], + "score": 0.918 + }, + { + "category_id": 1, + "poly": [ + 292, + 1781, + 1300, + 1781, + 1300, + 1819, + 292, + 1819 + ], + "score": 0.914 + }, + { + "category_id": 8, + "poly": [ + 316, + 1231, + 1381, + 1231, + 1381, + 1314, + 316, + 1314 + ], + "score": 0.913 + }, + { + "category_id": 1, + "poly": [ + 293, + 1604, + 832, + 1604, + 832, + 1638, + 293, + 1638 + ], + "score": 0.913 + }, + { + "category_id": 9, + "poly": [ + 1351, + 715, + 1400, + 715, + 1400, + 747, + 1351, + 747 + ], + "score": 0.905 + }, + { + "category_id": 1, + "poly": [ + 298, + 1564, + 521, + 1564, + 521, + 1597, + 298, + 1597 + ], + "score": 0.904 + }, + { + "category_id": 1, + "poly": [ + 296, + 539, + 948, + 539, + 948, + 576, + 296, + 576 + ], + "score": 0.901 + }, + { + "category_id": 0, + "poly": [ + 299, + 432, + 700, + 432, + 700, + 470, + 299, + 470 + ], + "score": 0.899 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1663, + 1401, + 1663, + 1401, + 1695, + 1351, + 1695 + ], + "score": 0.888 + }, + { + "category_id": 1, + "poly": [ + 298, + 501, + 522, + 501, + 522, + 533, + 298, + 533 + ], + "score": 0.874 + }, + { + "category_id": 2, + "poly": [ + 836, + 2087, + 865, + 2087, + 865, + 2113, + 836, + 2113 + ], + "score": 0.869 + }, + { + "category_id": 2, + "poly": [ + 1374, + 1500, + 1404, + 1500, + 1404, + 1529, + 1374, + 1529 + ], + "score": 0.83 + }, + { + "category_id": 2, + "poly": [ + 1374, + 358, + 1404, + 358, + 1404, + 388, + 1374, + 388 + ], + "score": 0.826 + }, + { + "category_id": 14, + "poly": [ + 648, + 686, + 1048, + 686, + 1048, + 781, + 648, + 781 + ], + "score": 0.95, + "latex": "p _ { \\vec { \\alpha } } ^ { j } ( X ) = \\sum _ { i _ { 2 } , \\ldots , i _ { K } = 1 } ^ { n } x _ { j } ^ { \\alpha _ { 1 } } x _ { i _ { 2 } } ^ { \\alpha _ { 2 } } \\cdot \\cdot \\cdot x _ { i _ { k } } ^ { \\alpha _ { k } }" + }, + { + "category_id": 13, + "poly": [ + 965, + 637, + 1124, + 637, + 1124, + 680, + 965, + 680 + ], + "score": 0.95, + "latex": "p _ { \\vec { \\alpha } } = \\stackrel { \\cdot } { ( } p _ { \\vec { \\alpha } } ^ { j } ) _ { j = 1 } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 504, + 1187, + 644, + 1187, + 644, + 1225, + 504, + 1225 + ], + "score": 0.94, + "latex": "p = ( p _ { j } ) _ { j = 1 } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 1171, + 1938, + 1225, + 1938, + 1225, + 1974, + 1171, + 1974 + ], + "score": 0.94, + "latex": "Q ^ { ( \\vec { r } ) }" + }, + { + "category_id": 13, + "poly": [ + 721, + 791, + 843, + 791, + 843, + 830, + 721, + 830 + ], + "score": 0.93, + "latex": "\\alpha _ { k } ~ \\in ~ \\mathbb { N } _ { + } ^ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 374, + 1326, + 712, + 1326, + 712, + 1369, + 374, + 1369 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\Lambda _ { \\vec { r } } = \\sum _ { \\vec { \\alpha } | | \\alpha _ { k } | = r _ { k } } \\Lambda _ { \\vec { \\alpha } , \\vec { r } } \\circ \\iota _ { T } ^ { - 1 } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 434, + 1401, + 1264, + 1401, + 1264, + 1492, + 434, + 1492 + ], + "score": 0.93, + "latex": "p ( X ) = p ( X - \\frac { 1 } { n } X 1 _ { n } \\mathbf { 1 } _ { n } ^ { T } ) = \\sum _ { T \\leq D } \\sum _ { \\vec { r } \\in \\Sigma _ { T } } \\hat { \\Lambda } _ { \\vec { r } } \\left( \\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac { 1 } { n } X 1 _ { n } \\mathbf { 1 } _ { n } ^ { T } ) \\right) ." + }, + { + "category_id": 13, + "poly": [ + 796, + 1908, + 994, + 1908, + 994, + 1939, + 796, + 1939 + ], + "score": 0.93, + "latex": "D - 1 \\geq D ^ { \\prime } \\geq 0" + }, + { + "category_id": 14, + "poly": [ + 321, + 1232, + 1372, + 1232, + 1372, + 1315, + 321, + 1315 + ], + "score": 0.93, + "latex": "p _ { j } ( X ) = \\sum _ { T \\leq D } \\sum _ { \\vec { r } \\in \\Sigma _ { T } } \\sum _ { \\vec { \\alpha } | | \\alpha _ { k } | = r _ { k } } \\Lambda _ { \\vec { \\alpha } , \\vec { r } } \\left( Q _ { j } ^ { ( \\vec { r } ) } ( X ) \\right) = \\sum _ { T \\leq D } \\sum _ { \\vec { r } \\in \\Sigma _ { T } } \\Lambda _ { \\vec { r } } \\left( \\iota \\circ Q _ { j } ^ { ( \\vec { r } ) } ( X ) \\right) , j = 1 , \\dots , n ," + }, + { + "category_id": 14, + "poly": [ + 519, + 1066, + 1179, + 1066, + 1179, + 1162, + 519, + 1162 + ], + "score": 0.92, + "latex": "Q _ { j } ^ { ( \\vec { r } ) } ( X ) = \\sum _ { i _ { 2 } , \\ldots , i _ { K } = 1 } ^ { n } x _ { j } ^ { \\otimes r _ { 1 } } \\otimes x _ { i _ { 2 } } ^ { \\otimes r _ { 2 } } \\otimes x _ { i _ { 3 } } ^ { \\otimes r _ { 3 } } \\otimes \\ldots \\otimes x _ { i _ { K } } ^ { \\otimes r _ { K } } ." + }, + { + "category_id": 14, + "poly": [ + 570, + 271, + 1128, + 271, + 1128, + 347, + 570, + 347 + ], + "score": 0.92, + "latex": "\\tilde { \\Lambda } _ { k } ( w ) = \\int _ { \\mathrm { S O ( 3 ) } } \\rho _ { 2 } ( R ^ { - 1 } ) \\circ \\Lambda _ { k } \\left( \\rho _ { 1 } ( R ) w \\right) d \\nu ( R ) ." + }, + { + "category_id": 13, + "poly": [ + 297, + 1862, + 468, + 1862, + 468, + 1895, + 297, + 1895 + ], + "score": 0.92, + "latex": "X \\mapsto 1 _ { n } \\in \\mathcal { T } _ { 0 } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 1185, + 1784, + 1290, + 1784, + 1290, + 1818, + 1185, + 1818 + ], + "score": 0.92, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )" + }, + { + "category_id": 14, + "poly": [ + 708, + 940, + 989, + 940, + 989, + 987, + 708, + 987 + ], + "score": 0.92, + "latex": "p _ { \\vec { \\alpha } } ^ { j } ( X ) = \\Lambda _ { \\vec { \\alpha } , \\vec { r } ^ { } ^ { \\mathrm { ~ O ~ } } } Q _ { j } ^ { \\vec { r } } ( X )" + }, + { + "category_id": 13, + "poly": [ + 991, + 866, + 1131, + 866, + 1131, + 902, + 991, + 902 + ], + "score": 0.92, + "latex": "\\vec { r } = ( r _ { k } ) _ { k = 1 } ^ { K }" + }, + { + "category_id": 13, + "poly": [ + 542, + 542, + 639, + 542, + 639, + 574, + 542, + 574 + ], + "score": 0.92, + "latex": "D \\in { \\mathbb { N } } _ { + }" + }, + { + "category_id": 13, + "poly": [ + 1201, + 826, + 1394, + 826, + 1394, + 868, + 1201, + 868 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\sum _ { k = 1 } ^ { K } \\left| \\alpha _ { k } \\right| \\leq D } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 776, + 1325, + 819, + 1325, + 819, + 1364, + 776, + 1364 + ], + "score": 0.92, + "latex": "\\iota _ { T } ^ { - 1 }" + }, + { + "category_id": 13, + "poly": [ + 376, + 793, + 600, + 793, + 600, + 827, + 376, + 827 + ], + "score": 0.92, + "latex": "\\vec { \\alpha } \\ : = \\ : ( \\alpha _ { 1 } , \\ldots , \\alpha _ { K } )" + }, + { + "category_id": 13, + "poly": [ + 659, + 357, + 854, + 357, + 854, + 393, + 659, + 393 + ], + "score": 0.92, + "latex": "\\mathcal { F } _ { C } ( \\mathcal { F } _ { \\mathrm { f e a t } } , \\mathcal { F } _ { \\mathrm { p o o l } } )" + }, + { + "category_id": 13, + "poly": [ + 679, + 1943, + 768, + 1943, + 768, + 1973, + 679, + 1973 + ], + "score": 0.92, + "latex": "T \\leq D" + }, + { + "category_id": 13, + "poly": [ + 506, + 900, + 679, + 900, + 679, + 932, + 506, + 932 + ], + "score": 0.91, + "latex": "\\Lambda _ { \\vec { \\alpha } , \\vec { r } } : \\mathcal { T } _ { T } \\to \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1941, + 566, + 1941, + 566, + 1974, + 298, + 1974 + ], + "score": 0.91, + "latex": "\\vec { r } = ( r _ { 1 } , \\ldots , r _ { k } ) \\in \\Sigma _ { T }" + }, + { + "category_id": 13, + "poly": [ + 529, + 996, + 771, + 996, + 771, + 1058, + 529, + 1058 + ], + "score": 0.91, + "latex": "Q ^ { \\vec { r } } = \\left( Q _ { j } ^ { ( \\vec { r } ) } ( X ) \\right) _ { j = 1 } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 1232, + 231, + 1353, + 231, + 1353, + 262, + 1232, + 262 + ], + "score": 0.91, + "latex": "w \\in W _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 838, + 1786, + 898, + 1786, + 898, + 1815, + 838, + 1815 + ], + "score": 0.91, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 1138, + 605, + 1324, + 605, + 1324, + 639, + 1138, + 639 + ], + "score": 0.9, + "latex": "p : \\mathbb { R } ^ { 3 \\times n } \\mathbb { R } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 372, + 225, + 408, + 225, + 408, + 261, + 372, + 261 + ], + "score": 0.9, + "latex": "\\tilde { \\Lambda } _ { k }" + }, + { + "category_id": 13, + "poly": [ + 948, + 231, + 1014, + 231, + 1014, + 262, + 948, + 262 + ], + "score": 0.9, + "latex": "W _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 898, + 359, + 984, + 359, + 984, + 387, + 898, + 387 + ], + "score": 0.9, + "latex": "C = K" + }, + { + "category_id": 13, + "poly": [ + 873, + 1189, + 929, + 1189, + 929, + 1218, + 873, + 1218 + ], + "score": 0.9, + "latex": "\\leq D" + }, + { + "category_id": 13, + "poly": [ + 1028, + 1189, + 1062, + 1189, + 1062, + 1218, + 1028, + 1218 + ], + "score": 0.9, + "latex": "S _ { n }" + }, + { + "category_id": 14, + "poly": [ + 389, + 1645, + 1309, + 1645, + 1309, + 1711, + 389, + 1711 + ], + "score": 0.9, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } } = \\big \\{ \\iota \\circ \\pi _ { V } \\circ f ^ { 1 } \\circ f ^ { 2 } \\circ \\ldots \\circ f ^ { T } \\circ \\mathrm { e x t } ( X - \\frac { 1 } { n } X \\boldsymbol { 1 } _ { n } \\boldsymbol { 1 } _ { n } ^ { T } ) | f ^ { j } \\in \\mathcal { F } _ { m i n } , T \\leq D \\big \\} ." + }, + { + "category_id": 13, + "poly": [ + 367, + 1974, + 661, + 1974, + 661, + 2005, + 367, + 2005 + ], + "score": 0.9, + "latex": "{ \\mathcal { F } } _ { \\mathrm { f e a t } } ( D - 1 ) \\subseteq { \\mathcal { F } } _ { \\mathrm { f e a t } } ( D )" + }, + { + "category_id": 13, + "poly": [ + 845, + 1863, + 1074, + 1863, + 1074, + 1895, + 845, + 1895 + ], + "score": 0.9, + "latex": "\\pi _ { V } \\circ \\mathrm { e x t } \\in \\mathcal { F } _ { \\mathrm { f e a t } } ( 0 )" + }, + { + "category_id": 13, + "poly": [ + 361, + 1500, + 405, + 1500, + 405, + 1531, + 361, + 1531 + ], + "score": 0.9, + "latex": "\\mathcal { Q } _ { D }" + }, + { + "category_id": 13, + "poly": [ + 823, + 1833, + 900, + 1833, + 900, + 1860, + 823, + 1860 + ], + "score": 0.89, + "latex": "D = 0" + }, + { + "category_id": 13, + "poly": [ + 1287, + 1941, + 1393, + 1941, + 1393, + 1974, + 1287, + 1974 + ], + "score": 0.89, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )" + }, + { + "category_id": 13, + "poly": [ + 1047, + 231, + 1093, + 231, + 1093, + 262, + 1047, + 262 + ], + "score": 0.89, + "latex": "W _ { T }" + }, + { + "category_id": 13, + "poly": [ + 446, + 2005, + 529, + 2005, + 529, + 2030, + 446, + 2030 + ], + "score": 0.89, + "latex": "T = D" + }, + { + "category_id": 13, + "poly": [ + 1233, + 795, + 1267, + 795, + 1267, + 824, + 1233, + 824 + ], + "score": 0.89, + "latex": "S _ { n }" + }, + { + "category_id": 13, + "poly": [ + 1085, + 827, + 1118, + 827, + 1118, + 866, + 1085, + 866 + ], + "score": 0.89, + "latex": "p _ { \\vec { \\alpha } } ^ { j }" + }, + { + "category_id": 13, + "poly": [ + 359, + 1720, + 419, + 1720, + 419, + 1749, + 359, + 1749 + ], + "score": 0.89, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 731, + 543, + 774, + 543, + 774, + 572, + 731, + 572 + ], + "score": 0.89, + "latex": "\\mathcal { Q } _ { D }" + }, + { + "category_id": 13, + "poly": [ + 1309, + 1975, + 1393, + 1975, + 1393, + 2001, + 1309, + 2001 + ], + "score": 0.88, + "latex": "T < D" + }, + { + "category_id": 13, + "poly": [ + 338, + 643, + 373, + 643, + 373, + 673, + 338, + 673 + ], + "score": 0.88, + "latex": "S _ { n }" + }, + { + "category_id": 13, + "poly": [ + 615, + 1607, + 658, + 1607, + 658, + 1636, + 615, + 1636 + ], + "score": 0.88, + "latex": "\\mathcal { Q } _ { D }" + }, + { + "category_id": 13, + "poly": [ + 700, + 1909, + 734, + 1909, + 734, + 1937, + 700, + 1937 + ], + "score": 0.88, + "latex": "D ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 976, + 1832, + 1013, + 1832, + 1013, + 1862, + 976, + 1862 + ], + "score": 0.87, + "latex": "\\mathcal { Q } _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 557, + 833, + 617, + 833, + 617, + 863, + 557, + 863 + ], + "score": 0.87, + "latex": "\\leq D" + }, + { + "category_id": 13, + "poly": [ + 408, + 866, + 674, + 866, + 674, + 900, + 408, + 900 + ], + "score": 0.86, + "latex": "r _ { k } = | \\alpha _ { k } | , k = 1 , \\ldots K" + }, + { + "category_id": 13, + "poly": [ + 842, + 870, + 871, + 870, + 871, + 898, + 842, + 898 + ], + "score": 0.85, + "latex": "r _ { k }" + }, + { + "category_id": 13, + "poly": [ + 937, + 1786, + 964, + 1786, + 964, + 1812, + 937, + 1812 + ], + "score": 0.83, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 584, + 363, + 602, + 363, + 602, + 391, + 584, + 391 + ], + "score": 0.82, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 690, + 1833, + 716, + 1833, + 716, + 1859, + 690, + 1859 + ], + "score": 0.82, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 910, + 868, + 932, + 868, + 932, + 895, + 910, + 895 + ], + "score": 0.82, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 435, + 1501, + 461, + 1501, + 461, + 1527, + 435, + 1527 + ], + "score": 0.82, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 1301, + 1336, + 1319, + 1336, + 1319, + 1362, + 1301, + 1362 + ], + "score": 0.81, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 805, + 543, + 831, + 543, + 831, + 569, + 805, + 569 + ], + "score": 0.81, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 448, + 1720, + 475, + 1720, + 475, + 1746, + 448, + 1746 + ], + "score": 0.81, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 1274, + 1910, + 1299, + 1910, + 1299, + 1936, + 1274, + 1936 + ], + "score": 0.81, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 1240, + 1336, + 1253, + 1336, + 1253, + 1357, + 1240, + 1357 + ], + "score": 0.7, + "latex": "\\iota" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 815.0, + 72.0, + 815.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 429.0, + 702.0, + 429.0, + 702.0, + 474.0, + 296.0, + 474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 871.0, + 2085.0, + 871.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1380.0, + 1505.0, + 1402.0, + 1505.0, + 1402.0, + 1528.0, + 1380.0, + 1528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 363.0, + 1403.0, + 363.0, + 1403.0, + 390.0, + 1378.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 790.0, + 375.0, + 790.0, + 375.0, + 829.0, + 295.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 790.0, + 720.0, + 790.0, + 720.0, + 829.0, + 601.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 844.0, + 790.0, + 1232.0, + 790.0, + 1232.0, + 829.0, + 844.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1268.0, + 790.0, + 1406.0, + 790.0, + 1406.0, + 829.0, + 1268.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 822.0, + 556.0, + 822.0, + 556.0, + 877.0, + 290.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 618.0, + 822.0, + 1084.0, + 822.0, + 1084.0, + 877.0, + 618.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 822.0, + 1200.0, + 822.0, + 1200.0, + 877.0, + 1119.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1395.0, + 822.0, + 1408.0, + 822.0, + 1408.0, + 877.0, + 1395.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 854.0, + 407.0, + 854.0, + 407.0, + 913.0, + 288.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 675.0, + 854.0, + 841.0, + 854.0, + 841.0, + 913.0, + 675.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 854.0, + 909.0, + 854.0, + 909.0, + 913.0, + 872.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 933.0, + 854.0, + 990.0, + 854.0, + 990.0, + 913.0, + 933.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 854.0, + 1412.0, + 854.0, + 1412.0, + 913.0, + 1132.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 897.0, + 505.0, + 897.0, + 505.0, + 932.0, + 294.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 897.0, + 791.0, + 897.0, + 791.0, + 932.0, + 680.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1905.0, + 699.0, + 1905.0, + 699.0, + 1942.0, + 294.0, + 1942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 1905.0, + 795.0, + 1905.0, + 795.0, + 1942.0, + 735.0, + 1942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 995.0, + 1905.0, + 1273.0, + 1905.0, + 1273.0, + 1942.0, + 995.0, + 1942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 1905.0, + 1405.0, + 1905.0, + 1405.0, + 1942.0, + 1300.0, + 1942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1938.0, + 297.0, + 1938.0, + 297.0, + 1978.0, + 292.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.0, + 1938.0, + 678.0, + 1938.0, + 678.0, + 1978.0, + 567.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 1938.0, + 1170.0, + 1938.0, + 1170.0, + 1978.0, + 769.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1226.0, + 1938.0, + 1286.0, + 1938.0, + 1286.0, + 1978.0, + 1226.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 1938.0, + 1404.0, + 1938.0, + 1404.0, + 1978.0, + 1394.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1969.0, + 366.0, + 1969.0, + 366.0, + 2010.0, + 294.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 662.0, + 1969.0, + 1308.0, + 1969.0, + 1308.0, + 2010.0, + 662.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 1969.0, + 1406.0, + 1969.0, + 1406.0, + 2010.0, + 1394.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 2000.0, + 445.0, + 2000.0, + 445.0, + 2036.0, + 292.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 530.0, + 2000.0, + 807.0, + 2000.0, + 807.0, + 2036.0, + 530.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 603.0, + 1137.0, + 603.0, + 1137.0, + 643.0, + 292.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 603.0, + 1406.0, + 603.0, + 1406.0, + 643.0, + 1325.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 635.0, + 337.0, + 635.0, + 337.0, + 681.0, + 292.0, + 681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 635.0, + 964.0, + 635.0, + 964.0, + 681.0, + 374.0, + 681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.0, + 635.0, + 1256.0, + 635.0, + 1256.0, + 681.0, + 1125.0, + 681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1830.0, + 689.0, + 1830.0, + 689.0, + 1866.0, + 296.0, + 1866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1830.0, + 822.0, + 1830.0, + 822.0, + 1866.0, + 717.0, + 1866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 1830.0, + 975.0, + 1830.0, + 975.0, + 1866.0, + 901.0, + 1866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1014.0, + 1830.0, + 1403.0, + 1830.0, + 1403.0, + 1866.0, + 1014.0, + 1866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 1859.0, + 844.0, + 1859.0, + 844.0, + 1898.0, + 469.0, + 1898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 1859.0, + 1083.0, + 1859.0, + 1083.0, + 1898.0, + 1075.0, + 1898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1315.0, + 373.0, + 1315.0, + 373.0, + 1381.0, + 288.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 713.0, + 1315.0, + 775.0, + 1315.0, + 775.0, + 1381.0, + 713.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 820.0, + 1315.0, + 1239.0, + 1315.0, + 1239.0, + 1381.0, + 820.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1254.0, + 1315.0, + 1300.0, + 1315.0, + 1300.0, + 1381.0, + 1254.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1320.0, + 1315.0, + 1413.0, + 1315.0, + 1413.0, + 1381.0, + 1320.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1363.0, + 587.0, + 1363.0, + 587.0, + 1397.0, + 297.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 283.0, + 978.0, + 528.0, + 978.0, + 528.0, + 1066.0, + 283.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 772.0, + 978.0, + 1078.0, + 978.0, + 1078.0, + 1066.0, + 772.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.75, + 1035.5, + 764.75, + 1035.5, + 764.75, + 1053.5, + 743.75, + 1053.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 355.0, + 583.0, + 355.0, + 583.0, + 395.0, + 297.0, + 395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 603.0, + 355.0, + 658.0, + 355.0, + 658.0, + 395.0, + 603.0, + 395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 855.0, + 355.0, + 897.0, + 355.0, + 897.0, + 395.0, + 855.0, + 395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 355.0, + 1130.0, + 355.0, + 1130.0, + 395.0, + 985.0, + 395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1179.0, + 503.0, + 1179.0, + 503.0, + 1229.0, + 294.0, + 1229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 645.0, + 1179.0, + 872.0, + 1179.0, + 872.0, + 1229.0, + 645.0, + 1229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 930.0, + 1179.0, + 1027.0, + 1179.0, + 1027.0, + 1229.0, + 930.0, + 1229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1063.0, + 1179.0, + 1399.0, + 1179.0, + 1399.0, + 1229.0, + 1063.0, + 1229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1491.0, + 360.0, + 1491.0, + 360.0, + 1541.0, + 294.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1491.0, + 434.0, + 1491.0, + 434.0, + 1541.0, + 406.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 1491.0, + 582.0, + 1491.0, + 582.0, + 1541.0, + 462.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1712.0, + 358.0, + 1712.0, + 358.0, + 1759.0, + 294.0, + 1759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 420.0, + 1712.0, + 447.0, + 1712.0, + 447.0, + 1759.0, + 420.0, + 1759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1712.0, + 596.0, + 1712.0, + 596.0, + 1759.0, + 476.0, + 1759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 223.0, + 371.0, + 223.0, + 371.0, + 271.0, + 293.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 409.0, + 223.0, + 947.0, + 223.0, + 947.0, + 271.0, + 409.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1015.0, + 223.0, + 1046.0, + 223.0, + 1046.0, + 271.0, + 1015.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 223.0, + 1231.0, + 223.0, + 1231.0, + 271.0, + 1094.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 223.0, + 1392.0, + 223.0, + 1392.0, + 271.0, + 1354.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1778.0, + 837.0, + 1778.0, + 837.0, + 1824.0, + 292.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 1778.0, + 936.0, + 1778.0, + 936.0, + 1824.0, + 899.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 1778.0, + 1184.0, + 1778.0, + 1184.0, + 1824.0, + 965.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1291.0, + 1778.0, + 1303.0, + 1778.0, + 1303.0, + 1824.0, + 1291.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1602.0, + 614.0, + 1602.0, + 614.0, + 1642.0, + 295.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 1602.0, + 832.0, + 1602.0, + 832.0, + 1642.0, + 659.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1562.0, + 524.0, + 1562.0, + 524.0, + 1601.0, + 295.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 534.0, + 541.0, + 534.0, + 541.0, + 582.0, + 294.0, + 582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 640.0, + 534.0, + 730.0, + 534.0, + 730.0, + 582.0, + 640.0, + 582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 775.0, + 534.0, + 804.0, + 534.0, + 804.0, + 582.0, + 775.0, + 582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 534.0, + 952.0, + 534.0, + 952.0, + 582.0, + 832.0, + 582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 499.0, + 524.0, + 499.0, + 524.0, + 539.0, + 296.0, + 539.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 13, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 365, + 784, + 1405, + 784, + 1405, + 957, + 365, + 957 + ], + "score": 0.977 + }, + { + "category_id": 8, + "poly": [ + 342, + 1780, + 1351, + 1780, + 1351, + 1986, + 342, + 1986 + ], + "score": 0.967 + }, + { + "category_id": 8, + "poly": [ + 498, + 969, + 1298, + 969, + 1298, + 1211, + 498, + 1211 + ], + "score": 0.966 + }, + { + "category_id": 8, + "poly": [ + 522, + 468, + 1273, + 468, + 1273, + 612, + 522, + 612 + ], + "score": 0.965 + }, + { + "category_id": 8, + "poly": [ + 608, + 1477, + 1088, + 1477, + 1088, + 1568, + 608, + 1568 + ], + "score": 0.959 + }, + { + "category_id": 1, + "poly": [ + 396, + 372, + 1406, + 372, + 1406, + 450, + 396, + 450 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 362, + 225, + 1401, + 225, + 1401, + 296, + 362, + 296 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 297, + 1696, + 1402, + 1696, + 1402, + 1763, + 297, + 1763 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 295, + 1586, + 1402, + 1586, + 1402, + 1657, + 295, + 1657 + ], + "score": 0.944 + }, + { + "category_id": 8, + "poly": [ + 631, + 312, + 1168, + 312, + 1168, + 355, + 631, + 355 + ], + "score": 0.942 + }, + { + "category_id": 2, + "poly": [ + 298, + 73, + 814, + 73, + 814, + 106, + 298, + 106 + ], + "score": 0.922 + }, + { + "category_id": 1, + "poly": [ + 296, + 1422, + 873, + 1422, + 873, + 1456, + 296, + 1456 + ], + "score": 0.921 + }, + { + "category_id": 1, + "poly": [ + 395, + 1226, + 1380, + 1226, + 1380, + 1267, + 395, + 1267 + ], + "score": 0.918 + }, + { + "category_id": 1, + "poly": [ + 295, + 2000, + 1159, + 2000, + 1159, + 2038, + 295, + 2038 + ], + "score": 0.918 + }, + { + "category_id": 1, + "poly": [ + 396, + 626, + 475, + 626, + 475, + 656, + 396, + 656 + ], + "score": 0.916 + }, + { + "category_id": 1, + "poly": [ + 397, + 726, + 867, + 726, + 867, + 769, + 397, + 769 + ], + "score": 0.91 + }, + { + "category_id": 1, + "poly": [ + 297, + 1379, + 602, + 1379, + 602, + 1412, + 297, + 1412 + ], + "score": 0.909 + }, + { + "category_id": 9, + "poly": [ + 1350, + 670, + 1401, + 670, + 1401, + 702, + 1350, + 702 + ], + "score": 0.893 + }, + { + "category_id": 9, + "poly": [ + 1351, + 318, + 1401, + 318, + 1401, + 350, + 1351, + 350 + ], + "score": 0.886 + }, + { + "category_id": 8, + "poly": [ + 531, + 656, + 1262, + 656, + 1262, + 717, + 531, + 717 + ], + "score": 0.873 + }, + { + "category_id": 2, + "poly": [ + 836, + 2087, + 865, + 2087, + 865, + 2113, + 836, + 2113 + ], + "score": 0.869 + }, + { + "category_id": 2, + "poly": [ + 1374, + 2003, + 1404, + 2003, + 1404, + 2032, + 1374, + 2032 + ], + "score": 0.84 + }, + { + "category_id": 2, + "poly": [ + 1373, + 1304, + 1404, + 1304, + 1404, + 1334, + 1373, + 1334 + ], + "score": 0.821 + }, + { + "category_id": 14, + "poly": [ + 497, + 966, + 1297, + 966, + 1297, + 1214, + 497, + 1214 + ], + "score": 0.95, + "latex": "\\begin{array} { r l } & { \\tilde { V } _ { j } ( \\bar { X } , Q ^ { ( \\tilde { r } ) } ( \\bar { X } ) ) = \\displaystyle \\sum _ { j = 1 } ^ { n } \\displaystyle \\sum _ { i _ { 3 } , \\ldots , i _ { K } = 1 } ^ { n } \\bar { x } _ { j } \\otimes \\bar { x } _ { j } ^ { \\otimes ( r _ { 2 } - 1 ) } \\otimes \\bar { x } _ { i _ { 3 } } ^ { \\otimes r _ { 2 } } \\otimes \\ldots \\otimes \\bar { x } _ { i _ { K } } ^ { \\otimes r _ { K } } } \\\\ & { \\qquad = \\displaystyle \\sum _ { i _ { 2 } , i _ { 3 } , \\ldots , i _ { K } = 1 } ^ { n } \\bar { x } _ { i _ { 2 } } ^ { \\otimes r _ { 2 } } \\otimes \\bar { x } _ { i _ { 3 } } ^ { \\otimes r _ { 2 } } \\otimes \\ldots \\otimes \\bar { x } _ { i _ { K } } ^ { \\otimes r _ { K } } } \\\\ & { \\qquad = Q _ { j } ^ { ( \\tilde { r } ) } ( \\bar { X } ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1111, + 370, + 1263, + 370, + 1263, + 413, + 1111, + 413 + ], + "score": 0.94, + "latex": "\\tilde { V } = ( \\tilde { V } _ { j } ) _ { j = 1 } ^ { n }" + }, + { + "category_id": 14, + "poly": [ + 348, + 1780, + 1354, + 1780, + 1354, + 1988, + 348, + 1988 + ], + "score": 0.94, + "latex": "\\begin{array} { l } { \\displaystyle p ( X ) = \\sum _ { \\vec { r } ^ { \\prime } | | \\vec { r } | | _ { 1 } \\le D } \\hat { \\Lambda } _ { \\vec { r } } \\bigg ( \\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac 1 n X \\mathbf { 1 } _ { n } \\mathbf { 1 } _ { n } ^ { T } ) \\bigg ) = \\sum _ { \\vec { r } ^ { \\prime } | | \\vec { r } ^ { \\prime } | | _ { 1 } \\le D } \\hat { \\Lambda } _ { \\vec { r } } \\bigg ( \\sum _ { k = 1 } ^ { K _ { \\vec { r } } } \\iota \\circ \\hat { A } _ { k , \\vec { r } } f _ { k , \\vec { r } } ( X ) \\bigg ) } \\\\ { = \\sum _ { \\vec { r } ^ { \\prime } | | \\vec { r } | | _ { 1 } \\le D } \\sum _ { k = 1 } ^ { K _ { \\vec { r } } } \\hat { \\Lambda } _ { k , \\vec { r } } ( f _ { k , \\vec { r } } ( X ) ) } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 520, + 465, + 1272, + 465, + 1272, + 618, + 520, + 618 + ], + "score": 0.93, + "latex": "\\begin{array} { r l } & { \\tilde { V } _ { j } ( \\bar { X } , Q ^ { ( \\tilde { r } ) } ( \\bar { X } ) ) = \\displaystyle \\sum _ { i _ { 2 } , \\ldots , i _ { K } = 1 } ^ { n } \\bar { x } _ { j } \\otimes \\bar { x } _ { j } ^ { \\otimes ( r _ { 1 } - 1 ) } \\otimes \\bar { x } _ { i _ { 2 } } ^ { \\otimes r _ { 2 } } \\otimes \\ldots \\otimes \\bar { x } _ { i _ { K } } ^ { \\otimes r _ { K } } } \\\\ & { \\qquad = Q _ { j } ^ { ( \\tilde { r } ) } ( \\bar { X } ) . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 609, + 1472, + 1088, + 1472, + 1088, + 1569, + 609, + 1569 + ], + "score": 0.93, + "latex": "\\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac { 1 } { n } X 1 _ { n } 1 _ { n } ^ { T } ) = \\sum _ { k = 1 } ^ { K } \\hat { A } _ { k } f _ { k } ( X ) ," + }, + { + "category_id": 13, + "poly": [ + 1212, + 846, + 1366, + 846, + 1366, + 888, + 1212, + 888 + ], + "score": 0.93, + "latex": "\\tilde { V } = \\tilde { ( } \\tilde { V } _ { j } ) _ { j = 1 } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 499, + 884, + 780, + 884, + 780, + 929, + 499, + 929 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\tilde { V } _ { j } ( X , V ) = \\sum _ { j } x _ { j } \\otimes V _ { j } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 841, + 263, + 964, + 263, + 964, + 294, + 841, + 294 + ], + "score": 0.93, + "latex": "f _ { 2 } , \\ldots , f _ { D }" + }, + { + "category_id": 13, + "poly": [ + 862, + 414, + 1030, + 414, + 1030, + 446, + 862, + 446 + ], + "score": 0.93, + "latex": "\\theta _ { 1 } = 1 , \\theta _ { 2 } = 0" + }, + { + "category_id": 13, + "poly": [ + 397, + 409, + 634, + 409, + 634, + 450, + 397, + 450 + ], + "score": 0.93, + "latex": "\\tilde { V } _ { j } ( X , V ) = x _ { j } \\otimes V _ { j }" + }, + { + "category_id": 13, + "poly": [ + 820, + 1586, + 1105, + 1586, + 1105, + 1628, + 820, + 1628 + ], + "score": 0.92, + "latex": "{ \\hat { A } } _ { k } : W _ { \\mathrm { f e a t } } ^ { n } \\to ( W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } } ) ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 1044, + 227, + 1365, + 227, + 1365, + 264, + 1044, + 264 + ], + "score": 0.92, + "latex": "\\iota \\circ Q ^ { ( \\tilde { r } ) } ( \\bar { X } ) \\in \\mathcal { F } _ { \\mathrm { f e a t } } ( D - 1 )" + }, + { + "category_id": 13, + "poly": [ + 590, + 231, + 871, + 231, + 871, + 264, + 590, + 264 + ], + "score": 0.92, + "latex": "\\tilde { r } = ( r _ { 1 } - 1 , r _ { 2 } , \\dots , r _ { K } )" + }, + { + "category_id": 13, + "poly": [ + 493, + 2003, + 730, + 2003, + 730, + 2037, + 493, + 2037 + ], + "score": 0.92, + "latex": "\\Lambda _ { k , \\vec { r } } = \\Lambda _ { \\vec { r } } \\circ \\iota \\circ A _ { k , \\vec { r } } ." + }, + { + "category_id": 13, + "poly": [ + 1009, + 889, + 1177, + 889, + 1177, + 922, + 1009, + 922 + ], + "score": 0.92, + "latex": "\\theta _ { 1 } = 0 , \\theta _ { 2 } = 1" + }, + { + "category_id": 13, + "poly": [ + 962, + 1228, + 1376, + 1228, + 1376, + 1267, + 962, + 1267 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac { 1 } { n } X 1 _ { n } \\mathbf { 1 } _ { n } ^ { T } ) \\in \\mathcal { F } _ { \\mathrm { f e a t } } ( D ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 637, + 850, + 767, + 850, + 767, + 883, + 637, + 883 + ], + "score": 0.92, + "latex": "f _ { 1 } \\in \\mathcal { F } _ { m i n }" + }, + { + "category_id": 13, + "poly": [ + 968, + 818, + 1092, + 818, + 1092, + 849, + 968, + 849 + ], + "score": 0.92, + "latex": "f _ { 2 } , \\ldots , f _ { D }" + }, + { + "category_id": 13, + "poly": [ + 543, + 376, + 669, + 376, + 669, + 408, + 543, + 408 + ], + "score": 0.91, + "latex": "f _ { 1 } \\in \\mathcal { F } _ { m i n }" + }, + { + "category_id": 13, + "poly": [ + 422, + 788, + 501, + 788, + 501, + 817, + 422, + 817 + ], + "score": 0.91, + "latex": "r _ { 1 } = 0" + }, + { + "category_id": 13, + "poly": [ + 422, + 232, + 502, + 232, + 502, + 261, + 422, + 261 + ], + "score": 0.91, + "latex": "r _ { 1 } > 0" + }, + { + "category_id": 14, + "poly": [ + 629, + 309, + 1169, + 309, + 1169, + 353, + 629, + 353 + ], + "score": 0.91, + "latex": "\\iota \\circ \\pi _ { V } \\circ f _ { 2 } \\circ . . . \\circ f _ { D } \\circ \\mathrm { e x t } ( \\bar { X } ) = \\iota \\circ Q ^ { ( \\tilde { r } ) } ( \\bar { X } ) ." + }, + { + "category_id": 13, + "poly": [ + 925, + 2004, + 985, + 2004, + 985, + 2035, + 925, + 2035 + ], + "score": 0.9, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 1115, + 787, + 1394, + 787, + 1394, + 820, + 1115, + 820 + ], + "score": 0.9, + "latex": "\\tilde { r } = ( r _ { 2 } - 1 , r _ { 3 } , \\dots , r _ { K } )" + }, + { + "category_id": 13, + "poly": [ + 890, + 1701, + 934, + 1701, + 934, + 1730, + 890, + 1730 + ], + "score": 0.9, + "latex": "\\mathcal { Q } _ { D }" + }, + { + "category_id": 13, + "poly": [ + 459, + 1625, + 495, + 1625, + 495, + 1654, + 459, + 1654 + ], + "score": 0.89, + "latex": "A _ { k }" + }, + { + "category_id": 13, + "poly": [ + 981, + 788, + 1060, + 788, + 1060, + 817, + 981, + 817 + ], + "score": 0.89, + "latex": "r _ { 2 } > 0" + }, + { + "category_id": 13, + "poly": [ + 516, + 1590, + 763, + 1590, + 763, + 1625, + 516, + 1625 + ], + "score": 0.88, + "latex": "A _ { k } : W _ { \\mathrm { f e a t } } \\to W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } }" + }, + { + "category_id": 13, + "poly": [ + 627, + 1424, + 671, + 1424, + 671, + 1454, + 627, + 1454 + ], + "score": 0.88, + "latex": "\\mathcal { Q } _ { D }" + }, + { + "category_id": 13, + "poly": [ + 964, + 1701, + 990, + 1701, + 990, + 1727, + 964, + 1727 + ], + "score": 0.84, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 890, + 1732, + 917, + 1732, + 917, + 1757, + 890, + 1757 + ], + "score": 0.83, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 1015, + 2004, + 1041, + 2004, + 1041, + 2030, + 1015, + 2030 + ], + "score": 0.83, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 759, + 1735, + 777, + 1735, + 777, + 1762, + 759, + 1762 + ], + "score": 0.82, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 956, + 376, + 982, + 376, + 982, + 403, + 956, + 403 + ], + "score": 0.82, + "latex": "V" + }, + { + "category_id": 13, + "poly": [ + 1056, + 852, + 1082, + 852, + 1082, + 879, + 1056, + 879 + ], + "score": 0.79, + "latex": "V" + }, + { + "category_id": 13, + "poly": [ + 374, + 1593, + 502, + 1593, + 502, + 1624, + 374, + 1624 + ], + "score": 0.73, + "latex": "f _ { k } \\in \\mathcal { F } _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 562, + 1625, + 622, + 1625, + 622, + 1655, + 562, + 1655 + ], + "score": 0.71, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } }" + }, + { + "category_id": 13, + "poly": [ + 445, + 726, + 861, + 726, + 861, + 769, + 445, + 769 + ], + "score": 0.67, + "latex": "\\begin{array} { r } { \\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac { 1 } { n } X 1 _ { n } 1 _ { n } ^ { T } ) \\in \\mathcal { F } _ { \\mathrm { f e a t } } ( D ) . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 437, + 650, + 1266, + 650, + 1266, + 773, + 437, + 773 + ], + "score": 0.66, + "latex": "\\begin{array} { c } { \\displaystyle { \\imath \\circ \\pi _ { V } \\circ f _ { 1 } \\circ f _ { 2 } \\circ . . . \\circ f _ { D } \\circ \\mathrm { e x t } ( X - \\frac 1 n X 1 _ { n } \\mathbf { 1 } _ { n } ^ { T } ) = \\imath \\circ Q ^ { ( \\vec { r } ) } ( \\bar { X } ) . } } \\\\ { \\displaystyle { \\mathrm { } } } \\\\ { \\displaystyle { \\mathrm { } \\mathrm { } } \\iota \\circ Q ^ { ( \\vec { r } ) } ( X - \\frac 1 n X 1 _ { n } \\mathbf { 1 } _ { n } ^ { T } ) \\in \\mathcal { F } _ { \\mathrm { f e a t } } ( D ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 652, + 1625, + 678, + 1625, + 678, + 1651, + 652, + 1651 + ], + "score": 0.56, + "latex": "D" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 815.0, + 72.0, + 815.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1380.0, + 2008.0, + 1402.0, + 2008.0, + 1402.0, + 2031.0, + 1380.0, + 2031.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1377.0, + 1308.0, + 1403.0, + 1308.0, + 1403.0, + 1336.0, + 1377.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 782.0, + 421.0, + 782.0, + 421.0, + 823.0, + 358.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 502.0, + 782.0, + 980.0, + 782.0, + 980.0, + 823.0, + 502.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1061.0, + 782.0, + 1114.0, + 782.0, + 1114.0, + 823.0, + 1061.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1395.0, + 782.0, + 1405.0, + 782.0, + 1405.0, + 823.0, + 1395.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 813.0, + 967.0, + 813.0, + 967.0, + 855.0, + 392.0, + 855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 813.0, + 1406.0, + 813.0, + 1406.0, + 855.0, + 1093.0, + 855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 845.0, + 636.0, + 845.0, + 636.0, + 890.0, + 393.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 845.0, + 1055.0, + 845.0, + 1055.0, + 890.0, + 768.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1083.0, + 845.0, + 1211.0, + 845.0, + 1211.0, + 890.0, + 1083.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1367.0, + 845.0, + 1409.0, + 845.0, + 1409.0, + 890.0, + 1367.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 886.0, + 498.0, + 886.0, + 498.0, + 928.0, + 395.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 781.0, + 886.0, + 1008.0, + 886.0, + 1008.0, + 928.0, + 781.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1178.0, + 886.0, + 1406.0, + 886.0, + 1406.0, + 928.0, + 1178.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 922.0, + 497.0, + 922.0, + 497.0, + 957.0, + 393.0, + 957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 364.0, + 542.0, + 364.0, + 542.0, + 418.0, + 389.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 364.0, + 955.0, + 364.0, + 955.0, + 418.0, + 670.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 364.0, + 1110.0, + 364.0, + 1110.0, + 418.0, + 983.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1264.0, + 364.0, + 1406.0, + 364.0, + 1406.0, + 418.0, + 1264.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 635.0, + 411.0, + 861.0, + 411.0, + 861.0, + 452.0, + 635.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 411.0, + 1370.0, + 411.0, + 1370.0, + 452.0, + 1031.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 220.0, + 421.0, + 220.0, + 421.0, + 270.0, + 358.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 220.0, + 589.0, + 220.0, + 589.0, + 270.0, + 503.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 220.0, + 1043.0, + 220.0, + 1043.0, + 270.0, + 872.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1366.0, + 220.0, + 1405.0, + 220.0, + 1405.0, + 270.0, + 1366.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 259.0, + 840.0, + 259.0, + 840.0, + 297.0, + 394.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 259.0, + 1076.0, + 259.0, + 1076.0, + 297.0, + 965.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1698.0, + 889.0, + 1698.0, + 889.0, + 1735.0, + 295.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 1698.0, + 963.0, + 1698.0, + 963.0, + 1735.0, + 935.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 991.0, + 1698.0, + 1404.0, + 1698.0, + 1404.0, + 1735.0, + 991.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1730.0, + 758.0, + 1730.0, + 758.0, + 1766.0, + 295.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 778.0, + 1730.0, + 889.0, + 1730.0, + 889.0, + 1766.0, + 778.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1730.0, + 1116.0, + 1730.0, + 1116.0, + 1766.0, + 918.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 1576.0, + 373.0, + 1576.0, + 373.0, + 1643.0, + 286.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1576.0, + 515.0, + 1576.0, + 515.0, + 1643.0, + 503.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 1576.0, + 819.0, + 1576.0, + 819.0, + 1643.0, + 764.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1106.0, + 1576.0, + 1413.0, + 1576.0, + 1413.0, + 1643.0, + 1106.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1619.0, + 458.0, + 1619.0, + 458.0, + 1660.0, + 294.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 1619.0, + 561.0, + 1619.0, + 561.0, + 1660.0, + 496.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 623.0, + 1619.0, + 651.0, + 1619.0, + 651.0, + 1660.0, + 623.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 679.0, + 1619.0, + 797.0, + 1619.0, + 797.0, + 1660.0, + 679.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1420.0, + 626.0, + 1420.0, + 626.0, + 1460.0, + 294.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 1420.0, + 874.0, + 1420.0, + 874.0, + 1460.0, + 672.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 1220.0, + 961.0, + 1220.0, + 961.0, + 1274.0, + 389.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1377.0, + 1220.0, + 1388.0, + 1220.0, + 1388.0, + 1274.0, + 1377.0, + 1274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1994.0, + 492.0, + 1994.0, + 492.0, + 2044.0, + 293.0, + 2044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 731.0, + 1994.0, + 924.0, + 1994.0, + 924.0, + 2044.0, + 731.0, + 2044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 1994.0, + 1014.0, + 1994.0, + 1014.0, + 2044.0, + 986.0, + 2044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 1994.0, + 1162.0, + 1994.0, + 1162.0, + 2044.0, + 1042.0, + 2044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 623.0, + 477.0, + 623.0, + 477.0, + 658.0, + 392.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 721.0, + 436.0, + 721.0, + 436.0, + 773.0, + 393.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1378.0, + 603.0, + 1378.0, + 603.0, + 1417.0, + 295.0, + 1417.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 14, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1288, + 1404, + 1288, + 1404, + 1417, + 297, + 1417 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 296, + 880, + 1406, + 880, + 1406, + 1037, + 296, + 1037 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 297, + 337, + 1405, + 337, + 1405, + 452, + 297, + 452 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 297, + 665, + 1405, + 665, + 1405, + 761, + 297, + 761 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 298, + 1074, + 1402, + 1074, + 1402, + 1159, + 298, + 1159 + ], + "score": 0.969 + }, + { + "category_id": 1, + "poly": [ + 299, + 1774, + 1406, + 1774, + 1406, + 1870, + 299, + 1870 + ], + "score": 0.968 + }, + { + "category_id": 8, + "poly": [ + 733, + 474, + 967, + 474, + 967, + 564, + 733, + 564 + ], + "score": 0.958 + }, + { + "category_id": 8, + "poly": [ + 663, + 1182, + 1033, + 1182, + 1033, + 1273, + 663, + 1273 + ], + "score": 0.957 + }, + { + "category_id": 8, + "poly": [ + 715, + 1954, + 982, + 1954, + 982, + 2043, + 715, + 2043 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 291, + 778, + 1404, + 778, + 1404, + 845, + 291, + 845 + ], + "score": 0.95 + }, + { + "category_id": 1, + "poly": [ + 296, + 1882, + 1400, + 1882, + 1400, + 1949, + 296, + 1949 + ], + "score": 0.946 + }, + { + "category_id": 1, + "poly": [ + 295, + 580, + 941, + 580, + 941, + 627, + 295, + 627 + ], + "score": 0.933 + }, + { + "category_id": 2, + "poly": [ + 298, + 73, + 814, + 73, + 814, + 106, + 298, + 106 + ], + "score": 0.914 + }, + { + "category_id": 0, + "poly": [ + 300, + 224, + 701, + 224, + 701, + 263, + 300, + 263 + ], + "score": 0.914 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1211, + 1400, + 1211, + 1400, + 1243, + 1351, + 1243 + ], + "score": 0.904 + }, + { + "category_id": 9, + "poly": [ + 1351, + 501, + 1401, + 501, + 1401, + 533, + 1351, + 533 + ], + "score": 0.901 + }, + { + "category_id": 1, + "poly": [ + 293, + 1499, + 1364, + 1499, + 1364, + 1534, + 293, + 1534 + ], + "score": 0.894 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1980, + 1400, + 1980, + 1400, + 2011, + 1351, + 2011 + ], + "score": 0.894 + }, + { + "category_id": 2, + "poly": [ + 836, + 2087, + 866, + 2087, + 866, + 2113, + 836, + 2113 + ], + "score": 0.873 + }, + { + "category_id": 1, + "poly": [ + 298, + 296, + 521, + 296, + 521, + 329, + 298, + 329 + ], + "score": 0.861 + }, + { + "category_id": 1, + "poly": [ + 300, + 1543, + 719, + 1543, + 719, + 1577, + 300, + 1577 + ], + "score": 0.847 + }, + { + "category_id": 2, + "poly": [ + 1374, + 1430, + 1402, + 1430, + 1402, + 1459, + 1374, + 1459 + ], + "score": 0.762 + }, + { + "category_id": 1, + "poly": [ + 358, + 1601, + 1403, + 1601, + 1403, + 1738, + 358, + 1738 + ], + "score": 0.631 + }, + { + "category_id": 14, + "poly": [ + 717, + 1949, + 981, + 1949, + 981, + 2043, + 717, + 2043 + ], + "score": 0.95, + "latex": "p ( X ) = \\sum _ { k = 1 } ^ { K } \\hat { A } _ { k } f _ { k } ( X ) ," + }, + { + "category_id": 14, + "poly": [ + 664, + 1179, + 1036, + 1179, + 1036, + 1275, + 664, + 1275 + ], + "score": 0.95, + "latex": "\\Lambda _ { j } ( V _ { 1 } , \\dots , V _ { K _ { 1 } } ) = \\sum _ { i = 1 } ^ { K _ { 1 } } \\Lambda _ { i j } ( V _ { i } ) ." + }, + { + "category_id": 14, + "poly": [ + 732, + 470, + 968, + 470, + 968, + 566, + 732, + 566 + ], + "score": 0.94, + "latex": "\\Lambda _ { j } ( V ) = \\sum _ { i = 1 } ^ { K _ { 1 } } M _ { i j } V _ { i }" + }, + { + "category_id": 13, + "poly": [ + 606, + 408, + 726, + 408, + 726, + 454, + 606, + 454 + ], + "score": 0.94, + "latex": "\\ell _ { i } ^ { ( 1 ) } \\neq \\ell _ { j } ^ { ( 2 ) }" + }, + { + "category_id": 13, + "poly": [ + 1024, + 1332, + 1145, + 1332, + 1145, + 1379, + 1024, + 1379 + ], + "score": 0.94, + "latex": "\\ell _ { i } ^ { ( 1 ) } = \\ell _ { j } ^ { ( 2 ) }" + }, + { + "category_id": 13, + "poly": [ + 1267, + 1775, + 1373, + 1775, + 1373, + 1809, + 1267, + 1809 + ], + "score": 0.94, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )" + }, + { + "category_id": 13, + "poly": [ + 583, + 1806, + 688, + 1806, + 688, + 1839, + 583, + 1839 + ], + "score": 0.93, + "latex": "\\mathcal { F } _ { \\mathrm { f e a t } } ( D )" + }, + { + "category_id": 13, + "poly": [ + 672, + 1382, + 772, + 1382, + 772, + 1417, + 672, + 1417 + ], + "score": 0.93, + "latex": "M _ { i j } = 0" + }, + { + "category_id": 13, + "poly": [ + 371, + 582, + 516, + 582, + 516, + 622, + 371, + 622 + ], + "score": 0.93, + "latex": "V = ( V _ { i } ) _ { i = 1 } ^ { K _ { 1 } }" + }, + { + "category_id": 13, + "poly": [ + 1016, + 1886, + 1104, + 1886, + 1104, + 1920, + 1016, + 1920 + ], + "score": 0.92, + "latex": "\\langle \\mathcal { V } , \\mathcal { T } _ { T } \\rangle" + }, + { + "category_id": 13, + "poly": [ + 425, + 884, + 617, + 884, + 617, + 916, + 425, + 916 + ], + "score": 0.92, + "latex": "\\Lambda : W _ { \\ell _ { 1 } } \\to W _ { \\ell _ { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1916, + 481, + 1916, + 481, + 1951, + 298, + 1951 + ], + "score": 0.92, + "latex": "p : \\mathbb { R } ^ { 3 \\times n } \\overset { \\cdot } { } \\mathcal { T } _ { T } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 911, + 1606, + 1101, + 1606, + 1101, + 1642, + 911, + 1642 + ], + "score": 0.92, + "latex": "p : \\mathbb { R } ^ { 3 \\times n } W _ { T } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 731, + 945, + 827, + 945, + 827, + 975, + 731, + 975 + ], + "score": 0.92, + "latex": "\\ell _ { 1 } = \\ell _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 570, + 585, + 691, + 585, + 691, + 631, + 570, + 631 + ], + "score": 0.92, + "latex": "V _ { i } \\in W _ { \\ell _ { i } ^ { ( 1 ) } }" + }, + { + "category_id": 13, + "poly": [ + 528, + 1544, + 710, + 1544, + 710, + 1579, + 528, + 1579 + ], + "score": 0.92, + "latex": "n \\in \\mathbb { N } , l _ { T } \\in \\mathbb { N } _ { + } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 463, + 946, + 559, + 946, + 559, + 975, + 463, + 975 + ], + "score": 0.92, + "latex": "\\ell _ { 1 } = \\ell _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 833, + 782, + 1031, + 782, + 1031, + 814, + 833, + 814 + ], + "score": 0.92, + "latex": "\\Lambda : W _ { \\ell _ { 1 } } \\to W _ { \\ell _ { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 297, + 811, + 385, + 811, + 385, + 844, + 297, + 844 + ], + "score": 0.92, + "latex": "\\ell _ { 1 } \\neq \\ell _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 591, + 1292, + 653, + 1292, + 653, + 1325, + 591, + 1325 + ], + "score": 0.92, + "latex": "W _ { l ^ { ( 1 ) } }" + }, + { + "category_id": 13, + "poly": [ + 1291, + 1604, + 1392, + 1604, + 1392, + 1648, + 1291, + 1648 + ], + "score": 0.92, + "latex": "\\mathcal { F } _ { C ( D ) , D } ^ { \\mathrm { T F N } }" + }, + { + "category_id": 13, + "poly": [ + 963, + 915, + 1043, + 915, + 1043, + 946, + 963, + 946 + ], + "score": 0.92, + "latex": "\\Lambda \\neq 0" + }, + { + "category_id": 13, + "poly": [ + 389, + 414, + 490, + 414, + 490, + 450, + 389, + 450 + ], + "score": 0.91, + "latex": "M _ { i j } = 0" + }, + { + "category_id": 13, + "poly": [ + 774, + 586, + 933, + 586, + 933, + 619, + 774, + 619 + ], + "score": 0.91, + "latex": "i = 1 , \\ldots , K _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1204, + 1338, + 1400, + 1338, + 1400, + 1376, + 1204, + 1376 + ], + "score": 0.91, + "latex": "\\Lambda _ { i j } ( V _ { i } ) = M _ { i j } V _ { i }" + }, + { + "category_id": 13, + "poly": [ + 1234, + 1291, + 1404, + 1291, + 1404, + 1335, + 1234, + 1335 + ], + "score": 0.91, + "latex": "\\Lambda _ { i j } : W _ { \\ell _ { i } ^ { ( 1 ) } } \\to" + }, + { + "category_id": 13, + "poly": [ + 567, + 1695, + 752, + 1695, + 752, + 1737, + 567, + 1737 + ], + "score": 0.91, + "latex": "\\cup _ { D \\in \\mathbb { N } _ { + } } \\mathcal { F } _ { C ( D ) , D } ^ { \\mathrm { T F N } }" + }, + { + "category_id": 13, + "poly": [ + 795, + 1289, + 1037, + 1289, + 1037, + 1325, + 795, + 1325 + ], + "score": 0.91, + "latex": "( 0 , \\ldots , 0 , V _ { i } , 0 \\ldots , 0 )" + }, + { + "category_id": 13, + "poly": [ + 450, + 1383, + 499, + 1383, + 499, + 1417, + 450, + 1417 + ], + "score": 0.91, + "latex": "M _ { i j }" + }, + { + "category_id": 13, + "poly": [ + 1118, + 945, + 1166, + 945, + 1166, + 978, + 1118, + 978 + ], + "score": 0.91, + "latex": "W _ { \\ell _ { 1 } }" + }, + { + "category_id": 13, + "poly": [ + 1333, + 1837, + 1389, + 1837, + 1389, + 1871, + 1333, + 1871 + ], + "score": 0.91, + "latex": "\\mathcal { G } _ { J , D }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1340, + 361, + 1340, + 361, + 1387, + 297, + 1387 + ], + "score": 0.91, + "latex": "W _ { \\ell _ { j } ^ { ( 2 ) } }" + }, + { + "category_id": 13, + "poly": [ + 679, + 976, + 767, + 976, + 767, + 1005, + 679, + 1005 + ], + "score": 0.91, + "latex": "\\Lambda - \\lambda I" + }, + { + "category_id": 13, + "poly": [ + 684, + 1884, + 839, + 1884, + 839, + 1914, + 684, + 1914 + ], + "score": 0.9, + "latex": "\\mathbb { R } ^ { 3 \\times n } Y ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 442, + 1608, + 539, + 1608, + 539, + 1640, + 442, + 1640 + ], + "score": 0.9, + "latex": "D \\in { \\mathbb { N } } _ { + }" + }, + { + "category_id": 13, + "poly": [ + 554, + 917, + 603, + 917, + 603, + 947, + 554, + 947 + ], + "score": 0.9, + "latex": "W _ { \\ell _ { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 472, + 338, + 718, + 338, + 718, + 383, + 472, + 383 + ], + "score": 0.9, + "latex": "l ^ { ( 1 ) } = ( \\ell _ { 1 } ^ { ( 1 ) } , \\dots , \\ell _ { K _ { 1 } } ^ { ( 1 ) } )" + }, + { + "category_id": 13, + "poly": [ + 1152, + 347, + 1374, + 347, + 1374, + 380, + 1152, + 380 + ], + "score": 0.9, + "latex": "\\boldsymbol { \\Lambda } = \\left( \\Lambda _ { 1 } , \\ldots , \\Lambda _ { K _ { 2 } } \\right)" + }, + { + "category_id": 13, + "poly": [ + 449, + 916, + 498, + 916, + 498, + 947, + 449, + 947 + ], + "score": 0.9, + "latex": "W _ { \\ell _ { 1 } }" + }, + { + "category_id": 13, + "poly": [ + 903, + 1118, + 1137, + 1118, + 1137, + 1163, + 903, + 1163 + ], + "score": 0.89, + "latex": "W _ { \\ell _ { i } ^ { ( 1 ) } } , i = 1 , \\dots , K _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 772, + 340, + 1017, + 340, + 1017, + 381, + 772, + 381 + ], + "score": 0.89, + "latex": "l ^ { ( 2 ) } = ( \\ell _ { 1 } ^ { ( 2 ) } , \\dots , \\ell _ { K _ { 2 } } ^ { ( 2 ) } )" + }, + { + "category_id": 13, + "poly": [ + 1214, + 381, + 1320, + 381, + 1320, + 408, + 1214, + 408 + ], + "score": 0.89, + "latex": "K _ { 1 } \\times K _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 727, + 379, + 787, + 379, + 787, + 410, + 727, + 410 + ], + "score": 0.88, + "latex": "W _ { l ^ { ( 1 ) } }" + }, + { + "category_id": 13, + "poly": [ + 940, + 1078, + 1171, + 1078, + 1171, + 1116, + 940, + 1116 + ], + "score": 0.88, + "latex": "\\Lambda _ { j } : W _ { l ^ { ( 1 ) } } \\to W _ { \\ell _ { \\vec { \\cdot } } ^ { ( 2 ) } }" + }, + { + "category_id": 13, + "poly": [ + 444, + 812, + 516, + 812, + 516, + 840, + 444, + 840 + ], + "score": 0.88, + "latex": "\\Lambda = 0" + }, + { + "category_id": 13, + "poly": [ + 1282, + 976, + 1404, + 976, + 1404, + 1005, + 1282, + 1005 + ], + "score": 0.88, + "latex": "\\Lambda - \\lambda I =" + }, + { + "category_id": 13, + "poly": [ + 843, + 381, + 904, + 381, + 904, + 410, + 843, + 410 + ], + "score": 0.88, + "latex": "W _ { l ^ { ( 2 ) } }" + }, + { + "category_id": 13, + "poly": [ + 1176, + 1121, + 1216, + 1121, + 1216, + 1154, + 1176, + 1154 + ], + "score": 0.88, + "latex": "\\Lambda _ { i j }" + }, + { + "category_id": 13, + "poly": [ + 835, + 1120, + 869, + 1120, + 869, + 1154, + 835, + 1154 + ], + "score": 0.87, + "latex": "\\Lambda _ { j }" + }, + { + "category_id": 13, + "poly": [ + 325, + 1888, + 350, + 1888, + 350, + 1916, + 325, + 1916 + ], + "score": 0.84, + "latex": "\\mathcal { V }" + }, + { + "category_id": 13, + "poly": [ + 1257, + 1808, + 1283, + 1808, + 1283, + 1834, + 1257, + 1834 + ], + "score": 0.82, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 863, + 1807, + 904, + 1807, + 904, + 1835, + 863, + 1835 + ], + "score": 0.81, + "latex": "2 D" + }, + { + "category_id": 13, + "poly": [ + 989, + 1838, + 1010, + 1838, + 1010, + 1865, + 989, + 1865 + ], + "score": 0.81, + "latex": "J" + }, + { + "category_id": 13, + "poly": [ + 599, + 1668, + 623, + 1668, + 623, + 1694, + 599, + 1694 + ], + "score": 0.8, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 616, + 1609, + 640, + 1609, + 640, + 1635, + 616, + 1635 + ], + "score": 0.8, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 1105, + 916, + 1127, + 916, + 1127, + 942, + 1105, + 942 + ], + "score": 0.79, + "latex": "\\Lambda" + }, + { + "category_id": 13, + "poly": [ + 1210, + 1609, + 1237, + 1609, + 1237, + 1635, + 1210, + 1635 + ], + "score": 0.78, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 298, + 1808, + 324, + 1808, + 324, + 1834, + 298, + 1834 + ], + "score": 0.76, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 423, + 977, + 443, + 977, + 443, + 1003, + 423, + 1003 + ], + "score": 0.75, + "latex": "\\lambda" + }, + { + "category_id": 13, + "poly": [ + 1231, + 885, + 1254, + 885, + 1254, + 911, + 1231, + 911 + ], + "score": 0.75, + "latex": "\\Lambda" + }, + { + "category_id": 13, + "poly": [ + 1260, + 946, + 1283, + 946, + 1283, + 973, + 1260, + 973 + ], + "score": 0.74, + "latex": "\\Lambda" + }, + { + "category_id": 13, + "poly": [ + 298, + 415, + 331, + 415, + 331, + 445, + 298, + 445 + ], + "score": 0.69, + "latex": "M" + }, + { + "category_id": 13, + "poly": [ + 650, + 813, + 673, + 813, + 673, + 839, + 650, + 839 + ], + "score": 0.43, + "latex": "\\Lambda" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 815.0, + 72.0, + 815.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 223.0, + 702.0, + 223.0, + 702.0, + 267.0, + 295.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2125.0, + 832.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1380.0, + 1435.0, + 1401.0, + 1435.0, + 1401.0, + 1458.0, + 1380.0, + 1458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1280.0, + 590.0, + 1280.0, + 590.0, + 1338.0, + 289.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 654.0, + 1280.0, + 794.0, + 1280.0, + 794.0, + 1338.0, + 654.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 1280.0, + 1233.0, + 1280.0, + 1233.0, + 1338.0, + 1038.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1324.0, + 296.0, + 1324.0, + 296.0, + 1387.0, + 288.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1324.0, + 1023.0, + 1324.0, + 1023.0, + 1387.0, + 362.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1146.0, + 1324.0, + 1203.0, + 1324.0, + 1203.0, + 1387.0, + 1146.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 1324.0, + 1411.0, + 1324.0, + 1411.0, + 1387.0, + 1401.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1378.0, + 449.0, + 1378.0, + 449.0, + 1422.0, + 292.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 500.0, + 1378.0, + 671.0, + 1378.0, + 671.0, + 1422.0, + 500.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 773.0, + 1378.0, + 1386.0, + 1378.0, + 1386.0, + 1422.0, + 773.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 881.0, + 424.0, + 881.0, + 424.0, + 918.0, + 295.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 618.0, + 881.0, + 1230.0, + 881.0, + 1230.0, + 918.0, + 618.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 881.0, + 1407.0, + 881.0, + 1407.0, + 918.0, + 1255.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 913.0, + 448.0, + 913.0, + 448.0, + 950.0, + 294.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 913.0, + 553.0, + 913.0, + 553.0, + 950.0, + 499.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 604.0, + 913.0, + 962.0, + 913.0, + 962.0, + 950.0, + 604.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 913.0, + 1104.0, + 913.0, + 1104.0, + 950.0, + 1044.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 913.0, + 1406.0, + 913.0, + 1406.0, + 950.0, + 1128.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 942.0, + 462.0, + 942.0, + 462.0, + 980.0, + 293.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 942.0, + 730.0, + 942.0, + 730.0, + 980.0, + 560.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 942.0, + 1117.0, + 942.0, + 1117.0, + 980.0, + 828.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1167.0, + 942.0, + 1259.0, + 942.0, + 1259.0, + 980.0, + 1167.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1284.0, + 942.0, + 1406.0, + 942.0, + 1406.0, + 980.0, + 1284.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 973.0, + 422.0, + 973.0, + 422.0, + 1011.0, + 294.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 444.0, + 973.0, + 678.0, + 973.0, + 678.0, + 1011.0, + 444.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 973.0, + 1281.0, + 973.0, + 1281.0, + 1011.0, + 768.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1005.0, + 327.0, + 1005.0, + 327.0, + 1039.0, + 294.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1374.0, + 1008.0, + 1403.0, + 1008.0, + 1403.0, + 1036.0, + 1374.0, + 1036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 321.0, + 471.0, + 321.0, + 471.0, + 405.0, + 288.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 321.0, + 726.0, + 321.0, + 726.0, + 405.0, + 719.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1018.0, + 321.0, + 1151.0, + 321.0, + 1151.0, + 405.0, + 1018.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1375.0, + 321.0, + 1420.0, + 321.0, + 1420.0, + 405.0, + 1375.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 376.0, + 726.0, + 376.0, + 726.0, + 413.0, + 294.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 376.0, + 842.0, + 376.0, + 842.0, + 413.0, + 788.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 905.0, + 376.0, + 1213.0, + 376.0, + 1213.0, + 413.0, + 905.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1321.0, + 376.0, + 1406.0, + 376.0, + 1406.0, + 413.0, + 1321.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 399.0, + 297.0, + 399.0, + 297.0, + 459.0, + 288.0, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 399.0, + 388.0, + 399.0, + 388.0, + 459.0, + 332.0, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 399.0, + 605.0, + 399.0, + 605.0, + 459.0, + 491.0, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 727.0, + 399.0, + 855.0, + 399.0, + 855.0, + 459.0, + 727.0, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 708.0, + 425.0, + 726.0, + 425.0, + 726.0, + 435.5, + 708.0, + 435.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 664.0, + 1403.0, + 664.0, + 1403.0, + 703.0, + 294.0, + 703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 698.0, + 1402.0, + 698.0, + 1402.0, + 733.0, + 296.0, + 733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 726.0, + 1339.0, + 726.0, + 1339.0, + 762.0, + 295.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1074.0, + 939.0, + 1074.0, + 939.0, + 1116.0, + 294.0, + 1116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 1074.0, + 1403.0, + 1074.0, + 1403.0, + 1116.0, + 1172.0, + 1116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1118.0, + 834.0, + 1118.0, + 834.0, + 1157.0, + 294.0, + 1157.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 1118.0, + 902.0, + 1118.0, + 902.0, + 1157.0, + 870.0, + 1157.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1138.0, + 1118.0, + 1175.0, + 1118.0, + 1175.0, + 1157.0, + 1138.0, + 1157.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 1118.0, + 1379.0, + 1118.0, + 1379.0, + 1157.0, + 1217.0, + 1157.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1772.0, + 1266.0, + 1772.0, + 1266.0, + 1812.0, + 293.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1374.0, + 1772.0, + 1407.0, + 1772.0, + 1407.0, + 1812.0, + 1374.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1805.0, + 297.0, + 1805.0, + 297.0, + 1843.0, + 294.0, + 1843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1805.0, + 582.0, + 1805.0, + 582.0, + 1843.0, + 325.0, + 1843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 689.0, + 1805.0, + 862.0, + 1805.0, + 862.0, + 1843.0, + 689.0, + 1843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 905.0, + 1805.0, + 1256.0, + 1805.0, + 1256.0, + 1843.0, + 905.0, + 1843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1284.0, + 1805.0, + 1406.0, + 1805.0, + 1406.0, + 1843.0, + 1284.0, + 1843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1837.0, + 988.0, + 1837.0, + 988.0, + 1875.0, + 294.0, + 1875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1011.0, + 1837.0, + 1332.0, + 1837.0, + 1332.0, + 1875.0, + 1011.0, + 1875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1390.0, + 1837.0, + 1403.0, + 1837.0, + 1403.0, + 1875.0, + 1390.0, + 1875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 774.0, + 832.0, + 774.0, + 832.0, + 821.0, + 292.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 774.0, + 1409.0, + 774.0, + 1409.0, + 821.0, + 1032.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 386.0, + 810.0, + 443.0, + 810.0, + 443.0, + 847.0, + 386.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 810.0, + 649.0, + 810.0, + 649.0, + 847.0, + 517.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 810.0, + 1059.0, + 810.0, + 1059.0, + 847.0, + 674.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1878.0, + 324.0, + 1878.0, + 324.0, + 1925.0, + 290.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1878.0, + 683.0, + 1878.0, + 683.0, + 1925.0, + 351.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 1878.0, + 1015.0, + 1878.0, + 1015.0, + 1925.0, + 840.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1105.0, + 1878.0, + 1406.0, + 1878.0, + 1406.0, + 1925.0, + 1105.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 1908.0, + 297.0, + 1908.0, + 297.0, + 1954.0, + 287.0, + 1954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 1908.0, + 621.0, + 1908.0, + 621.0, + 1954.0, + 482.0, + 1954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 574.0, + 370.0, + 574.0, + 370.0, + 629.0, + 289.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 574.0, + 569.0, + 574.0, + 569.0, + 629.0, + 517.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 574.0, + 773.0, + 574.0, + 773.0, + 629.0, + 692.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 574.0, + 950.0, + 574.0, + 950.0, + 629.0, + 934.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1498.0, + 1364.0, + 1498.0, + 1364.0, + 1536.0, + 295.0, + 1536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 294.0, + 524.0, + 294.0, + 524.0, + 333.0, + 295.0, + 333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1538.0, + 527.0, + 1538.0, + 527.0, + 1584.0, + 295.0, + 1584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 1538.0, + 722.0, + 1538.0, + 722.0, + 1584.0, + 711.0, + 1584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1584.0, + 441.0, + 1584.0, + 441.0, + 1666.0, + 349.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 1584.0, + 615.0, + 1584.0, + 615.0, + 1666.0, + 540.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 1584.0, + 910.0, + 1584.0, + 910.0, + 1666.0, + 641.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1102.0, + 1584.0, + 1209.0, + 1584.0, + 1209.0, + 1666.0, + 1102.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1584.0, + 1290.0, + 1584.0, + 1290.0, + 1666.0, + 1238.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1393.0, + 1584.0, + 1414.0, + 1584.0, + 1414.0, + 1666.0, + 1393.0, + 1666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 1662.0, + 598.0, + 1662.0, + 598.0, + 1703.0, + 357.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 624.0, + 1662.0, + 1404.0, + 1662.0, + 1404.0, + 1703.0, + 624.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 1692.0, + 758.0, + 1692.0, + 758.0, + 1739.0, + 390.0, + 1739.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 15, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 468, + 1405, + 468, + 1405, + 702, + 296, + 702 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 296, + 224, + 1405, + 224, + 1405, + 459, + 296, + 459 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 296, + 1628, + 1404, + 1628, + 1404, + 1758, + 296, + 1758 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 297, + 1438, + 1404, + 1438, + 1404, + 1570, + 297, + 1570 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 300, + 969, + 1403, + 969, + 1403, + 1065, + 300, + 1065 + ], + "score": 0.962 + }, + { + "category_id": 8, + "poly": [ + 568, + 1916, + 1128, + 1916, + 1128, + 1999, + 568, + 1999 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 300, + 1826, + 1406, + 1826, + 1406, + 1916, + 300, + 1916 + ], + "score": 0.952 + }, + { + "category_id": 8, + "poly": [ + 648, + 1286, + 1051, + 1286, + 1051, + 1369, + 648, + 1369 + ], + "score": 0.951 + }, + { + "category_id": 8, + "poly": [ + 672, + 1764, + 1028, + 1764, + 1028, + 1807, + 672, + 1807 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 299, + 712, + 1399, + 712, + 1399, + 776, + 299, + 776 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 296, + 1220, + 1401, + 1220, + 1401, + 1283, + 296, + 1283 + ], + "score": 0.94 + }, + { + "category_id": 1, + "poly": [ + 296, + 1077, + 1403, + 1077, + 1403, + 1142, + 296, + 1142 + ], + "score": 0.939 + }, + { + "category_id": 1, + "poly": [ + 296, + 2001, + 527, + 2001, + 527, + 2038, + 296, + 2038 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 297, + 1371, + 532, + 1371, + 532, + 1408, + 297, + 1408 + ], + "score": 0.922 + }, + { + "category_id": 1, + "poly": [ + 299, + 1174, + 784, + 1174, + 784, + 1208, + 299, + 1208 + ], + "score": 0.921 + }, + { + "category_id": 2, + "poly": [ + 299, + 74, + 813, + 74, + 813, + 105, + 299, + 105 + ], + "score": 0.911 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1771, + 1400, + 1771, + 1400, + 1801, + 1352, + 1801 + ], + "score": 0.908 + }, + { + "category_id": 1, + "poly": [ + 297, + 788, + 705, + 788, + 705, + 823, + 297, + 823 + ], + "score": 0.901 + }, + { + "category_id": 2, + "poly": [ + 836, + 2087, + 865, + 2087, + 865, + 2112, + 836, + 2112 + ], + "score": 0.872 + }, + { + "category_id": 1, + "poly": [ + 360, + 850, + 717, + 850, + 717, + 944, + 360, + 944 + ], + "score": 0.87 + }, + { + "category_id": 2, + "poly": [ + 1374, + 2003, + 1403, + 2003, + 1403, + 2032, + 1374, + 2032 + ], + "score": 0.845 + }, + { + "category_id": 8, + "poly": [ + 617, + 1578, + 1078, + 1578, + 1078, + 1622, + 617, + 1622 + ], + "score": 0.833 + }, + { + "category_id": 8, + "poly": [ + 618, + 1578, + 1078, + 1578, + 1078, + 1623, + 618, + 1623 + ], + "score": 0.441 + }, + { + "category_id": 8, + "poly": [ + 360, + 850, + 717, + 850, + 717, + 944, + 360, + 944 + ], + "score": 0.113 + }, + { + "category_id": 14, + "poly": [ + 647, + 1279, + 1052, + 1279, + 1052, + 1371, + 647, + 1371 + ], + "score": 0.95, + "latex": "q _ { a } ( X ) = \\sum _ { b = 1 } ^ { n } ( { \\bar { x } } _ { a } - { \\bar { x } } _ { b } ) ^ { \\otimes s } \\otimes p _ { b } ( X )" + }, + { + "category_id": 13, + "poly": [ + 1109, + 387, + 1335, + 387, + 1335, + 426, + 1109, + 426 + ], + "score": 0.94, + "latex": "f ( X ) \\ = \\ \\hat { A } \\circ g ( X )" + }, + { + "category_id": 13, + "poly": [ + 474, + 662, + 754, + 662, + 754, + 701, + 474, + 701 + ], + "score": 0.94, + "latex": "\\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { T } \\rangle \\subseteq \\langle \\tilde { \\mathcal { G } } _ { J , D } , \\mathcal { T } _ { T } \\rangle" + }, + { + "category_id": 13, + "poly": [ + 1299, + 261, + 1400, + 261, + 1400, + 299, + 1299, + 299 + ], + "score": 0.93, + "latex": "Q ^ { ( \\vec { r } ) } ( \\bar { X } )" + }, + { + "category_id": 13, + "poly": [ + 539, + 790, + 695, + 790, + 695, + 823, + 539, + 823 + ], + "score": 0.93, + "latex": "J , D , T \\in \\mathbb { N } _ { + }" + }, + { + "category_id": 13, + "poly": [ + 506, + 1002, + 658, + 1002, + 658, + 1036, + 506, + 1036 + ], + "score": 0.93, + "latex": "\\langle \\mathcal { G } _ { J + 1 , D } , \\mathcal { T } _ { T } \\rangle" + }, + { + "category_id": 13, + "poly": [ + 764, + 1469, + 923, + 1469, + 923, + 1505, + 764, + 1505 + ], + "score": 0.93, + "latex": "\\| x \\| ^ { s - \\ell } Y _ { m } ^ { \\ell } ( \\bar { x } )" + }, + { + "category_id": 14, + "poly": [ + 566, + 1909, + 1129, + 1909, + 1129, + 2000, + 566, + 2000 + ], + "score": 0.93, + "latex": "\\sum _ { b = 1 } ^ { n } ( x _ { a } - x _ { b } ) \\otimes p _ { b } ( X ) = \\sum _ { b = 1 } ^ { n } ( \\bar { x } _ { a } - \\bar { x } _ { b } ) \\otimes p _ { b } ( X )" + }, + { + "category_id": 13, + "poly": [ + 1081, + 971, + 1204, + 971, + 1204, + 1005, + 1081, + 1005 + ], + "score": 0.93, + "latex": "\\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { T } \\rangle" + }, + { + "category_id": 13, + "poly": [ + 298, + 1535, + 507, + 1535, + 507, + 1572, + 298, + 1572 + ], + "score": 0.93, + "latex": "R ^ { ( \\ell ) } ( \\| x \\| ) = \\| x \\| ^ { s }" + }, + { + "category_id": 13, + "poly": [ + 358, + 296, + 499, + 296, + 499, + 330, + 358, + 330 + ], + "score": 0.93, + "latex": "\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { T } \\rangle" + }, + { + "category_id": 13, + "poly": [ + 767, + 226, + 946, + 226, + 946, + 265, + 767, + 265 + ], + "score": 0.93, + "latex": "\\hat { A } _ { k } : Y ^ { n } \\mathcal { T } _ { T } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 1130, + 530, + 1188, + 530, + 1188, + 569, + 1130, + 569 + ], + "score": 0.92, + "latex": "\\tilde { \\mathcal { G } } _ { J , D }" + }, + { + "category_id": 13, + "poly": [ + 539, + 626, + 597, + 626, + 597, + 664, + 539, + 664 + ], + "score": 0.92, + "latex": "\\tilde { \\mathcal { G } } _ { J , D }" + }, + { + "category_id": 13, + "poly": [ + 1249, + 625, + 1404, + 625, + 1404, + 665, + 1249, + 665 + ], + "score": 0.92, + "latex": "\\langle \\tilde { \\mathcal { G } } _ { J , D } , \\mathcal { T } _ { T } \\rangle \\subseteq" + }, + { + "category_id": 13, + "poly": [ + 299, + 665, + 422, + 665, + 422, + 701, + 299, + 701 + ], + "score": 0.92, + "latex": "\\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { T } \\rangle" + }, + { + "category_id": 13, + "poly": [ + 373, + 231, + 533, + 231, + 533, + 262, + 373, + 262 + ], + "score": 0.92, + "latex": "A _ { k } : Y \\mathcal { T } _ { T }" + }, + { + "category_id": 13, + "poly": [ + 585, + 296, + 683, + 296, + 683, + 327, + 585, + 327 + ], + "score": 0.92, + "latex": "\\vec { r } \\in \\Sigma _ { T }" + }, + { + "category_id": 13, + "poly": [ + 820, + 391, + 989, + 391, + 989, + 423, + 820, + 423 + ], + "score": 0.92, + "latex": "A : Y _ { 1 } Y _ { 2 }" + }, + { + "category_id": 14, + "poly": [ + 618, + 1577, + 1080, + 1577, + 1080, + 1621, + 618, + 1621 + ], + "score": 0.92, + "latex": "F _ { m } ^ { ( \\ell ) } ( x ) = \\| x \\| ^ { s } Y _ { m } ^ { \\ell } ( \\hat { x } ) = \\| x \\| ^ { s - \\ell } Y _ { m } ^ { \\ell } ( x ) ." + }, + { + "category_id": 13, + "poly": [ + 1146, + 1859, + 1202, + 1859, + 1202, + 1893, + 1146, + 1893 + ], + "score": 0.92, + "latex": "\\mathcal { G } _ { 1 , D }" + }, + { + "category_id": 13, + "poly": [ + 577, + 1223, + 709, + 1223, + 709, + 1255, + 577, + 1255 + ], + "score": 0.92, + "latex": "J , D \\in \\mathbb { N } _ { + }" + }, + { + "category_id": 13, + "poly": [ + 671, + 1725, + 852, + 1725, + 852, + 1759, + 671, + 1759 + ], + "score": 0.92, + "latex": "B _ { s } : W _ { l _ { D } } T _ { s }" + }, + { + "category_id": 13, + "poly": [ + 775, + 296, + 870, + 296, + 870, + 326, + 775, + 326 + ], + "score": 0.92, + "latex": "T \\leq D" + }, + { + "category_id": 13, + "poly": [ + 1013, + 1626, + 1321, + 1626, + 1321, + 1664, + 1013, + 1664 + ], + "score": 0.92, + "latex": "F ^ { s , D } = ( F ^ { ( 0 ) } , \\ldots , F ^ { ( D ) } )" + }, + { + "category_id": 13, + "poly": [ + 706, + 327, + 784, + 327, + 784, + 358, + 706, + 358 + ], + "score": 0.92, + "latex": "\\mathcal { \\scriptsize { N } } _ { 1 } , \\mathcal { \\scriptsize { N } } _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 628, + 1506, + 713, + 1506, + 713, + 1536, + 628, + 1536 + ], + "score": 0.92, + "latex": "s \\leq D" + }, + { + "category_id": 13, + "poly": [ + 477, + 356, + 677, + 356, + 677, + 390, + 477, + 390 + ], + "score": 0.91, + "latex": "\\langle y _ { i } , \\mathcal { T } _ { T } \\rangle , i = \\bar { 1 } , 2" + }, + { + "category_id": 13, + "poly": [ + 430, + 629, + 489, + 629, + 489, + 664, + 430, + 664 + ], + "score": 0.91, + "latex": "\\mathcal { G } _ { J , D }" + }, + { + "category_id": 14, + "poly": [ + 671, + 1764, + 1029, + 1764, + 1029, + 1806, + 671, + 1806 + ], + "score": 0.91, + "latex": "x ^ { \\otimes s } = B _ { s } ( F ^ { s , D } ( { \\boldsymbol { x } } ) ) , \\forall { \\boldsymbol { x } } \\in \\mathbb { R } ^ { 3 } ." + }, + { + "category_id": 13, + "poly": [ + 419, + 1631, + 518, + 1631, + 518, + 1659, + 419, + 1659 + ], + "score": 0.91, + "latex": "D \\ \\in \\ \\mathbb { N }" + }, + { + "category_id": 13, + "poly": [ + 576, + 1631, + 669, + 1631, + 669, + 1661, + 576, + 1661 + ], + "score": 0.91, + "latex": "s \\leq D" + }, + { + "category_id": 13, + "poly": [ + 345, + 265, + 433, + 265, + 433, + 297, + 345, + 297 + ], + "score": 0.91, + "latex": "f _ { k } \\in \\mathcal { V }" + }, + { + "category_id": 13, + "poly": [ + 1313, + 1829, + 1369, + 1829, + 1369, + 1861, + 1313, + 1861 + ], + "score": 0.91, + "latex": "\\mathcal { G } _ { J , D }" + }, + { + "category_id": 13, + "poly": [ + 819, + 1003, + 890, + 1003, + 890, + 1032, + 819, + 1032 + ], + "score": 0.91, + "latex": "J + 1" + }, + { + "category_id": 13, + "poly": [ + 516, + 393, + 610, + 393, + 610, + 424, + 516, + 424 + ], + "score": 0.9, + "latex": "g \\in \\mathcal { V } 1" + }, + { + "category_id": 13, + "poly": [ + 993, + 1222, + 1170, + 1222, + 1170, + 1257, + 993, + 1257 + ], + "score": 0.9, + "latex": "\\dot { \\boldsymbol { p } } \\in \\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { t } \\rangle" + }, + { + "category_id": 13, + "poly": [ + 297, + 1033, + 399, + 1033, + 399, + 1064, + 297, + 1064 + ], + "score": 0.9, + "latex": "1 , F = 0" + }, + { + "category_id": 13, + "poly": [ + 1170, + 1080, + 1248, + 1080, + 1248, + 1109, + 1170, + 1109 + ], + "score": 0.9, + "latex": "D + 1" + }, + { + "category_id": 13, + "poly": [ + 353, + 1372, + 524, + 1372, + 524, + 1407, + 353, + 1407 + ], + "score": 0.9, + "latex": "\\langle \\mathcal { G } _ { J + 1 , D } , \\mathcal { T } _ { t + s } \\rangle" + }, + { + "category_id": 13, + "poly": [ + 297, + 1693, + 396, + 1693, + 396, + 1727, + 297, + 1727 + ], + "score": 0.9, + "latex": "F ^ { s , D } ( x )" + }, + { + "category_id": 13, + "poly": [ + 842, + 1223, + 960, + 1223, + 960, + 1255, + 842, + 1255 + ], + "score": 0.9, + "latex": "t , s \\in \\mathbb { N } _ { + }" + }, + { + "category_id": 13, + "poly": [ + 298, + 423, + 532, + 423, + 532, + 458, + 298, + 458 + ], + "score": 0.9, + "latex": "\\langle { \\mathcal { V } } _ { 2 } , { \\mathcal { T } } _ { T } \\rangle \\subseteq \\langle { \\mathcal { V } } _ { 1 } , { \\mathcal { T } } _ { T } \\rangle" + }, + { + "category_id": 13, + "poly": [ + 298, + 391, + 393, + 391, + 393, + 424, + 298, + 424 + ], + "score": 0.9, + "latex": "f \\in \\mathcal { V } _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1228, + 1502, + 1280, + 1502, + 1280, + 1533, + 1228, + 1533 + ], + "score": 0.9, + "latex": "F ^ { ( \\ell ) }" + }, + { + "category_id": 13, + "poly": [ + 976, + 1472, + 1253, + 1472, + 1253, + 1503, + 976, + 1503 + ], + "score": 0.89, + "latex": "\\ell = s , s - 2 , s - 4 \\ldots" + }, + { + "category_id": 13, + "poly": [ + 617, + 1858, + 663, + 1858, + 663, + 1886, + 617, + 1886 + ], + "score": 0.89, + "latex": "x ^ { \\otimes s }" + }, + { + "category_id": 13, + "poly": [ + 614, + 1694, + 661, + 1694, + 661, + 1723, + 614, + 1723 + ], + "score": 0.89, + "latex": "x ^ { \\otimes s }" + }, + { + "category_id": 13, + "poly": [ + 353, + 2002, + 525, + 2002, + 525, + 2038, + 353, + 2038 + ], + "score": 0.89, + "latex": "\\langle \\mathcal { G } _ { J + 1 , D } , \\mathcal { T } _ { t + s } \\rangle" + }, + { + "category_id": 13, + "poly": [ + 298, + 1111, + 372, + 1111, + 372, + 1139, + 298, + 1139 + ], + "score": 0.88, + "latex": "D + 1" + }, + { + "category_id": 13, + "poly": [ + 1243, + 1693, + 1278, + 1693, + 1278, + 1722, + 1243, + 1722 + ], + "score": 0.88, + "latex": "\\mathbb { R } ^ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 1341, + 1003, + 1405, + 1003, + 1405, + 1034, + 1341, + 1034 + ], + "score": 0.87, + "latex": "\\theta _ { 0 } =" + }, + { + "category_id": 13, + "poly": [ + 1365, + 1438, + 1400, + 1438, + 1400, + 1468, + 1365, + 1468 + ], + "score": 0.86, + "latex": "\\mathbb { R } ^ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 1023, + 973, + 1044, + 973, + 1044, + 1004, + 1023, + 1004 + ], + "score": 0.85, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 298, + 1661, + 486, + 1661, + 486, + 1694, + 298, + 1694 + ], + "score": 0.84, + "latex": "F ^ { ( s ) } , F ^ { ( s - 2 ) } , \\ldots ." + }, + { + "category_id": 13, + "poly": [ + 916, + 1631, + 943, + 1631, + 943, + 1658, + 916, + 1658 + ], + "score": 0.83, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 808, + 1081, + 834, + 1081, + 834, + 1107, + 808, + 1107 + ], + "score": 0.83, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 430, + 1834, + 448, + 1834, + 448, + 1859, + 430, + 1859 + ], + "score": 0.81, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 865, + 567, + 891, + 567, + 891, + 593, + 865, + 593 + ], + "score": 0.79, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 877, + 1700, + 893, + 1700, + 893, + 1722, + 877, + 1722 + ], + "score": 0.78, + "latex": "s" + }, + { + "category_id": 13, + "poly": [ + 395, + 905, + 709, + 905, + 709, + 941, + 395, + 941 + ], + "score": 0.76, + "latex": "\\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { T } \\rangle \\subseteq \\langle \\mathcal { G } _ { J , D + 1 } , \\mathcal { T } _ { T } \\rangle ." + }, + { + "category_id": 13, + "poly": [ + 340, + 745, + 367, + 745, + 367, + 771, + 340, + 771 + ], + "score": 0.75, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 506, + 567, + 528, + 567, + 528, + 594, + 506, + 594 + ], + "score": 0.75, + "latex": "J" + }, + { + "category_id": 13, + "poly": [ + 369, + 852, + 713, + 852, + 713, + 890, + 369, + 890 + ], + "score": 0.75, + "latex": "l . \\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { T } \\rangle \\subseteq \\langle \\mathcal { G } _ { J + 1 , D } , \\mathcal { T } _ { T } \\rangle ." + }, + { + "category_id": 13, + "poly": [ + 957, + 716, + 977, + 716, + 977, + 742, + 957, + 742 + ], + "score": 0.74, + "latex": "J" + }, + { + "category_id": 13, + "poly": [ + 992, + 1448, + 1009, + 1448, + 1009, + 1468, + 992, + 1468 + ], + "score": 0.73, + "latex": "s" + }, + { + "category_id": 13, + "poly": [ + 551, + 1507, + 566, + 1507, + 566, + 1533, + 551, + 1533 + ], + "score": 0.68, + "latex": "\\ell" + }, + { + "category_id": 13, + "poly": [ + 1384, + 1230, + 1401, + 1230, + 1401, + 1255, + 1384, + 1255 + ], + "score": 0.65, + "latex": "q" + }, + { + "category_id": 13, + "poly": [ + 1238, + 327, + 1329, + 327, + 1329, + 359, + 1238, + 359 + ], + "score": 0.63, + "latex": "Y _ { 1 } ^ { n } , Y _ { 2 } ^ { n }" + }, + { + "category_id": 14, + "poly": [ + 370, + 852, + 713, + 852, + 713, + 889, + 370, + 889 + ], + "score": 0.4, + "latex": "l . \\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { T } \\rangle \\subseteq \\langle \\mathcal { G } _ { J + 1 , D } , \\mathcal { T } _ { T } \\rangle ." + }, + { + "category_id": 13, + "poly": [ + 1288, + 327, + 1328, + 327, + 1328, + 359, + 1288, + 359 + ], + "score": 0.4, + "latex": "Y _ { 2 } ^ { n }" + }, + { + "category_id": 14, + "poly": [ + 381, + 904, + 709, + 904, + 709, + 941, + 381, + 941 + ], + "score": 0.38, + "latex": ". \\langle \\mathcal { G } _ { J , D } , \\mathcal { T } _ { T } \\rangle \\subseteq \\langle \\mathcal { G } _ { J , D + 1 } , \\mathcal { T } _ { T } \\rangle ." + }, + { + "category_id": 13, + "poly": [ + 1239, + 327, + 1279, + 327, + 1279, + 359, + 1239, + 359 + ], + "score": 0.36, + "latex": "Y _ { 1 } ^ { n }" + }, + { + "category_id": 13, + "poly": [ + 1355, + 1472, + 1394, + 1472, + 1394, + 1501, + 1355, + 1501 + ], + "score": 0.33, + "latex": "\\mathrm { X u }" + }, + { + "category_id": 13, + "poly": [ + 496, + 1112, + 512, + 1112, + 512, + 1137, + 496, + 1137 + ], + "score": 0.26, + "latex": "0" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 72.0, + 815.0, + 72.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 2007.0, + 1403.0, + 2007.0, + 1403.0, + 2033.0, + 1378.0, + 2033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 468.0, + 1406.0, + 468.0, + 1406.0, + 507.0, + 294.0, + 507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 498.0, + 1406.0, + 498.0, + 1406.0, + 535.0, + 293.0, + 535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 531.0, + 1129.0, + 531.0, + 1129.0, + 571.0, + 293.0, + 571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1189.0, + 531.0, + 1406.0, + 531.0, + 1406.0, + 571.0, + 1189.0, + 571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 560.0, + 505.0, + 560.0, + 505.0, + 603.0, + 291.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 529.0, + 560.0, + 864.0, + 560.0, + 864.0, + 603.0, + 529.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 560.0, + 1406.0, + 560.0, + 1406.0, + 603.0, + 892.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 594.0, + 1405.0, + 594.0, + 1405.0, + 631.0, + 294.0, + 631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 628.0, + 429.0, + 628.0, + 429.0, + 668.0, + 293.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 628.0, + 538.0, + 628.0, + 538.0, + 668.0, + 490.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 598.0, + 628.0, + 1248.0, + 628.0, + 1248.0, + 668.0, + 598.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 666.0, + 473.0, + 666.0, + 473.0, + 702.0, + 423.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 666.0, + 1015.0, + 666.0, + 1015.0, + 702.0, + 755.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 225.0, + 372.0, + 225.0, + 372.0, + 267.0, + 295.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 225.0, + 766.0, + 225.0, + 766.0, + 267.0, + 534.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 947.0, + 225.0, + 1403.0, + 225.0, + 1403.0, + 267.0, + 947.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 256.0, + 344.0, + 256.0, + 344.0, + 304.0, + 290.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 256.0, + 1298.0, + 256.0, + 1298.0, + 304.0, + 434.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 256.0, + 1406.0, + 256.0, + 1406.0, + 304.0, + 1401.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 293.0, + 357.0, + 293.0, + 357.0, + 334.0, + 291.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 500.0, + 293.0, + 584.0, + 293.0, + 584.0, + 334.0, + 500.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 684.0, + 293.0, + 774.0, + 293.0, + 774.0, + 334.0, + 684.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 293.0, + 1406.0, + 293.0, + 1406.0, + 334.0, + 871.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 319.0, + 705.0, + 319.0, + 705.0, + 368.0, + 290.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 319.0, + 1237.0, + 319.0, + 1237.0, + 368.0, + 785.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1330.0, + 319.0, + 1411.0, + 319.0, + 1411.0, + 368.0, + 1330.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 352.0, + 476.0, + 352.0, + 476.0, + 395.0, + 291.0, + 395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 352.0, + 1405.0, + 352.0, + 1405.0, + 395.0, + 678.0, + 395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 389.0, + 297.0, + 389.0, + 297.0, + 428.0, + 293.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 389.0, + 515.0, + 389.0, + 515.0, + 428.0, + 394.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 611.0, + 389.0, + 819.0, + 389.0, + 819.0, + 428.0, + 611.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 389.0, + 1108.0, + 389.0, + 1108.0, + 428.0, + 990.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1336.0, + 389.0, + 1406.0, + 389.0, + 1406.0, + 428.0, + 1336.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 533.0, + 421.0, + 544.0, + 421.0, + 544.0, + 459.0, + 533.0, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1621.0, + 418.0, + 1621.0, + 418.0, + 1667.0, + 289.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 519.0, + 1621.0, + 575.0, + 1621.0, + 575.0, + 1667.0, + 519.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 1621.0, + 915.0, + 1621.0, + 915.0, + 1667.0, + 670.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 1621.0, + 1012.0, + 1621.0, + 1012.0, + 1667.0, + 944.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 1621.0, + 1409.0, + 1621.0, + 1409.0, + 1667.0, + 1322.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1655.0, + 297.0, + 1655.0, + 297.0, + 1705.0, + 291.0, + 1705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 487.0, + 1655.0, + 1410.0, + 1655.0, + 1410.0, + 1705.0, + 487.0, + 1705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1686.0, + 296.0, + 1686.0, + 296.0, + 1734.0, + 293.0, + 1734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 1686.0, + 613.0, + 1686.0, + 613.0, + 1734.0, + 397.0, + 1734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 662.0, + 1686.0, + 876.0, + 1686.0, + 876.0, + 1734.0, + 662.0, + 1734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 1686.0, + 1242.0, + 1686.0, + 1242.0, + 1734.0, + 894.0, + 1734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 1686.0, + 1407.0, + 1686.0, + 1407.0, + 1734.0, + 1279.0, + 1734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1724.0, + 670.0, + 1724.0, + 670.0, + 1761.0, + 295.0, + 1761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 1724.0, + 938.0, + 1724.0, + 938.0, + 1761.0, + 853.0, + 1761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1438.0, + 991.0, + 1438.0, + 991.0, + 1474.0, + 292.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1010.0, + 1438.0, + 1364.0, + 1438.0, + 1364.0, + 1474.0, + 1010.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 1438.0, + 1405.0, + 1438.0, + 1405.0, + 1474.0, + 1401.0, + 1474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1465.0, + 763.0, + 1465.0, + 763.0, + 1510.0, + 290.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 924.0, + 1465.0, + 975.0, + 1465.0, + 975.0, + 1510.0, + 924.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1254.0, + 1465.0, + 1354.0, + 1465.0, + 1354.0, + 1510.0, + 1254.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1395.0, + 1465.0, + 1407.0, + 1465.0, + 1407.0, + 1510.0, + 1395.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1497.0, + 550.0, + 1497.0, + 550.0, + 1545.0, + 290.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.0, + 1497.0, + 627.0, + 1497.0, + 627.0, + 1545.0, + 567.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 1497.0, + 1227.0, + 1497.0, + 1227.0, + 1545.0, + 714.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 1497.0, + 1406.0, + 1497.0, + 1406.0, + 1545.0, + 1281.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1532.0, + 297.0, + 1532.0, + 297.0, + 1578.0, + 294.0, + 1578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 1532.0, + 594.0, + 1532.0, + 594.0, + 1578.0, + 508.0, + 1578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 967.0, + 1022.0, + 967.0, + 1022.0, + 1007.0, + 293.0, + 1007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 967.0, + 1080.0, + 967.0, + 1080.0, + 1007.0, + 1045.0, + 1007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1205.0, + 967.0, + 1407.0, + 967.0, + 1407.0, + 1007.0, + 1205.0, + 1007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 999.0, + 505.0, + 999.0, + 505.0, + 1039.0, + 293.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 999.0, + 818.0, + 999.0, + 818.0, + 1039.0, + 659.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 999.0, + 1340.0, + 999.0, + 1340.0, + 1039.0, + 891.0, + 1039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 1030.0, + 413.0, + 1030.0, + 413.0, + 1066.0, + 400.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1825.0, + 429.0, + 1825.0, + 429.0, + 1863.0, + 294.0, + 1863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1825.0, + 1312.0, + 1825.0, + 1312.0, + 1863.0, + 449.0, + 1863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1370.0, + 1825.0, + 1406.0, + 1825.0, + 1406.0, + 1863.0, + 1370.0, + 1863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1854.0, + 616.0, + 1854.0, + 616.0, + 1894.0, + 291.0, + 1894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 664.0, + 1854.0, + 1145.0, + 1854.0, + 1145.0, + 1894.0, + 664.0, + 1894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1203.0, + 1854.0, + 1406.0, + 1854.0, + 1406.0, + 1894.0, + 1203.0, + 1894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1888.0, + 472.0, + 1888.0, + 472.0, + 1920.0, + 293.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 707.0, + 956.0, + 707.0, + 956.0, + 752.0, + 293.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 707.0, + 1404.0, + 707.0, + 1404.0, + 752.0, + 978.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 744.0, + 339.0, + 744.0, + 339.0, + 777.0, + 294.0, + 777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 744.0, + 971.0, + 744.0, + 971.0, + 777.0, + 368.0, + 777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1216.0, + 576.0, + 1216.0, + 576.0, + 1262.0, + 291.0, + 1262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 1216.0, + 841.0, + 1216.0, + 841.0, + 1262.0, + 710.0, + 1262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 1216.0, + 992.0, + 1216.0, + 992.0, + 1262.0, + 961.0, + 1262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 1216.0, + 1383.0, + 1216.0, + 1383.0, + 1262.0, + 1171.0, + 1262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1216.0, + 1407.0, + 1216.0, + 1407.0, + 1262.0, + 1402.0, + 1262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1250.0, + 420.0, + 1250.0, + 420.0, + 1292.0, + 291.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1076.0, + 807.0, + 1076.0, + 807.0, + 1113.0, + 294.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 835.0, + 1076.0, + 1169.0, + 1076.0, + 1169.0, + 1113.0, + 835.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1249.0, + 1076.0, + 1405.0, + 1076.0, + 1405.0, + 1113.0, + 1249.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1108.0, + 495.0, + 1108.0, + 495.0, + 1142.0, + 373.0, + 1142.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 1108.0, + 521.0, + 1108.0, + 521.0, + 1142.0, + 513.0, + 1142.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1375.0, + 1113.0, + 1404.0, + 1113.0, + 1404.0, + 1138.0, + 1375.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1994.0, + 352.0, + 1994.0, + 352.0, + 2045.0, + 291.0, + 2045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1994.0, + 531.0, + 1994.0, + 531.0, + 2045.0, + 526.0, + 2045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1366.0, + 352.0, + 1366.0, + 352.0, + 1413.0, + 291.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 1366.0, + 537.0, + 1366.0, + 537.0, + 1413.0, + 525.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1173.0, + 785.0, + 1173.0, + 785.0, + 1210.0, + 297.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 785.0, + 538.0, + 785.0, + 538.0, + 828.0, + 294.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 785.0, + 705.0, + 785.0, + 705.0, + 828.0, + 696.0, + 828.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 851.0, + 368.0, + 851.0, + 368.0, + 892.0, + 360.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 851.0, + 719.0, + 851.0, + 719.0, + 892.0, + 714.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 903.0, + 380.0, + 903.0, + 380.0, + 944.0, + 358.0, + 944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 903.0, + 716.0, + 903.0, + 716.0, + 944.0, + 710.0, + 944.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 16, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 295, + 1498, + 1405, + 1498, + 1405, + 1643, + 295, + 1643 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1079, + 1403, + 1079, + 1403, + 1242, + 298, + 1242 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 297, + 512, + 1405, + 512, + 1405, + 640, + 297, + 640 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 296, + 923, + 1405, + 923, + 1405, + 1070, + 296, + 1070 + ], + "score": 0.975 + }, + { + "category_id": 8, + "poly": [ + 467, + 1253, + 1228, + 1253, + 1228, + 1483, + 467, + 1483 + ], + "score": 0.969 + }, + { + "category_id": 8, + "poly": [ + 312, + 728, + 1383, + 728, + 1383, + 915, + 312, + 915 + ], + "score": 0.969 + }, + { + "category_id": 1, + "poly": [ + 296, + 308, + 1405, + 308, + 1405, + 437, + 296, + 437 + ], + "score": 0.965 + }, + { + "category_id": 8, + "poly": [ + 405, + 1873, + 1292, + 1873, + 1292, + 1964, + 405, + 1964 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 294, + 1971, + 1402, + 1971, + 1402, + 2036, + 294, + 2036 + ], + "score": 0.947 + }, + { + "category_id": 1, + "poly": [ + 294, + 1797, + 1404, + 1797, + 1404, + 1862, + 294, + 1862 + ], + "score": 0.945 + }, + { + "category_id": 8, + "poly": [ + 599, + 449, + 1099, + 449, + 1099, + 492, + 599, + 492 + ], + "score": 0.943 + }, + { + "category_id": 0, + "poly": [ + 297, + 1729, + 866, + 1729, + 866, + 1767, + 297, + 1767 + ], + "score": 0.932 + }, + { + "category_id": 8, + "poly": [ + 704, + 640, + 991, + 640, + 991, + 673, + 704, + 673 + ], + "score": 0.925 + }, + { + "category_id": 1, + "poly": [ + 293, + 227, + 1403, + 227, + 1403, + 297, + 293, + 297 + ], + "score": 0.925 + }, + { + "category_id": 2, + "poly": [ + 299, + 74, + 814, + 74, + 814, + 106, + 299, + 106 + ], + "score": 0.922 + }, + { + "category_id": 1, + "poly": [ + 298, + 678, + 1265, + 678, + 1265, + 717, + 298, + 717 + ], + "score": 0.919 + }, + { + "category_id": 2, + "poly": [ + 836, + 2087, + 865, + 2087, + 865, + 2113, + 836, + 2113 + ], + "score": 0.874 + }, + { + "category_id": 2, + "poly": [ + 1374, + 1656, + 1403, + 1656, + 1403, + 1685, + 1374, + 1685 + ], + "score": 0.775 + }, + { + "category_id": 13, + "poly": [ + 713, + 996, + 835, + 996, + 835, + 1038, + 713, + 1038 + ], + "score": 0.94, + "latex": "- Q _ { a } ^ { ( \\vec { r } ) } ( \\bar { X } )" + }, + { + "category_id": 13, + "poly": [ + 595, + 958, + 696, + 958, + 696, + 999, + 595, + 999 + ], + "score": 0.94, + "latex": "Q _ { a } ^ { ( \\vec { r } ) } ( \\bar { X } )" + }, + { + "category_id": 13, + "poly": [ + 1177, + 1493, + 1323, + 1493, + 1323, + 1536, + 1177, + 1536 + ], + "score": 0.94, + "latex": "P _ { D } Q _ { a } ^ { ( r ^ { \\prime } ) } ( \\bar { X } )" + }, + { + "category_id": 14, + "poly": [ + 467, + 1254, + 1228, + 1254, + 1228, + 1486, + 467, + 1486 + ], + "score": 0.93, + "latex": "\\begin{array} { r l } & { q _ { a } ( X ) = \\displaystyle \\sum _ { b = 1 } ^ { n } ( \\bar { x } _ { a } - \\bar { x } _ { b } ) ^ { \\otimes r _ { 1 } } \\otimes Q _ { b } ^ { ( \\bar { r } ) } ( \\bar { X } ) } \\\\ & { \\qquad = \\displaystyle \\sum _ { b = 1 } ^ { n } ( \\bar { x } _ { a } - \\bar { x } _ { b } ) ^ { \\otimes r _ { 1 } } \\otimes \\bar { x } _ { b } ^ { \\otimes r _ { 2 } } \\otimes \\displaystyle \\sum _ { i _ { 3 } , \\ldots , i _ { K } = 1 } ^ { n } \\bar { x } _ { i _ { 3 } } ^ { \\otimes r _ { 3 } } \\otimes \\ldots \\otimes \\bar { x } _ { i _ { K } } ^ { \\otimes r _ { K } } } \\\\ & { \\qquad = Q _ { a } ^ { ( \\bar { r } ) } ( \\bar { X } ) + \\mathrm { a d d i t i o n a l ~ t e r m s } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 504, + 1605, + 604, + 1605, + 604, + 1642, + 504, + 1642 + ], + "score": 0.93, + "latex": "Q ^ { ( \\vec { r } ) } ( \\bar { X } )" + }, + { + "category_id": 14, + "poly": [ + 405, + 1870, + 1291, + 1870, + 1291, + 1964, + 405, + 1964 + ], + "score": 0.93, + "latex": "\\tilde { V } _ { a } ( X , V ) = U ( l _ { f } , l _ { i } ) \\left( \\theta _ { 1 } \\sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \\otimes V _ { b } + \\theta _ { 2 } \\sum _ { b = 1 } ^ { n } F ( x _ { a } - x _ { b } ) \\otimes V _ { a } \\right) ," + }, + { + "category_id": 13, + "poly": [ + 882, + 1572, + 1024, + 1572, + 1024, + 1607, + 882, + 1607 + ], + "score": 0.93, + "latex": "\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { D } \\rangle" + }, + { + "category_id": 13, + "poly": [ + 1242, + 1000, + 1343, + 1000, + 1343, + 1038, + 1242, + 1038 + ], + "score": 0.93, + "latex": "Q ^ { ( \\vec { r } ) } ( \\bar { X } )" + }, + { + "category_id": 13, + "poly": [ + 352, + 924, + 551, + 924, + 551, + 961, + 352, + 961 + ], + "score": 0.93, + "latex": "\\langle \\mathcal { G } _ { 2 D - 1 , D - 1 } , \\mathcal { T } _ { D } \\rangle" + }, + { + "category_id": 13, + "poly": [ + 686, + 1608, + 829, + 1608, + 829, + 1643, + 686, + 1643 + ], + "score": 0.93, + "latex": "\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { D } \\rangle" + }, + { + "category_id": 13, + "poly": [ + 407, + 343, + 499, + 343, + 499, + 375, + 407, + 375 + ], + "score": 0.93, + "latex": "\\vec { r } \\in \\Sigma _ { T }" + }, + { + "category_id": 13, + "poly": [ + 1199, + 1564, + 1405, + 1564, + 1405, + 1607, + 1199, + 1607 + ], + "score": 0.93, + "latex": "P _ { D } Q _ { a } ^ { ( r ^ { \\prime } ) } ( \\bar { X } ) , a =" + }, + { + "category_id": 13, + "poly": [ + 531, + 1207, + 957, + 1207, + 957, + 1242, + 531, + 1242 + ], + "score": 0.93, + "latex": "q _ { a } \\in \\langle \\mathcal { G } _ { 2 D ^ { \\prime } + 1 , D ^ { \\prime } } , \\mathcal { T } _ { D } \\rangle \\subseteq \\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { D } \\rangle" + }, + { + "category_id": 14, + "poly": [ + 313, + 726, + 1382, + 726, + 1382, + 918, + 313, + 918 + ], + "score": 0.93, + "latex": "\\begin{array} { l } { \\displaystyle q _ { a } ( X ) = \\sum _ { b = 1 } ^ { n } ( \\bar { x } _ { a } - \\bar { x } _ { b } ) \\otimes Q _ { b } ^ { ( \\bar { r } ) } ( \\bar { X } ) = \\sum _ { b = 1 } ^ { n } ( \\bar { x } _ { a } - \\bar { x } _ { b } ) \\otimes \\bar { x } _ { b } ^ { \\otimes r _ { 2 } - 1 } \\otimes \\sum _ { i _ { 3 } , \\dots , i _ { K } = 1 } ^ { n } \\bar { x } _ { i _ { 3 } } ^ { \\otimes r _ { 3 } } \\otimes \\dots \\otimes \\bar { x } _ { i _ { K } } ^ { \\otimes r _ { K } } } \\\\ { \\displaystyle \\quad = \\left( \\bar { x } _ { a } \\otimes \\sum _ { b = 1 } ^ { n } Q _ { b } ^ { ( \\bar { r } ) } ( \\bar { X } ) \\right) - Q ^ { ( \\bar { r } ) } ( \\bar { X } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 773, + 343, + 913, + 343, + 913, + 378, + 773, + 378 + ], + "score": 0.93, + "latex": "\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { T } \\rangle" + }, + { + "category_id": 13, + "poly": [ + 457, + 2000, + 592, + 2000, + 592, + 2036, + 457, + 2036 + ], + "score": 0.93, + "latex": "l _ { 1 } \\stackrel { \\cdot } { = } [ 0 , 1 ] ^ { \\frac { \\cdot } { T } }" + }, + { + "category_id": 13, + "poly": [ + 837, + 545, + 889, + 545, + 889, + 580, + 837, + 580 + ], + "score": 0.92, + "latex": "Q ^ { ( \\vec { r } ) }" + }, + { + "category_id": 13, + "poly": [ + 979, + 925, + 1150, + 925, + 1150, + 961, + 979, + 961 + ], + "score": 0.92, + "latex": "\\langle \\mathcal { G } _ { 2 D - 1 , D } , \\mathcal { T } _ { D } \\rangle" + }, + { + "category_id": 13, + "poly": [ + 978, + 1534, + 1160, + 1534, + 1160, + 1565, + 978, + 1565 + ], + "score": 0.92, + "latex": "P _ { D } : \\mathcal { T } _ { D } \\mathcal { T } _ { D }" + }, + { + "category_id": 13, + "poly": [ + 655, + 341, + 708, + 341, + 708, + 377, + 655, + 377 + ], + "score": 0.92, + "latex": "Q ^ { ( \\vec { r } ) }" + }, + { + "category_id": 13, + "poly": [ + 704, + 1082, + 794, + 1082, + 794, + 1112, + 704, + 1112 + ], + "score": 0.92, + "latex": "\\vec { r } \\in \\Sigma _ { D }" + }, + { + "category_id": 13, + "poly": [ + 956, + 1173, + 1209, + 1173, + 1209, + 1210, + 956, + 1210 + ], + "score": 0.92, + "latex": "Q ^ { ( \\tilde { r } ) } \\in \\langle { \\mathcal { G } _ { 2 D ^ { \\prime } , D ^ { \\prime } } } , { \\mathcal { T } _ { D ^ { \\prime } } } \\rangle" + }, + { + "category_id": 13, + "poly": [ + 762, + 1567, + 822, + 1567, + 822, + 1606, + 762, + 1606 + ], + "score": 0.92, + "latex": "Q ^ { ( r ^ { \\prime } ) }" + }, + { + "category_id": 13, + "poly": [ + 473, + 406, + 554, + 406, + 554, + 436, + 473, + 436 + ], + "score": 0.92, + "latex": "\\vec { r } \\in \\Sigma _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 834, + 515, + 981, + 515, + 981, + 546, + 834, + 546 + ], + "score": 0.92, + "latex": "D > D ^ { \\prime } \\geq 0" + }, + { + "category_id": 13, + "poly": [ + 1056, + 1003, + 1146, + 1003, + 1146, + 1038, + 1056, + 1038 + ], + "score": 0.92, + "latex": "a \\in [ n ]" + }, + { + "category_id": 13, + "poly": [ + 1089, + 310, + 1225, + 310, + 1225, + 344, + 1089, + 344 + ], + "score": 0.92, + "latex": "D , T \\in \\mathbb { N } _ { + }" + }, + { + "category_id": 13, + "poly": [ + 603, + 548, + 692, + 548, + 692, + 579, + 603, + 579 + ], + "score": 0.92, + "latex": "\\vec { r } \\in \\Sigma _ { D }" + }, + { + "category_id": 13, + "poly": [ + 949, + 548, + 1092, + 548, + 1092, + 582, + 949, + 582 + ], + "score": 0.91, + "latex": "\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { D } \\rangle" + }, + { + "category_id": 13, + "poly": [ + 298, + 1533, + 399, + 1533, + 399, + 1565, + 298, + 1565 + ], + "score": 0.91, + "latex": "r ^ { \\prime } \\in \\Sigma _ { D }" + }, + { + "category_id": 13, + "poly": [ + 873, + 580, + 963, + 580, + 963, + 609, + 873, + 609 + ], + "score": 0.91, + "latex": "\\vec { r } \\in \\Sigma _ { D }" + }, + { + "category_id": 14, + "poly": [ + 598, + 446, + 1100, + 446, + 1100, + 490, + 598, + 490 + ], + "score": 0.91, + "latex": "Q ^ { ( \\vec { r } ) } ( \\bar { X } ) = 1 _ { n } = \\pi _ { V } \\circ \\mathrm { e x t } ( X ) \\in \\langle \\mathcal { G } _ { 0 , 0 } , \\mathcal { T } _ { 0 } \\rangle ." + }, + { + "category_id": 13, + "poly": [ + 374, + 1145, + 568, + 1145, + 568, + 1176, + 374, + 1176 + ], + "score": 0.91, + "latex": "\\tilde { \\boldsymbol { r } } = \\left( r _ { 2 } , \\ldots , r _ { K } \\right)" + }, + { + "category_id": 13, + "poly": [ + 298, + 1035, + 442, + 1035, + 442, + 1069, + 298, + 1069 + ], + "score": 0.91, + "latex": "\\langle \\mathcal { G } _ { 2 D , D } , \\mathcal { T } _ { D } \\rangle" + }, + { + "category_id": 13, + "poly": [ + 1218, + 1143, + 1404, + 1143, + 1404, + 1176, + 1218, + 1176 + ], + "score": 0.91, + "latex": "D ^ { \\prime } = D - r _ { 1 } ^ { \\prime } <" + }, + { + "category_id": 13, + "poly": [ + 877, + 377, + 962, + 377, + 962, + 403, + 877, + 403 + ], + "score": 0.91, + "latex": "T = D" + }, + { + "category_id": 13, + "poly": [ + 325, + 2003, + 372, + 2003, + 372, + 2036, + 325, + 2036 + ], + "score": 0.91, + "latex": "W _ { l _ { 1 } }" + }, + { + "category_id": 13, + "poly": [ + 344, + 260, + 606, + 260, + 606, + 296, + 344, + 296 + ], + "score": 0.91, + "latex": "\\theta _ { 0 } , R ^ { ( \\mathrm { 1 } ) } , \\ldots , R ^ { ( \\dot { D } ) } = 0" + }, + { + "category_id": 13, + "poly": [ + 850, + 678, + 1185, + 678, + 1185, + 718, + 850, + 718 + ], + "score": 0.9, + "latex": "Q ^ { ( \\tilde { r } ) } ( \\bar { X } ) \\in \\mathcal { G } _ { 2 ( D - 1 ) , D - 1 , D - 1 }" + }, + { + "category_id": 13, + "poly": [ + 418, + 1112, + 497, + 1112, + 497, + 1145, + 418, + 1145 + ], + "score": 0.9, + "latex": "r _ { 1 } ^ { \\prime } \\geq 1" + }, + { + "category_id": 13, + "poly": [ + 1034, + 967, + 1128, + 967, + 1128, + 997, + 1034, + 997 + ], + "score": 0.9, + "latex": "r _ { 1 } ~ = ~ 0" + }, + { + "category_id": 13, + "poly": [ + 1296, + 226, + 1402, + 226, + 1402, + 260, + 1296, + 260 + ], + "score": 0.9, + "latex": "R ^ { ( 0 ) } = 1" + }, + { + "category_id": 13, + "poly": [ + 298, + 610, + 368, + 610, + 368, + 642, + 298, + 642 + ], + "score": 0.9, + "latex": "\\Sigma _ { D - 1 }" + }, + { + "category_id": 13, + "poly": [ + 1305, + 310, + 1393, + 310, + 1393, + 342, + 1305, + 342 + ], + "score": 0.89, + "latex": "T \\leq D" + }, + { + "category_id": 13, + "poly": [ + 298, + 1609, + 392, + 1609, + 392, + 1641, + 298, + 1641 + ], + "score": 0.89, + "latex": "1 , \\ldots , n" + }, + { + "category_id": 13, + "poly": [ + 889, + 1534, + 917, + 1534, + 917, + 1566, + 889, + 1566 + ], + "score": 0.89, + "latex": "r _ { 1 } ^ { \\prime }" + }, + { + "category_id": 14, + "poly": [ + 704, + 637, + 992, + 637, + 992, + 674, + 704, + 674 + ], + "score": 0.89, + "latex": "\\tilde { r } = ( r _ { 2 } - 1 , r _ { 3 } , \\ldots , r _ { K } ) ." + }, + { + "category_id": 13, + "poly": [ + 740, + 515, + 774, + 515, + 774, + 543, + 740, + 543 + ], + "score": 0.88, + "latex": "D ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 298, + 407, + 375, + 407, + 375, + 434, + 298, + 434 + ], + "score": 0.87, + "latex": "D = 0" + }, + { + "category_id": 13, + "poly": [ + 1313, + 1113, + 1340, + 1113, + 1340, + 1144, + 1313, + 1144 + ], + "score": 0.86, + "latex": "r _ { 1 } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 1016, + 583, + 1100, + 583, + 1100, + 609, + 1016, + 609 + ], + "score": 0.85, + "latex": "r _ { 1 } = 0 ." + }, + { + "category_id": 13, + "poly": [ + 771, + 1978, + 810, + 1978, + 810, + 2000, + 771, + 2000 + ], + "score": 0.85, + "latex": "2 D" + }, + { + "category_id": 13, + "poly": [ + 685, + 1538, + 714, + 1538, + 714, + 1564, + 685, + 1564 + ], + "score": 0.84, + "latex": "r _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 297, + 968, + 330, + 968, + 330, + 996, + 297, + 996 + ], + "score": 0.84, + "latex": "\\textstyle { \\bar { x } } _ { a }" + }, + { + "category_id": 13, + "poly": [ + 693, + 583, + 720, + 583, + 720, + 609, + 693, + 609 + ], + "score": 0.84, + "latex": "r _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 608, + 1579, + 636, + 1579, + 636, + 1604, + 608, + 1604 + ], + "score": 0.84, + "latex": "r _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 457, + 1974, + 483, + 1974, + 483, + 2000, + 457, + 2000 + ], + "score": 0.84, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 657, + 683, + 683, + 683, + 683, + 709, + 657, + 709 + ], + "score": 0.83, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 1171, + 1114, + 1190, + 1114, + 1190, + 1140, + 1171, + 1140 + ], + "score": 0.83, + "latex": "\\vec { r }" + }, + { + "category_id": 13, + "poly": [ + 1320, + 377, + 1345, + 377, + 1345, + 402, + 1320, + 402 + ], + "score": 0.82, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 1099, + 264, + 1125, + 264, + 1125, + 291, + 1099, + 291 + ], + "score": 0.82, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 388, + 1009, + 418, + 1009, + 418, + 1036, + 388, + 1036 + ], + "score": 0.82, + "latex": "q _ { a }" + }, + { + "category_id": 13, + "poly": [ + 955, + 1083, + 981, + 1083, + 981, + 1109, + 955, + 1109 + ], + "score": 0.82, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 686, + 232, + 712, + 232, + 712, + 258, + 686, + 258 + ], + "score": 0.82, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 1323, + 518, + 1348, + 518, + 1348, + 543, + 1323, + 543 + ], + "score": 0.82, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 735, + 1144, + 754, + 1144, + 754, + 1170, + 735, + 1170 + ], + "score": 0.81, + "latex": "\\vec { r }" + }, + { + "category_id": 13, + "poly": [ + 1238, + 580, + 1255, + 580, + 1255, + 606, + 1238, + 606 + ], + "score": 0.81, + "latex": "\\tilde { r }" + }, + { + "category_id": 13, + "poly": [ + 766, + 1178, + 793, + 1178, + 793, + 1204, + 766, + 1204 + ], + "score": 0.8, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 298, + 1177, + 324, + 1177, + 324, + 1204, + 298, + 1204 + ], + "score": 0.79, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 941, + 972, + 960, + 972, + 960, + 993, + 941, + 993 + ], + "score": 0.72, + "latex": "a" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1729.0, + 867.0, + 1729.0, + 867.0, + 1771.0, + 293.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 815.0, + 72.0, + 815.0, + 109.0, + 296.0, + 109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1377.0, + 1660.0, + 1403.0, + 1660.0, + 1403.0, + 1687.0, + 1377.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1493.0, + 1176.0, + 1493.0, + 1176.0, + 1540.0, + 292.0, + 1540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1324.0, + 1493.0, + 1407.0, + 1493.0, + 1407.0, + 1540.0, + 1324.0, + 1540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1528.0, + 297.0, + 1528.0, + 297.0, + 1571.0, + 292.0, + 1571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 1528.0, + 684.0, + 1528.0, + 684.0, + 1571.0, + 400.0, + 1571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 1528.0, + 888.0, + 1528.0, + 888.0, + 1571.0, + 715.0, + 1571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 1528.0, + 977.0, + 1528.0, + 977.0, + 1571.0, + 918.0, + 1571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1161.0, + 1528.0, + 1405.0, + 1528.0, + 1405.0, + 1571.0, + 1161.0, + 1571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1565.0, + 607.0, + 1565.0, + 607.0, + 1614.0, + 292.0, + 1614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 637.0, + 1565.0, + 761.0, + 1565.0, + 761.0, + 1614.0, + 637.0, + 1614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 1565.0, + 881.0, + 1565.0, + 881.0, + 1614.0, + 823.0, + 1614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1025.0, + 1565.0, + 1198.0, + 1565.0, + 1198.0, + 1614.0, + 1025.0, + 1614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1605.0, + 297.0, + 1605.0, + 297.0, + 1643.0, + 294.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1605.0, + 503.0, + 1605.0, + 503.0, + 1643.0, + 393.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 605.0, + 1605.0, + 685.0, + 1605.0, + 685.0, + 1643.0, + 605.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 1605.0, + 1373.0, + 1605.0, + 1373.0, + 1643.0, + 830.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1082.0, + 703.0, + 1082.0, + 703.0, + 1115.0, + 294.0, + 1115.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 1082.0, + 954.0, + 1082.0, + 954.0, + 1115.0, + 795.0, + 1115.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 1082.0, + 1405.0, + 1082.0, + 1405.0, + 1115.0, + 982.0, + 1115.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1111.0, + 417.0, + 1111.0, + 417.0, + 1148.0, + 293.0, + 1148.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 1111.0, + 1170.0, + 1111.0, + 1170.0, + 1148.0, + 498.0, + 1148.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 1111.0, + 1312.0, + 1111.0, + 1312.0, + 1148.0, + 1191.0, + 1148.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1341.0, + 1111.0, + 1407.0, + 1111.0, + 1407.0, + 1148.0, + 1341.0, + 1148.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1142.0, + 373.0, + 1142.0, + 373.0, + 1178.0, + 294.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 569.0, + 1142.0, + 734.0, + 1142.0, + 734.0, + 1178.0, + 569.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 1142.0, + 1217.0, + 1142.0, + 1217.0, + 1178.0, + 755.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1173.0, + 297.0, + 1173.0, + 297.0, + 1213.0, + 294.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1173.0, + 765.0, + 1173.0, + 765.0, + 1213.0, + 325.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 794.0, + 1173.0, + 955.0, + 1173.0, + 955.0, + 1213.0, + 794.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 1173.0, + 1407.0, + 1173.0, + 1407.0, + 1213.0, + 1210.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1207.0, + 530.0, + 1207.0, + 530.0, + 1244.0, + 294.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 958.0, + 1207.0, + 1084.0, + 1207.0, + 1084.0, + 1244.0, + 958.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 513.0, + 739.0, + 513.0, + 739.0, + 550.0, + 295.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 775.0, + 513.0, + 833.0, + 513.0, + 833.0, + 550.0, + 775.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 513.0, + 1322.0, + 513.0, + 1322.0, + 550.0, + 982.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1349.0, + 513.0, + 1406.0, + 513.0, + 1406.0, + 550.0, + 1349.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 542.0, + 602.0, + 542.0, + 602.0, + 587.0, + 290.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 542.0, + 836.0, + 542.0, + 836.0, + 587.0, + 693.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 542.0, + 948.0, + 542.0, + 948.0, + 587.0, + 890.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 542.0, + 1406.0, + 542.0, + 1406.0, + 587.0, + 1093.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 577.0, + 692.0, + 577.0, + 692.0, + 613.0, + 294.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 577.0, + 872.0, + 577.0, + 872.0, + 613.0, + 721.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 577.0, + 1015.0, + 577.0, + 1015.0, + 613.0, + 964.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1101.0, + 577.0, + 1237.0, + 577.0, + 1237.0, + 613.0, + 1101.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1256.0, + 577.0, + 1405.0, + 577.0, + 1405.0, + 613.0, + 1256.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 607.0, + 495.0, + 607.0, + 495.0, + 645.0, + 369.0, + 645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 919.0, + 351.0, + 919.0, + 351.0, + 963.0, + 291.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 919.0, + 978.0, + 919.0, + 978.0, + 963.0, + 552.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1151.0, + 919.0, + 1407.0, + 919.0, + 1407.0, + 963.0, + 1151.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 957.0, + 296.0, + 957.0, + 296.0, + 1004.0, + 290.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 957.0, + 594.0, + 957.0, + 594.0, + 1004.0, + 331.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 957.0, + 940.0, + 957.0, + 940.0, + 1004.0, + 697.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 957.0, + 1033.0, + 957.0, + 1033.0, + 1004.0, + 961.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 957.0, + 1406.0, + 957.0, + 1406.0, + 1004.0, + 1129.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 995.0, + 387.0, + 995.0, + 387.0, + 1043.0, + 290.0, + 1043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 419.0, + 995.0, + 712.0, + 995.0, + 712.0, + 1043.0, + 419.0, + 1043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 995.0, + 1055.0, + 995.0, + 1055.0, + 1043.0, + 836.0, + 1043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 995.0, + 1241.0, + 995.0, + 1241.0, + 1043.0, + 1147.0, + 1043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1344.0, + 995.0, + 1407.0, + 995.0, + 1407.0, + 1043.0, + 1344.0, + 1043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 1031.0, + 455.0, + 1031.0, + 455.0, + 1072.0, + 443.0, + 1072.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 307.0, + 1088.0, + 307.0, + 1088.0, + 346.0, + 294.0, + 346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1226.0, + 307.0, + 1304.0, + 307.0, + 1304.0, + 346.0, + 1226.0, + 346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 307.0, + 1405.0, + 307.0, + 1405.0, + 346.0, + 1394.0, + 346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 338.0, + 406.0, + 338.0, + 406.0, + 383.0, + 291.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 500.0, + 338.0, + 654.0, + 338.0, + 654.0, + 383.0, + 500.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 709.0, + 338.0, + 772.0, + 338.0, + 772.0, + 383.0, + 709.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 338.0, + 1406.0, + 338.0, + 1406.0, + 383.0, + 914.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 374.0, + 876.0, + 374.0, + 876.0, + 409.0, + 293.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 374.0, + 1319.0, + 374.0, + 1319.0, + 409.0, + 963.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1346.0, + 374.0, + 1405.0, + 374.0, + 1405.0, + 409.0, + 1346.0, + 409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 403.0, + 472.0, + 403.0, + 472.0, + 439.0, + 376.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 403.0, + 858.0, + 403.0, + 858.0, + 439.0, + 555.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1971.0, + 456.0, + 1971.0, + 456.0, + 2007.0, + 296.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 484.0, + 1971.0, + 770.0, + 1971.0, + 770.0, + 2007.0, + 484.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 811.0, + 1971.0, + 1404.0, + 1971.0, + 1404.0, + 2007.0, + 811.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 2001.0, + 324.0, + 2001.0, + 324.0, + 2037.0, + 296.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 2001.0, + 456.0, + 2001.0, + 456.0, + 2037.0, + 373.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 593.0, + 2001.0, + 1404.0, + 2001.0, + 1404.0, + 2037.0, + 593.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1797.0, + 1405.0, + 1797.0, + 1405.0, + 1833.0, + 294.0, + 1833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1828.0, + 472.0, + 1828.0, + 472.0, + 1863.0, + 292.0, + 1863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 225.0, + 685.0, + 225.0, + 685.0, + 266.0, + 293.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 713.0, + 225.0, + 1295.0, + 225.0, + 1295.0, + 266.0, + 713.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 254.0, + 343.0, + 254.0, + 343.0, + 303.0, + 291.0, + 303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 607.0, + 254.0, + 1098.0, + 254.0, + 1098.0, + 303.0, + 607.0, + 303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1126.0, + 254.0, + 1409.0, + 254.0, + 1409.0, + 303.0, + 1126.0, + 303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 671.0, + 656.0, + 671.0, + 656.0, + 726.0, + 289.0, + 726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 684.0, + 671.0, + 849.0, + 671.0, + 849.0, + 726.0, + 684.0, + 726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1186.0, + 671.0, + 1272.0, + 671.0, + 1272.0, + 726.0, + 1186.0, + 726.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 17, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1316, + 1405, + 1316, + 1405, + 1472, + 297, + 1472 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 901, + 1406, + 901, + 1406, + 1036, + 298, + 1036 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 297, + 593, + 1405, + 593, + 1405, + 718, + 297, + 718 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 291, + 340, + 1402, + 340, + 1402, + 409, + 291, + 409 + ], + "score": 0.954 + }, + { + "category_id": 1, + "poly": [ + 297, + 1231, + 1401, + 1231, + 1401, + 1303, + 297, + 1303 + ], + "score": 0.952 + }, + { + "category_id": 1, + "poly": [ + 298, + 228, + 1402, + 228, + 1402, + 324, + 298, + 324 + ], + "score": 0.951 + }, + { + "category_id": 8, + "poly": [ + 421, + 1126, + 1277, + 1126, + 1277, + 1216, + 421, + 1216 + ], + "score": 0.944 + }, + { + "category_id": 8, + "poly": [ + 483, + 782, + 1213, + 782, + 1213, + 869, + 483, + 869 + ], + "score": 0.943 + }, + { + "category_id": 8, + "poly": [ + 653, + 423, + 1046, + 423, + 1046, + 464, + 653, + 464 + ], + "score": 0.94 + }, + { + "category_id": 8, + "poly": [ + 785, + 526, + 914, + 526, + 914, + 565, + 785, + 565 + ], + "score": 0.933 + }, + { + "category_id": 1, + "poly": [ + 296, + 476, + 1282, + 476, + 1282, + 515, + 296, + 515 + ], + "score": 0.928 + }, + { + "category_id": 1, + "poly": [ + 291, + 732, + 1304, + 732, + 1304, + 767, + 291, + 767 + ], + "score": 0.925 + }, + { + "category_id": 0, + "poly": [ + 298, + 1514, + 953, + 1514, + 953, + 1551, + 298, + 1551 + ], + "score": 0.916 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 813, + 75, + 813, + 105, + 299, + 105 + ], + "score": 0.912 + }, + { + "category_id": 1, + "poly": [ + 362, + 1672, + 1404, + 1672, + 1404, + 2034, + 362, + 2034 + ], + "score": 0.907 + }, + { + "category_id": 2, + "poly": [ + 837, + 2088, + 865, + 2088, + 865, + 2112, + 837, + 2112 + ], + "score": 0.854 + }, + { + "category_id": 1, + "poly": [ + 293, + 1050, + 1403, + 1050, + 1403, + 1117, + 293, + 1117 + ], + "score": 0.843 + }, + { + "category_id": 1, + "poly": [ + 299, + 1584, + 1400, + 1584, + 1400, + 1647, + 299, + 1647 + ], + "score": 0.818 + }, + { + "category_id": 13, + "poly": [ + 501, + 372, + 671, + 372, + 671, + 409, + 501, + 409 + ], + "score": 0.94, + "latex": "R ^ { ( 0 ) } ( \\| x \\| ) = 0" + }, + { + "category_id": 13, + "poly": [ + 798, + 477, + 967, + 477, + 967, + 515, + 798, + 515 + ], + "score": 0.93, + "latex": "R ^ { ( 0 ) } ( \\| x \\| ) = 1" + }, + { + "category_id": 14, + "poly": [ + 419, + 1123, + 1277, + 1123, + 1277, + 1216, + 419, + 1216 + ], + "score": 0.93, + "latex": "\\sum _ { b = 1 } ^ { n } ( x _ { b } - x _ { a } ) \\otimes V _ { b } = \\sum _ { b = 1 } ^ { n } ( { \\bar { x } } _ { b } - { \\bar { x } } _ { a } ) \\otimes V _ { b } = { \\tilde { V } } _ { a } ^ { { \\mathrm { t e n s o r } } , 2 } ( V , { \\bar { X } } ) - { \\bar { x } } _ { a } \\otimes \\sum _ { b = 1 } ^ { n } V _ { b }" + }, + { + "category_id": 13, + "poly": [ + 722, + 372, + 920, + 372, + 920, + 410, + 722, + 410 + ], + "score": 0.93, + "latex": "R ^ { ( 1 ) } ( \\| x \\| ) = \\| x \\|" + }, + { + "category_id": 13, + "poly": [ + 986, + 1230, + 1275, + 1230, + 1275, + 1269, + 986, + 1269 + ], + "score": 0.93, + "latex": "\\textstyle \\theta _ { 1 } = { \\frac { 1 } { n } } , F = F _ { 1 } , \\theta _ { 2 } = 0" + }, + { + "category_id": 13, + "poly": [ + 1067, + 1317, + 1401, + 1317, + 1401, + 1351, + 1067, + 1351 + ], + "score": 0.93, + "latex": "F _ { I d } ( x ) = ( 0 , x ) \\in W _ { 0 } \\oplus W _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 501, + 1002, + 800, + 1002, + 800, + 1036, + 501, + 1036 + ], + "score": 0.93, + "latex": "\\bar { \\theta } _ { 1 } = 0 , \\theta _ { 2 } = 1 / n , F = F _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1019, + 476, + 1187, + 476, + 1187, + 515, + 1019, + 515 + ], + "score": 0.92, + "latex": "R ^ { ( 1 ) } ( \\| x \\| ) = 0" + }, + { + "category_id": 13, + "poly": [ + 1182, + 337, + 1400, + 337, + 1400, + 375, + 1182, + 375 + ], + "score": 0.92, + "latex": "F _ { I d } = ( F ^ { ( 0 ) } , F ^ { ( 1 ) } )" + }, + { + "category_id": 13, + "poly": [ + 1084, + 1054, + 1394, + 1054, + 1394, + 1087, + 1084, + 1087 + ], + "score": 0.92, + "latex": "\\theta _ { 1 } = - 1 , F = F _ { I d } , \\theta _ { 2 } = 0" + }, + { + "category_id": 13, + "poly": [ + 346, + 1348, + 670, + 1348, + 670, + 1382, + 346, + 1382 + ], + "score": 0.91, + "latex": "F _ { 1 } ( x ) = ( 1 , 0 ) \\in W _ { 0 } \\oplus W _ { 1 }" + }, + { + "category_id": 14, + "poly": [ + 484, + 779, + 1214, + 779, + 1214, + 871, + 484, + 871 + ], + "score": 0.91, + "latex": "\\tilde { V } _ { a } ^ { \\mathrm { t e n s o r } , 1 } ( \\bar { X } , V ) = \\bar { x } _ { a } \\otimes V _ { a } \\mathrm { a n d } \\tilde { V } _ { a } ^ { \\mathrm { t e n s o r } , 2 } ( \\bar { X } , V ) = \\sum _ { b = 1 } ^ { n } \\bar { x } _ { b } \\otimes V _ { b } ." + }, + { + "category_id": 14, + "poly": [ + 651, + 421, + 1047, + 421, + 1047, + 463, + 651, + 463 + ], + "score": 0.91, + "latex": "F _ { I d } ( x ) = \\| x \\| Y ^ { 1 } ( \\hat { x } ) = \\| x \\| \\hat { x } = x ." + }, + { + "category_id": 13, + "poly": [ + 503, + 1836, + 561, + 1836, + 561, + 1868, + 503, + 1868 + ], + "score": 0.91, + "latex": "\\theta _ { 0 } V _ { a }" + }, + { + "category_id": 13, + "poly": [ + 748, + 939, + 1063, + 939, + 1063, + 975, + 748, + 975 + ], + "score": 0.9, + "latex": "\\tilde { V } _ { a } ^ { \\mathrm { t e n s o r , 1 } } ( \\bar { X } , V ) = \\bar { x } _ { a } \\otimes V _ { a }" + }, + { + "category_id": 13, + "poly": [ + 769, + 906, + 1108, + 906, + 1108, + 939, + 769, + 939 + ], + "score": 0.9, + "latex": "\\theta _ { 1 } = 0 , \\theta _ { 2 } = 1 / n , F = F _ { I d }" + }, + { + "category_id": 13, + "poly": [ + 714, + 1940, + 764, + 1940, + 764, + 1970, + 714, + 1970 + ], + "score": 0.9, + "latex": "R ^ { ( \\ell ) }" + }, + { + "category_id": 14, + "poly": [ + 783, + 526, + 916, + 526, + 916, + 564, + 783, + 564 + ], + "score": 0.89, + "latex": "F _ { 1 } ( x ) = 1 ." + }, + { + "category_id": 13, + "poly": [ + 299, + 1267, + 476, + 1267, + 476, + 1305, + 299, + 1305 + ], + "score": 0.89, + "latex": "\\tilde { V } _ { a } ^ { \\mathrm { t e n s o r } , 2 } ( V , \\bar { X } )" + }, + { + "category_id": 13, + "poly": [ + 570, + 902, + 672, + 902, + 672, + 940, + 570, + 940 + ], + "score": 0.89, + "latex": "\\tilde { V } _ { a } ^ { \\mathrm { t e n s o r , 1 } }" + }, + { + "category_id": 13, + "poly": [ + 557, + 482, + 589, + 482, + 589, + 511, + 557, + 511 + ], + "score": 0.88, + "latex": "F _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 595, + 1049, + 697, + 1049, + 697, + 1088, + 595, + 1088 + ], + "score": 0.87, + "latex": "\\tilde { V } _ { a } ^ { \\mathrm { t e n s o r } , 2 }" + }, + { + "category_id": 13, + "poly": [ + 448, + 1441, + 475, + 1441, + 475, + 1467, + 448, + 1467 + ], + "score": 0.84, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 751, + 628, + 778, + 628, + 778, + 654, + 751, + 654 + ], + "score": 0.84, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 714, + 658, + 741, + 658, + 741, + 684, + 714, + 684 + ], + "score": 0.83, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 796, + 1410, + 837, + 1410, + 837, + 1438, + 796, + 1438 + ], + "score": 0.81, + "latex": "2 D" + }, + { + "category_id": 13, + "poly": [ + 298, + 688, + 324, + 688, + 324, + 714, + 298, + 714 + ], + "score": 0.8, + "latex": "D" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1514.0, + 955.0, + 1514.0, + 955.0, + 1555.0, + 293.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1314.0, + 1066.0, + 1314.0, + 1066.0, + 1353.0, + 294.0, + 1353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1347.0, + 345.0, + 1347.0, + 345.0, + 1385.0, + 294.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 671.0, + 1347.0, + 1405.0, + 1347.0, + 1405.0, + 1385.0, + 671.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1378.0, + 1402.0, + 1378.0, + 1402.0, + 1411.0, + 295.0, + 1411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1407.0, + 795.0, + 1407.0, + 795.0, + 1443.0, + 294.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 1407.0, + 1405.0, + 1407.0, + 1405.0, + 1443.0, + 838.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1439.0, + 447.0, + 1439.0, + 447.0, + 1473.0, + 294.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1439.0, + 1372.0, + 1439.0, + 1372.0, + 1473.0, + 476.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 899.0, + 569.0, + 899.0, + 569.0, + 943.0, + 292.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 673.0, + 899.0, + 768.0, + 899.0, + 768.0, + 943.0, + 673.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 899.0, + 1407.0, + 899.0, + 1407.0, + 943.0, + 1109.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 932.0, + 747.0, + 932.0, + 747.0, + 980.0, + 291.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 932.0, + 1408.0, + 932.0, + 1408.0, + 980.0, + 1064.0, + 980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 973.0, + 1404.0, + 973.0, + 1404.0, + 1008.0, + 295.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1003.0, + 500.0, + 1003.0, + 500.0, + 1037.0, + 295.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 801.0, + 1003.0, + 812.0, + 1003.0, + 812.0, + 1037.0, + 801.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 596.0, + 1403.0, + 596.0, + 1403.0, + 629.0, + 296.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 625.0, + 750.0, + 625.0, + 750.0, + 660.0, + 294.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 625.0, + 1405.0, + 625.0, + 1405.0, + 660.0, + 779.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 652.0, + 713.0, + 652.0, + 713.0, + 694.0, + 292.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 652.0, + 1405.0, + 652.0, + 1405.0, + 694.0, + 742.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 686.0, + 701.0, + 686.0, + 701.0, + 722.0, + 325.0, + 722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 334.0, + 1181.0, + 334.0, + 1181.0, + 378.0, + 295.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 334.0, + 1404.0, + 334.0, + 1404.0, + 378.0, + 1401.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 371.0, + 500.0, + 371.0, + 500.0, + 411.0, + 294.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 371.0, + 721.0, + 371.0, + 721.0, + 411.0, + 672.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 371.0, + 1031.0, + 371.0, + 1031.0, + 411.0, + 921.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1229.0, + 985.0, + 1229.0, + 985.0, + 1271.0, + 293.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1276.0, + 1229.0, + 1406.0, + 1229.0, + 1406.0, + 1271.0, + 1276.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 1254.0, + 298.0, + 1254.0, + 298.0, + 1314.0, + 286.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 1254.0, + 910.0, + 1254.0, + 910.0, + 1314.0, + 477.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 229.0, + 1406.0, + 229.0, + 1406.0, + 264.0, + 294.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 260.0, + 1406.0, + 260.0, + 1406.0, + 296.0, + 293.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 288.0, + 535.0, + 288.0, + 535.0, + 330.0, + 292.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 470.0, + 556.0, + 470.0, + 556.0, + 521.0, + 291.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 470.0, + 797.0, + 470.0, + 797.0, + 521.0, + 590.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 968.0, + 470.0, + 1018.0, + 470.0, + 1018.0, + 521.0, + 968.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 470.0, + 1282.0, + 470.0, + 1282.0, + 521.0, + 1188.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 729.0, + 1309.0, + 729.0, + 1309.0, + 772.0, + 293.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1671.0, + 1402.0, + 1671.0, + 1402.0, + 1709.0, + 361.0, + 1709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1705.0, + 1404.0, + 1705.0, + 1404.0, + 1737.0, + 393.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1734.0, + 1403.0, + 1734.0, + 1403.0, + 1770.0, + 394.0, + 1770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1766.0, + 1405.0, + 1766.0, + 1405.0, + 1800.0, + 394.0, + 1800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1796.0, + 694.0, + 1796.0, + 694.0, + 1830.0, + 395.0, + 1830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1834.0, + 502.0, + 1834.0, + 502.0, + 1873.0, + 362.0, + 1873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1834.0, + 1404.0, + 1834.0, + 1404.0, + 1873.0, + 562.0, + 1873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1866.0, + 1406.0, + 1866.0, + 1406.0, + 1904.0, + 393.0, + 1904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1898.0, + 1254.0, + 1898.0, + 1254.0, + 1935.0, + 393.0, + 1935.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 1938.0, + 713.0, + 1938.0, + 713.0, + 1978.0, + 356.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 1938.0, + 1406.0, + 1938.0, + 1406.0, + 1978.0, + 765.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1973.0, + 1405.0, + 1973.0, + 1405.0, + 2007.0, + 394.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 2002.0, + 1406.0, + 2002.0, + 1406.0, + 2039.0, + 394.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1044.0, + 594.0, + 1044.0, + 594.0, + 1093.0, + 290.0, + 1093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 1044.0, + 1083.0, + 1044.0, + 1083.0, + 1093.0, + 698.0, + 1093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1395.0, + 1044.0, + 1408.0, + 1044.0, + 1408.0, + 1093.0, + 1395.0, + 1093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1081.0, + 406.0, + 1081.0, + 406.0, + 1119.0, + 292.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1584.0, + 1402.0, + 1584.0, + 1402.0, + 1619.0, + 294.0, + 1619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1614.0, + 940.0, + 1614.0, + 940.0, + 1651.0, + 293.0, + 1651.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 18, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 395, + 228, + 1403, + 228, + 1403, + 416, + 395, + 416 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 1514, + 1404, + 1514, + 1404, + 1645, + 297, + 1645 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 1934, + 1403, + 1934, + 1403, + 2035, + 298, + 2035 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 297, + 1393, + 1402, + 1393, + 1402, + 1492, + 297, + 1492 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 297, + 1675, + 1403, + 1675, + 1403, + 1775, + 297, + 1775 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 298, + 892, + 1403, + 892, + 1403, + 986, + 298, + 986 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 296, + 525, + 1402, + 525, + 1402, + 600, + 296, + 600 + ], + "score": 0.958 + }, + { + "category_id": 8, + "poly": [ + 665, + 616, + 1034, + 616, + 1034, + 706, + 665, + 706 + ], + "score": 0.958 + }, + { + "category_id": 1, + "poly": [ + 297, + 720, + 1400, + 720, + 1400, + 783, + 297, + 783 + ], + "score": 0.95 + }, + { + "category_id": 8, + "poly": [ + 501, + 986, + 1195, + 986, + 1195, + 1025, + 501, + 1025 + ], + "score": 0.943 + }, + { + "category_id": 8, + "poly": [ + 597, + 1343, + 1100, + 1343, + 1100, + 1385, + 597, + 1385 + ], + "score": 0.942 + }, + { + "category_id": 8, + "poly": [ + 620, + 1787, + 1077, + 1787, + 1077, + 1826, + 620, + 1826 + ], + "score": 0.938 + }, + { + "category_id": 1, + "poly": [ + 299, + 1029, + 1408, + 1029, + 1408, + 1119, + 299, + 1119 + ], + "score": 0.935 + }, + { + "category_id": 1, + "poly": [ + 296, + 1160, + 837, + 1160, + 837, + 1194, + 296, + 1194 + ], + "score": 0.935 + }, + { + "category_id": 8, + "poly": [ + 651, + 836, + 1044, + 836, + 1044, + 880, + 651, + 880 + ], + "score": 0.934 + }, + { + "category_id": 1, + "poly": [ + 298, + 798, + 639, + 798, + 639, + 829, + 298, + 829 + ], + "score": 0.929 + }, + { + "category_id": 1, + "poly": [ + 297, + 1833, + 1028, + 1833, + 1028, + 1870, + 297, + 1870 + ], + "score": 0.928 + }, + { + "category_id": 1, + "poly": [ + 298, + 1292, + 717, + 1292, + 717, + 1328, + 298, + 1328 + ], + "score": 0.925 + }, + { + "category_id": 8, + "poly": [ + 658, + 1884, + 1041, + 1884, + 1041, + 1921, + 658, + 1921 + ], + "score": 0.925 + }, + { + "category_id": 0, + "poly": [ + 299, + 457, + 703, + 457, + 703, + 494, + 299, + 494 + ], + "score": 0.916 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 813, + 75, + 813, + 105, + 299, + 105 + ], + "score": 0.904 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1249, + 1400, + 1249, + 1400, + 1279, + 1351, + 1279 + ], + "score": 0.897 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1347, + 1400, + 1347, + 1400, + 1378, + 1352, + 1378 + ], + "score": 0.895 + }, + { + "category_id": 9, + "poly": [ + 1352, + 990, + 1400, + 990, + 1400, + 1020, + 1352, + 1020 + ], + "score": 0.886 + }, + { + "category_id": 2, + "poly": [ + 834, + 2087, + 866, + 2087, + 866, + 2113, + 834, + 2113 + ], + "score": 0.876 + }, + { + "category_id": 8, + "poly": [ + 549, + 1120, + 1147, + 1120, + 1147, + 1154, + 549, + 1154 + ], + "score": 0.86 + }, + { + "category_id": 8, + "poly": [ + 314, + 1206, + 1379, + 1206, + 1379, + 1283, + 314, + 1283 + ], + "score": 0.855 + }, + { + "category_id": 14, + "poly": [ + 665, + 611, + 1033, + 611, + 1033, + 709, + 665, + 709 + ], + "score": 0.94, + "latex": "\\Lambda ( w _ { 0 } , . . . , w _ { D } ) = \\sum _ { T = 0 } ^ { D } \\Lambda _ { T } ( w _ { T } ) ," + }, + { + "category_id": 13, + "poly": [ + 423, + 1835, + 576, + 1835, + 576, + 1871, + 423, + 1871 + ], + "score": 0.94, + "latex": "\\rho ( R ) = R ^ { \\otimes D }" + }, + { + "category_id": 13, + "poly": [ + 359, + 559, + 576, + 559, + 576, + 602, + 359, + 602 + ], + "score": 0.93, + "latex": "W _ { \\mathrm { f e a t } } ^ { \\mathcal { T } } = \\bigoplus _ { T = 0 } ^ { D } \\mathcal { T } _ { T }" + }, + { + "category_id": 13, + "poly": [ + 784, + 1610, + 910, + 1610, + 910, + 1642, + 784, + 1642 + ], + "score": 0.93, + "latex": "{ \\mathcal { T } } _ { 2 } = { \\overline { { W } } } ^ { \\otimes 2 }" + }, + { + "category_id": 13, + "poly": [ + 663, + 1969, + 766, + 1969, + 766, + 2006, + 663, + 2006 + ], + "score": 0.93, + "latex": "\\tilde { \\Lambda } \\circ \\tilde { \\rho } ( R )" + }, + { + "category_id": 13, + "poly": [ + 819, + 1972, + 923, + 1972, + 923, + 2006, + 819, + 2006 + ], + "score": 0.92, + "latex": "\\Lambda \\circ \\rho ( R )" + }, + { + "category_id": 14, + "poly": [ + 314, + 1203, + 1382, + 1203, + 1382, + 1286, + 314, + 1286 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\tilde { \\Lambda } _ { k } ( w _ { 1 } , \\dots , w _ { D } ) = \\operatorname* { d e t } ( w _ { 1 } , w _ { 2 } , w _ { 3 } ) \\times \\dots \\times \\operatorname* { d e t } ( w _ { 3 k - 2 } , w _ { 3 k - 1 } , w _ { 3 k } ) \\times \\langle w _ { 3 k + 1 } , w _ { 3 k + 2 } \\rangle \\times \\dots } \\\\ { \\times \\langle w _ { D - 1 } , w _ { D } \\rangle , \\qquad ( 2 \\tilde { \\lambda } _ { 1 } - \\tilde { \\lambda } _ { 1 } ) \\times \\dots , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 385, + 1294, + 600, + 1294, + 600, + 1328, + 385, + 1328 + ], + "score": 0.92, + "latex": "( k , \\sigma ) \\in K _ { D } \\times S _ { D }" + }, + { + "category_id": 13, + "poly": [ + 928, + 529, + 994, + 529, + 994, + 563, + 928, + 563 + ], + "score": 0.92, + "latex": "\\mathcal { F } _ { \\mathrm { p o o l } }" + }, + { + "category_id": 13, + "poly": [ + 730, + 925, + 888, + 925, + 888, + 957, + 730, + 957 + ], + "score": 0.92, + "latex": "\\tilde { \\Lambda } : W ^ { D } \\to \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 835, + 1546, + 891, + 1546, + 891, + 1584, + 835, + 1584 + ], + "score": 0.92, + "latex": "\\tilde { \\Lambda } _ { k , \\sigma }" + }, + { + "category_id": 13, + "poly": [ + 1199, + 1614, + 1295, + 1614, + 1295, + 1645, + 1199, + 1645 + ], + "score": 0.92, + "latex": "| \\lambda _ { 2 } | = 2" + }, + { + "category_id": 13, + "poly": [ + 1346, + 1426, + 1400, + 1426, + 1400, + 1463, + 1346, + 1463 + ], + "score": 0.91, + "latex": "\\tilde { \\Lambda } _ { k , \\sigma }" + }, + { + "category_id": 13, + "poly": [ + 535, + 1428, + 915, + 1428, + 915, + 1462, + 535, + 1462 + ], + "score": 0.91, + "latex": "\\lambda _ { D } = \\{ \\Lambda _ { k , \\sigma } | \\left( k , \\sigma \\right) \\in K _ { D } \\times S _ { D } \\}" + }, + { + "category_id": 13, + "poly": [ + 335, + 752, + 478, + 752, + 478, + 782, + 335, + 782 + ], + "score": 0.91, + "latex": "\\Lambda : \\mathcal { T } _ { D } \\to \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 792, + 1707, + 963, + 1707, + 963, + 1740, + 792, + 1740 + ], + "score": 0.91, + "latex": "\\tilde { \\Lambda } : \\dot { W } ^ { D } \\to \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 397, + 322, + 463, + 322, + 463, + 356, + 397, + 356 + ], + "score": 0.91, + "latex": "\\mathcal { F } _ { C , D }" + }, + { + "category_id": 13, + "poly": [ + 342, + 1162, + 437, + 1162, + 437, + 1192, + 342, + 1192 + ], + "score": 0.91, + "latex": "k \\in K _ { D }" + }, + { + "category_id": 13, + "poly": [ + 529, + 1058, + 597, + 1058, + 597, + 1088, + 529, + 1088 + ], + "score": 0.91, + "latex": "W ^ { \\otimes D }" + }, + { + "category_id": 14, + "poly": [ + 599, + 1339, + 1101, + 1339, + 1101, + 1384, + 599, + 1384 + ], + "score": 0.91, + "latex": "\\tilde { \\Lambda } _ { k , \\sigma } ( w _ { 1 } , \\ldots , w _ { D } ) = \\tilde { \\Lambda } _ { k } ( w _ { \\sigma ( 1 ) } , \\ldots , w _ { \\sigma ( D ) } )" + }, + { + "category_id": 14, + "poly": [ + 618, + 1786, + 1079, + 1786, + 1079, + 1824, + 618, + 1824 + ], + "score": 0.91, + "latex": "\\tilde { \\rho } ( R ) ( w _ { 1 } , \\ldots , w _ { D } ) = ( R w _ { 1 } , \\ldots , R w _ { D } ) ." + }, + { + "category_id": 13, + "poly": [ + 298, + 925, + 431, + 925, + 431, + 956, + 298, + 956 + ], + "score": 0.91, + "latex": "W ^ { \\otimes D } \\to \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 726, + 1835, + 794, + 1835, + 794, + 1866, + 726, + 1866 + ], + "score": 0.9, + "latex": "W ^ { \\otimes D }" + }, + { + "category_id": 13, + "poly": [ + 1158, + 1581, + 1393, + 1581, + 1393, + 1614, + 1158, + 1614 + ], + "score": 0.9, + "latex": "\\langle w _ { 1 } , w _ { 2 } \\rangle = \\langle w _ { 2 } , w _ { 1 } \\rangle" + }, + { + "category_id": 13, + "poly": [ + 298, + 1710, + 487, + 1710, + 487, + 1740, + 298, + 1740 + ], + "score": 0.9, + "latex": "\\Lambda : W ^ { \\otimes D } \\to \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1551, + 353, + 1551, + 353, + 1584, + 298, + 1584 + ], + "score": 0.9, + "latex": "\\Lambda _ { k , \\sigma }" + }, + { + "category_id": 14, + "poly": [ + 658, + 1883, + 1040, + 1883, + 1040, + 1920, + 658, + 1920 + ], + "score": 0.9, + "latex": "( w _ { 1 } , \\dots , w _ { D } ) \\mapsto w _ { 1 } \\otimes w _ { 2 } \\dots w _ { D }" + }, + { + "category_id": 13, + "poly": [ + 372, + 1034, + 477, + 1034, + 477, + 1061, + 372, + 1061 + ], + "score": 0.9, + "latex": "e _ { 1 } , e _ { 2 } , e _ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 1197, + 1741, + 1249, + 1741, + 1249, + 1771, + 1197, + 1771 + ], + "score": 0.89, + "latex": "W ^ { D }" + }, + { + "category_id": 13, + "poly": [ + 1342, + 1061, + 1392, + 1061, + 1392, + 1088, + 1342, + 1088 + ], + "score": 0.89, + "latex": "W ^ { D }" + }, + { + "category_id": 14, + "poly": [ + 652, + 836, + 1041, + 836, + 1041, + 879, + 652, + 879 + ], + "score": 0.89, + "latex": "W = \\mathbb { R } ^ { 3 } \\operatorname { a n d } W ^ { \\otimes D } \\cong \\mathbb { R } ^ { 3 ^ { D } } = \\mathcal { T } _ { D } ." + }, + { + "category_id": 13, + "poly": [ + 414, + 1583, + 453, + 1583, + 453, + 1612, + 414, + 1612 + ], + "score": 0.89, + "latex": "\\lambda _ { D }" + }, + { + "category_id": 13, + "poly": [ + 428, + 723, + 467, + 723, + 467, + 751, + 428, + 751 + ], + "score": 0.88, + "latex": "\\Lambda _ { T }" + }, + { + "category_id": 13, + "poly": [ + 972, + 1548, + 1009, + 1548, + 1009, + 1580, + 972, + 1580 + ], + "score": 0.88, + "latex": "3 ^ { D }" + }, + { + "category_id": 13, + "poly": [ + 907, + 1028, + 941, + 1028, + 941, + 1057, + 907, + 1057 + ], + "score": 0.87, + "latex": "\\mathbb { R } ^ { 3 }" + }, + { + "category_id": 14, + "poly": [ + 504, + 983, + 1193, + 983, + 1193, + 1025, + 504, + 1025 + ], + "score": 0.87, + "latex": "\\tilde { \\Lambda } ( e _ { i _ { 1 } } , \\ldots , e _ { i _ { D } } ) = \\Lambda ( e _ { i _ { 1 } } \\otimes \\ldots \\otimes e _ { i _ { D } } ) , \\forall ( i _ { 1 } , \\ldots , i _ { D } ) \\in [ 3 ] ^ { D } ," + }, + { + "category_id": 13, + "poly": [ + 969, + 1395, + 1007, + 1395, + 1007, + 1425, + 969, + 1425 + ], + "score": 0.87, + "latex": "\\mathcal { T } _ { D }" + }, + { + "category_id": 13, + "poly": [ + 750, + 1933, + 772, + 1933, + 772, + 1964, + 750, + 1964 + ], + "score": 0.87, + "latex": "\\tilde { \\Lambda }" + }, + { + "category_id": 13, + "poly": [ + 1232, + 1969, + 1254, + 1969, + 1254, + 2000, + 1232, + 2000 + ], + "score": 0.86, + "latex": "\\tilde { \\Lambda }" + }, + { + "category_id": 13, + "poly": [ + 1215, + 1937, + 1350, + 1937, + 1350, + 1971, + 1215, + 1971 + ], + "score": 0.85, + "latex": "R \\in \\mathrm { S O ( 3 ) }" + }, + { + "category_id": 13, + "poly": [ + 1314, + 924, + 1335, + 924, + 1335, + 956, + 1314, + 956 + ], + "score": 0.84, + "latex": "\\hat { \\Lambda }" + }, + { + "category_id": 13, + "poly": [ + 946, + 567, + 969, + 567, + 969, + 593, + 946, + 593 + ], + "score": 0.84, + "latex": "\\Lambda" + }, + { + "category_id": 13, + "poly": [ + 627, + 293, + 652, + 293, + 652, + 319, + 627, + 319 + ], + "score": 0.83, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 559, + 2004, + 580, + 2004, + 580, + 2030, + 559, + 2030 + ], + "score": 0.82, + "latex": "\\Lambda" + }, + { + "category_id": 13, + "poly": [ + 826, + 1938, + 848, + 1938, + 848, + 1965, + 826, + 1965 + ], + "score": 0.82, + "latex": "\\Lambda" + }, + { + "category_id": 13, + "poly": [ + 1017, + 929, + 1040, + 929, + 1040, + 956, + 1017, + 956 + ], + "score": 0.81, + "latex": "\\Lambda" + }, + { + "category_id": 13, + "poly": [ + 554, + 297, + 573, + 297, + 573, + 323, + 554, + 323 + ], + "score": 0.81, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 607, + 566, + 630, + 566, + 630, + 593, + 607, + 593 + ], + "score": 0.8, + "latex": "\\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 685, + 354, + 713, + 354, + 713, + 381, + 685, + 381 + ], + "score": 0.79, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 1039, + 1396, + 1063, + 1396, + 1063, + 1422, + 1039, + 1422 + ], + "score": 0.76, + "latex": "\\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 552, + 1119, + 837, + 1119, + 837, + 1155, + 552, + 1155 + ], + "score": 0.75, + "latex": "K _ { D } = \\left\\{ k \\in \\mathbb { N } _ { + } | D - 3 k \\right\\}" + }, + { + "category_id": 13, + "poly": [ + 1204, + 326, + 1224, + 326, + 1224, + 354, + 1204, + 354 + ], + "score": 0.75, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 1360, + 895, + 1404, + 895, + 1404, + 923, + 1360, + 923 + ], + "score": 0.71, + "latex": "\\Lambda :" + }, + { + "category_id": 13, + "poly": [ + 937, + 359, + 953, + 359, + 953, + 380, + 937, + 380 + ], + "score": 0.65, + "latex": "\\epsilon" + }, + { + "category_id": 14, + "poly": [ + 548, + 1117, + 1145, + 1117, + 1145, + 1156, + 548, + 1156 + ], + "score": 0.61, + "latex": "K _ { D } = \\{ k \\in \\mathbb { N } _ { + } | D - 3 k { \\mathrm { i s ~ e v e n ~ a n d ~ n o n } } { \\mathrm { - n e g a t i v e . ~ } } \\}" + }, + { + "category_id": 13, + "poly": [ + 532, + 1680, + 547, + 1680, + 547, + 1705, + 532, + 1705 + ], + "score": 0.55, + "latex": "^ { l }" + }, + { + "category_id": 13, + "poly": [ + 1129, + 1120, + 1146, + 1120, + 1146, + 1155, + 1129, + 1155 + ], + "score": 0.53, + "latex": "\\}" + }, + { + "category_id": 13, + "poly": [ + 326, + 2003, + 401, + 2003, + 401, + 2036, + 326, + 2036 + ], + "score": 0.37, + "latex": "\\mathrm { S O ( 3 ) }" + }, + { + "category_id": 13, + "poly": [ + 437, + 1612, + 511, + 1612, + 511, + 1645, + 437, + 1645 + ], + "score": 0.35, + "latex": "\\mathrm { S O ( 3 ) }" + }, + { + "category_id": 13, + "poly": [ + 496, + 721, + 570, + 721, + 570, + 755, + 496, + 755 + ], + "score": 0.34, + "latex": "\\mathrm { S O ( 3 ) }" + }, + { + "category_id": 13, + "poly": [ + 731, + 1463, + 764, + 1463, + 764, + 1487, + 731, + 1487 + ], + "score": 0.32, + "latex": "2 6" + }, + { + "category_id": 13, + "poly": [ + 1094, + 527, + 1169, + 527, + 1169, + 561, + 1094, + 561 + ], + "score": 0.25, + "latex": "\\mathrm { S O ( 3 ) }" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 453.0, + 708.0, + 453.0, + 708.0, + 500.0, + 294.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 2083.0, + 872.0, + 2083.0, + 872.0, + 2124.0, + 828.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 231.0, + 1406.0, + 231.0, + 1406.0, + 266.0, + 393.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 261.0, + 1406.0, + 261.0, + 1406.0, + 296.0, + 393.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 292.0, + 553.0, + 292.0, + 553.0, + 326.0, + 393.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 292.0, + 626.0, + 292.0, + 626.0, + 326.0, + 574.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 292.0, + 1406.0, + 292.0, + 1406.0, + 326.0, + 653.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 322.0, + 396.0, + 322.0, + 396.0, + 357.0, + 393.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 322.0, + 1203.0, + 322.0, + 1203.0, + 357.0, + 464.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1225.0, + 322.0, + 1406.0, + 322.0, + 1406.0, + 357.0, + 1225.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 352.0, + 684.0, + 352.0, + 684.0, + 387.0, + 393.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 352.0, + 936.0, + 352.0, + 936.0, + 387.0, + 714.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 954.0, + 352.0, + 1406.0, + 352.0, + 1406.0, + 387.0, + 954.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 383.0, + 1151.0, + 383.0, + 1151.0, + 418.0, + 393.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1513.0, + 1404.0, + 1513.0, + 1404.0, + 1552.0, + 296.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1547.0, + 297.0, + 1547.0, + 297.0, + 1585.0, + 292.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1547.0, + 834.0, + 1547.0, + 834.0, + 1585.0, + 354.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 1547.0, + 971.0, + 1547.0, + 971.0, + 1585.0, + 892.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1010.0, + 1547.0, + 1407.0, + 1547.0, + 1407.0, + 1585.0, + 1010.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1581.0, + 413.0, + 1581.0, + 413.0, + 1618.0, + 294.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 454.0, + 1581.0, + 1157.0, + 1581.0, + 1157.0, + 1618.0, + 454.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1394.0, + 1581.0, + 1406.0, + 1581.0, + 1406.0, + 1618.0, + 1394.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1607.0, + 436.0, + 1607.0, + 436.0, + 1648.0, + 292.0, + 1648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 1607.0, + 783.0, + 1607.0, + 783.0, + 1648.0, + 512.0, + 1648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 911.0, + 1607.0, + 1198.0, + 1607.0, + 1198.0, + 1648.0, + 911.0, + 1648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1296.0, + 1607.0, + 1309.0, + 1607.0, + 1309.0, + 1648.0, + 1296.0, + 1648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1933.0, + 749.0, + 1933.0, + 749.0, + 1972.0, + 292.0, + 1972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 773.0, + 1933.0, + 825.0, + 1933.0, + 825.0, + 1972.0, + 773.0, + 1972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 1933.0, + 1214.0, + 1933.0, + 1214.0, + 1972.0, + 849.0, + 1972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1351.0, + 1933.0, + 1405.0, + 1933.0, + 1405.0, + 1972.0, + 1351.0, + 1972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1969.0, + 662.0, + 1969.0, + 662.0, + 2008.0, + 293.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 1969.0, + 818.0, + 1969.0, + 818.0, + 2008.0, + 767.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 924.0, + 1969.0, + 1231.0, + 1969.0, + 1231.0, + 2008.0, + 924.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 1969.0, + 1405.0, + 1969.0, + 1405.0, + 2008.0, + 1255.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2002.0, + 325.0, + 2002.0, + 325.0, + 2036.0, + 293.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 2002.0, + 558.0, + 2002.0, + 558.0, + 2036.0, + 402.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 2002.0, + 593.0, + 2002.0, + 593.0, + 2036.0, + 581.0, + 2036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1393.0, + 968.0, + 1393.0, + 968.0, + 1431.0, + 295.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1008.0, + 1393.0, + 1038.0, + 1393.0, + 1038.0, + 1431.0, + 1008.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 1393.0, + 1406.0, + 1393.0, + 1406.0, + 1431.0, + 1064.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1423.0, + 534.0, + 1423.0, + 534.0, + 1469.0, + 292.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 1423.0, + 1345.0, + 1423.0, + 1345.0, + 1469.0, + 916.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1459.0, + 730.0, + 1459.0, + 730.0, + 1493.0, + 293.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 1459.0, + 768.0, + 1459.0, + 768.0, + 1493.0, + 765.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1675.0, + 531.0, + 1675.0, + 531.0, + 1714.0, + 295.0, + 1714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 1675.0, + 1404.0, + 1675.0, + 1404.0, + 1714.0, + 548.0, + 1714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1705.0, + 297.0, + 1705.0, + 297.0, + 1748.0, + 292.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 488.0, + 1705.0, + 791.0, + 1705.0, + 791.0, + 1748.0, + 488.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 1705.0, + 1407.0, + 1705.0, + 1407.0, + 1748.0, + 964.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1737.0, + 1196.0, + 1737.0, + 1196.0, + 1780.0, + 294.0, + 1780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1250.0, + 1737.0, + 1402.0, + 1737.0, + 1402.0, + 1780.0, + 1250.0, + 1780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 893.0, + 1359.0, + 893.0, + 1359.0, + 927.0, + 296.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 920.0, + 297.0, + 920.0, + 297.0, + 966.0, + 292.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 920.0, + 729.0, + 920.0, + 729.0, + 966.0, + 432.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 920.0, + 1016.0, + 920.0, + 1016.0, + 966.0, + 889.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 920.0, + 1313.0, + 920.0, + 1313.0, + 966.0, + 1041.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1336.0, + 920.0, + 1405.0, + 920.0, + 1405.0, + 966.0, + 1336.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 957.0, + 350.0, + 957.0, + 350.0, + 992.0, + 292.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 521.0, + 927.0, + 521.0, + 927.0, + 568.0, + 292.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 995.0, + 521.0, + 1093.0, + 521.0, + 1093.0, + 568.0, + 995.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1170.0, + 521.0, + 1407.0, + 521.0, + 1407.0, + 568.0, + 1170.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 553.0, + 358.0, + 553.0, + 358.0, + 612.0, + 291.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 553.0, + 606.0, + 553.0, + 606.0, + 612.0, + 577.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 631.0, + 553.0, + 945.0, + 553.0, + 945.0, + 612.0, + 631.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 970.0, + 553.0, + 1133.0, + 553.0, + 1133.0, + 612.0, + 970.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 716.0, + 427.0, + 716.0, + 427.0, + 755.0, + 295.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 468.0, + 716.0, + 495.0, + 716.0, + 495.0, + 755.0, + 468.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 571.0, + 716.0, + 1402.0, + 716.0, + 1402.0, + 755.0, + 571.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 750.0, + 334.0, + 750.0, + 334.0, + 783.0, + 296.0, + 783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 750.0, + 490.0, + 750.0, + 490.0, + 783.0, + 479.0, + 783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1027.0, + 371.0, + 1027.0, + 371.0, + 1063.0, + 294.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 1027.0, + 906.0, + 1027.0, + 906.0, + 1063.0, + 478.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 1027.0, + 1406.0, + 1027.0, + 1406.0, + 1063.0, + 942.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1054.0, + 528.0, + 1054.0, + 528.0, + 1094.0, + 292.0, + 1094.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 598.0, + 1054.0, + 1341.0, + 1054.0, + 1341.0, + 1094.0, + 598.0, + 1094.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1393.0, + 1054.0, + 1405.0, + 1054.0, + 1405.0, + 1094.0, + 1393.0, + 1094.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1091.0, + 481.0, + 1091.0, + 481.0, + 1125.0, + 296.0, + 1125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1159.0, + 341.0, + 1159.0, + 341.0, + 1196.0, + 295.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 1159.0, + 838.0, + 1159.0, + 838.0, + 1196.0, + 438.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 795.0, + 639.0, + 795.0, + 639.0, + 833.0, + 294.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1824.0, + 422.0, + 1824.0, + 422.0, + 1882.0, + 291.0, + 1882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 1824.0, + 725.0, + 1824.0, + 725.0, + 1882.0, + 577.0, + 1882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 1824.0, + 1030.0, + 1824.0, + 1030.0, + 1882.0, + 795.0, + 1882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1291.0, + 384.0, + 1291.0, + 384.0, + 1331.0, + 296.0, + 1331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 1291.0, + 719.0, + 1291.0, + 719.0, + 1331.0, + 601.0, + 1331.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 19, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 425, + 886, + 1274, + 886, + 1274, + 1138, + 425, + 1138 + ], + "score": 0.98, + "html": "
Model variant Mean l1error
Order 0irreps 111.2 ± 0.3
Order 0-1 ireps 82.1 ± 0.4
Order 0-2 irreps 62.1 ± 1.5
Order 0-3 irreps 53.2 ± 1.9
Order 0-4 irreps 51.2 ± 0.4
Order O-3 irreps without non-linearity 53.1 ± 0.8
Order O-3 irreps + self-interaction only in the final layer 55.2 ± 0.3
" + }, + { + "category_id": 1, + "poly": [ + 299, + 1910, + 1402, + 1910, + 1402, + 2035, + 299, + 2035 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 298, + 493, + 1405, + 493, + 1405, + 587, + 298, + 587 + ], + "score": 0.969 + }, + { + "category_id": 1, + "poly": [ + 302, + 1657, + 1398, + 1657, + 1398, + 1752, + 302, + 1752 + ], + "score": 0.964 + }, + { + "category_id": 1, + "poly": [ + 298, + 1783, + 1403, + 1783, + 1403, + 1876, + 298, + 1876 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 297, + 227, + 1402, + 227, + 1402, + 323, + 297, + 323 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 297, + 1174, + 1398, + 1174, + 1398, + 1237, + 297, + 1237 + ], + "score": 0.945 + }, + { + "category_id": 8, + "poly": [ + 476, + 452, + 1222, + 452, + 1222, + 489, + 476, + 489 + ], + "score": 0.934 + }, + { + "category_id": 6, + "poly": [ + 298, + 799, + 1402, + 799, + 1402, + 885, + 298, + 885 + ], + "score": 0.932 + }, + { + "category_id": 8, + "poly": [ + 486, + 339, + 1211, + 339, + 1211, + 378, + 486, + 378 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 296, + 390, + 1403, + 390, + 1403, + 451, + 296, + 451 + ], + "score": 0.924 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 813, + 75, + 813, + 105, + 299, + 105 + ], + "score": 0.906 + }, + { + "category_id": 0, + "poly": [ + 299, + 697, + 564, + 697, + 564, + 733, + 299, + 733 + ], + "score": 0.904 + }, + { + "category_id": 9, + "poly": [ + 1352, + 453, + 1399, + 453, + 1399, + 483, + 1352, + 483 + ], + "score": 0.89 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 863, + 2088, + 863, + 2112, + 835, + 2112 + ], + "score": 0.834 + }, + { + "category_id": 2, + "poly": [ + 1375, + 602, + 1402, + 602, + 1402, + 631, + 1375, + 631 + ], + "score": 0.811 + }, + { + "category_id": 1, + "poly": [ + 361, + 1265, + 1404, + 1265, + 1404, + 1623, + 361, + 1623 + ], + "score": 0.765 + }, + { + "category_id": 13, + "poly": [ + 809, + 529, + 945, + 529, + 945, + 556, + 809, + 556 + ], + "score": 0.9, + "latex": "w _ { 1 } , \\ldots , w _ { D }" + }, + { + "category_id": 13, + "poly": [ + 599, + 228, + 651, + 228, + 651, + 259, + 599, + 259 + ], + "score": 0.9, + "latex": "W ^ { D }" + }, + { + "category_id": 13, + "poly": [ + 707, + 1971, + 767, + 1971, + 767, + 2001, + 707, + 2001 + ], + "score": 0.9, + "latex": "1 0 ^ { - 3 }" + }, + { + "category_id": 13, + "poly": [ + 1037, + 556, + 1075, + 556, + 1075, + 585, + 1037, + 585 + ], + "score": 0.9, + "latex": "\\lambda _ { D }" + }, + { + "category_id": 13, + "poly": [ + 1246, + 228, + 1297, + 228, + 1297, + 259, + 1246, + 259 + ], + "score": 0.9, + "latex": "W ^ { D }" + }, + { + "category_id": 13, + "poly": [ + 1124, + 1944, + 1152, + 1944, + 1152, + 1973, + 1124, + 1973 + ], + "score": 0.87, + "latex": "\\ell _ { 1 }" + }, + { + "category_id": 14, + "poly": [ + 480, + 450, + 1222, + 450, + 1222, + 488, + 480, + 488 + ], + "score": 0.86, + "latex": "\\operatorname * { d e t } \\bigl ( w _ { i _ { 1 } } , w _ { i _ { 2 } } , w _ { i _ { 3 } } \\bigr ) \\operatorname * { d e t } \\bigl ( w _ { i _ { 4 } } , w _ { i _ { 5 } } , w _ { i _ { 6 } } \\bigr ) \\ldots \\langle w _ { j _ { 1 } } , w _ { j _ { 2 } } \\rangle \\langle w _ { j _ { 3 } } , w _ { j _ { 4 } } \\rangle \\ldots ." + }, + { + "category_id": 13, + "poly": [ + 784, + 233, + 802, + 233, + 802, + 262, + 784, + 262 + ], + "score": 0.85, + "latex": "\\tilde { \\rho }" + }, + { + "category_id": 13, + "poly": [ + 297, + 860, + 340, + 860, + 340, + 890, + 297, + 890 + ], + "score": 0.84, + "latex": "1 0 ^ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 770, + 1946, + 842, + 1946, + 842, + 1974, + 770, + 1974 + ], + "score": 0.83, + "latex": "\\epsilon _ { h o m o }" + }, + { + "category_id": 14, + "poly": [ + 488, + 338, + 1202, + 338, + 1202, + 377, + 488, + 377 + ], + "score": 0.83, + "latex": "\\begin{array} { r } { \\operatorname* { d e t } ( w _ { i _ { 1 } } , w _ { i _ { 2 } } , w _ { i _ { 3 } } ) \\mathrm { ~ a n d ~ } \\langle w _ { j _ { 1 } } , w _ { j _ { 2 } } \\rangle , \\mathrm { ~ w h e r e ~ } i _ { 1 } , i _ { 2 } , i _ { 3 } , j _ { 1 } , j _ { 2 } \\in [ D ] . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 325, + 263, + 343, + 263, + 343, + 294, + 325, + 294 + ], + "score": 0.77, + "latex": "\\tilde { \\rho }" + }, + { + "category_id": 13, + "poly": [ + 902, + 835, + 974, + 835, + 974, + 862, + 902, + 862 + ], + "score": 0.64, + "latex": "\\epsilon _ { h o m o }" + }, + { + "category_id": 13, + "poly": [ + 628, + 1105, + 651, + 1105, + 651, + 1128, + 628, + 1128 + ], + "score": 0.6, + "latex": "^ +" + }, + { + "category_id": 13, + "poly": [ + 959, + 341, + 1200, + 341, + 1200, + 375, + 959, + 375 + ], + "score": 0.52, + "latex": "i _ { 1 } , i _ { 2 } , i _ { 3 } , j _ { 1 } , j _ { 2 } \\in [ D ]" + }, + { + "category_id": 13, + "poly": [ + 452, + 390, + 527, + 390, + 527, + 425, + 452, + 425 + ], + "score": 0.3, + "latex": "\\mathrm { S O ( 3 ) }" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 800.0, + 1404.0, + 800.0, + 1404.0, + 833.0, + 294.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 828.0, + 901.0, + 828.0, + 901.0, + 866.0, + 291.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 975.0, + 828.0, + 1404.0, + 828.0, + 1404.0, + 866.0, + 975.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 856.0, + 296.0, + 856.0, + 296.0, + 893.0, + 292.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 856.0, + 356.0, + 856.0, + 356.0, + 893.0, + 341.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 695.0, + 570.0, + 695.0, + 570.0, + 737.0, + 294.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 869.0, + 2084.0, + 869.0, + 2125.0, + 829.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1380.0, + 607.0, + 1401.0, + 607.0, + 1401.0, + 629.0, + 1380.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1910.0, + 1407.0, + 1910.0, + 1407.0, + 1947.0, + 295.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1941.0, + 769.0, + 1941.0, + 769.0, + 1981.0, + 293.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 843.0, + 1941.0, + 1123.0, + 1941.0, + 1123.0, + 1981.0, + 843.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1153.0, + 1941.0, + 1408.0, + 1941.0, + 1408.0, + 1981.0, + 1153.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1971.0, + 706.0, + 1971.0, + 706.0, + 2008.0, + 294.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 1971.0, + 1407.0, + 1971.0, + 1407.0, + 2008.0, + 768.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2002.0, + 1323.0, + 2002.0, + 1323.0, + 2038.0, + 293.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 491.0, + 1405.0, + 491.0, + 1405.0, + 529.0, + 295.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 524.0, + 808.0, + 524.0, + 808.0, + 562.0, + 293.0, + 562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 946.0, + 524.0, + 1405.0, + 524.0, + 1405.0, + 562.0, + 946.0, + 562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 553.0, + 1036.0, + 553.0, + 1036.0, + 591.0, + 293.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1076.0, + 553.0, + 1088.0, + 553.0, + 1088.0, + 591.0, + 1076.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1658.0, + 1404.0, + 1658.0, + 1404.0, + 1692.0, + 296.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1685.0, + 1403.0, + 1685.0, + 1403.0, + 1725.0, + 293.0, + 1725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1717.0, + 1053.0, + 1717.0, + 1053.0, + 1756.0, + 296.0, + 1756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1779.0, + 1404.0, + 1779.0, + 1404.0, + 1822.0, + 293.0, + 1822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1813.0, + 1404.0, + 1813.0, + 1404.0, + 1852.0, + 293.0, + 1852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1846.0, + 572.0, + 1846.0, + 572.0, + 1877.0, + 296.0, + 1877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 227.0, + 598.0, + 227.0, + 598.0, + 265.0, + 293.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 227.0, + 783.0, + 227.0, + 783.0, + 265.0, + 652.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 803.0, + 227.0, + 1245.0, + 227.0, + 1245.0, + 265.0, + 803.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1298.0, + 227.0, + 1406.0, + 227.0, + 1406.0, + 265.0, + 1298.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 259.0, + 324.0, + 259.0, + 324.0, + 298.0, + 291.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 259.0, + 1403.0, + 259.0, + 1403.0, + 298.0, + 344.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 292.0, + 688.0, + 292.0, + 688.0, + 326.0, + 293.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1170.0, + 1405.0, + 1170.0, + 1405.0, + 1212.0, + 295.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1204.0, + 1084.0, + 1204.0, + 1084.0, + 1240.0, + 295.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 387.0, + 451.0, + 387.0, + 451.0, + 428.0, + 293.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 528.0, + 387.0, + 1406.0, + 387.0, + 1406.0, + 428.0, + 528.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 419.0, + 402.0, + 419.0, + 402.0, + 456.0, + 293.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1266.0, + 1402.0, + 1266.0, + 1402.0, + 1300.0, + 362.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1296.0, + 1404.0, + 1296.0, + 1404.0, + 1332.0, + 394.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1327.0, + 805.0, + 1327.0, + 805.0, + 1364.0, + 394.0, + 1364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1369.0, + 1404.0, + 1369.0, + 1404.0, + 1402.0, + 361.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1396.0, + 1403.0, + 1396.0, + 1403.0, + 1438.0, + 392.0, + 1438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1427.0, + 1405.0, + 1427.0, + 1405.0, + 1465.0, + 392.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1462.0, + 598.0, + 1462.0, + 598.0, + 1490.0, + 394.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1500.0, + 1406.0, + 1500.0, + 1406.0, + 1537.0, + 362.0, + 1537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1531.0, + 1406.0, + 1531.0, + 1406.0, + 1568.0, + 394.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1562.0, + 1406.0, + 1562.0, + 1406.0, + 1599.0, + 393.0, + 1599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 1589.0, + 469.0, + 1589.0, + 469.0, + 1633.0, + 391.0, + 1633.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 20, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 416, + 824, + 416, + 824, + 968, + 298, + 968 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 229, + 1404, + 229, + 1404, + 384, + 297, + 384 + ], + "score": 0.976 + }, + { + "category_id": 3, + "poly": [ + 857, + 433, + 1337, + 433, + 1337, + 794, + 857, + 794 + ], + "score": 0.968 + }, + { + "category_id": 4, + "poly": [ + 845, + 834, + 1403, + 834, + 1403, + 987, + 845, + 987 + ], + "score": 0.963 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 812, + 76, + 812, + 104, + 300, + 104 + ], + "score": 0.894 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 865, + 2088, + 865, + 2112, + 835, + 2112 + ], + "score": 0.845 + }, + { + "category_id": 13, + "poly": [ + 976, + 836, + 1005, + 836, + 1005, + 866, + 976, + 866 + ], + "score": 0.83, + "latex": "\\ell _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 440, + 354, + 458, + 354, + 458, + 380, + 440, + 380 + ], + "score": 0.78, + "latex": "d" + }, + { + "category_id": 13, + "poly": [ + 1110, + 322, + 1156, + 322, + 1156, + 350, + 1110, + 350 + ], + "score": 0.76, + "latex": "1 6 d" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 442.0, + 924.0, + 442.0, + 924.0, + 485.0, + 870.0, + 485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 931.0, + 449.0, + 946.0, + 449.0, + 946.0, + 466.0, + 931.0, + 466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1149.0, + 445.0, + 1327.0, + 445.0, + 1327.0, + 478.0, + 1149.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 487.0, + 924.0, + 487.0, + 924.0, + 530.0, + 871.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 533.0, + 924.0, + 533.0, + 924.0, + 584.0, + 881.0, + 584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 855.0, + 568.0, + 883.0, + 568.0, + 883.0, + 623.0, + 855.0, + 623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 576.0, + 922.0, + 576.0, + 922.0, + 624.0, + 883.0, + 624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 625.0, + 923.0, + 625.0, + 923.0, + 669.0, + 883.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 884.0, + 672.0, + 921.0, + 672.0, + 921.0, + 713.0, + 884.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 884.0, + 717.0, + 922.0, + 717.0, + 922.0, + 754.0, + 884.0, + 754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 748.0, + 948.0, + 748.0, + 948.0, + 774.0, + 928.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1025.0, + 750.0, + 1041.0, + 750.0, + 1041.0, + 770.0, + 1025.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1115.0, + 744.0, + 1138.0, + 744.0, + 1138.0, + 773.0, + 1115.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1211.0, + 748.0, + 1231.0, + 748.0, + 1231.0, + 772.0, + 1211.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1309.0, + 753.0, + 1322.0, + 753.0, + 1322.0, + 767.0, + 1309.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 763.0, + 1204.0, + 763.0, + 1204.0, + 799.0, + 1051.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 834.0, + 975.0, + 834.0, + 975.0, + 868.0, + 846.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1006.0, + 834.0, + 1404.0, + 834.0, + 1404.0, + 868.0, + 1006.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 844.0, + 865.0, + 1405.0, + 865.0, + 1405.0, + 898.0, + 844.0, + 898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 898.0, + 1402.0, + 898.0, + 1402.0, + 928.0, + 846.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 927.0, + 1403.0, + 927.0, + 1403.0, + 958.0, + 846.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 843.0, + 957.0, + 907.0, + 957.0, + 907.0, + 991.0, + 843.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2124.0, + 831.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 416.0, + 826.0, + 416.0, + 826.0, + 453.0, + 296.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 447.0, + 826.0, + 447.0, + 826.0, + 481.0, + 295.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 476.0, + 828.0, + 476.0, + 828.0, + 516.0, + 293.0, + 516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 511.0, + 826.0, + 511.0, + 826.0, + 542.0, + 295.0, + 542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 539.0, + 827.0, + 539.0, + 827.0, + 574.0, + 294.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 569.0, + 827.0, + 569.0, + 827.0, + 602.0, + 296.0, + 602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 598.0, + 828.0, + 598.0, + 828.0, + 635.0, + 294.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 630.0, + 828.0, + 630.0, + 828.0, + 664.0, + 294.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 660.0, + 827.0, + 660.0, + 827.0, + 695.0, + 294.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 694.0, + 827.0, + 694.0, + 827.0, + 726.0, + 295.0, + 726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 718.0, + 828.0, + 718.0, + 828.0, + 758.0, + 294.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 753.0, + 826.0, + 753.0, + 826.0, + 786.0, + 293.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 784.0, + 826.0, + 784.0, + 826.0, + 815.0, + 295.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 813.0, + 826.0, + 813.0, + 826.0, + 845.0, + 295.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 844.0, + 827.0, + 844.0, + 827.0, + 878.0, + 294.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 874.0, + 826.0, + 874.0, + 826.0, + 909.0, + 294.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 905.0, + 825.0, + 905.0, + 825.0, + 938.0, + 294.0, + 938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 937.0, + 819.0, + 937.0, + 819.0, + 968.0, + 294.0, + 968.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 231.0, + 1405.0, + 231.0, + 1405.0, + 264.0, + 296.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 259.0, + 1402.0, + 259.0, + 1402.0, + 297.0, + 295.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 291.0, + 1406.0, + 291.0, + 1406.0, + 326.0, + 294.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 320.0, + 1109.0, + 320.0, + 1109.0, + 356.0, + 294.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1157.0, + 320.0, + 1406.0, + 320.0, + 1406.0, + 356.0, + 1157.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 352.0, + 439.0, + 352.0, + 439.0, + 389.0, + 295.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 459.0, + 352.0, + 1106.0, + 352.0, + 1106.0, + 389.0, + 459.0, + 389.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 21, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/train/IhiU6AJYpDs/IhiU6AJYpDs.md b/parse/train/IhiU6AJYpDs/IhiU6AJYpDs.md new file mode 100644 index 0000000000000000000000000000000000000000..f8e32636cdff29b6b1218eeda7745b1c96b380a5 --- /dev/null +++ b/parse/train/IhiU6AJYpDs/IhiU6AJYpDs.md @@ -0,0 +1,415 @@ +# Online Robust Reinforcement Learning with Model Uncertainty + +Yue Wang +University at Buffalo +Buffalo, NY 14228 +ywang294@buffalo.edu + +Shaofeng Zou University at Buffalo Buffalo, NY 14228 szou3@buffalo.edu + +# Abstract + +Robust reinforcement learning (RL) is to find a policy that optimizes the worstcase performance over an uncertainty set of MDPs. In this paper, we focus on model-free robust RL, where the uncertainty set is defined to be centering at a misspecified MDP that generates a single sample trajectory sequentially, and is assumed to be unknown. We develop a sample-based approach to estimate the unknown uncertainty set, and design robust Q-learning algorithm (tabular case) and robust TDC algorithm (function approximation setting), which can be implemented in an online and incremental fashion. For the robust Q-learning algorithm, we prove that it converges to the optimal robust Q function, and for the robust TDC algorithm, we prove that it converges asymptotically to some stationary points. Unlike the results in [Roy et al., 2017], our algorithms do not need any additional conditions on the discount factor to guarantee the convergence. We further characterize the finite-time error bounds of the two algorithms, and show that both the robust Qlearning and robust TDC algorithms converge as fast as their vanilla counterparts (within a constant factor). Our numerical experiments further demonstrate the robustness of our algorithms. Our approach can be readily extended to robustify many other algorithms, e.g., TD, SARSA, and other GTD algorithms. + +# 1 Introduction + +Existing studies on Markov decision process (MDP) and reinforcement learning (RL) [Sutton and Barto, 2018] mostly rely on the crucial assumption that the environment on which a learned policy will be deployed is the same one that was used to generate the policy, which is often violated in practice – e.g., the simulator may be different from the true environment, and the MDP may evolve over time. Due to such model deviation, the actual performance of the learned policy can significantly degrade. To address this problem, the framework of robust MDP was formulated in [Bagnell et al., 2001, Nilim and El Ghaoui, 2004, Iyengar, 2005], where the transition kernel of the MDP is not fixed and lies in an uncertainty set, and the goal is to learn a policy that performs well under the worst-case MDP in the uncertainty set. In [Bagnell et al., 2001, Nilim and El Ghaoui, 2004, Iyengar, 2005], it was assumed that the uncertainty set is known beforehand, i.e., model-based approach, and dynamic programming can be used to find the optimal robust policy. + +The model-based approach, however, requires a model of the uncertainty set known beforehand, and needs a large memory to store the model when the state and action spaces are large, which make it less applicable for many practical scenarios. This motivates the study in this paper, model-free robust RL with model uncertainty, which is to learn a robust policy using a single sample trajectory from a misspecified MDP, e.g., a simulator and a similar environment in which samples are easier to collect than in the target environment where the policy is going to be deployed. The major challenge lies in that the transition kernel of the misspecified MDP is not given beforehand, and thus, the uncertainty set and the optimal robust policy need to be learned simultaneously using sequentially observed data from the misspecified MDP. Moreover, robust RL learns the value function of the worst-case MDP in the uncertainty set which is different from the misspecified MDP that generates samples. This is similar to the off-policy learning, which we refer to as the "off-transition-kernel" setting. Therefore, the learning may be unstable and could diverge especially when function approximation is used [Baird, 1995]. + +In this paper, we develop a model-free approach for robust RL with model uncertainty. Our major contributions in this paper are summarized as follows. + +• Motivated by empirical studies of adversarial training in RL [Huang et al., 2017, Kos and Song, 2017, Lin et al., 2017, Pattanaik et al., 2018, Mandlekar et al., 2017] and the $R$ -contamination model in robust detection (called $\epsilon$ -contamination model in [Huber, 1965]), we design the uncertainty set using the $R$ -contamination model (see (4) for the details). We then develop an approach to estimate the unknown uncertainty set using only the current sample, which does not incur any additional memory cost. Unlike the approach in [Roy et al., 2017], where the uncertainty set is relaxed to one not depending on the misspecified MDP that generates samples so that an online algorithm can be constructed, our approach does not need to relax the uncertainty set. + +• We develop a robust Q-learning algorithm for the tabular case, which can be implemented in an online and incremental fashion, and has the same memory cost as the vanilla Q-learning algorithm. We show that our robust Q-learning algorithm converges asymptotically, and further characterize its finite-time error bound. Unlike the results in [Roy et al., 2017] where a stringent condition on the discount factor (which is due to the relaxation of the uncertainty set, and prevents the use of a discount factor close to 1 in practice) is needed to guarantee the convergence, our algorithm converges without the need of such condition. Furthermore, our robust Q-learning algorithm converges as fast as the vanilla Q-learning algorithm [Li et al., 2020] (within a constant factor), while being robust to model uncertainty. + +• We generalize our approach to the case with function approximation (for large state/action space). We investigate the robust policy evaluation problem, i.e., evaluate a given policy under the worstcase MDP in the uncertainty set. As mentioned before, the robust RL problem is essentially "off-transition-kernel", and therefore non-robust methods with function approximation may diverge [Baird, 1995] (also see our experiments). We develop a novel extension of the gradient TD (GTD) method [Maei et al., 2010, Maei, 2011, Sutton et al., 2008] to robust RL. Our approach introduces a novel smoothed robust Bellman operator to construct the smoothed mean-squared projected robust Bellman error (MSPRBE). Using our uncertainty set design and online sample-based estimation, we develop a two time-scale robust TDC algorithm. We further characterize its convergence and finite-time error bound. + +• We conduct numerical experiments to validate the robustness of our approach. In our experiments, our robust Q-learning algorithm achieves a much higher reward than the vanilla Q-learning algorithm when being trained on a misspecified MDP; and our robust TDC algorithm converges much faster than the vanilla TDC algorithm, and the vanilla TDC algorithm may even diverge. + +# 1.1 Related Work + +Model-Based Robust MDP. The framework of robust MDP was investigated in [Iyengar, 2005, Nilim and El Ghaoui, 2004, Bagnell et al., 2001, Satia and Lave Jr, 1973, Wiesemann et al., 2013], where the transition kernel is assumed to be in some uncertainty set, and the problem can be solved by dynamic programming. This approach was further extended to the case with function approximation in [Tamar et al., 2014]. However, these studies are model-based, which assume beforehand knowledge of the uncertainty set. In this paper, we investigate the model-free setting, where the uncertainty set is a set of MDPs centered around some unknown Markov transition kernel from which a single sample trajectory can be sequentially observed. + +Adversarial Robust RL. It was shown in [Iyengar, 2005] that the robust MDP problem is equivalent to a zero-sum game between the agent and the nature. Motivated by this fact, the adversarial training approach, where an adversary perturbs the state transition, was studied in [Vinitsky et al., 2020, Pinto et al., 2017, Abdullah et al., 2019, Hou et al., 2020, Rajeswaran et al., 2017, Atkeson and Morimoto, 2003, Morimoto and Doya, 2005]. This method relies on a simulator, where the state transition can be modified in an arbitrary way. Another approach is to modify the current state through adversarial samples, which is more heuristic, e.g., [Huang et al., 2017, Kos and Song, 2017, Lin et al., 2017, Pattanaik et al., 2018, Mandlekar et al., 2017]. Despite the empirical success of these approaches, theoretical performance guarantees, e.g., convergence to the optimal robust policy and convergence rate, are yet to be established. The main difference lies in that during the training, our approach does not need to manipulate the state transition of the MDP. More importantly, we develop the asymptotic convergence to the optimal robust policy and further characterize the finite-time error bound. In [Lim et al., 2013], the scenario where some unknown parts of the state space can have arbitrary transitions while other parts are purely stochastic was studied. Adaptive algorithm to adversarial behavior was designed, and its regret bound is shown to be similar to the purely stochastic case. In [Zhang et al., 2020a], the robust adversarial RL problem for the special linear quadratic case was investigated. + +Model-free Robust RL. In [Roy et al., 2017, Badrinath and Kalathil, 2021] model-free RL with model uncertainty was studied, where in order to construct an algorithm that can be implemented in an online and incremental fashion, the uncertainty set was firstly relaxed by dropping the dependency on the misspecified MDP that generates the samples (centroid of the uncertainty set). Such a relaxation is pessimistic since the relaxed uncertainty set is not centered at the misspecified MDP anymore (which is usually similar to the target MDP), making the robustness to the relaxed uncertainty set not well-justified. Such a relaxation will further incur a stringent condition on the discounted factor to guarantee the convergence, which prevents the use of a discount factor close to 1 in practice. Moreover, only asymptotic convergence was established in [Roy et al., 2017]. In this paper, we do not relax the uncertainty set, and instead propose an online approach to estimate it. Our algorithms converge without the need of the condition on the discount factor. We also provides finite-time error bounds for our algorithms. The multi-agent RL robust to reward uncertainty was investigated in [Zhang et al., 2020b], where the reward uncertainty set is known, but the transition kernel is fixed. + +Finite-time Error Bound for RL Algorithms. For the tubular case, Q-learning has been studied intensively, e.g., in [Even-Dar et al., 2003, Beck and Srikant, 2012, Qu and Wierman, 2020, Li et al., 2020, Wainwright, 2019, Li et al., 2021]. TD with function approximation were studied in [Dalal, Gal and Szörényi, Balázs and Thoppe, Gugan and Mannor, Shie, 2018, Bhandari et al., 2018, Srikant and Ying, 2019, Cai et al., 2019, Sun et al., 2020]. Q-learning and SARSA with linear function approximation were investigated in [Zou et al., 2019, Chen et al., 2019]. The finite-time error bounds for the gradient TD algorithms [Maei et al., 2010, Sutton et al., 2009, Maei et al., 2010] were further developed recently in [Dalal et al., 2018, Liu et al., 2015, Gupta et al., 2019, Xu et al., 2019, Dalal et al., 2020, Kaledin et al., 2020, Ma et al., 2020, Wang and Zou, 2020, Ma et al., 2021, Doan, 2021]. There are also finite-time error bounds on the policy gradient methods and actor critic methods, e.g., [Wang et al., 2020, Yang et al., 2019, Kumar et al., 2019, Qiu et al., 2019, Wu et al., 2020, Cen et al., 2020, Bhandari and Russo, 2019, Agarwal et al., 2021, Mei et al., 2020]. We note that these studies are for the non-robust RL algorithms, and in this paper, we design robust RL algorithms, and characterize their finite-time error bounds. + +# 2 Preliminaries + +Markov Decision Process. An MDP can be characterized by a tuple $( \mathcal { S } , \mathcal { A } , \mathsf { P } , c , \gamma )$ , where S and $\mathcal { A }$ are the state and action spaces, $\mathsf { P } = \{ p _ { s } ^ { a } \in \varDelta _ { | \mathcal { S } | } , a \in \mathcal { A } , s \in \mathcal { S } \}$ is the transition kernel1, $c$ is the cost function, and $\gamma \in [ 0 , 1 )$ is the discount factor. Specifically, $p _ { s } ^ { a }$ denotes the distribution of the next state if taking action $a$ at state $s$ . Let $p _ { s } ^ { a } = \{ p _ { s , s ^ { \prime } } ^ { a } \} _ { s ^ { \prime } \in \mathcal S }$ , where $p _ { s , s ^ { \prime } } ^ { a }$ denotes the probability that the environment transits to state $s ^ { \prime }$ if taking action $a$ at state $s$ . The cost of taking action $a$ at state $s$ is given by $c ( s , a )$ . A stationary policy $\pi$ is a mapping from S to a distribution over $\mathcal A$ . At each time $t$ , an agent takes an action $A _ { t } \in \mathcal A$ at state $S _ { t } \in \mathcal S$ . The environment then transits to the next state $S _ { t + 1 }$ with probability $p _ { S _ { t } , S _ { t + 1 } } ^ { A _ { t } }$ , and the agent receives cost given by $c ( S _ { t } , A _ { t } )$ . The value function of a policy by follo $\pi$ staing $s \in \mathcal { S }$ fined as the expected accumulatand the goal is to find the policy discounted costthat minimizes $\pi \colon \bar { \mathbb { E } } \left[ \sum _ { t = 0 } ^ { \infty } \dot { \gamma } ^ { t } c ( S _ { t } , A _ { t } ) | S _ { 0 } = s , \pi \right]$ $\pi$ +$s \in \mathcal { S }$ + +Robust Markov Decision Process. In the robust case, the transition kernel is not fixed and lies in some uncertainty set. Denote the transition kernel at time $t$ by $\mathsf { P } _ { t }$ , and let $\kappa = ( \mathsf { P } _ { 0 } , \mathsf { P } _ { 1 } , \ldots )$ , where $\mathsf { P } _ { t } \in \mathbf { P } , \forall t \ge 0$ , and $\mathbf { P }$ is the uncertainty set of the transition kernel. The sequence $\kappa$ can be viewed as the policy of the nature, and is adversarially chosen by the nature [Bagnell et al., 2001, Nilim and + +El Ghaoui, 2004, Iyengar, 2005]. Define the robust value function of a policy $\pi$ as the worst-case expected accumulated discounted cost following a fixed policy $\pi$ over all transition kernels in the uncertainty set: + +$$ +V ^ { \pi } ( s ) = \operatorname* { m a x } _ { \kappa } \mathbb { E } _ { \kappa } \left[ \sum _ { t = 0 } ^ { \infty } \gamma ^ { t } c ( S _ { t } , A _ { t } ) | S _ { 0 } = s , \pi \right] , +$$ + +where $\mathbb { E } _ { \kappa }$ denotes the expectation when the state transits according to $\kappa$ . Similarly, define the robust action-value function for a policy $\begin{array} { r } { \pi \colon Q ^ { \pi } ( s , a ) = \operatorname* { m a x } _ { \kappa } \mathbb { E } _ { \kappa } \left[ \sum _ { t = 0 } ^ { \infty } \gamma ^ { t } c ( S _ { t } , A _ { t } ) | \dot { S _ { 0 } } = s , A _ { 0 } = a , \pi \right] } \end{array}$ . The goal of robust RL is to find the optimal robust policy $\pi ^ { * }$ that minimizes the worst-case accumulated discounted cost: + +$$ +\pi ^ { * } = \arg \operatorname* { m i n } _ { \pi } V ^ { \pi } ( s ) , \forall s \in \mathcal { S } . +$$ + +We also denote $V ^ { \pi ^ { * } }$ and $Q ^ { \pi ^ { * } }$ by $V ^ { * }$ and $Q ^ { * }$ , respectively, and $\begin{array} { r } { V ^ { * } ( s ) = \operatorname* { m i n } _ { a \in \mathcal { A } } Q ^ { * } ( s , a ) } \end{array}$ . + +Note that a transition kernel is a collection of conditional distributions. Therefore, the uncertainty set $\mathbf { P }$ of the transition kernel can be equivalently written as a collection of $\mathcal { P } _ { s } ^ { a }$ for all $s \in \mathcal { S } , a \in \mathcal { A }$ , where $\mathcal { P } _ { s } ^ { a }$ is a set of conditional distributions $p _ { s } ^ { a }$ over the state space S. Denote by $\sigma _ { \mathcal { P } } ( v ) \triangleq \operatorname* { m a x } _ { p \in \mathcal { P } } ( p ^ { \top } v )$ the support function of vector $v$ over a set of probability distributions $\mathcal { P }$ . For robust MDP, the following robust analogue of the Bellman recursion was provided in [Nilim and El Ghaoui, 2004, Iyengar, 2005]. + +Theorem 1. [Nilim and El Ghaoui, 2004] The following perfect duality condition holds for all $s \in \mathcal { S }$ : + +$$ +\operatorname* { m i n } _ { \pi } \operatorname* { m a x } _ { \kappa } \mathbb { E } _ { \kappa } \left[ \sum _ { t = 0 } ^ { \infty } \gamma ^ { t } c ( S _ { t } , A _ { t } ) \big | \pi , S _ { 0 } = s \right] = \operatorname* { m a x } _ { \kappa } \operatorname* { m i n } _ { \pi } \mathbb { E } _ { \kappa } \left[ \sum _ { t = 0 } ^ { \infty } \gamma ^ { t } c ( S _ { t } , A _ { t } ) \big | \pi , S _ { 0 } = s \right] . +$$ + +The optimal robust value function $V ^ { * }$ satisfies $\begin{array} { r } { V ^ { * } ( s ) = \operatorname* { m i n } _ { a \in \mathcal { A } } \bigl ( c ( s , a ) + \gamma \sigma _ { \mathscr { P } _ { s } ^ { a } } ( V ^ { * } ) \bigr ) } \end{array}$ , and the optimal robust action-value function $Q ^ { * }$ satisfies $Q ^ { * } ( s , a ) = c ( s , a ) + \gamma \sigma _ { \mathscr P _ { s } ^ { a } } ( V ^ { * } )$ . + +Define the robust Bellman operator $\mathbf { T }$ by $\begin{array} { r } { \mathbf { T } Q ( s , a ) = c ( s , a ) + \gamma \sigma _ { \mathcal { P } _ { s } ^ { a } } ( \operatorname* { m i n } _ { a \in \mathcal { A } } Q ( s , a ) ) } \end{array}$ . It was shown in [Nilim and El Ghaoui, 2004, Iyengar, 2005] that $\mathbf { T }$ is a contraction and its fixed point is the optimal robust $Q ^ { * }$ . When the uncertainty set is known, so that $\sigma _ { \mathcal P _ { s } ^ { a } }$ can be computed exactly, $V ^ { * }$ and $Q ^ { * }$ can be solved by dynamic programming [Iyengar, 2005, Nilim and El Ghaoui, 2004]. + +# 3 R-Contamination Model For Uncertainty Set Construction + +In this section, we construct the uncertainty set using the $R$ -contamination model. + +Let $\mathsf { P } = \{ p _ { s } ^ { a } , s \in \mathscr { S } , a \in \mathscr { A } \}$ be the centroid of the uncertainty set, i.e., the transition kernel that generates the sample trajectory, and $\mathsf { P }$ is unknown. For example, P can be the simulator at hand, which may not be exactly accurate; and $\mathsf { P }$ can be the transition kernel of environment 1, from which we can take samples to learn a policy that will be deployed in a similar environment 2. The goal is to learn a policy using samples from $\mathsf { P }$ that performs well when applied to a perturbed MDP from P. + +Motivated by empirical studies of adversarial training in RL [Huang et al., 2017, Kos and Song, 2017, Lin et al., 2017, Pattanaik et al., 2018, Mandlekar et al., 2017] and the $R$ -contamination model in robust detection [Huber, 1965], we use the $R$ -contamination model to define the uncertainty set: + +$$ +\begin{array} { r } { \mathfrak { P } _ { s } ^ { a } = \big \{ ( 1 - R ) p _ { s } ^ { a } + R q | q \in \mathcal { A } _ { | \mathcal { S } | } \big \} , s \in \mathcal { S } , a \in \mathcal { A } , \mathrm { f o r ~ s o m e ~ } 0 \le R \le 1 . } \end{array} +$$ + +Here, $p _ { s } ^ { a }$ is the centroid of the uncertainty set $\mathcal { P } _ { s } ^ { a }$ at $( s , a )$ , which is unknown, and $R$ is the design parameter of the uncertainty set, which measures the size of the uncertainty set, and is assumed to be known in the algorithm. We then let $\mathbf { P } = \bigotimes _ { s \in \mathcal { S } , a \in \mathcal { A } } \mathcal { P } _ { s } ^ { a }$ . + +Remark 1. $R$ -contamination model is closely related to other uncertainty set models like total variation and $K L$ -divergence. It can be shown that $R$ -contamination set certered at $p$ is a subset of total variation ball : $\{ ( 1 - R ) p + R q | q \in \varDelta _ { | \mathcal { S } | } \} \subset \{ q \in \varDelta _ { | \mathcal { S } | } | d _ { T V } ( p , q ) \leq R \}$ . Hence the total variation uncertainty set is less conservative than our $R$ -contamination uncertainty set. KL-divergence moreover can be related to total variation using Pinsker’s inequality, i.e., $\begin{array} { r } { d _ { T V } ( p , q ) \leq \sqrt { \frac 1 2 d _ { K L } ( p , q ) } } \end{array}$ . + +# 4 Tabular Case: Robust Q-Learning + +In this section, we focus on the tabular case with finite state and action spaces. We focus on the asynchronous setting where a single sample trajectory is available with Markovian noise. We will develop an efficient approach to estimate the unknown uncertainty set $\mathbf { P }$ , and further the support function $\sigma _ { \mathcal { P } _ { s } ^ { a } } ( \cdot )$ , and then design our robust Q-learning algorithm. + +We propose an efficient and data-driven approach to estimate the unknown $p _ { s } ^ { a }$ and thus the unknown uncertainty set $\mathcal { P } _ { s } ^ { a }$ for any $s \in \mathcal { S }$ and $a \in { \mathcal { A } }$ . Specifically, denote the sample at $t$ -th time step by $O _ { t } = ( s _ { t } , a _ { t } , s _ { t + 1 } )$ . We then use $O _ { t }$ to obtain the maximum likelihood estimate (MLE) $\hat { p } _ { t } \triangleq \mathbb { 1 } _ { s _ { t + 1 } }$ of the transition kernel $p _ { s _ { t } } ^ { a _ { t } }$ , where $\mathbb { 1 } _ { s _ { t + 1 } }$ is a probability distribution taking probability 1 at $s _ { t + 1 }$ and 0 at other states. This is an unbiased estimate of the transition kernel $p _ { s _ { t } } ^ { a _ { t } }$ conditioning on $S _ { t } = s _ { t }$ and $A _ { t } = a _ { t }$ . We then design a sample-based estimate $\hat { \mathcal { P } } _ { t } \triangleq \left\{ ( 1 - R ) \hat { p } _ { t } + R q | q \in \Delta _ { | \mathcal { S } | } \right\}$ of the uncertainty set $\mathcal { P } _ { s _ { t } } ^ { a _ { t } }$ . Using the sample-based uncertainty set $\hat { \mathcal { P } } _ { t }$ , we construct the following robust Q-learning algorithm in Algorithm 1. For any $t$ , $\sigma _ { \hat { \mathcal { P } } _ { t } } ( V _ { t } )$ can be easily computed: $\sigma _ { \hat { \Phi } _ { t } } ( V _ { t } ) =$ + +# Algorithm 1 Robust Q-Learning + +Initialization: $T$ , $Q _ { 0 } ( s , a )$ for all $( s , a ) \in \mathcal { S } \times \mathcal { A }$ , behavior policy $\pi _ { b }$ , $s _ { 0 }$ , step size $\alpha _ { t }$ + +1: for $t = 0 , 1 , 2 , . . . , T - 1$ do +2: Choose $a _ { t }$ according to $\pi _ { b } ( \cdot | s _ { t } )$ +3: Observe $s _ { t + 1 }$ and $c _ { t }$ +4: $\begin{array} { r } { V _ { t } ( s ) \operatorname* { m i n } _ { a \in \mathcal { A } } Q _ { t } ( s , a ) , \forall s \in \mathfrak { s } } \end{array}$ +5: Qt+1( $s _ { t } , a _ { t } ) \gets ( 1 - \alpha _ { t } ) Q _ { t } ( s _ { t } , a _ { t } ) + \alpha _ { t } ( c _ { t } + \gamma \sigma _ { \hat { \mathcal P } _ { t } } ( V _ { t } ) )$ +6: $Q _ { t + 1 } ( s , a ) \gets Q _ { t } ( s , a )$ for $( s , a ) \neq ( s _ { t } , a _ { t } )$ +7: end for + +Output: $Q _ { T }$ + +$\begin{array} { r } { R \operatorname* { m a x } _ { s \in \mathcal { S } } V _ { t } ( s ) + ( 1 - R ) V _ { t } ( s _ { t + 1 } ) } \end{array}$ . Hence the update in Algorithm 1 (line 5) can be written as + +$$ +Q _ { t + 1 } ( s _ { t } , a _ { t } ) \gets ( 1 - \alpha _ { t } ) Q _ { t } ( s _ { t } , a _ { t } ) + \alpha _ { t } ( c _ { t } + \gamma R \operatorname* { m a x } _ { s \in \mathcal { S } } V _ { t } ( s ) + \gamma ( 1 - R ) V _ { t } ( s _ { t + 1 } ) ) . +$$ + +Compared to the model-based approach, our approach is model-free. It does not require the prior knowledge of the uncertainty set, i.e., the knowledge of $p _ { s } ^ { a } , \forall s \in \mathcal { S } , a \in \mathcal { A }$ . Furthermore, the memory requirement of our algorithm is $| \mathcal { S } | \times | \mathcal { A } |$ (used to store the Q-table), and unlike the modelbased approach it does not need a table of size $| \dot { \mathcal { S } } | ^ { 2 } | \mathcal { A } |$ to store $p _ { s } ^ { a } , \forall s \in \mathcal { S } , a \in \mathcal { A }$ , which could be problematic if the state space is large. Moreover, our algorithm does not involve a relaxation of the uncertainty set like the one in [Roy et al., 2017], which will incur a stringent condition on the discount factor to guarantee the convergence. As will be shown below, the convergence of our Algorithm 1 does not require any condition on the discount factor. + +We show in the following theorem that the robust Q-learning algorithm converges asymptotically to the optimal robust action-value function $Q ^ { * }$ . + +Theorem 2. (Asymptotic Convergence) If step sizes $\alpha _ { t }$ satisfy that $\textstyle \sum _ { t = 0 } ^ { \infty } \alpha _ { t } = \infty$ and $\textstyle \sum _ { t = 0 } ^ { \infty } \alpha _ { t } ^ { 2 } < \infty$ then $Q _ { t } \to Q ^ { * }$ as with probability . + +To further establish the finite-time error bound for our robust Q-learning algorithm in Algorithm 1, we make the following assumption that is commonly used in the analysis of vanilla Q-learning. + +Assumption 1. The Markov chain induced by the behavior policy $\pi _ { b }$ and the transition kernel $p _ { s } ^ { a } , \forall s \in \mathcal { S } , a \in \mathcal { A }$ is uniformly ergodic. + +Let $\mu _ { \pi _ { b } }$ denote the stationary distribution over $\mathcal { S } \times \mathcal { A }$ induced by $\pi _ { b }$ and $p _ { s } ^ { a } , \forall s \in \mathcal { S } , a \in \mathcal { A }$ . We then further define $\begin{array} { r } { \mu _ { \mathrm { m i n } } = \operatorname* { m i n } _ { ( s , a ) \in \mathcal { S } \times \mathcal { A } } \mu _ { \pi _ { b } } ( s , a ) } \end{array}$ . This quantity characterizes how many samples are needed to visit every state-action pair sufficiently often. Define the following mixing time of the induced Markov chain: $\begin{array} { r } { t _ { \mathrm { m i x } } = \operatorname* { m i n } \left. t : \operatorname* { m a x } _ { s \in \mathcal { S } } \mathsf { \bar { d } _ { \mathrm { T V } } } ( \mu _ { \pi } , P ( s _ { t } = \cdot | s _ { 0 } = s ) ) \breve { \leq } \frac { 1 } { 4 } \right. } \end{array}$ , where $d _ { \mathrm { T V } }$ is the total variation distance. + +The following theorem establishes the finite-time error bound of our robust Q-learning algorithm. + +Theorem 3. (Finite-Time Error Bound) There exist some positive constants $c _ { 0 }$ and $c _ { 1 }$ such that for any $\delta < 1$ , any $\begin{array} { r } { \epsilon < \frac { 1 } { 1 - \gamma } } \end{array}$ , any $T$ satisfying + +$$ +T \geq c _ { 0 } \left( \frac { 1 } { \mu _ { m i n } ( 1 - \gamma ) ^ { 5 } \epsilon ^ { 2 } } + \frac { t _ { m i x } } { \mu _ { m i n } ( 1 - \gamma ) } \right) \log { \left( \frac { T | \mathcal { S } | | \mathcal { A } | } { \delta } \right) } \log { \left( \frac { 1 } { \epsilon ( 1 - \gamma ) ^ { 2 } } \right) } , +$$ + +and step size αt = $\begin{array} { r } { \alpha _ { t } = \frac { c _ { 1 } } { \log \left( \frac { T | \mathcal { S } | | \mathcal { A } | } { \delta } \right) } \operatorname* { m i n } \left( \frac { 1 } { t _ { m i x } } , \frac { \epsilon ^ { 2 } ( 1 - \gamma ) ^ { 4 } } { \gamma ^ { 2 } } \right) , \forall t \geq 0 } \end{array}$ we have with probability at least $1 - 6 \delta$ , $\| Q _ { T } - Q ^ { * } \| _ { \infty } \leq 3 \epsilon$ . + +From the theorem, we can see that to guarantee an $\epsilon$ -accurate estimate, a sample size $\begin{array} { r } { \tilde { \mathcal { O } } ( \frac { 1 } { \mu _ { \mathrm { m i n } } ( 1 - \gamma ) ^ { 5 } \epsilon ^ { 2 } } + } \end{array}$ tmixµmin(1−γ) ) (up to some logarithmic terms) is needed. This complexity matches with the one for the vanilla Q-learning in [Li et al., 2020] (within a constant factor), while our algorithm also guarantees robustness to MDP model uncertainty. Our algorithm design and analysis can be readily extended to robustify TD and SARSA. The variance-reduction technique [Wainwright, 2019] can also be combined with our robust Q-learning algorithm to further improve the dependency on $( 1 - \gamma )$ . + +# 5 Function Approximation: Robust TDC + +In this section, we investigate the case where the state and action spaces can be large or even continuous. A popular approach is to approximate the value function using a parameterized function, e.g., linear function and neural network. In this section, we focus on the case with linear function approximation to illustrate the main idea of designing robust RL algorithms. Our approach can be extended to non-linear (smooth) function approximation using techniques in, e.g., [Cai et al., 2019, Bhatnagar et al., 2009, Wai et al., 2019, Wang et al., 2021]. + +We focus on the problem of robust policy evaluation, i.e., estimate the robust value function $V ^ { \pi }$ defined in (1) for a given policy $\pi$ under the worst-case MDP transition kernel in the uncertainty set. Note that for robust RL with model uncertainty, any policy evaluation problem can be viewed as "off-transition-kernel", as it is to evaluate the value function under the worst-case MDP using samples from a different MDP. Since the TD algorithm with function approximation may diverge under off-policy training [Baird, 1995] and importance sampling cannot be applied here due to unknown transition kernel, in this paper we generalize the GTD method [Maei et al., 2010, Maei, 2011] to the robust setting. + +Let $\left\{ \phi ^ { ( i ) } : \mathbb { S } \to \mathbb { R } , i = 1 , \ldots , N \right\}$ be a set of $N$ fixed base functions, where $N \ll | \mathcal S | | \mathcal A |$ . In particular, we approximate the robust value function using a linear combination of $\begin{array} { r } { \sum _ { i = 1 } ^ { N } \theta ^ { i } \phi _ { s } ^ { ( i ) } = \phi _ { s } ^ { \top } \theta } \end{array}$ , where $\boldsymbol { \theta } \in \mathbb { R } ^ { N }$ is the weight vector. $\phi ^ { ( i ) }$ ’s: $V _ { \theta } ( s ) =$ + +Define the following robust Bellman operator for a given policy $\pi$ : + +$$ +\begin{array} { r l } & { \mathrm { \bf T } _ { \pi } V ( s ) \triangleq \mathbb E _ { A \sim \pi ( \cdot | s ) } [ c ( s , A ) + \gamma \sigma _ { \mathcal { P } _ { s } ^ { A } } ( V ) ] } \\ & { \quad \quad \quad = \mathbb E _ { A \sim \pi ( \cdot | s ) } \left[ c ( s , A ) + \gamma ( 1 - R ) \displaystyle \sum _ { s ^ { \prime } \in \mathcal { S } } p _ { s , s ^ { \prime } } ^ { A } V ( s ^ { \prime } ) + \gamma R \operatorname* { m a x } _ { s ^ { \prime } \in \mathcal { S } } V ( s ^ { \prime } ) \right] . } \end{array} +$$ + +We then define the mean squared projected robust Bellman error (MSPRBE) as + +$$ +\begin{array} { r } { \mathrm { M S P R B E } ( \theta ) = \| \mathbf { I I T } _ { \pi } V _ { \theta } - V _ { \theta } \| _ { \mu _ { \pi } } ^ { 2 } , } \end{array} +$$ + +where $\begin{array} { r } { \| v \| _ { \mu _ { \pi } } ^ { 2 } = \int v ^ { 2 } ( s ) \mu _ { \pi } ( d s ) } \end{array}$ , $\mu _ { \pi }$ is the stationary distribution induced by $\pi$ , and $\mathbf { I I }$ is a projection onto the linear function space w.r.t. $\| \cdot \| _ { \mu _ { \pi } }$ . We will develop a two time-scale gradient-based approach to minimize the MSPRBE. However, it can be seen that $\operatorname* { m a x } _ { s } V _ { \theta } ( s )$ in (7) is not smooth in $\theta$ , which is troublesome in both algorithm design and analysis. To solve this issue, we introduce the following smoothed robust Bellman operator $\hat { \mathbf { T } } _ { \pi }$ by smoothing the max with a LSE(LogSumExp): + +$$ +\hat { \mathbf { T } } _ { \pi } V ( s ) = \mathbb { E } _ { A \sim \pi ( \cdot | s ) } \left[ c ( s , A ) + \gamma ( 1 - R ) \sum _ { s ^ { \prime } \in \mathcal { S } } p _ { s , s ^ { \prime } } ^ { A } V ( s ^ { \prime } ) + \gamma R \cdot \mathrm { L S E } ( V ) \right] , +$$ + +where $\begin{array} { r } { \mathrm { L S E } ( V ) = \frac { \log \left( \sum _ { s } e ^ { \varrho V ( s ) } \right) } { \varrho } } \end{array}$ is the LogSumExp w.r.t. $V$ with a parameter $\varrho > 0$ . Note that when $\varrho \to \infty$ , the smoothed robust Bellman operator $\hat { \mathbf { T } } _ { \pi } \to \mathbf { T } _ { \pi }$ . The LSE operator can also be replaced by some other operator that approximates the max operator and is smooth, e.g., mellow-max [Asadi and Littman, 2017]. In the following, we first show that the fixed point of $\hat { \mathbf { T } } _ { \pi }$ exists for any $\varrho$ , and the fixed points converge to the one of $\mathbf { T } _ { \pi }$ for large $\varrho$ . + +Theorem 4. (1). For any $\varrho$ , $\hat { \mathbf { T } } _ { \pi }$ has a fixed point. + +(2). Let $V _ { 1 }$ and $V _ { 2 }$ be the fixed points of $\hat { \mathbf { T } } _ { \pi }$ and $\mathbf { T } _ { \pi }$ , respectively. Then + +$$ +\| V _ { 1 } - V _ { 2 } \| _ { \infty } \leq \frac { \gamma R } { 1 - \gamma } \frac { \log | \mathcal { S } | } { \varrho } 0 , a s \varrho \infty . +$$ + +We then denote by $J ( \theta )$ the smoothed MSPRBE with the LSE operator, and the goal is: + +$$ +\operatorname* { m i n } _ { \theta } J ( \theta ) = \operatorname* { m i n } _ { \theta } \Big \| \mathbf { H } \hat { \mathbf { T } } _ { \pi } V _ { \theta } - V _ { \theta } \Big \| _ { \mu _ { \pi } } ^ { 2 } . +$$ + +# 5.1 Algorithm Development + +In the following, we develop the robust TDC algorithm to solve the problem in (11). We will first derive the gradient of the smoothed MSPRBE, $J ( \theta )$ , and then design a two time-scale update rule using the weight doubling trick in [Sutton et al., 2009] to solve the double sampling problem. Define $\delta _ { s , a , s ^ { \prime } } ( \theta ) \ \triangleq \ c ( s , a ) + \gamma ( 1 - R ) V _ { \theta } ( s ^ { \prime } ) + \gamma R \mathrm { L S E } ( V _ { \theta } ) - V _ { \theta } ( s )$ , where $\operatorname { L S E } ( V _ { \theta } )$ is the LogSumExp function w.r.t. $V _ { \theta } = \theta ^ { \top } \phi$ . Denote by $C \triangleq \mathbb { E } _ { \mu _ { \pi } } \left[ \phi _ { S } ^ { \intercal } \phi _ { S } \right]$ . Then, $\mathbb { E } _ { \mu } [ \delta _ { S , A , S ^ { \prime } } ( \theta ) \phi _ { S } ] \ = \ \varPhi ^ { \top } D \left( \hat { \mathbf { T } } _ { \pi } V _ { \theta } - V _ { \theta } \right)$ , where $\textit { D } = \ \mathrm { d i a g } ( \mu _ { \pi } ( s _ { 1 } ) , \mu _ { \pi } ( s _ { 2 } ) , . . . , \mu _ { \pi } ( s _ { | \mathcal { S } | } ) )$ and $\boldsymbol { \varPhi } = ( \phi _ { s _ { 1 } } , \phi _ { s _ { 2 } } , . . . , \phi _ { s _ { | \mathcal { S } | } } ) ^ { \top } \in \dot { \mathbb { R } } ^ { | \mathcal { S } | \times N }$ . We know that $\begin{array} { r } { \pmb { \Pi } ^ { \top } D \pmb { \Pi } = D ^ { \top } \pmb { \varPhi } ( \pmb { \varPhi } ^ { \top } D \pmb { \varPhi } ) ^ { - 1 } \pmb { \varPhi } ^ { \top } D } \end{array}$ from [Maei, 2011]. Hence we have + +$$ +\begin{array} { r } { \boldsymbol J ( \theta ) = \Big \| \mathbf { \hat { I } } \mathbf { \hat { T } } _ { \boldsymbol \pi } V _ { \theta } - V _ { \theta } \Big \| _ { \mu _ { \boldsymbol \pi } } ^ { 2 } = \mathbb { E } _ { \mu _ { \boldsymbol \pi } } [ \delta _ { S , A , S ^ { \prime } } ( \theta ) \phi _ { S } ] ^ { \top } \boldsymbol C ^ { - 1 } \mathbb { E } _ { \mu _ { \boldsymbol \pi } } [ \delta _ { S , A , S ^ { \prime } } ( \theta ) \phi _ { S } ] . } \end{array} +$$ + +Then, its gradient can be written as: + +$$ +\begin{array} { l } { \displaystyle - \frac { 1 } { 2 } \nabla J ( \theta ) = - \mathbb { E } _ { \mu _ { \pi } } \big [ ( \nabla \delta _ { S , A , S ^ { \prime } } ( \theta ) ) \phi _ { S } \big ] ^ { \top } C ^ { - 1 } \mathbb { E } _ { \mu _ { \pi } } \big [ \delta _ { S , A , S ^ { \prime } } ( \theta ) \phi _ { S } \big ] } \\ { = \mathbb { E } _ { \mu _ { \pi } } \big [ \delta _ { S , A , S ^ { \prime } } ( \theta ) \phi _ { S } \big ] - \gamma \mathbb { E } _ { \mu _ { \pi } } \bigg [ \bigg ( ( 1 - R ) \phi _ { S ^ { \prime } } + R \cdot \nabla \mathrm { L S E } ( V _ { \theta } ) \bigg ) \phi _ { S } ^ { \top } \bigg ] \omega ( \theta ) , } \end{array} +$$ + +where $\omega ( \theta ) = C ^ { - 1 } \mathbb { E } _ { \mu _ { \pi } } [ \delta _ { S , A , S ^ { \prime } } ( \theta ) \phi _ { S } ]$ . It can be seen that to obtain an unbiased estimate of $\nabla J ( \theta )$ , two independent samples are needed as there exists a multiplication of two expectations, which is not applicable when there is only one sample trajectory. We then utilize the weight doubling trick in [Sutton et al., 2009], and design the robust TDC algorithm in Algorithm 2. Specifically, we introduce a fast time scale to estimate $\omega ( \theta )$ , and a slow time scale to estimate $\nabla J ( \theta )$ . Denote the projection by $\Pi _ { K } ( x ) \triangleq \arg \operatorname* { m i n } _ { \| y \| \leq K } \| y - x \|$ for any $\boldsymbol { x } \in \mathbb { R } ^ { N }$ . Our robust TDC algorithm in Algorithm 2 can be implemented in an online and incremental fashion. If the uncertainty set becomes a singleton, i.e., $R = 0$ , then Algorithm 2 reduces to the vanilla TDC algorithm. + +# 5.2 Finite-Time Error Bound of Robust TDC + +Unlike the vanilla TDC algorithm, $J ( \theta )$ here is non-convex. Therefore, we are interested in the convergence to stationary points, i.e., the rate of $\| \nabla J ( \theta ) \| 0$ . We first make some standard assumptions which are commonly used in RL algorithm analysis, e.g., [Wang and Zou, 2020, Kaledin et al., 2020, Xu et al., 2019, Srikant and Ying, 2019, Bhandari et al., 2018]. + +Assumption 2 (Bounded feature). $\| \phi _ { s } \| _ { 2 } \leq 1 , \forall s \in \mathcal { S }$ . + +Assumption 3 (Bounded cost function). $| c ( s , a ) | \leq c _ { \mathrm { m a x } } , \forall s \in \mathcal { S }$ and $a \in { \mathcal { A } }$ . + +Assumption 4 (Problem solvability). The matrix $C = \mathbb { E } _ { \mu _ { \pi } } [ \phi _ { S } \phi _ { S } ^ { \top } ]$ is non-singular with $\lambda > 0$ being its smallest eigenvalue. + +Input: $T , \alpha , \beta , \varrho , \phi _ { i }$ for $i = 1 , . . . , N$ , projection radius $K$ +Initialization: θ0,w0, s0 +1: Choose $W \sim \mathrm { U n i f o r m } ( 0 , 1 , . . . , T - 1 )$ +2: for $t = 0 , 1 , 2 , . . . , W - 1$ do +3: Take action according to $\pi ( \cdot | s _ { t } )$ and observe $s _ { t + 1 }$ and $c _ { t }$ +4: $\phi _ { t } \phi _ { s _ { t } }$ +7: 5: 6: $\begin{array} { r l } & { \ddot { \delta _ { t } } ( \theta _ { t } ) \gets c _ { t } + \gamma ( 1 - R ) V _ { \theta _ { t } } ( s _ { t + 1 } ) + \gamma R \frac { \log ( \sum _ { s } e ^ { \varrho \theta ^ { \top } \phi _ { s } } ) } { \varrho } - V _ { \theta _ { t } } ( s _ { t } ) } \\ & { \theta _ { t + 1 } \gets \Pi _ { K } \left( \theta _ { t } + \alpha \left( \delta _ { t } ( \theta _ { t } ) \phi _ { t } - \gamma \bigg ( ( 1 - R ) \phi _ { t + 1 } + R \sum _ { s \in \mathcal { S } } \left( \frac { e ^ { \sigma V _ { \theta } ( s ) } \phi _ { s } } { \sum _ { j \in \mathcal { S } } e ^ { \sigma V _ { \theta } ( j ) } } \right) \bigg ) \phi _ { t } ^ { \top } \omega _ { t } \right) \right) } \\ & { \omega _ { t + 1 } \gets \Pi _ { K } ( \omega _ { t } + \beta ( \delta _ { t } ( \theta _ { t } ) - \phi _ { t } ^ { \top } \omega _ { t } ) \phi _ { t } ) } \end{array}$ + +8: end for + +Output: $\theta _ { W }$ + +Assumption 5 (Geometric uniform ergodicity). There exist some constants $m > 0$ and $\rho \in ( 0 , 1 )$ such that for any $t > 0$ , $\begin{array} { r } { \operatorname* { m a x } _ { s \in \mathcal { S } } d _ { T V } ( \bar { \mathbb { P } } ( s _ { t } | s _ { 0 } = s ) , \mu _ { \pi } ) \leq m \rho ^ { t } } \end{array}$ . + +In the following theorem, we characterize the finite-time error bound for the convergence of our robust TDC algorithm. Here we only provide the order of the bounds in terms of $T$ . The explicit bounds can be found in (129) in Appendix D.3. + +Theorem 5. Consider the following step-sizes: $\begin{array} { r } { \beta = \mathcal { O } \left( \frac { 1 } { T ^ { b } } \right) } \end{array}$ , and $\begin{array} { r } { \alpha = \mathcal { O } \left( \frac { 1 } { T ^ { a } } \right) } \end{array}$ , where $\begin{array} { r } { \frac { 1 } { 2 } < a \leq 1 } \end{array}$ and $0 < b \leq a$ . Then we have that + +$$ +\mathbb { E } [ \| \nabla J ( \theta _ { W } ) \| ^ { 2 } ] = \odot \left( \frac { 1 } { T \alpha } + \alpha \log ( 1 / \alpha ) + \frac { 1 } { T \beta } + \beta \log ( 1 / \beta ) \right) , +$$ + +If we further let $a = b = 0 . 5$ , then $\begin{array} { r } { \mathbb { E } [ \| \nabla J ( \theta _ { W } ) \| ^ { 2 } ] = \odot \left( \frac { \log T } { \sqrt { T } } \right) . } \end{array}$ + +The robust TDC has a matching complexity with the vanilla TDC with non-linear function approximation [Wang et al., 2021], but provides the additional robustness to model uncertainty. It does not need to relax the uncertainty set like in [Roy et al., 2017], and our convergence results do not need a condition on the discount factor. + +# 6 Experiments + +# 6.1 Robust Q-Learning + +In this section, we compare our robust Q-learning with the vanilla non-robust Q-learning. We use OpenAI gym framework [Brockman et al., 2016], and consider two different problems: Frozen lake and Cart-Pole. One more example of the taxi problem is given in the appendix. To demonstrate the robustness, the policy is learned in a perturbed MDP, and is then tested on the true unperturbed MDP. Specifically, during the training, we set a probability $p$ such that after the agent takes an action, with probability $p$ , the state transition is uniformly over S, and with probability $1 - p$ the state transition is according to the true unperturbed transition kernel. The behavior policy for all the experiments below is set to be a uniform distribution over the action space given any state, i.e., $\begin{array} { r } { \pi _ { b } ( a | s ) ^ { \bullet } = \frac { 1 } { | \mathcal { A } | } } \end{array}$ for any $s \in \mathcal { S }$ and $a \in { \mathcal { A } }$ . We then evaluate the performance of the obtained policy in the unperturbed environment. At each time t, the policy we evaluate is the greedy-policy w.r.t. the current estimate of the Q-function, i.e., $\pi _ { t } ( s ) = \arg \operatorname* { m a x } _ { a } Q _ { t } ( s , a )$ . A Monte-Carlo method with horizon 100 is used to evaluate the accumulated discounted reward of the learned policy on the unperturbed MDP. We take the average over 30 trajectories. More details are provided in the appendix. + +In Figure 1 and Figure 2, we plot the accumulated discounted reward of both algorithms under different $p$ and $R$ for both problems. The upper and lower envelopes of the curves correspond to the 95 and 5 percentiles of the 30 trajectories, respectively. It can be seen that overall our robust Q-learning algorithm achieves a much higher reward than the vanilla Q-learning. This demonstrates the robustness of our robust Q-learning algorithm to model uncertainty. Moreover, as $p$ and $R$ getting larger, i.e., as the MDP that we learn the policy deviates from the MDP we test the policy, the advantage of our robust Q-learning algorithm is getting more significant compared to the vanilla Q-learning algorithm. + +![](images/f6e7e9e2735f3d2f3cb6a78235a2c13acfd85270e0532cc9f312127eab2de53d.jpg) +Figure 1: FrozenLake-v0: robust Q-learning v.s. non-robust Q-learning. + +![](images/a43d2928aff120beef997e93a1a33000db2a33693b2a692ee8c4a435bdabda59.jpg) +Figure 2: CartPole-v0: robust Q-learning v.s. non-robust Q-learning. + +# 6.2 Robust TDC with Linear Function Approximation + +In this section we compare our robust TDC with the vanilla non-robust TDC with linear function approximation on the $4 \times 4$ Frozen Lake problem. The problem setting is the same as the one in Section 6.1. More details about the experiment setup are provided in the appendix. + +We implement the two algorithms using samples from the perturbed MDP both for 30 times, and obtain 30 sequences of $\{ \theta _ { t } ^ { i } \} _ { t = 1 } ^ { \infty }$ , $i = 1 , 2 , . . . , 3 0$ . We then compute the squared gradient norm $\| \nabla J ( \theta ) \| ^ { 2 }$ on the true unperturbed MDP, and see whether $\{ \theta _ { t } ^ { i } \} _ { t = 1 } ^ { \infty }$ converges to some stationary points on the true unperturbed MDP. In Fig. 3, we plot the average squared gradient norm $\| \nabla J ( \theta ) \| ^ { 2 }$ for different $p$ and $R$ . The upper and lower envelops are the 95 and 5 percentiles of the 30 curves. It can be seen that our robust TDC converges much faster than vanilla TDC, and as the model mismatch between the training and test MDPs enlarges, the vanilla TDC may diverge (Fig. 3(c)), while our robust TDC still converges to some stationary point. Also, the robust TDC has a much smaller variance, which indicates a much stable behavior under model uncertainty. + +![](images/88a93ae259a06e7bc8fa9f4ad18dbea7ab033fd1a750f80210fcc36ef5a44b84.jpg) +Figure 3: FrozenLake-v0: $\| \nabla J ( \theta ) \| ^ { 2 }$ of robust TDC and non-robust TDC. + +# 6.3 Comparison with The Adversarial Training Approach + +We also compare our robust Q-learning with Robust Adversarial Reinforcement Learning (RARL) in [Pinto et al., 2017]. To apply their algorithm to our problem setting, we model the nature as an adversarial player, and its goal is to minimize the reward that the agent receives. The action space $\mathcal { A } _ { a d }$ of the nature is set to be the state space $\mathcal { A } _ { a d } \triangleq \mathcal { S }$ . Then the perturbed training environment can be viewed as an adversarial model: both the agent and the adversary take actions $a _ { a } , a _ { a d }$ , then the environment will transit to state $a _ { a d }$ with probability $R$ and transit following the unperturbed MDP $p _ { s } ^ { a _ { a } }$ with probability $1 - R$ . The goal of the maximize its accumulated reward, while the goal of the natural is to minimize it. + +Following the RARL algorithm [Pinto et al., 2017], in each iteration of the training, we first fix the adversarial policy and use Q-learning to optimize the agent’s policy and obtain the Q-table $Q _ { t }$ . Then we fix the agent’s policy and optimize the adversarial policy. + +After each training iteration, we test the performance of the greedy policies w.r.t. Q-tables obtained from robust Q-learning and RARL. The testing environment is set to be the worst-case, i.e., after the agent takes an action, the environment transits to the state which has the minimal value function $( \mathrm { a r g m i n } _ { s \in \mathcal { S } } V _ { t } ( s ) )$ with probability $p$ . We plot the accumulated discounted rewards of both algorithms against number of training iterations under different parameters. We set $\alpha = 0 . 2$ and $\gamma = 0 . 9$ . It can be seen from Fig. 4 that our robust Q-learning achieves a higher accumulative reward, and thus is more robust that the RARL algorithm in [Pinto et al., 2017]. Also our robust Q-learning is more stable during training, i.e., the variance is smaller. + +![](images/05ae19de592c7a368e9a827c529316238af525b6bd10c6a6b56e811a7ad5cc4c.jpg) +Figure 4: Taxi-v3: robust Q-learning v.s. RARL. + +# 7 Conclusion + +In this paper, we develop a novel approach for solving model-free robust RL problems with model uncertainty. Our algorithms can be implemented in an online and incremental fashion, do not require additional memory than their non-robust counterparts. We theoretically proved the convergence of our algorithms under no additional assumption on the discount factor, and further characterized their finite-time error bounds, which match with their non-robust counterparts (within a constant factor). Our approach can be readily extended to robustify TD, SARSA and other GTD algorithms. + +Limitations: It is also of future interest to investigate robustness to reward uncertainty, and other types of uncertainty sets, e.g., ones defined by KL divergence, Wasserstein distance and total variation. Negative societal impact: To the best of the authors’ knowledge, this study does not have any potential negative impact on the society. + +# 8 Acknowledgment + +The work of Y. Wang and S. Zou was supported by the National Science Foundation under Grants CCF-2106560 and CCF- 2007783. + +# References + +Mohammed Amin Abdullah, Hang Ren, Haitham Bou Ammar, Vladimir Milenkovic, Rui Luo, Mingtian Zhang, and Jun Wang. Wasserstein robust reinforcement learning. arXiv preprint arXiv:1907.13196, 2019. + +Alekh Agarwal, Sham M Kakade, Jason D Lee, and Gaurav Mahajan. On the theory of policy gradient methods: Optimality, approximation, and distribution shift. Journal of Machine Learning Research, 22(98):1–76, 2021. + +Kavosh Asadi and Michael L Littman. An alternative softmax operator for reinforcement learning. In Proc. International Conference on Machine Learning (ICML), volume 70, pages 243–252. JMLR, 2017. + +Christopher G Atkeson and Jun Morimoto. Nonparametric representation of policies and value functions: A trajectory-based approach. In Proc. Advances in Neural Information Processing Systems (NIPS), pages 1643–1650, 2003. + +Kishan Panaganti Badrinath and Dileep Kalathil. Robust reinforcement learning using least squares policy iteration with provable performance guarantees. In Proc. International Conference on Machine Learning (ICML), pages 511–520. PMLR, 2021. + +J Andrew Bagnell, Andrew Y Ng, and Jeff G Schneider. Solving uncertain markov decision. 09 2001. + +Leemon Baird. Residual algorithms: Reinforcement learning with function approximation. In Machine Learning Proceedings, pages 30–37. Elsevier, 1995. + +Carolyn L Beck and Rayadurgam Srikant. Error bounds for constant step-size Q-learning. Systems & control letters, 61(12):1203–1208, 2012. + +Jalaj Bhandari and Daniel Russo. Global optimality guarantees for policy gradient methods. arXiv preprint arXiv:1906.01786, 2019. + +Jalaj Bhandari, Daniel Russo, and Raghav Singal. A finite time analysis of temporal difference learning with linear function approximation. In Proc. Annual Conference on Learning Theory (CoLT), pages 1691–1692. PMLR, 2018. + +Shalabh Bhatnagar, Doina Precup, David Silver, Richard S Sutton, Hamid Maei, and Csaba Szepesvári. Convergent temporal-difference learning with arbitrary smooth function approximation. In Proc. Advances in Neural Information Processing Systems (NIPS), volume 22, pages 1204–1212, 2009. + +Vivek S Borkar and Sean P Meyn. The ODE method for convergence of stochastic approximation and reinforcement learning. SIAM Journal on Control and Optimization, 38(2):447–469, 2000. + +Greg Brockman, Vicki Cheung, Ludwig Pettersson, Jonas Schneider, John Schulman, Jie Tang, and Wojciech Zaremba. OpenAI Gym. arXiv preprint arXiv:1606.01540, 2016. + +Qi Cai, Zhuoran Yang, Jason D Lee, and Zhaoran Wang. Neural temporal-difference learning converges to global optima. In Proc. Advances in Neural Information Processing Systems (NeurIPS), pages 11312–11322, 2019. + +Shicong Cen, Chen Cheng, Yuxin Chen, Yuting Wei, and Yuejie Chi. Fast global convergence of natural policy gradient methods with entropy regularization. arXiv preprint arXiv:2007.06558, 2020. + +Zaiwei Chen, Sheng Zhang, Thinh T Doan, Siva Theja Maguluri, and John-Paul Clarke. Performance of Q-learning with linear function approuimation: Stability and finite-time analysis. arXiv preprint arXiv:1905.11425, 2019. + +Gal Dalal, Balázs Szörényi, Gugan Thoppe, and Shie Mannor. Finite sample analysis of two-timescale stochastic approximation with applications to reinforcement learning. Proceedings of Machine Learning Research, 75:1–35, 2018. + +Gal Dalal, Balázs Szörényi, and Gugan Thoppe. A tale of two-timescale reinforcement learning with the tightest finite-time bound. In Proc. Conference on Artificial Intelligence (AAAI), pages 3701–3708, 2020. + +Dalal, Gal and Szörényi, Balázs and Thoppe, Gugan and Mannor, Shie. Finite sample analyses for TD(0) with function approximation. In Proc. AAAI Conference on Artificial Intelligence (AAAI), pages 6144–6160, 2018. + +Thinh T Doan. Finite-time analysis and restarting scheme for linear two-time-scale stochastic approximation. SIAM Journal on Control and Optimization, 59(4):2798–2819, 2021. + +Eyal Even-Dar, Yishay Mansour, and Peter Bartlett. Learning rates for Q-learning. Journal of machine learning Research, 5(1), 2003. + +Harsh Gupta, R Srikant, and Lei Ying. Finite-time performance bounds and adaptive learning rate selection for two time-scale reinforcement learning. In Proc. Advances in Neural Information Processing Systems (NeurIPS), pages 4706–4715, 2019. + +Linfang Hou, Liang Pang, Xin Hong, Yanyan Lan, Zhiming Ma, and Dawei Yin. Robust reinforcement learning with Wasserstein constraint. arXiv preprint arXiv:2006.00945, 2020. + +Sandy Huang, Nicolas Papernot, Ian Goodfellow, Yan Duan, and Pieter Abbeel. Adversarial attacks on neural network policies. In Proc. International Conference on Learning Representations (ICLR), 2017. + +Peter J Huber. A robust version of the probability ratio test. The Annals of Mathematical Statistics, pages 1753–1758, 1965. + +Garud N Iyengar. Robust dynamic programming. Mathematics of Operations Research, 30(2): 257–280, 2005. + +Maxim Kaledin, Eric Moulines, Alexey Naumov, Vladislav Tadic, and Hoi-To Wai. Finite time analysis of linear two-timescale stochastic approximation with Markovian noise. In Proc. Annual Conference on Learning Theory (CoLT), pages 2144–2203. PMLR, 2020. + +Jernej Kos and Dawn Song. Delving into adversarial attacks on deep policies. In Proc. International Conference on Learning Representations (ICLR), 2017. + +Harshat Kumar, Alec Koppel, and Alejandro Ribeiro. On the sample complexity of actor-critic method for reinforcement learning with function approximation. arXiv preprint arXiv:1910.08412, 2019. + +Gen Li, Yuting Wei, Yuejie Chi, Yuantao Gu, and Yuxin Chen. Sample complexity of asynchronous Q-learning: Sharper analysis and variance reduction. In Proc. Advances in Neural Information Processing Systems (NeurIPS), 2020. + +Gen Li, Changxiao Cai, Yuxin Chen, Yuantao Gu, Yuting Wei, and Yuejie Chi. Is Q-learning minimax optimal? A tight sample complexity analysis. arXiv preprint arXiv:2102.06548, 2021. + +Shiau Hong Lim, Huan Xu, and Shie Mannor. Reinforcement learning in robust Markov decision processes. In Proc. Advances in Neural Information Processing Systems (NIPS), pages 701–709, 2013. + +Yen-Chen Lin, Zhang-Wei Hong, Yuan-Hong Liao, Meng-Li Shih, Ming-Yu Liu, and Min Sun. Tactics of adversarial attack on deep reinforcement learning agents. In Proc. International Joint Conferences on Artificial Intelligence (IJCAI), pages 3756–3762, 2017. + +Bo Liu, Ji Liu, Mohammad Ghavamzadeh, Sridhar Mahadevan, and Marek Petrik. Finite-sample analysis of proximal gradient TD algorithms. In Proc. International Conference on Uncertainty in Artificial Intelligence (UAI), pages 504–513. Citeseer, 2015. + +Shaocong Ma, Yi Zhou, and Shaofeng Zou. Variance-reduced off-policy TDC learning: Nonasymptotic convergence analysis. In Proc. Advances in Neural Information Processing Systems (NeurIPS), volume 33, pages 14796–14806, 2020. + +Shaocong Ma, Yi Zhou, and Shaofeng Zou. Greedy-GQ with variance reduction: Finite-time analysis and improved complexity. In Proc. International Conference on Learning Representations (ICLR), 2021. + +Hamid Reza Maei. Gradient temporal-difference learning algorithms. Thesis, University of Alberta, 2011. + +Hamid Reza Maei, Csaba Szepesvári, Shalabh Bhatnagar, and Richard S Sutton. Toward off-policy learning control with function approximation. In Proc. International Conference on Machine Learning (ICML), pages 719–726, 2010. + +Ajay Mandlekar, Yuke Zhu, Animesh Garg, Li Fei-Fei, and Silvio Savarese. Adversarially robust policy learning: Active construction of physically-plausible perturbations. In 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 3932–3939. IEEE, 2017. + +Jincheng Mei, Chenjun Xiao, Csaba Szepesvari, and Dale Schuurmans. On the global convergence rates of softmax policy gradient methods. In Proc. International Conference on Machine Learning (ICML), pages 6820–6829. PMLR, 2020. + +Jun Morimoto and Kenji Doya. Robust reinforcement learning. Neural computation, 17(2):335–359, 2005. + +Arnab Nilim and Laurent El Ghaoui. Robustness in Markov decision problems with uncertain transition matrices. In Proc. Advances in Neural Information Processing Systems (NIPS), pages 839–846, 2004. + +Anay Pattanaik, Zhenyi Tang, Shuijing Liu, Gautham Bommannan, and Girish Chowdhary. Robust deep reinforcement learning with adversarial attacks. In Proc. International Conference on Autonomous Agents and MultiAgent Systems, pages 2040–2042, 2018. + +Lerrel Pinto, James Davidson, Rahul Sukthankar, and Abhinav Gupta. Robust adversarial reinforcement learning. In Proc. International Conference on Machine Learning (ICML), pages 2817–2826. PMLR, 2017. + +Shuang Qiu, Zhuoran Yang, Jieping Ye, and Zhaoran Wang. On the finite-time convergence of actor-critic algorithm. In Proc. Optimization Foundations for Reinforcement Learning Workshop at Advances in Neural Information Processing Systems (NeurIPS), 2019. + +Guannan Qu and Adam Wierman. Finite-time analysis of asynchronous stochastic approximation and Q-learning. In Proc. Annual Conference on Learning Theory (CoLT), pages 3185–3205. PMLR, 2020. + +Aravind Rajeswaran, Sarvjeet Ghotra, Balaraman Ravindran, and Sergey Levine. Epopt: Learning robust neural network policies using model ensembles. In Proc. International Conference on Learning Representations (ICLR), 2017. + +Aurko Roy, Huan Xu, and Sebastian Pokutta. Reinforcement learning under model mismatch. In Proc. Advances in Neural Information Processing Systems (NIPS), pages 3046–3055, 2017. + +Jay K Satia and Roy E Lave Jr. Markovian decision processes with uncertain transition probabilities. Operations Research, 21(3):728–740, 1973. + +R. Srikant and Lei Ying. Finite-time error bounds for linear stochastic approximation and TD learning. In Proc. Annual Conference on Learning Theory (CoLT), pages 2803–2830, 2019. + +Jun Sun, Gang Wang, Georgios B Giannakis, Qinmin Yang, and Zaiyue Yang. Finite-sample analysis of decentralized temporal-difference learning with linear function approximation. In Proc. International Conference on Artifical Intelligence and Statistics (AISTATS), 2020. + +Richard S. Sutton and Andrew G. Barto. Reinforcement Learning: An Introduction, Second Edition. The MIT Press, Cambridge, Massachusetts, 2018. + +Richard S Sutton, Csaba Szepesvári, and Hamid Reza Maei. A convergent O(n) algorithm for off-policy temporal-difference learning with linear function approximation. In Proc. Advances in Neural Information Processing Systems (NIPS), volume 21, pages 1609–1616. MIT Press, 2008. + +Richard S Sutton, Hamid Reza Maei, Doina Precup, Shalabh Bhatnagar, David Silver, Csaba Szepesvári, and Eric Wiewiora. Fast gradient-descent methods for temporal-difference learning with linear function approximation. In Proc. International Conference on Machine Learning (ICML), pages 993–1000, 2009. + +Aviv Tamar, Shie Mannor, and Huan Xu. Scaling up robust MDPs using function approximation. In Proc. International Conference on Machine Learning (ICML), pages 181–189. PMLR, 2014. + +Eugene Vinitsky, Yuqing Du, Kanaad Parvate, Kathy Jang, Pieter Abbeel, and Alexandre Bayen. Robust reinforcement learning using adversarial populations. arXiv preprint arXiv:2008.01825, 2020. + +Hoi-To Wai, Mingyi Hong, Zhuoran Yang, Zhaoran Wang, and Kexin Tang. Variance reduced policy evaluation with smooth function approximation. In Proc. Advances in Neural Information Processing Systems (NeurIPS), volume 32, pages 5784–5795, 2019. + +Martin J Wainwright. Variance-reduced Q-learning is minimax optimal. arXiv preprint arXiv:1906.04697, 2019. + +Lingxiao Wang, Qi Cai, Zhuoran Yang, and Zhaoran Wang. Neural policy gradient methods: Global optimality and rates of convergence. In Proc. International Conference on Learning Representations (ICLR), 2020. + +Yue Wang and Shaofeng Zou. Finite-sample analysis of Greedy-GQ with linear function approximation under Markovian noise. In Proc. International Conference on Uncertainty in Artificial Intelligence (UAI), pages 11–20. PMLR, 2020. + +Yue Wang, Shaofeng Zou, and Yi Zhou. Non-asymptotic analysis for two time-scale TDC with general smooth function approximation. In Proc. Advances in Neural Information Processing Systems (NeurIPS), volume 34, 2021. + +Wolfram Wiesemann, Daniel Kuhn, and Berç Rustem. Robust Markov decision processes. Mathematics of Operations Research, 38(1):153–183, 2013. + +Yue Wu, Weitong Zhang, Pan Xu, and Quanquan Gu. A finite time analysis of two time-scale actor critic methods. arXiv preprint arXiv:2005.01350, 2020. + +Tengyu Xu and Yingbin Liang. Sample complexity bounds for two timescale value-based reinforcement learning algorithms. In Proc. International Conference on Artifical Intelligence and Statistics (AISTATS), pages 811–819. PMLR, 2021. + +Tengyu Xu, Shaofeng Zou, and Yingbin Liang. Two time-scale off-policy TD learning: Nonasymptotic analysis over Markovian samples. In Proc. Advances in Neural Information Processing Systems (NeurIPS), pages 10633–10643, 2019. + +Zhuoran Yang, Yongxin Chen, Mingyi Hong, and Zhaoran Wang. Provably global convergence of actor-critic: A case for linear quadratic regulator with ergodic cost. In Proc. Advances in Neural Information Processing Systems (NeurIPS), pages 8353–8365, 2019. + +Kaiqing Zhang, Bin Hu, and Tamer Basar. On the stability and convergence of robust adversarial reinforcement learning: A case study on linear quadratic systems. In Proc. Advances in Neural Information Processing Systems (NeurIPS), volume 33, 2020a. + +Kaiqing Zhang, Tao Sun, Yunzhe Tao, Sahika Genc, Sunil Mallya, and Tamer Basar. Robust multiagent reinforcement learning with model uncertainty. In Proc. Advances in Neural Information Processing Systems (NeurIPS), volume 33, 2020b. + +Shaofeng Zou, Tengyu Xu, and Yingbin Liang. Finite-sample analysis for SARSA with linear function approximation. In Proc. Advances in Neural Information Processing Systems (NeurIPS), pages 8665–8675, 2019. \ No newline at end of file diff --git a/parse/train/IhiU6AJYpDs/IhiU6AJYpDs_content_list.json b/parse/train/IhiU6AJYpDs/IhiU6AJYpDs_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..bff75e9eaf3923d6cf771225c7200b8e9d5764e1 --- /dev/null +++ b/parse/train/IhiU6AJYpDs/IhiU6AJYpDs_content_list.json @@ -0,0 +1,2132 @@ +[ + { + "type": "text", + "text": "Online Robust Reinforcement Learning with Model Uncertainty ", + "text_level": 1, + "bbox": [ + 187, + 122, + 812, + 172 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Yue Wang \nUniversity at Buffalo \nBuffalo, NY 14228 \nywang294@buffalo.edu ", + "bbox": [ + 281, + 226, + 455, + 281 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Shaofeng Zou University at Buffalo Buffalo, NY 14228 szou3@buffalo.edu ", + "bbox": [ + 570, + 227, + 715, + 281 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Abstract ", + "text_level": 1, + "bbox": [ + 462, + 318, + 535, + 334 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Robust reinforcement learning (RL) is to find a policy that optimizes the worstcase performance over an uncertainty set of MDPs. In this paper, we focus on model-free robust RL, where the uncertainty set is defined to be centering at a misspecified MDP that generates a single sample trajectory sequentially, and is assumed to be unknown. We develop a sample-based approach to estimate the unknown uncertainty set, and design robust Q-learning algorithm (tabular case) and robust TDC algorithm (function approximation setting), which can be implemented in an online and incremental fashion. For the robust Q-learning algorithm, we prove that it converges to the optimal robust Q function, and for the robust TDC algorithm, we prove that it converges asymptotically to some stationary points. Unlike the results in [Roy et al., 2017], our algorithms do not need any additional conditions on the discount factor to guarantee the convergence. We further characterize the finite-time error bounds of the two algorithms, and show that both the robust Qlearning and robust TDC algorithms converge as fast as their vanilla counterparts (within a constant factor). Our numerical experiments further demonstrate the robustness of our algorithms. Our approach can be readily extended to robustify many other algorithms, e.g., TD, SARSA, and other GTD algorithms. ", + "bbox": [ + 233, + 349, + 766, + 584 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 Introduction ", + "text_level": 1, + "bbox": [ + 174, + 612, + 310, + 628 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Existing studies on Markov decision process (MDP) and reinforcement learning (RL) [Sutton and Barto, 2018] mostly rely on the crucial assumption that the environment on which a learned policy will be deployed is the same one that was used to generate the policy, which is often violated in practice – e.g., the simulator may be different from the true environment, and the MDP may evolve over time. Due to such model deviation, the actual performance of the learned policy can significantly degrade. To address this problem, the framework of robust MDP was formulated in [Bagnell et al., 2001, Nilim and El Ghaoui, 2004, Iyengar, 2005], where the transition kernel of the MDP is not fixed and lies in an uncertainty set, and the goal is to learn a policy that performs well under the worst-case MDP in the uncertainty set. In [Bagnell et al., 2001, Nilim and El Ghaoui, 2004, Iyengar, 2005], it was assumed that the uncertainty set is known beforehand, i.e., model-based approach, and dynamic programming can be used to find the optimal robust policy. ", + "bbox": [ + 174, + 643, + 825, + 796 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "The model-based approach, however, requires a model of the uncertainty set known beforehand, and needs a large memory to store the model when the state and action spaces are large, which make it less applicable for many practical scenarios. This motivates the study in this paper, model-free robust RL with model uncertainty, which is to learn a robust policy using a single sample trajectory from a misspecified MDP, e.g., a simulator and a similar environment in which samples are easier to collect than in the target environment where the policy is going to be deployed. The major challenge lies in that the transition kernel of the misspecified MDP is not given beforehand, and thus, the uncertainty set and the optimal robust policy need to be learned simultaneously using sequentially observed data from the misspecified MDP. Moreover, robust RL learns the value function of the worst-case MDP in the uncertainty set which is different from the misspecified MDP that generates samples. This is similar to the off-policy learning, which we refer to as the \"off-transition-kernel\" setting. Therefore, the learning may be unstable and could diverge especially when function approximation is used [Baird, 1995]. ", + "bbox": [ + 174, + 803, + 825, + 900 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 92, + 825, + 174 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "In this paper, we develop a model-free approach for robust RL with model uncertainty. Our major contributions in this paper are summarized as follows. ", + "bbox": [ + 173, + 181, + 823, + 208 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "• Motivated by empirical studies of adversarial training in RL [Huang et al., 2017, Kos and Song, 2017, Lin et al., 2017, Pattanaik et al., 2018, Mandlekar et al., 2017] and the $R$ -contamination model in robust detection (called $\\epsilon$ -contamination model in [Huber, 1965]), we design the uncertainty set using the $R$ -contamination model (see (4) for the details). We then develop an approach to estimate the unknown uncertainty set using only the current sample, which does not incur any additional memory cost. Unlike the approach in [Roy et al., 2017], where the uncertainty set is relaxed to one not depending on the misspecified MDP that generates samples so that an online algorithm can be constructed, our approach does not need to relax the uncertainty set. ", + "bbox": [ + 173, + 220, + 825, + 332 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "• We develop a robust Q-learning algorithm for the tabular case, which can be implemented in an online and incremental fashion, and has the same memory cost as the vanilla Q-learning algorithm. We show that our robust Q-learning algorithm converges asymptotically, and further characterize its finite-time error bound. Unlike the results in [Roy et al., 2017] where a stringent condition on the discount factor (which is due to the relaxation of the uncertainty set, and prevents the use of a discount factor close to 1 in practice) is needed to guarantee the convergence, our algorithm converges without the need of such condition. Furthermore, our robust Q-learning algorithm converges as fast as the vanilla Q-learning algorithm [Li et al., 2020] (within a constant factor), while being robust to model uncertainty. ", + "bbox": [ + 174, + 337, + 825, + 462 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "• We generalize our approach to the case with function approximation (for large state/action space). We investigate the robust policy evaluation problem, i.e., evaluate a given policy under the worstcase MDP in the uncertainty set. As mentioned before, the robust RL problem is essentially \"off-transition-kernel\", and therefore non-robust methods with function approximation may diverge [Baird, 1995] (also see our experiments). We develop a novel extension of the gradient TD (GTD) method [Maei et al., 2010, Maei, 2011, Sutton et al., 2008] to robust RL. Our approach introduces a novel smoothed robust Bellman operator to construct the smoothed mean-squared projected robust Bellman error (MSPRBE). Using our uncertainty set design and online sample-based estimation, we develop a two time-scale robust TDC algorithm. We further characterize its convergence and finite-time error bound. ", + "bbox": [ + 173, + 468, + 825, + 606 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "• We conduct numerical experiments to validate the robustness of our approach. In our experiments, our robust Q-learning algorithm achieves a much higher reward than the vanilla Q-learning algorithm when being trained on a misspecified MDP; and our robust TDC algorithm converges much faster than the vanilla TDC algorithm, and the vanilla TDC algorithm may even diverge. ", + "bbox": [ + 174, + 611, + 826, + 666 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "1.1 Related Work ", + "text_level": 1, + "bbox": [ + 174, + 684, + 310, + 699 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Model-Based Robust MDP. The framework of robust MDP was investigated in [Iyengar, 2005, Nilim and El Ghaoui, 2004, Bagnell et al., 2001, Satia and Lave Jr, 1973, Wiesemann et al., 2013], where the transition kernel is assumed to be in some uncertainty set, and the problem can be solved by dynamic programming. This approach was further extended to the case with function approximation in [Tamar et al., 2014]. However, these studies are model-based, which assume beforehand knowledge of the uncertainty set. In this paper, we investigate the model-free setting, where the uncertainty set is a set of MDPs centered around some unknown Markov transition kernel from which a single sample trajectory can be sequentially observed. ", + "bbox": [ + 173, + 710, + 825, + 821 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Adversarial Robust RL. It was shown in [Iyengar, 2005] that the robust MDP problem is equivalent to a zero-sum game between the agent and the nature. Motivated by this fact, the adversarial training approach, where an adversary perturbs the state transition, was studied in [Vinitsky et al., 2020, Pinto et al., 2017, Abdullah et al., 2019, Hou et al., 2020, Rajeswaran et al., 2017, Atkeson and Morimoto, 2003, Morimoto and Doya, 2005]. This method relies on a simulator, where the state transition can be modified in an arbitrary way. Another approach is to modify the current state through adversarial samples, which is more heuristic, e.g., [Huang et al., 2017, Kos and Song, 2017, Lin et al., 2017, Pattanaik et al., 2018, Mandlekar et al., 2017]. Despite the empirical success of these approaches, theoretical performance guarantees, e.g., convergence to the optimal robust policy and convergence rate, are yet to be established. The main difference lies in that during the training, our approach does not need to manipulate the state transition of the MDP. More importantly, we develop the asymptotic convergence to the optimal robust policy and further characterize the finite-time error bound. In [Lim et al., 2013], the scenario where some unknown parts of the state space can have arbitrary transitions while other parts are purely stochastic was studied. Adaptive algorithm to adversarial behavior was designed, and its regret bound is shown to be similar to the purely stochastic case. In [Zhang et al., 2020a], the robust adversarial RL problem for the special linear quadratic case was investigated. ", + "bbox": [ + 174, + 828, + 825, + 911 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 90, + 825, + 229 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Model-free Robust RL. In [Roy et al., 2017, Badrinath and Kalathil, 2021] model-free RL with model uncertainty was studied, where in order to construct an algorithm that can be implemented in an online and incremental fashion, the uncertainty set was firstly relaxed by dropping the dependency on the misspecified MDP that generates the samples (centroid of the uncertainty set). Such a relaxation is pessimistic since the relaxed uncertainty set is not centered at the misspecified MDP anymore (which is usually similar to the target MDP), making the robustness to the relaxed uncertainty set not well-justified. Such a relaxation will further incur a stringent condition on the discounted factor to guarantee the convergence, which prevents the use of a discount factor close to 1 in practice. Moreover, only asymptotic convergence was established in [Roy et al., 2017]. In this paper, we do not relax the uncertainty set, and instead propose an online approach to estimate it. Our algorithms converge without the need of the condition on the discount factor. We also provides finite-time error bounds for our algorithms. The multi-agent RL robust to reward uncertainty was investigated in [Zhang et al., 2020b], where the reward uncertainty set is known, but the transition kernel is fixed. ", + "bbox": [ + 173, + 236, + 825, + 415 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Finite-time Error Bound for RL Algorithms. For the tubular case, Q-learning has been studied intensively, e.g., in [Even-Dar et al., 2003, Beck and Srikant, 2012, Qu and Wierman, 2020, Li et al., 2020, Wainwright, 2019, Li et al., 2021]. TD with function approximation were studied in [Dalal, Gal and Szörényi, Balázs and Thoppe, Gugan and Mannor, Shie, 2018, Bhandari et al., 2018, Srikant and Ying, 2019, Cai et al., 2019, Sun et al., 2020]. Q-learning and SARSA with linear function approximation were investigated in [Zou et al., 2019, Chen et al., 2019]. The finite-time error bounds for the gradient TD algorithms [Maei et al., 2010, Sutton et al., 2009, Maei et al., 2010] were further developed recently in [Dalal et al., 2018, Liu et al., 2015, Gupta et al., 2019, Xu et al., 2019, Dalal et al., 2020, Kaledin et al., 2020, Ma et al., 2020, Wang and Zou, 2020, Ma et al., 2021, Doan, 2021]. There are also finite-time error bounds on the policy gradient methods and actor critic methods, e.g., [Wang et al., 2020, Yang et al., 2019, Kumar et al., 2019, Qiu et al., 2019, Wu et al., 2020, Cen et al., 2020, Bhandari and Russo, 2019, Agarwal et al., 2021, Mei et al., 2020]. We note that these studies are for the non-robust RL algorithms, and in this paper, we design robust RL algorithms, and characterize their finite-time error bounds. ", + "bbox": [ + 173, + 421, + 826, + 614 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "2 Preliminaries ", + "text_level": 1, + "bbox": [ + 174, + 633, + 318, + 651 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Markov Decision Process. An MDP can be characterized by a tuple $( \\mathcal { S } , \\mathcal { A } , \\mathsf { P } , c , \\gamma )$ , where S and $\\mathcal { A }$ are the state and action spaces, $\\mathsf { P } = \\{ p _ { s } ^ { a } \\in \\varDelta _ { | \\mathcal { S } | } , a \\in \\mathcal { A } , s \\in \\mathcal { S } \\}$ is the transition kernel1, $c$ is the cost function, and $\\gamma \\in [ 0 , 1 )$ is the discount factor. Specifically, $p _ { s } ^ { a }$ denotes the distribution of the next state if taking action $a$ at state $s$ . Let $p _ { s } ^ { a } = \\{ p _ { s , s ^ { \\prime } } ^ { a } \\} _ { s ^ { \\prime } \\in \\mathcal S }$ , where $p _ { s , s ^ { \\prime } } ^ { a }$ denotes the probability that the environment transits to state $s ^ { \\prime }$ if taking action $a$ at state $s$ . The cost of taking action $a$ at state $s$ is given by $c ( s , a )$ . A stationary policy $\\pi$ is a mapping from S to a distribution over $\\mathcal A$ . At each time $t$ , an agent takes an action $A _ { t } \\in \\mathcal A$ at state $S _ { t } \\in \\mathcal S$ . The environment then transits to the next state $S _ { t + 1 }$ with probability $p _ { S _ { t } , S _ { t + 1 } } ^ { A _ { t } }$ , and the agent receives cost given by $c ( S _ { t } , A _ { t } )$ . The value function of a policy by follo $\\pi$ staing $s \\in \\mathcal { S }$ fined as the expected accumulatand the goal is to find the policy discounted costthat minimizes $\\pi \\colon \\bar { \\mathbb { E } } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\dot { \\gamma } ^ { t } c ( S _ { t } , A _ { t } ) | S _ { 0 } = s , \\pi \\right]$ $\\pi$ \n$s \\in \\mathcal { S }$ ", + "bbox": [ + 174, + 665, + 825, + 827 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Robust Markov Decision Process. In the robust case, the transition kernel is not fixed and lies in some uncertainty set. Denote the transition kernel at time $t$ by $\\mathsf { P } _ { t }$ , and let $\\kappa = ( \\mathsf { P } _ { 0 } , \\mathsf { P } _ { 1 } , \\ldots )$ , where $\\mathsf { P } _ { t } \\in \\mathbf { P } , \\forall t \\ge 0$ , and $\\mathbf { P }$ is the uncertainty set of the transition kernel. The sequence $\\kappa$ can be viewed as the policy of the nature, and is adversarially chosen by the nature [Bagnell et al., 2001, Nilim and ", + "bbox": [ + 174, + 832, + 825, + 888 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "El Ghaoui, 2004, Iyengar, 2005]. Define the robust value function of a policy $\\pi$ as the worst-case expected accumulated discounted cost following a fixed policy $\\pi$ over all transition kernels in the uncertainty set: ", + "bbox": [ + 174, + 90, + 825, + 133 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/5c8479616301ab2d49dcead2362db851758cb5ec733965b24b83759529ec5c2b.jpg", + "text": "$$\nV ^ { \\pi } ( s ) = \\operatorname* { m a x } _ { \\kappa } \\mathbb { E } _ { \\kappa } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } c ( S _ { t } , A _ { t } ) | S _ { 0 } = s , \\pi \\right] ,\n$$", + "text_format": "latex", + "bbox": [ + 338, + 140, + 656, + 183 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "where $\\mathbb { E } _ { \\kappa }$ denotes the expectation when the state transits according to $\\kappa$ . Similarly, define the robust action-value function for a policy $\\begin{array} { r } { \\pi \\colon Q ^ { \\pi } ( s , a ) = \\operatorname* { m a x } _ { \\kappa } \\mathbb { E } _ { \\kappa } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } c ( S _ { t } , A _ { t } ) | \\dot { S _ { 0 } } = s , A _ { 0 } = a , \\pi \\right] } \\end{array}$ . The goal of robust RL is to find the optimal robust policy $\\pi ^ { * }$ that minimizes the worst-case accumulated discounted cost: ", + "bbox": [ + 173, + 189, + 826, + 244 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/323b5dbc7086c2a15a4b350fd9583f048099d6491412754c182179252321aee0.jpg", + "text": "$$\n\\pi ^ { * } = \\arg \\operatorname* { m i n } _ { \\pi } V ^ { \\pi } ( s ) , \\forall s \\in \\mathcal { S } .\n$$", + "text_format": "latex", + "bbox": [ + 400, + 252, + 598, + 275 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "We also denote $V ^ { \\pi ^ { * } }$ and $Q ^ { \\pi ^ { * } }$ by $V ^ { * }$ and $Q ^ { * }$ , respectively, and $\\begin{array} { r } { V ^ { * } ( s ) = \\operatorname* { m i n } _ { a \\in \\mathcal { A } } Q ^ { * } ( s , a ) } \\end{array}$ . ", + "bbox": [ + 173, + 282, + 766, + 299 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Note that a transition kernel is a collection of conditional distributions. Therefore, the uncertainty set $\\mathbf { P }$ of the transition kernel can be equivalently written as a collection of $\\mathcal { P } _ { s } ^ { a }$ for all $s \\in \\mathcal { S } , a \\in \\mathcal { A }$ , where $\\mathcal { P } _ { s } ^ { a }$ is a set of conditional distributions $p _ { s } ^ { a }$ over the state space S. Denote by $\\sigma _ { \\mathcal { P } } ( v ) \\triangleq \\operatorname* { m a x } _ { p \\in \\mathcal { P } } ( p ^ { \\top } v )$ the support function of vector $v$ over a set of probability distributions $\\mathcal { P }$ . For robust MDP, the following robust analogue of the Bellman recursion was provided in [Nilim and El Ghaoui, 2004, Iyengar, 2005]. ", + "bbox": [ + 174, + 304, + 826, + 391 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Theorem 1. [Nilim and El Ghaoui, 2004] The following perfect duality condition holds for all $s \\in \\mathcal { S }$ : ", + "bbox": [ + 173, + 393, + 825, + 410 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/e93045239e0b7e332eb486ed1d6c8b17f4379e21d3033a60baeafcd33b64fda6.jpg", + "text": "$$\n\\operatorname* { m i n } _ { \\pi } \\operatorname* { m a x } _ { \\kappa } \\mathbb { E } _ { \\kappa } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } c ( S _ { t } , A _ { t } ) \\big | \\pi , S _ { 0 } = s \\right] = \\operatorname* { m a x } _ { \\kappa } \\operatorname* { m i n } _ { \\pi } \\mathbb { E } _ { \\kappa } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } c ( S _ { t } , A _ { t } ) \\big | \\pi , S _ { 0 } = s \\right] .\n$$", + "text_format": "latex", + "bbox": [ + 196, + 415, + 781, + 458 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "The optimal robust value function $V ^ { * }$ satisfies $\\begin{array} { r } { V ^ { * } ( s ) = \\operatorname* { m i n } _ { a \\in \\mathcal { A } } \\bigl ( c ( s , a ) + \\gamma \\sigma _ { \\mathscr { P } _ { s } ^ { a } } ( V ^ { * } ) \\bigr ) } \\end{array}$ , and the optimal robust action-value function $Q ^ { * }$ satisfies $Q ^ { * } ( s , a ) = c ( s , a ) + \\gamma \\sigma _ { \\mathscr P _ { s } ^ { a } } ( V ^ { * } )$ . ", + "bbox": [ + 173, + 464, + 825, + 497 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Define the robust Bellman operator $\\mathbf { T }$ by $\\begin{array} { r } { \\mathbf { T } Q ( s , a ) = c ( s , a ) + \\gamma \\sigma _ { \\mathcal { P } _ { s } ^ { a } } ( \\operatorname* { m i n } _ { a \\in \\mathcal { A } } Q ( s , a ) ) } \\end{array}$ . It was shown in [Nilim and El Ghaoui, 2004, Iyengar, 2005] that $\\mathbf { T }$ is a contraction and its fixed point is the optimal robust $Q ^ { * }$ . When the uncertainty set is known, so that $\\sigma _ { \\mathcal P _ { s } ^ { a } }$ can be computed exactly, $V ^ { * }$ and $Q ^ { * }$ can be solved by dynamic programming [Iyengar, 2005, Nilim and El Ghaoui, 2004]. ", + "bbox": [ + 174, + 507, + 825, + 565 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3 R-Contamination Model For Uncertainty Set Construction ", + "text_level": 1, + "bbox": [ + 174, + 583, + 696, + 602 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "In this section, we construct the uncertainty set using the $R$ -contamination model. ", + "bbox": [ + 173, + 614, + 709, + 630 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Let $\\mathsf { P } = \\{ p _ { s } ^ { a } , s \\in \\mathscr { S } , a \\in \\mathscr { A } \\}$ be the centroid of the uncertainty set, i.e., the transition kernel that generates the sample trajectory, and $\\mathsf { P }$ is unknown. For example, P can be the simulator at hand, which may not be exactly accurate; and $\\mathsf { P }$ can be the transition kernel of environment 1, from which we can take samples to learn a policy that will be deployed in a similar environment 2. The goal is to learn a policy using samples from $\\mathsf { P }$ that performs well when applied to a perturbed MDP from P. ", + "bbox": [ + 173, + 635, + 826, + 707 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Motivated by empirical studies of adversarial training in RL [Huang et al., 2017, Kos and Song, 2017, Lin et al., 2017, Pattanaik et al., 2018, Mandlekar et al., 2017] and the $R$ -contamination model in robust detection [Huber, 1965], we use the $R$ -contamination model to define the uncertainty set: ", + "bbox": [ + 173, + 712, + 826, + 753 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/65c42db17c4a64629886dd7a5cba73fa31dcb0cb47e4877794a019ea1569cb7c.jpg", + "text": "$$\n\\begin{array} { r } { \\mathfrak { P } _ { s } ^ { a } = \\big \\{ ( 1 - R ) p _ { s } ^ { a } + R q | q \\in \\mathcal { A } _ { | \\mathcal { S } | } \\big \\} , s \\in \\mathcal { S } , a \\in \\mathcal { A } , \\mathrm { f o r ~ s o m e ~ } 0 \\le R \\le 1 . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 259, + 760, + 738, + 779 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Here, $p _ { s } ^ { a }$ is the centroid of the uncertainty set $\\mathcal { P } _ { s } ^ { a }$ at $( s , a )$ , which is unknown, and $R$ is the design parameter of the uncertainty set, which measures the size of the uncertainty set, and is assumed to be known in the algorithm. We then let $\\mathbf { P } = \\bigotimes _ { s \\in \\mathcal { S } , a \\in \\mathcal { A } } \\mathcal { P } _ { s } ^ { a }$ . ", + "bbox": [ + 173, + 786, + 825, + 830 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Remark 1. $R$ -contamination model is closely related to other uncertainty set models like total variation and $K L$ -divergence. It can be shown that $R$ -contamination set certered at $p$ is a subset of total variation ball : $\\{ ( 1 - R ) p + R q | q \\in \\varDelta _ { | \\mathcal { S } | } \\} \\subset \\{ q \\in \\varDelta _ { | \\mathcal { S } | } | d _ { T V } ( p , q ) \\leq R \\}$ . Hence the total variation uncertainty set is less conservative than our $R$ -contamination uncertainty set. KL-divergence moreover can be related to total variation using Pinsker’s inequality, i.e., $\\begin{array} { r } { d _ { T V } ( p , q ) \\leq \\sqrt { \\frac 1 2 d _ { K L } ( p , q ) } } \\end{array}$ . ", + "bbox": [ + 173, + 834, + 826, + 915 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "4 Tabular Case: Robust Q-Learning ", + "text_level": 1, + "bbox": [ + 174, + 88, + 493, + 107 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "In this section, we focus on the tabular case with finite state and action spaces. We focus on the asynchronous setting where a single sample trajectory is available with Markovian noise. We will develop an efficient approach to estimate the unknown uncertainty set $\\mathbf { P }$ , and further the support function $\\sigma _ { \\mathcal { P } _ { s } ^ { a } } ( \\cdot )$ , and then design our robust Q-learning algorithm. ", + "bbox": [ + 173, + 121, + 825, + 178 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "We propose an efficient and data-driven approach to estimate the unknown $p _ { s } ^ { a }$ and thus the unknown uncertainty set $\\mathcal { P } _ { s } ^ { a }$ for any $s \\in \\mathcal { S }$ and $a \\in { \\mathcal { A } }$ . Specifically, denote the sample at $t$ -th time step by $O _ { t } = ( s _ { t } , a _ { t } , s _ { t + 1 } )$ . We then use $O _ { t }$ to obtain the maximum likelihood estimate (MLE) $\\hat { p } _ { t } \\triangleq \\mathbb { 1 } _ { s _ { t + 1 } }$ of the transition kernel $p _ { s _ { t } } ^ { a _ { t } }$ , where $\\mathbb { 1 } _ { s _ { t + 1 } }$ is a probability distribution taking probability 1 at $s _ { t + 1 }$ and 0 at other states. This is an unbiased estimate of the transition kernel $p _ { s _ { t } } ^ { a _ { t } }$ conditioning on $S _ { t } = s _ { t }$ and $A _ { t } = a _ { t }$ . We then design a sample-based estimate $\\hat { \\mathcal { P } } _ { t } \\triangleq \\left\\{ ( 1 - R ) \\hat { p } _ { t } + R q | q \\in \\Delta _ { | \\mathcal { S } | } \\right\\}$ of the uncertainty set $\\mathcal { P } _ { s _ { t } } ^ { a _ { t } }$ . Using the sample-based uncertainty set $\\hat { \\mathcal { P } } _ { t }$ , we construct the following robust Q-learning algorithm in Algorithm 1. For any $t$ , $\\sigma _ { \\hat { \\mathcal { P } } _ { t } } ( V _ { t } )$ can be easily computed: $\\sigma _ { \\hat { \\Phi } _ { t } } ( V _ { t } ) =$ ", + "bbox": [ + 173, + 183, + 825, + 308 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Algorithm 1 Robust Q-Learning ", + "text_level": 1, + "bbox": [ + 174, + 325, + 390, + 340 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Initialization: $T$ , $Q _ { 0 } ( s , a )$ for all $( s , a ) \\in \\mathcal { S } \\times \\mathcal { A }$ , behavior policy $\\pi _ { b }$ , $s _ { 0 }$ , step size $\\alpha _ { t }$ ", + "bbox": [ + 174, + 343, + 732, + 359 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "1: for $t = 0 , 1 , 2 , . . . , T - 1$ do \n2: Choose $a _ { t }$ according to $\\pi _ { b } ( \\cdot | s _ { t } )$ \n3: Observe $s _ { t + 1 }$ and $c _ { t }$ \n4: $\\begin{array} { r } { V _ { t } ( s ) \\operatorname* { m i n } _ { a \\in \\mathcal { A } } Q _ { t } ( s , a ) , \\forall s \\in \\mathfrak { s } } \\end{array}$ \n5: Qt+1( $s _ { t } , a _ { t } ) \\gets ( 1 - \\alpha _ { t } ) Q _ { t } ( s _ { t } , a _ { t } ) + \\alpha _ { t } ( c _ { t } + \\gamma \\sigma _ { \\hat { \\mathcal P } _ { t } } ( V _ { t } ) )$ \n6: $Q _ { t + 1 } ( s , a ) \\gets Q _ { t } ( s , a )$ for $( s , a ) \\neq ( s _ { t } , a _ { t } )$ \n7: end for ", + "bbox": [ + 179, + 361, + 596, + 453 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "", + "bbox": [ + 181, + 446, + 254, + 458 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Output: $Q _ { T }$ ", + "bbox": [ + 174, + 462, + 258, + 477 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "$\\begin{array} { r } { R \\operatorname* { m a x } _ { s \\in \\mathcal { S } } V _ { t } ( s ) + ( 1 - R ) V _ { t } ( s _ { t + 1 } ) } \\end{array}$ . Hence the update in Algorithm 1 (line 5) can be written as ", + "bbox": [ + 169, + 493, + 802, + 510 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/ebc655b9749ff56b32e6e65d9da89184ff1f6cd737704bf092f1717d68c980c8.jpg", + "text": "$$\nQ _ { t + 1 } ( s _ { t } , a _ { t } ) \\gets ( 1 - \\alpha _ { t } ) Q _ { t } ( s _ { t } , a _ { t } ) + \\alpha _ { t } ( c _ { t } + \\gamma R \\operatorname* { m a x } _ { s \\in \\mathcal { S } } V _ { t } ( s ) + \\gamma ( 1 - R ) V _ { t } ( s _ { t + 1 } ) ) .\n$$", + "text_format": "latex", + "bbox": [ + 220, + 517, + 776, + 541 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Compared to the model-based approach, our approach is model-free. It does not require the prior knowledge of the uncertainty set, i.e., the knowledge of $p _ { s } ^ { a } , \\forall s \\in \\mathcal { S } , a \\in \\mathcal { A }$ . Furthermore, the memory requirement of our algorithm is $| \\mathcal { S } | \\times | \\mathcal { A } |$ (used to store the Q-table), and unlike the modelbased approach it does not need a table of size $| \\dot { \\mathcal { S } } | ^ { 2 } | \\mathcal { A } |$ to store $p _ { s } ^ { a } , \\forall s \\in \\mathcal { S } , a \\in \\mathcal { A }$ , which could be problematic if the state space is large. Moreover, our algorithm does not involve a relaxation of the uncertainty set like the one in [Roy et al., 2017], which will incur a stringent condition on the discount factor to guarantee the convergence. As will be shown below, the convergence of our Algorithm 1 does not require any condition on the discount factor. ", + "bbox": [ + 173, + 555, + 825, + 667 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "We show in the following theorem that the robust Q-learning algorithm converges asymptotically to the optimal robust action-value function $Q ^ { * }$ . ", + "bbox": [ + 174, + 672, + 823, + 702 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Theorem 2. (Asymptotic Convergence) If step sizes $\\alpha _ { t }$ satisfy that $\\textstyle \\sum _ { t = 0 } ^ { \\infty } \\alpha _ { t } = \\infty$ and $\\textstyle \\sum _ { t = 0 } ^ { \\infty } \\alpha _ { t } ^ { 2 } < \\infty$ then $Q _ { t } \\to Q ^ { * }$ as with probability . ", + "bbox": [ + 176, + 705, + 820, + 736 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "To further establish the finite-time error bound for our robust Q-learning algorithm in Algorithm 1, we make the following assumption that is commonly used in the analysis of vanilla Q-learning. ", + "bbox": [ + 176, + 747, + 821, + 776 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Assumption 1. The Markov chain induced by the behavior policy $\\pi _ { b }$ and the transition kernel $p _ { s } ^ { a } , \\forall s \\in \\mathcal { S } , a \\in \\mathcal { A }$ is uniformly ergodic. ", + "bbox": [ + 173, + 780, + 823, + 809 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Let $\\mu _ { \\pi _ { b } }$ denote the stationary distribution over $\\mathcal { S } \\times \\mathcal { A }$ induced by $\\pi _ { b }$ and $p _ { s } ^ { a } , \\forall s \\in \\mathcal { S } , a \\in \\mathcal { A }$ . We then further define $\\begin{array} { r } { \\mu _ { \\mathrm { m i n } } = \\operatorname* { m i n } _ { ( s , a ) \\in \\mathcal { S } \\times \\mathcal { A } } \\mu _ { \\pi _ { b } } ( s , a ) } \\end{array}$ . This quantity characterizes how many samples are needed to visit every state-action pair sufficiently often. Define the following mixing time of the induced Markov chain: $\\begin{array} { r } { t _ { \\mathrm { m i x } } = \\operatorname* { m i n } \\left. t : \\operatorname* { m a x } _ { s \\in \\mathcal { S } } \\mathsf { \\bar { d } _ { \\mathrm { T V } } } ( \\mu _ { \\pi } , P ( s _ { t } = \\cdot | s _ { 0 } = s ) ) \\breve { \\leq } \\frac { 1 } { 4 } \\right. } \\end{array}$ , where $d _ { \\mathrm { T V } }$ is the total variation distance. ", + "bbox": [ + 174, + 820, + 825, + 890 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "The following theorem establishes the finite-time error bound of our robust Q-learning algorithm. ", + "bbox": [ + 168, + 896, + 808, + 911 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Theorem 3. (Finite-Time Error Bound) There exist some positive constants $c _ { 0 }$ and $c _ { 1 }$ such that for any $\\delta < 1$ , any $\\begin{array} { r } { \\epsilon < \\frac { 1 } { 1 - \\gamma } } \\end{array}$ , any $T$ satisfying ", + "bbox": [ + 169, + 90, + 825, + 122 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/9ec28bd95e4de19b60114b1e15ae076fa40d0282d19b6b1f9240d27b88488068.jpg", + "text": "$$\nT \\geq c _ { 0 } \\left( \\frac { 1 } { \\mu _ { m i n } ( 1 - \\gamma ) ^ { 5 } \\epsilon ^ { 2 } } + \\frac { t _ { m i x } } { \\mu _ { m i n } ( 1 - \\gamma ) } \\right) \\log { \\left( \\frac { T | \\mathcal { S } | | \\mathcal { A } | } { \\delta } \\right) } \\log { \\left( \\frac { 1 } { \\epsilon ( 1 - \\gamma ) ^ { 2 } } \\right) } ,\n$$", + "text_format": "latex", + "bbox": [ + 240, + 131, + 754, + 166 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "and step size αt = $\\begin{array} { r } { \\alpha _ { t } = \\frac { c _ { 1 } } { \\log \\left( \\frac { T | \\mathcal { S } | | \\mathcal { A } | } { \\delta } \\right) } \\operatorname* { m i n } \\left( \\frac { 1 } { t _ { m i x } } , \\frac { \\epsilon ^ { 2 } ( 1 - \\gamma ) ^ { 4 } } { \\gamma ^ { 2 } } \\right) , \\forall t \\geq 0 } \\end{array}$ we have with probability at least $1 - 6 \\delta$ , $\\| Q _ { T } - Q ^ { * } \\| _ { \\infty } \\leq 3 \\epsilon$ . ", + "bbox": [ + 173, + 172, + 826, + 217 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "From the theorem, we can see that to guarantee an $\\epsilon$ -accurate estimate, a sample size $\\begin{array} { r } { \\tilde { \\mathcal { O } } ( \\frac { 1 } { \\mu _ { \\mathrm { m i n } } ( 1 - \\gamma ) ^ { 5 } \\epsilon ^ { 2 } } + } \\end{array}$ tmixµmin(1−γ) ) (up to some logarithmic terms) is needed. This complexity matches with the one for the vanilla Q-learning in [Li et al., 2020] (within a constant factor), while our algorithm also guarantees robustness to MDP model uncertainty. Our algorithm design and analysis can be readily extended to robustify TD and SARSA. The variance-reduction technique [Wainwright, 2019] can also be combined with our robust Q-learning algorithm to further improve the dependency on $( 1 - \\gamma )$ . ", + "bbox": [ + 173, + 226, + 826, + 320 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "5 Function Approximation: Robust TDC ", + "text_level": 1, + "bbox": [ + 174, + 340, + 531, + 358 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "In this section, we investigate the case where the state and action spaces can be large or even continuous. A popular approach is to approximate the value function using a parameterized function, e.g., linear function and neural network. In this section, we focus on the case with linear function approximation to illustrate the main idea of designing robust RL algorithms. Our approach can be extended to non-linear (smooth) function approximation using techniques in, e.g., [Cai et al., 2019, Bhatnagar et al., 2009, Wai et al., 2019, Wang et al., 2021]. ", + "bbox": [ + 173, + 372, + 826, + 457 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "We focus on the problem of robust policy evaluation, i.e., estimate the robust value function $V ^ { \\pi }$ defined in (1) for a given policy $\\pi$ under the worst-case MDP transition kernel in the uncertainty set. Note that for robust RL with model uncertainty, any policy evaluation problem can be viewed as \"off-transition-kernel\", as it is to evaluate the value function under the worst-case MDP using samples from a different MDP. Since the TD algorithm with function approximation may diverge under off-policy training [Baird, 1995] and importance sampling cannot be applied here due to unknown transition kernel, in this paper we generalize the GTD method [Maei et al., 2010, Maei, 2011] to the robust setting. ", + "bbox": [ + 173, + 462, + 826, + 574 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Let $\\left\\{ \\phi ^ { ( i ) } : \\mathbb { S } \\to \\mathbb { R } , i = 1 , \\ldots , N \\right\\}$ be a set of $N$ fixed base functions, where $N \\ll | \\mathcal S | | \\mathcal A |$ . In particular, we approximate the robust value function using a linear combination of $\\begin{array} { r } { \\sum _ { i = 1 } ^ { N } \\theta ^ { i } \\phi _ { s } ^ { ( i ) } = \\phi _ { s } ^ { \\top } \\theta } \\end{array}$ , where $\\boldsymbol { \\theta } \\in \\mathbb { R } ^ { N }$ is the weight vector. $\\phi ^ { ( i ) }$ ’s: $V _ { \\theta } ( s ) =$ ", + "bbox": [ + 173, + 579, + 825, + 631 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Define the following robust Bellman operator for a given policy $\\pi$ : ", + "bbox": [ + 173, + 636, + 611, + 651 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/9852dcd21269061985da10dcabeac8e4caafd9fd4a4cbfc50d42023c0a2ae526.jpg", + "text": "$$\n\\begin{array} { r l } & { \\mathrm { \\bf T } _ { \\pi } V ( s ) \\triangleq \\mathbb E _ { A \\sim \\pi ( \\cdot | s ) } [ c ( s , A ) + \\gamma \\sigma _ { \\mathcal { P } _ { s } ^ { A } } ( V ) ] } \\\\ & { \\quad \\quad \\quad = \\mathbb E _ { A \\sim \\pi ( \\cdot | s ) } \\left[ c ( s , A ) + \\gamma ( 1 - R ) \\displaystyle \\sum _ { s ^ { \\prime } \\in \\mathcal { S } } p _ { s , s ^ { \\prime } } ^ { A } V ( s ^ { \\prime } ) + \\gamma R \\operatorname* { m a x } _ { s ^ { \\prime } \\in \\mathcal { S } } V ( s ^ { \\prime } ) \\right] . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 238, + 657, + 759, + 724 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "We then define the mean squared projected robust Bellman error (MSPRBE) as ", + "bbox": [ + 173, + 731, + 692, + 746 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/7fdf581c44a14004534015be69003a637c4f9373f971dff5a0e1cb387e623e39.jpg", + "text": "$$\n\\begin{array} { r } { \\mathrm { M S P R B E } ( \\theta ) = \\| \\mathbf { I I T } _ { \\pi } V _ { \\theta } - V _ { \\theta } \\| _ { \\mu _ { \\pi } } ^ { 2 } , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 377, + 753, + 619, + 775 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "where $\\begin{array} { r } { \\| v \\| _ { \\mu _ { \\pi } } ^ { 2 } = \\int v ^ { 2 } ( s ) \\mu _ { \\pi } ( d s ) } \\end{array}$ , $\\mu _ { \\pi }$ is the stationary distribution induced by $\\pi$ , and $\\mathbf { I I }$ is a projection onto the linear function space w.r.t. $\\| \\cdot \\| _ { \\mu _ { \\pi } }$ . We will develop a two time-scale gradient-based approach to minimize the MSPRBE. However, it can be seen that $\\operatorname* { m a x } _ { s } V _ { \\theta } ( s )$ in (7) is not smooth in $\\theta$ , which is troublesome in both algorithm design and analysis. To solve this issue, we introduce the following smoothed robust Bellman operator $\\hat { \\mathbf { T } } _ { \\pi }$ by smoothing the max with a LSE(LogSumExp): ", + "bbox": [ + 174, + 782, + 825, + 859 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/7543aac30207ffdd163f7a3ab81924732889d4c9ebbac9f5e3be94660e8a76c2.jpg", + "text": "$$\n\\hat { \\mathbf { T } } _ { \\pi } V ( s ) = \\mathbb { E } _ { A \\sim \\pi ( \\cdot | s ) } \\left[ c ( s , A ) + \\gamma ( 1 - R ) \\sum _ { s ^ { \\prime } \\in \\mathcal { S } } p _ { s , s ^ { \\prime } } ^ { A } V ( s ^ { \\prime } ) + \\gamma R \\cdot \\mathrm { L S E } ( V ) \\right] ,\n$$", + "text_format": "latex", + "bbox": [ + 243, + 866, + 753, + 910 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "where $\\begin{array} { r } { \\mathrm { L S E } ( V ) = \\frac { \\log \\left( \\sum _ { s } e ^ { \\varrho V ( s ) } \\right) } { \\varrho } } \\end{array}$ is the LogSumExp w.r.t. $V$ with a parameter $\\varrho > 0$ . Note that when $\\varrho \\to \\infty$ , the smoothed robust Bellman operator $\\hat { \\mathbf { T } } _ { \\pi } \\to \\mathbf { T } _ { \\pi }$ . The LSE operator can also be replaced by some other operator that approximates the max operator and is smooth, e.g., mellow-max [Asadi and Littman, 2017]. In the following, we first show that the fixed point of $\\hat { \\mathbf { T } } _ { \\pi }$ exists for any $\\varrho$ , and the fixed points converge to the one of $\\mathbf { T } _ { \\pi }$ for large $\\varrho$ . ", + "bbox": [ + 173, + 87, + 826, + 172 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Theorem 4. (1). For any $\\varrho$ , $\\hat { \\mathbf { T } } _ { \\pi }$ has a fixed point. ", + "bbox": [ + 173, + 178, + 498, + 194 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "(2). Let $V _ { 1 }$ and $V _ { 2 }$ be the fixed points of $\\hat { \\mathbf { T } } _ { \\pi }$ and $\\mathbf { T } _ { \\pi }$ , respectively. Then ", + "bbox": [ + 171, + 199, + 642, + 217 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/ad911689bf0003f0de2b6c5368d62b6e5ae5d9dd29ce3c7f51bc819eed442bbf.jpg", + "text": "$$\n\\| V _ { 1 } - V _ { 2 } \\| _ { \\infty } \\leq \\frac { \\gamma R } { 1 - \\gamma } \\frac { \\log | \\mathcal { S } | } { \\varrho } 0 , a s \\varrho \\infty .\n$$", + "text_format": "latex", + "bbox": [ + 341, + 222, + 656, + 257 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "We then denote by $J ( \\theta )$ the smoothed MSPRBE with the LSE operator, and the goal is: ", + "bbox": [ + 173, + 268, + 746, + 285 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/fa3206f9b868d359d6943578e15679723de365442f3536e6f3308bb5498eeceb.jpg", + "text": "$$\n\\operatorname* { m i n } _ { \\theta } J ( \\theta ) = \\operatorname* { m i n } _ { \\theta } \\Big \\| \\mathbf { H } \\hat { \\mathbf { T } } _ { \\pi } V _ { \\theta } - V _ { \\theta } \\Big \\| _ { \\mu _ { \\pi } } ^ { 2 } .\n$$", + "text_format": "latex", + "bbox": [ + 372, + 291, + 622, + 324 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "5.1 Algorithm Development ", + "text_level": 1, + "bbox": [ + 174, + 337, + 380, + 353 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "In the following, we develop the robust TDC algorithm to solve the problem in (11). We will first derive the gradient of the smoothed MSPRBE, $J ( \\theta )$ , and then design a two time-scale update rule using the weight doubling trick in [Sutton et al., 2009] to solve the double sampling problem. Define $\\delta _ { s , a , s ^ { \\prime } } ( \\theta ) \\ \\triangleq \\ c ( s , a ) + \\gamma ( 1 - R ) V _ { \\theta } ( s ^ { \\prime } ) + \\gamma R \\mathrm { L S E } ( V _ { \\theta } ) - V _ { \\theta } ( s )$ , where $\\operatorname { L S E } ( V _ { \\theta } )$ is the LogSumExp function w.r.t. $V _ { \\theta } = \\theta ^ { \\top } \\phi$ . Denote by $C \\triangleq \\mathbb { E } _ { \\mu _ { \\pi } } \\left[ \\phi _ { S } ^ { \\intercal } \\phi _ { S } \\right]$ . Then, $\\mathbb { E } _ { \\mu } [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } ] \\ = \\ \\varPhi ^ { \\top } D \\left( \\hat { \\mathbf { T } } _ { \\pi } V _ { \\theta } - V _ { \\theta } \\right)$ , where $\\textit { D } = \\ \\mathrm { d i a g } ( \\mu _ { \\pi } ( s _ { 1 } ) , \\mu _ { \\pi } ( s _ { 2 } ) , . . . , \\mu _ { \\pi } ( s _ { | \\mathcal { S } | } ) )$ and $\\boldsymbol { \\varPhi } = ( \\phi _ { s _ { 1 } } , \\phi _ { s _ { 2 } } , . . . , \\phi _ { s _ { | \\mathcal { S } | } } ) ^ { \\top } \\in \\dot { \\mathbb { R } } ^ { | \\mathcal { S } | \\times N }$ . We know that $\\begin{array} { r } { \\pmb { \\Pi } ^ { \\top } D \\pmb { \\Pi } = D ^ { \\top } \\pmb { \\varPhi } ( \\pmb { \\varPhi } ^ { \\top } D \\pmb { \\varPhi } ) ^ { - 1 } \\pmb { \\varPhi } ^ { \\top } D } \\end{array}$ from [Maei, 2011]. Hence we have ", + "bbox": [ + 171, + 362, + 826, + 492 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/65951d26b16d119c3bba00a86bde2fc36e1073e08e65ed93c7a1b72420890a6a.jpg", + "text": "$$\n\\begin{array} { r } { \\boldsymbol J ( \\theta ) = \\Big \\| \\mathbf { \\hat { I } } \\mathbf { \\hat { T } } _ { \\boldsymbol \\pi } V _ { \\theta } - V _ { \\theta } \\Big \\| _ { \\mu _ { \\boldsymbol \\pi } } ^ { 2 } = \\mathbb { E } _ { \\mu _ { \\boldsymbol \\pi } } [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } ] ^ { \\top } \\boldsymbol C ^ { - 1 } \\mathbb { E } _ { \\mu _ { \\boldsymbol \\pi } } [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } ] . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 251, + 497, + 746, + 530 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Then, its gradient can be written as: ", + "bbox": [ + 174, + 535, + 408, + 550 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/f48da04f35475f740a3388ea84655013bc612fe39b6277c8a6300ce42221782b.jpg", + "text": "$$\n\\begin{array} { l } { \\displaystyle - \\frac { 1 } { 2 } \\nabla J ( \\theta ) = - \\mathbb { E } _ { \\mu _ { \\pi } } \\big [ ( \\nabla \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) ) \\phi _ { S } \\big ] ^ { \\top } C ^ { - 1 } \\mathbb { E } _ { \\mu _ { \\pi } } \\big [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } \\big ] } \\\\ { = \\mathbb { E } _ { \\mu _ { \\pi } } \\big [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } \\big ] - \\gamma \\mathbb { E } _ { \\mu _ { \\pi } } \\bigg [ \\bigg ( ( 1 - R ) \\phi _ { S ^ { \\prime } } + R \\cdot \\nabla \\mathrm { L S E } ( V _ { \\theta } ) \\bigg ) \\phi _ { S } ^ { \\top } \\bigg ] \\omega ( \\theta ) , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 250, + 555, + 746, + 622 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "where $\\omega ( \\theta ) = C ^ { - 1 } \\mathbb { E } _ { \\mu _ { \\pi } } [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } ]$ . It can be seen that to obtain an unbiased estimate of $\\nabla J ( \\theta )$ , two independent samples are needed as there exists a multiplication of two expectations, which is not applicable when there is only one sample trajectory. We then utilize the weight doubling trick in [Sutton et al., 2009], and design the robust TDC algorithm in Algorithm 2. Specifically, we introduce a fast time scale to estimate $\\omega ( \\theta )$ , and a slow time scale to estimate $\\nabla J ( \\theta )$ . Denote the projection by $\\Pi _ { K } ( x ) \\triangleq \\arg \\operatorname* { m i n } _ { \\| y \\| \\leq K } \\| y - x \\|$ for any $\\boldsymbol { x } \\in \\mathbb { R } ^ { N }$ . Our robust TDC algorithm in Algorithm 2 can be implemented in an online and incremental fashion. If the uncertainty set becomes a singleton, i.e., $R = 0$ , then Algorithm 2 reduces to the vanilla TDC algorithm. ", + "bbox": [ + 173, + 627, + 826, + 743 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "5.2 Finite-Time Error Bound of Robust TDC ", + "text_level": 1, + "bbox": [ + 173, + 757, + 500, + 773 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Unlike the vanilla TDC algorithm, $J ( \\theta )$ here is non-convex. Therefore, we are interested in the convergence to stationary points, i.e., the rate of $\\| \\nabla J ( \\theta ) \\| 0$ . We first make some standard assumptions which are commonly used in RL algorithm analysis, e.g., [Wang and Zou, 2020, Kaledin et al., 2020, Xu et al., 2019, Srikant and Ying, 2019, Bhandari et al., 2018]. ", + "bbox": [ + 173, + 784, + 826, + 839 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Assumption 2 (Bounded feature). $\\| \\phi _ { s } \\| _ { 2 } \\leq 1 , \\forall s \\in \\mathcal { S }$ . ", + "bbox": [ + 171, + 843, + 535, + 859 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Assumption 3 (Bounded cost function). $| c ( s , a ) | \\leq c _ { \\mathrm { m a x } } , \\forall s \\in \\mathcal { S }$ and $a \\in { \\mathcal { A } }$ . ", + "bbox": [ + 171, + 862, + 681, + 878 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Assumption 4 (Problem solvability). The matrix $C = \\mathbb { E } _ { \\mu _ { \\pi } } [ \\phi _ { S } \\phi _ { S } ^ { \\top } ]$ is non-singular with $\\lambda > 0$ being its smallest eigenvalue. ", + "bbox": [ + 173, + 882, + 825, + 912 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Input: $T , \\alpha , \\beta , \\varrho , \\phi _ { i }$ for $i = 1 , . . . , N$ , projection radius $K$ \nInitialization: θ0,w0, s0 \n1: Choose $W \\sim \\mathrm { U n i f o r m } ( 0 , 1 , . . . , T - 1 )$ \n2: for $t = 0 , 1 , 2 , . . . , W - 1$ do \n3: Take action according to $\\pi ( \\cdot | s _ { t } )$ and observe $s _ { t + 1 }$ and $c _ { t }$ \n4: $\\phi _ { t } \\phi _ { s _ { t } }$ \n7: 5: 6: $\\begin{array} { r l } & { \\ddot { \\delta _ { t } } ( \\theta _ { t } ) \\gets c _ { t } + \\gamma ( 1 - R ) V _ { \\theta _ { t } } ( s _ { t + 1 } ) + \\gamma R \\frac { \\log ( \\sum _ { s } e ^ { \\varrho \\theta ^ { \\top } \\phi _ { s } } ) } { \\varrho } - V _ { \\theta _ { t } } ( s _ { t } ) } \\\\ & { \\theta _ { t + 1 } \\gets \\Pi _ { K } \\left( \\theta _ { t } + \\alpha \\left( \\delta _ { t } ( \\theta _ { t } ) \\phi _ { t } - \\gamma \\bigg ( ( 1 - R ) \\phi _ { t + 1 } + R \\sum _ { s \\in \\mathcal { S } } \\left( \\frac { e ^ { \\sigma V _ { \\theta } ( s ) } \\phi _ { s } } { \\sum _ { j \\in \\mathcal { S } } e ^ { \\sigma V _ { \\theta } ( j ) } } \\right) \\bigg ) \\phi _ { t } ^ { \\top } \\omega _ { t } \\right) \\right) } \\\\ & { \\omega _ { t + 1 } \\gets \\Pi _ { K } ( \\omega _ { t } + \\beta ( \\delta _ { t } ( \\theta _ { t } ) - \\phi _ { t } ^ { \\top } \\omega _ { t } ) \\phi _ { t } ) } \\end{array}$ ", + "bbox": [ + 173, + 108, + 800, + 270 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "8: end for ", + "bbox": [ + 181, + 267, + 254, + 280 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Output: $\\theta _ { W }$ ", + "bbox": [ + 174, + 282, + 258, + 296 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Assumption 5 (Geometric uniform ergodicity). There exist some constants $m > 0$ and $\\rho \\in ( 0 , 1 )$ such that for any $t > 0$ , $\\begin{array} { r } { \\operatorname* { m a x } _ { s \\in \\mathcal { S } } d _ { T V } ( \\bar { \\mathbb { P } } ( s _ { t } | s _ { 0 } = s ) , \\mu _ { \\pi } ) \\leq m \\rho ^ { t } } \\end{array}$ . ", + "bbox": [ + 173, + 325, + 821, + 356 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "In the following theorem, we characterize the finite-time error bound for the convergence of our robust TDC algorithm. Here we only provide the order of the bounds in terms of $T$ . The explicit bounds can be found in (129) in Appendix D.3. ", + "bbox": [ + 174, + 364, + 825, + 406 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Theorem 5. Consider the following step-sizes: $\\begin{array} { r } { \\beta = \\mathcal { O } \\left( \\frac { 1 } { T ^ { b } } \\right) } \\end{array}$ , and $\\begin{array} { r } { \\alpha = \\mathcal { O } \\left( \\frac { 1 } { T ^ { a } } \\right) } \\end{array}$ , where $\\begin{array} { r } { \\frac { 1 } { 2 } < a \\leq 1 } \\end{array}$ and $0 < b \\leq a$ . Then we have that ", + "bbox": [ + 173, + 410, + 825, + 439 + ], + "page_idx": 7 + }, + { + "type": "equation", + "img_path": "images/a654cd452a7e1f7e9a634b553d55226b856e45fcf1332c2009f8149f8fab6ff9.jpg", + "text": "$$\n\\mathbb { E } [ \\| \\nabla J ( \\theta _ { W } ) \\| ^ { 2 } ] = \\odot \\left( \\frac { 1 } { T \\alpha } + \\alpha \\log ( 1 / \\alpha ) + \\frac { 1 } { T \\beta } + \\beta \\log ( 1 / \\beta ) \\right) ,\n$$", + "text_format": "latex", + "bbox": [ + 281, + 445, + 714, + 479 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "If we further let $a = b = 0 . 5$ , then $\\begin{array} { r } { \\mathbb { E } [ \\| \\nabla J ( \\theta _ { W } ) \\| ^ { 2 } ] = \\odot \\left( \\frac { \\log T } { \\sqrt { T } } \\right) . } \\end{array}$ ", + "bbox": [ + 173, + 486, + 599, + 512 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "The robust TDC has a matching complexity with the vanilla TDC with non-linear function approximation [Wang et al., 2021], but provides the additional robustness to model uncertainty. It does not need to relax the uncertainty set like in [Roy et al., 2017], and our convergence results do not need a condition on the discount factor. ", + "bbox": [ + 173, + 521, + 826, + 577 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "6 Experiments ", + "text_level": 1, + "bbox": [ + 174, + 597, + 312, + 613 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "6.1 Robust Q-Learning ", + "text_level": 1, + "bbox": [ + 174, + 627, + 349, + 642 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "In this section, we compare our robust Q-learning with the vanilla non-robust Q-learning. We use OpenAI gym framework [Brockman et al., 2016], and consider two different problems: Frozen lake and Cart-Pole. One more example of the taxi problem is given in the appendix. To demonstrate the robustness, the policy is learned in a perturbed MDP, and is then tested on the true unperturbed MDP. Specifically, during the training, we set a probability $p$ such that after the agent takes an action, with probability $p$ , the state transition is uniformly over S, and with probability $1 - p$ the state transition is according to the true unperturbed transition kernel. The behavior policy for all the experiments below is set to be a uniform distribution over the action space given any state, i.e., $\\begin{array} { r } { \\pi _ { b } ( a | s ) ^ { \\bullet } = \\frac { 1 } { | \\mathcal { A } | } } \\end{array}$ for any $s \\in \\mathcal { S }$ and $a \\in { \\mathcal { A } }$ . We then evaluate the performance of the obtained policy in the unperturbed environment. At each time t, the policy we evaluate is the greedy-policy w.r.t. the current estimate of the Q-function, i.e., $\\pi _ { t } ( s ) = \\arg \\operatorname* { m a x } _ { a } Q _ { t } ( s , a )$ . A Monte-Carlo method with horizon 100 is used to evaluate the accumulated discounted reward of the learned policy on the unperturbed MDP. We take the average over 30 trajectories. More details are provided in the appendix. ", + "bbox": [ + 173, + 652, + 826, + 835 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "In Figure 1 and Figure 2, we plot the accumulated discounted reward of both algorithms under different $p$ and $R$ for both problems. The upper and lower envelopes of the curves correspond to the 95 and 5 percentiles of the 30 trajectories, respectively. It can be seen that overall our robust Q-learning algorithm achieves a much higher reward than the vanilla Q-learning. This demonstrates the robustness of our robust Q-learning algorithm to model uncertainty. Moreover, as $p$ and $R$ getting larger, i.e., as the MDP that we learn the policy deviates from the MDP we test the policy, the advantage of our robust Q-learning algorithm is getting more significant compared to the vanilla Q-learning algorithm. ", + "bbox": [ + 174, + 842, + 825, + 911 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/f6e7e9e2735f3d2f3cb6a78235a2c13acfd85270e0532cc9f312127eab2de53d.jpg", + "image_caption": [ + "Figure 1: FrozenLake-v0: robust Q-learning v.s. non-robust Q-learning. " + ], + "image_footnote": [], + "bbox": [ + 192, + 107, + 803, + 231 + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/a43d2928aff120beef997e93a1a33000db2a33693b2a692ee8c4a435bdabda59.jpg", + "image_caption": [ + "Figure 2: CartPole-v0: robust Q-learning v.s. non-robust Q-learning. " + ], + "image_footnote": [], + "bbox": [ + 189, + 289, + 802, + 414 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 467, + 825, + 510 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "6.2 Robust TDC with Linear Function Approximation ", + "text_level": 1, + "bbox": [ + 173, + 526, + 563, + 541 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "In this section we compare our robust TDC with the vanilla non-robust TDC with linear function approximation on the $4 \\times 4$ Frozen Lake problem. The problem setting is the same as the one in Section 6.1. More details about the experiment setup are provided in the appendix. ", + "bbox": [ + 176, + 551, + 823, + 593 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "We implement the two algorithms using samples from the perturbed MDP both for 30 times, and obtain 30 sequences of $\\{ \\theta _ { t } ^ { i } \\} _ { t = 1 } ^ { \\infty }$ , $i = 1 , 2 , . . . , 3 0$ . We then compute the squared gradient norm $\\| \\nabla J ( \\theta ) \\| ^ { 2 }$ on the true unperturbed MDP, and see whether $\\{ \\theta _ { t } ^ { i } \\} _ { t = 1 } ^ { \\infty }$ converges to some stationary points on the true unperturbed MDP. In Fig. 3, we plot the average squared gradient norm $\\| \\nabla J ( \\theta ) \\| ^ { 2 }$ for different $p$ and $R$ . The upper and lower envelops are the 95 and 5 percentiles of the 30 curves. It can be seen that our robust TDC converges much faster than vanilla TDC, and as the model mismatch between the training and test MDPs enlarges, the vanilla TDC may diverge (Fig. 3(c)), while our robust TDC still converges to some stationary point. Also, the robust TDC has a much smaller variance, which indicates a much stable behavior under model uncertainty. ", + "bbox": [ + 173, + 599, + 825, + 724 + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/88a93ae259a06e7bc8fa9f4ad18dbea7ab033fd1a750f80210fcc36ef5a44b84.jpg", + "image_caption": [ + "Figure 3: FrozenLake-v0: $\\| \\nabla J ( \\theta ) \\| ^ { 2 }$ of robust TDC and non-robust TDC. " + ], + "image_footnote": [], + "bbox": [ + 191, + 756, + 802, + 881 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "6.3 Comparison with The Adversarial Training Approach ", + "text_level": 1, + "bbox": [ + 173, + 90, + 588, + 106 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "We also compare our robust Q-learning with Robust Adversarial Reinforcement Learning (RARL) in [Pinto et al., 2017]. To apply their algorithm to our problem setting, we model the nature as an adversarial player, and its goal is to minimize the reward that the agent receives. The action space $\\mathcal { A } _ { a d }$ of the nature is set to be the state space $\\mathcal { A } _ { a d } \\triangleq \\mathcal { S }$ . Then the perturbed training environment can be viewed as an adversarial model: both the agent and the adversary take actions $a _ { a } , a _ { a d }$ , then the environment will transit to state $a _ { a d }$ with probability $R$ and transit following the unperturbed MDP $p _ { s } ^ { a _ { a } }$ with probability $1 - R$ . The goal of the maximize its accumulated reward, while the goal of the natural is to minimize it. ", + "bbox": [ + 173, + 116, + 825, + 229 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Following the RARL algorithm [Pinto et al., 2017], in each iteration of the training, we first fix the adversarial policy and use Q-learning to optimize the agent’s policy and obtain the Q-table $Q _ { t }$ . Then we fix the agent’s policy and optimize the adversarial policy. ", + "bbox": [ + 174, + 236, + 825, + 277 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "After each training iteration, we test the performance of the greedy policies w.r.t. Q-tables obtained from robust Q-learning and RARL. The testing environment is set to be the worst-case, i.e., after the agent takes an action, the environment transits to the state which has the minimal value function $( \\mathrm { a r g m i n } _ { s \\in \\mathcal { S } } V _ { t } ( s ) )$ with probability $p$ . We plot the accumulated discounted rewards of both algorithms against number of training iterations under different parameters. We set $\\alpha = 0 . 2$ and $\\gamma = 0 . 9$ . It can be seen from Fig. 4 that our robust Q-learning achieves a higher accumulative reward, and thus is more robust that the RARL algorithm in [Pinto et al., 2017]. Also our robust Q-learning is more stable during training, i.e., the variance is smaller. ", + "bbox": [ + 173, + 284, + 826, + 395 + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/05ae19de592c7a368e9a827c529316238af525b6bd10c6a6b56e811a7ad5cc4c.jpg", + "image_caption": [ + "Figure 4: Taxi-v3: robust Q-learning v.s. RARL. " + ], + "image_footnote": [], + "bbox": [ + 197, + 425, + 790, + 546 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "7 Conclusion ", + "text_level": 1, + "bbox": [ + 174, + 608, + 299, + 625 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "In this paper, we develop a novel approach for solving model-free robust RL problems with model uncertainty. Our algorithms can be implemented in an online and incremental fashion, do not require additional memory than their non-robust counterparts. We theoretically proved the convergence of our algorithms under no additional assumption on the discount factor, and further characterized their finite-time error bounds, which match with their non-robust counterparts (within a constant factor). Our approach can be readily extended to robustify TD, SARSA and other GTD algorithms. ", + "bbox": [ + 174, + 628, + 825, + 712 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Limitations: It is also of future interest to investigate robustness to reward uncertainty, and other types of uncertainty sets, e.g., ones defined by KL divergence, Wasserstein distance and total variation. Negative societal impact: To the best of the authors’ knowledge, this study does not have any potential negative impact on the society. ", + "bbox": [ + 174, + 712, + 826, + 767 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "8 Acknowledgment ", + "text_level": 1, + "bbox": [ + 174, + 786, + 349, + 803 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "The work of Y. Wang and S. Zou was supported by the National Science Foundation under Grants CCF-2106560 and CCF- 2007783. ", + "bbox": [ + 174, + 818, + 823, + 844 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "References ", + "text_level": 1, + "bbox": [ + 174, + 92, + 253, + 106 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Mohammed Amin Abdullah, Hang Ren, Haitham Bou Ammar, Vladimir Milenkovic, Rui Luo, Mingtian Zhang, and Jun Wang. Wasserstein robust reinforcement learning. arXiv preprint arXiv:1907.13196, 2019. ", + "bbox": [ + 174, + 112, + 826, + 154 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Alekh Agarwal, Sham M Kakade, Jason D Lee, and Gaurav Mahajan. On the theory of policy gradient methods: Optimality, approximation, and distribution shift. Journal of Machine Learning Research, 22(98):1–76, 2021. ", + "bbox": [ + 171, + 165, + 825, + 208 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Kavosh Asadi and Michael L Littman. An alternative softmax operator for reinforcement learning. In Proc. International Conference on Machine Learning (ICML), volume 70, pages 243–252. JMLR, 2017. ", + "bbox": [ + 173, + 218, + 825, + 261 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Christopher G Atkeson and Jun Morimoto. Nonparametric representation of policies and value functions: A trajectory-based approach. In Proc. Advances in Neural Information Processing Systems (NIPS), pages 1643–1650, 2003. ", + "bbox": [ + 176, + 271, + 823, + 314 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Kishan Panaganti Badrinath and Dileep Kalathil. Robust reinforcement learning using least squares policy iteration with provable performance guarantees. In Proc. International Conference on Machine Learning (ICML), pages 511–520. PMLR, 2021. ", + "bbox": [ + 174, + 325, + 821, + 367 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "J Andrew Bagnell, Andrew Y Ng, and Jeff G Schneider. Solving uncertain markov decision. 09 2001. ", + "bbox": [ + 169, + 377, + 825, + 395 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Leemon Baird. Residual algorithms: Reinforcement learning with function approximation. In Machine Learning Proceedings, pages 30–37. Elsevier, 1995. ", + "bbox": [ + 169, + 404, + 823, + 433 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Carolyn L Beck and Rayadurgam Srikant. Error bounds for constant step-size Q-learning. Systems & control letters, 61(12):1203–1208, 2012. ", + "bbox": [ + 169, + 443, + 825, + 473 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Jalaj Bhandari and Daniel Russo. Global optimality guarantees for policy gradient methods. arXiv preprint arXiv:1906.01786, 2019. ", + "bbox": [ + 173, + 483, + 823, + 512 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Jalaj Bhandari, Daniel Russo, and Raghav Singal. A finite time analysis of temporal difference learning with linear function approximation. In Proc. Annual Conference on Learning Theory (CoLT), pages 1691–1692. PMLR, 2018. ", + "bbox": [ + 173, + 522, + 823, + 565 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Shalabh Bhatnagar, Doina Precup, David Silver, Richard S Sutton, Hamid Maei, and Csaba Szepesvári. Convergent temporal-difference learning with arbitrary smooth function approximation. In Proc. Advances in Neural Information Processing Systems (NIPS), volume 22, pages 1204–1212, 2009. ", + "bbox": [ + 176, + 575, + 825, + 619 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Vivek S Borkar and Sean P Meyn. The ODE method for convergence of stochastic approximation and reinforcement learning. SIAM Journal on Control and Optimization, 38(2):447–469, 2000. ", + "bbox": [ + 171, + 628, + 823, + 659 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Greg Brockman, Vicki Cheung, Ludwig Pettersson, Jonas Schneider, John Schulman, Jie Tang, and Wojciech Zaremba. OpenAI Gym. arXiv preprint arXiv:1606.01540, 2016. ", + "bbox": [ + 173, + 669, + 823, + 698 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Qi Cai, Zhuoran Yang, Jason D Lee, and Zhaoran Wang. Neural temporal-difference learning converges to global optima. In Proc. Advances in Neural Information Processing Systems (NeurIPS), pages 11312–11322, 2019. ", + "bbox": [ + 173, + 708, + 826, + 751 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Shicong Cen, Chen Cheng, Yuxin Chen, Yuting Wei, and Yuejie Chi. Fast global convergence of natural policy gradient methods with entropy regularization. arXiv preprint arXiv:2007.06558, 2020. ", + "bbox": [ + 173, + 762, + 825, + 804 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Zaiwei Chen, Sheng Zhang, Thinh T Doan, Siva Theja Maguluri, and John-Paul Clarke. Performance of Q-learning with linear function approuimation: Stability and finite-time analysis. arXiv preprint arXiv:1905.11425, 2019. ", + "bbox": [ + 171, + 815, + 825, + 858 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Gal Dalal, Balázs Szörényi, Gugan Thoppe, and Shie Mannor. Finite sample analysis of two-timescale stochastic approximation with applications to reinforcement learning. Proceedings of Machine Learning Research, 75:1–35, 2018. ", + "bbox": [ + 174, + 869, + 826, + 911 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Gal Dalal, Balázs Szörényi, and Gugan Thoppe. A tale of two-timescale reinforcement learning with the tightest finite-time bound. In Proc. Conference on Artificial Intelligence (AAAI), pages 3701–3708, 2020. ", + "bbox": [ + 173, + 90, + 823, + 133 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Dalal, Gal and Szörényi, Balázs and Thoppe, Gugan and Mannor, Shie. Finite sample analyses for TD(0) with function approximation. In Proc. AAAI Conference on Artificial Intelligence (AAAI), pages 6144–6160, 2018. ", + "bbox": [ + 174, + 142, + 823, + 185 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Thinh T Doan. Finite-time analysis and restarting scheme for linear two-time-scale stochastic approximation. SIAM Journal on Control and Optimization, 59(4):2798–2819, 2021. ", + "bbox": [ + 171, + 194, + 823, + 223 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Eyal Even-Dar, Yishay Mansour, and Peter Bartlett. Learning rates for Q-learning. Journal of machine learning Research, 5(1), 2003. ", + "bbox": [ + 173, + 231, + 823, + 261 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Harsh Gupta, R Srikant, and Lei Ying. Finite-time performance bounds and adaptive learning rate selection for two time-scale reinforcement learning. In Proc. Advances in Neural Information Processing Systems (NeurIPS), pages 4706–4715, 2019. ", + "bbox": [ + 173, + 268, + 825, + 313 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Linfang Hou, Liang Pang, Xin Hong, Yanyan Lan, Zhiming Ma, and Dawei Yin. Robust reinforcement learning with Wasserstein constraint. arXiv preprint arXiv:2006.00945, 2020. ", + "bbox": [ + 173, + 320, + 821, + 349 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Sandy Huang, Nicolas Papernot, Ian Goodfellow, Yan Duan, and Pieter Abbeel. Adversarial attacks on neural network policies. In Proc. International Conference on Learning Representations (ICLR), 2017. ", + "bbox": [ + 173, + 358, + 826, + 400 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Peter J Huber. A robust version of the probability ratio test. The Annals of Mathematical Statistics, pages 1753–1758, 1965. ", + "bbox": [ + 173, + 409, + 825, + 439 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Garud N Iyengar. Robust dynamic programming. Mathematics of Operations Research, 30(2): 257–280, 2005. ", + "bbox": [ + 173, + 446, + 823, + 477 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Maxim Kaledin, Eric Moulines, Alexey Naumov, Vladislav Tadic, and Hoi-To Wai. Finite time analysis of linear two-timescale stochastic approximation with Markovian noise. In Proc. Annual Conference on Learning Theory (CoLT), pages 2144–2203. PMLR, 2020. ", + "bbox": [ + 173, + 484, + 826, + 529 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Jernej Kos and Dawn Song. Delving into adversarial attacks on deep policies. In Proc. International Conference on Learning Representations (ICLR), 2017. ", + "bbox": [ + 173, + 536, + 823, + 565 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Harshat Kumar, Alec Koppel, and Alejandro Ribeiro. On the sample complexity of actor-critic method for reinforcement learning with function approximation. arXiv preprint arXiv:1910.08412, 2019. ", + "bbox": [ + 173, + 574, + 826, + 616 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Gen Li, Yuting Wei, Yuejie Chi, Yuantao Gu, and Yuxin Chen. Sample complexity of asynchronous Q-learning: Sharper analysis and variance reduction. In Proc. Advances in Neural Information Processing Systems (NeurIPS), 2020. ", + "bbox": [ + 174, + 625, + 823, + 667 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Gen Li, Changxiao Cai, Yuxin Chen, Yuantao Gu, Yuting Wei, and Yuejie Chi. Is Q-learning minimax optimal? A tight sample complexity analysis. arXiv preprint arXiv:2102.06548, 2021. ", + "bbox": [ + 169, + 676, + 823, + 707 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Shiau Hong Lim, Huan Xu, and Shie Mannor. Reinforcement learning in robust Markov decision processes. In Proc. Advances in Neural Information Processing Systems (NIPS), pages 701–709, 2013. ", + "bbox": [ + 174, + 714, + 825, + 757 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Yen-Chen Lin, Zhang-Wei Hong, Yuan-Hong Liao, Meng-Li Shih, Ming-Yu Liu, and Min Sun. Tactics of adversarial attack on deep reinforcement learning agents. In Proc. International Joint Conferences on Artificial Intelligence (IJCAI), pages 3756–3762, 2017. ", + "bbox": [ + 176, + 765, + 825, + 809 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Bo Liu, Ji Liu, Mohammad Ghavamzadeh, Sridhar Mahadevan, and Marek Petrik. Finite-sample analysis of proximal gradient TD algorithms. In Proc. International Conference on Uncertainty in Artificial Intelligence (UAI), pages 504–513. Citeseer, 2015. ", + "bbox": [ + 174, + 816, + 823, + 861 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Shaocong Ma, Yi Zhou, and Shaofeng Zou. Variance-reduced off-policy TDC learning: Nonasymptotic convergence analysis. In Proc. Advances in Neural Information Processing Systems (NeurIPS), volume 33, pages 14796–14806, 2020. ", + "bbox": [ + 174, + 868, + 826, + 911 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Shaocong Ma, Yi Zhou, and Shaofeng Zou. Greedy-GQ with variance reduction: Finite-time analysis and improved complexity. In Proc. International Conference on Learning Representations (ICLR), 2021. ", + "bbox": [ + 174, + 90, + 825, + 133 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Hamid Reza Maei. Gradient temporal-difference learning algorithms. Thesis, University of Alberta, 2011. ", + "bbox": [ + 173, + 143, + 823, + 172 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Hamid Reza Maei, Csaba Szepesvári, Shalabh Bhatnagar, and Richard S Sutton. Toward off-policy learning control with function approximation. In Proc. International Conference on Machine Learning (ICML), pages 719–726, 2010. ", + "bbox": [ + 173, + 183, + 826, + 226 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Ajay Mandlekar, Yuke Zhu, Animesh Garg, Li Fei-Fei, and Silvio Savarese. Adversarially robust policy learning: Active construction of physically-plausible perturbations. In 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 3932–3939. IEEE, 2017. ", + "bbox": [ + 174, + 236, + 826, + 291 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Jincheng Mei, Chenjun Xiao, Csaba Szepesvari, and Dale Schuurmans. On the global convergence rates of softmax policy gradient methods. In Proc. International Conference on Machine Learning (ICML), pages 6820–6829. PMLR, 2020. ", + "bbox": [ + 174, + 301, + 823, + 344 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Jun Morimoto and Kenji Doya. Robust reinforcement learning. Neural computation, 17(2):335–359, 2005. ", + "bbox": [ + 171, + 354, + 825, + 385 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Arnab Nilim and Laurent El Ghaoui. Robustness in Markov decision problems with uncertain transition matrices. In Proc. Advances in Neural Information Processing Systems (NIPS), pages 839–846, 2004. ", + "bbox": [ + 174, + 393, + 826, + 438 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Anay Pattanaik, Zhenyi Tang, Shuijing Liu, Gautham Bommannan, and Girish Chowdhary. Robust deep reinforcement learning with adversarial attacks. In Proc. International Conference on Autonomous Agents and MultiAgent Systems, pages 2040–2042, 2018. ", + "bbox": [ + 174, + 446, + 825, + 491 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Lerrel Pinto, James Davidson, Rahul Sukthankar, and Abhinav Gupta. Robust adversarial reinforcement learning. In Proc. International Conference on Machine Learning (ICML), pages 2817–2826. PMLR, 2017. ", + "bbox": [ + 173, + 500, + 826, + 542 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Shuang Qiu, Zhuoran Yang, Jieping Ye, and Zhaoran Wang. On the finite-time convergence of actor-critic algorithm. In Proc. Optimization Foundations for Reinforcement Learning Workshop at Advances in Neural Information Processing Systems (NeurIPS), 2019. ", + "bbox": [ + 174, + 553, + 825, + 597 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Guannan Qu and Adam Wierman. Finite-time analysis of asynchronous stochastic approximation and Q-learning. In Proc. Annual Conference on Learning Theory (CoLT), pages 3185–3205. PMLR, 2020. ", + "bbox": [ + 174, + 606, + 826, + 648 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Aravind Rajeswaran, Sarvjeet Ghotra, Balaraman Ravindran, and Sergey Levine. Epopt: Learning robust neural network policies using model ensembles. In Proc. International Conference on Learning Representations (ICLR), 2017. ", + "bbox": [ + 173, + 659, + 826, + 702 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Aurko Roy, Huan Xu, and Sebastian Pokutta. Reinforcement learning under model mismatch. In Proc. Advances in Neural Information Processing Systems (NIPS), pages 3046–3055, 2017. ", + "bbox": [ + 173, + 712, + 825, + 742 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Jay K Satia and Roy E Lave Jr. Markovian decision processes with uncertain transition probabilities. Operations Research, 21(3):728–740, 1973. ", + "bbox": [ + 169, + 751, + 825, + 780 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "R. Srikant and Lei Ying. Finite-time error bounds for linear stochastic approximation and TD learning. In Proc. Annual Conference on Learning Theory (CoLT), pages 2803–2830, 2019. ", + "bbox": [ + 169, + 790, + 826, + 820 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Jun Sun, Gang Wang, Georgios B Giannakis, Qinmin Yang, and Zaiyue Yang. Finite-sample analysis of decentralized temporal-difference learning with linear function approximation. In Proc. International Conference on Artifical Intelligence and Statistics (AISTATS), 2020. ", + "bbox": [ + 174, + 829, + 823, + 872 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Richard S. Sutton and Andrew G. Barto. Reinforcement Learning: An Introduction, Second Edition. The MIT Press, Cambridge, Massachusetts, 2018. ", + "bbox": [ + 174, + 882, + 823, + 911 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Richard S Sutton, Csaba Szepesvári, and Hamid Reza Maei. A convergent O(n) algorithm for off-policy temporal-difference learning with linear function approximation. In Proc. Advances in Neural Information Processing Systems (NIPS), volume 21, pages 1609–1616. MIT Press, 2008. ", + "bbox": [ + 174, + 90, + 823, + 133 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Richard S Sutton, Hamid Reza Maei, Doina Precup, Shalabh Bhatnagar, David Silver, Csaba Szepesvári, and Eric Wiewiora. Fast gradient-descent methods for temporal-difference learning with linear function approximation. In Proc. International Conference on Machine Learning (ICML), pages 993–1000, 2009. ", + "bbox": [ + 173, + 141, + 825, + 198 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Aviv Tamar, Shie Mannor, and Huan Xu. Scaling up robust MDPs using function approximation. In Proc. International Conference on Machine Learning (ICML), pages 181–189. PMLR, 2014. ", + "bbox": [ + 174, + 207, + 821, + 237 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Eugene Vinitsky, Yuqing Du, Kanaad Parvate, Kathy Jang, Pieter Abbeel, and Alexandre Bayen. Robust reinforcement learning using adversarial populations. arXiv preprint arXiv:2008.01825, 2020. ", + "bbox": [ + 173, + 244, + 825, + 286 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Hoi-To Wai, Mingyi Hong, Zhuoran Yang, Zhaoran Wang, and Kexin Tang. Variance reduced policy evaluation with smooth function approximation. In Proc. Advances in Neural Information Processing Systems (NeurIPS), volume 32, pages 5784–5795, 2019. ", + "bbox": [ + 173, + 295, + 826, + 338 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Martin J Wainwright. Variance-reduced Q-learning is minimax optimal. arXiv preprint arXiv:1906.04697, 2019. ", + "bbox": [ + 173, + 347, + 825, + 376 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Lingxiao Wang, Qi Cai, Zhuoran Yang, and Zhaoran Wang. Neural policy gradient methods: Global optimality and rates of convergence. In Proc. International Conference on Learning Representations (ICLR), 2020. ", + "bbox": [ + 173, + 383, + 825, + 426 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Yue Wang and Shaofeng Zou. Finite-sample analysis of Greedy-GQ with linear function approximation under Markovian noise. In Proc. International Conference on Uncertainty in Artificial Intelligence (UAI), pages 11–20. PMLR, 2020. ", + "bbox": [ + 174, + 435, + 826, + 478 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Yue Wang, Shaofeng Zou, and Yi Zhou. Non-asymptotic analysis for two time-scale TDC with general smooth function approximation. In Proc. Advances in Neural Information Processing Systems (NeurIPS), volume 34, 2021. ", + "bbox": [ + 173, + 486, + 825, + 529 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Wolfram Wiesemann, Daniel Kuhn, and Berç Rustem. Robust Markov decision processes. Mathematics of Operations Research, 38(1):153–183, 2013. ", + "bbox": [ + 174, + 537, + 823, + 568 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Yue Wu, Weitong Zhang, Pan Xu, and Quanquan Gu. A finite time analysis of two time-scale actor critic methods. arXiv preprint arXiv:2005.01350, 2020. ", + "bbox": [ + 173, + 575, + 823, + 604 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Tengyu Xu and Yingbin Liang. Sample complexity bounds for two timescale value-based reinforcement learning algorithms. In Proc. International Conference on Artifical Intelligence and Statistics (AISTATS), pages 811–819. PMLR, 2021. ", + "bbox": [ + 174, + 613, + 823, + 656 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Tengyu Xu, Shaofeng Zou, and Yingbin Liang. Two time-scale off-policy TD learning: Nonasymptotic analysis over Markovian samples. In Proc. Advances in Neural Information Processing Systems (NeurIPS), pages 10633–10643, 2019. ", + "bbox": [ + 174, + 664, + 823, + 707 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Zhuoran Yang, Yongxin Chen, Mingyi Hong, and Zhaoran Wang. Provably global convergence of actor-critic: A case for linear quadratic regulator with ergodic cost. In Proc. Advances in Neural Information Processing Systems (NeurIPS), pages 8353–8365, 2019. ", + "bbox": [ + 176, + 714, + 825, + 758 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Kaiqing Zhang, Bin Hu, and Tamer Basar. On the stability and convergence of robust adversarial reinforcement learning: A case study on linear quadratic systems. In Proc. Advances in Neural Information Processing Systems (NeurIPS), volume 33, 2020a. ", + "bbox": [ + 174, + 766, + 823, + 809 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Kaiqing Zhang, Tao Sun, Yunzhe Tao, Sahika Genc, Sunil Mallya, and Tamer Basar. Robust multiagent reinforcement learning with model uncertainty. In Proc. Advances in Neural Information Processing Systems (NeurIPS), volume 33, 2020b. ", + "bbox": [ + 174, + 818, + 823, + 861 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Shaofeng Zou, Tengyu Xu, and Yingbin Liang. Finite-sample analysis for SARSA with linear function approximation. In Proc. Advances in Neural Information Processing Systems (NeurIPS), pages 8665–8675, 2019. ", + "bbox": [ + 174, + 868, + 826, + 911 + ], + "page_idx": 13 + } +] \ No newline at end of file diff --git a/parse/train/IhiU6AJYpDs/IhiU6AJYpDs_middle.json b/parse/train/IhiU6AJYpDs/IhiU6AJYpDs_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..883049af472c2514084c72816271915ea394ab8c --- /dev/null +++ b/parse/train/IhiU6AJYpDs/IhiU6AJYpDs_middle.json @@ -0,0 +1,44641 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 115, + 97, + 497, + 137 + ], + "lines": [ + { + "bbox": [ + 115, + 97, + 497, + 118 + ], + "spans": [ + { + "bbox": [ + 115, + 97, + 497, + 118 + ], + "score": 1.0, + "content": "Online Robust Reinforcement Learning with Model", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 260, + 116, + 352, + 140 + ], + "spans": [ + { + "bbox": [ + 260, + 116, + 352, + 140 + ], + "score": 1.0, + "content": "Uncertainty", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 172, + 179, + 279, + 223 + ], + "lines": [ + { + "bbox": [ + 202, + 177, + 250, + 194 + ], + "spans": [ + { + "bbox": [ + 202, + 177, + 250, + 194 + ], + "score": 1.0, + "content": "Yue Wang", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 182, + 190, + 270, + 202 + ], + "spans": [ + { + "bbox": [ + 182, + 190, + 270, + 202 + ], + "score": 1.0, + "content": "University at Buffalo", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 185, + 201, + 265, + 212 + ], + "spans": [ + { + "bbox": [ + 185, + 201, + 265, + 212 + ], + "score": 1.0, + "content": "Buffalo, NY 14228", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 171, + 212, + 279, + 224 + ], + "spans": [ + { + "bbox": [ + 171, + 212, + 279, + 224 + ], + "score": 1.0, + "content": "ywang294@buffalo.edu", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 5.0 + }, + { + "type": "text", + "bbox": [ + 349, + 180, + 438, + 223 + ], + "lines": [ + { + "bbox": [ + 362, + 179, + 425, + 191 + ], + "spans": [ + { + "bbox": [ + 362, + 179, + 425, + 191 + ], + "score": 1.0, + "content": "Shaofeng Zou", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 350, + 190, + 437, + 202 + ], + "spans": [ + { + "bbox": [ + 350, + 190, + 437, + 202 + ], + "score": 1.0, + "content": "University at Buffalo", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 353, + 201, + 434, + 212 + ], + "spans": [ + { + "bbox": [ + 353, + 201, + 434, + 212 + ], + "score": 1.0, + "content": "Buffalo, NY 14228", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 348, + 212, + 439, + 223 + ], + "spans": [ + { + "bbox": [ + 348, + 212, + 439, + 223 + ], + "score": 1.0, + "content": "szou3@buffalo.edu", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6.0 + }, + { + "type": "title", + "bbox": [ + 283, + 252, + 328, + 265 + ], + "lines": [ + { + "bbox": [ + 281, + 251, + 330, + 267 + ], + "spans": [ + { + "bbox": [ + 281, + 251, + 330, + 267 + ], + "score": 1.0, + "content": "Abstract", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 143, + 277, + 469, + 463 + ], + "lines": [ + { + "bbox": [ + 141, + 277, + 471, + 291 + ], + "spans": [ + { + "bbox": [ + 141, + 277, + 471, + 291 + ], + "score": 1.0, + "content": "Robust reinforcement learning (RL) is to find a policy that optimizes the worst-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 288, + 470, + 302 + ], + "spans": [ + { + "bbox": [ + 141, + 288, + 470, + 302 + ], + "score": 1.0, + "content": "case performance over an uncertainty set of MDPs. In this paper, we focus on", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 299, + 470, + 312 + ], + "spans": [ + { + "bbox": [ + 141, + 299, + 470, + 312 + ], + "score": 1.0, + "content": "model-free robust RL, where the uncertainty set is defined to be centering at a", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 310, + 470, + 324 + ], + "spans": [ + { + "bbox": [ + 141, + 310, + 470, + 324 + ], + "score": 1.0, + "content": "misspecified MDP that generates a single sample trajectory sequentially, and is", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 321, + 469, + 333 + ], + "spans": [ + { + "bbox": [ + 141, + 321, + 469, + 333 + ], + "score": 1.0, + "content": "assumed to be unknown. We develop a sample-based approach to estimate the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 332, + 470, + 344 + ], + "spans": [ + { + "bbox": [ + 141, + 332, + 470, + 344 + ], + "score": 1.0, + "content": "unknown uncertainty set, and design robust Q-learning algorithm (tabular case) and", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 343, + 469, + 356 + ], + "spans": [ + { + "bbox": [ + 141, + 343, + 469, + 356 + ], + "score": 1.0, + "content": "robust TDC algorithm (function approximation setting), which can be implemented", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 353, + 469, + 366 + ], + "spans": [ + { + "bbox": [ + 141, + 353, + 469, + 366 + ], + "score": 1.0, + "content": "in an online and incremental fashion. For the robust Q-learning algorithm, we prove", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 365, + 470, + 378 + ], + "spans": [ + { + "bbox": [ + 141, + 365, + 470, + 378 + ], + "score": 1.0, + "content": "that it converges to the optimal robust Q function, and for the robust TDC algorithm,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 376, + 470, + 389 + ], + "spans": [ + { + "bbox": [ + 141, + 376, + 470, + 389 + ], + "score": 1.0, + "content": "we prove that it converges asymptotically to some stationary points. Unlike the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 387, + 469, + 399 + ], + "spans": [ + { + "bbox": [ + 141, + 387, + 469, + 399 + ], + "score": 1.0, + "content": "results in [Roy et al., 2017], our algorithms do not need any additional conditions", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 398, + 470, + 411 + ], + "spans": [ + { + "bbox": [ + 141, + 398, + 470, + 411 + ], + "score": 1.0, + "content": "on the discount factor to guarantee the convergence. We further characterize the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 141, + 408, + 470, + 421 + ], + "spans": [ + { + "bbox": [ + 141, + 408, + 470, + 421 + ], + "score": 1.0, + "content": "finite-time error bounds of the two algorithms, and show that both the robust Q-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 420, + 470, + 432 + ], + "spans": [ + { + "bbox": [ + 141, + 420, + 470, + 432 + ], + "score": 1.0, + "content": "learning and robust TDC algorithms converge as fast as their vanilla counterparts", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 141, + 430, + 469, + 442 + ], + "spans": [ + { + "bbox": [ + 141, + 430, + 469, + 442 + ], + "score": 1.0, + "content": "(within a constant factor). Our numerical experiments further demonstrate the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 141, + 441, + 469, + 454 + ], + "spans": [ + { + "bbox": [ + 141, + 441, + 469, + 454 + ], + "score": 1.0, + "content": "robustness of our algorithms. Our approach can be readily extended to robustify", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 141, + 453, + 423, + 465 + ], + "spans": [ + { + "bbox": [ + 141, + 453, + 423, + 465 + ], + "score": 1.0, + "content": "many other algorithms, e.g., TD, SARSA, and other GTD algorithms.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 19 + }, + { + "type": "title", + "bbox": [ + 107, + 485, + 190, + 498 + ], + "lines": [ + { + "bbox": [ + 105, + 484, + 192, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 192, + 501 + ], + "score": 1.0, + "content": "1 Introduction", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 510, + 505, + 631 + ], + "lines": [ + { + "bbox": [ + 105, + 510, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 505, + 523 + ], + "score": 1.0, + "content": "Existing studies on Markov decision process (MDP) and reinforcement learning (RL) [Sutton and", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 520, + 505, + 535 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 505, + 535 + ], + "score": 1.0, + "content": "Barto, 2018] mostly rely on the crucial assumption that the environment on which a learned policy", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 533, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 505, + 545 + ], + "score": 1.0, + "content": "will be deployed is the same one that was used to generate the policy, which is often violated in", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 543, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 505, + 556 + ], + "score": 1.0, + "content": "practice – e.g., the simulator may be different from the true environment, and the MDP may evolve", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 554, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 505, + 567 + ], + "score": 1.0, + "content": "over time. Due to such model deviation, the actual performance of the learned policy can significantly", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 565, + 506, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 506, + 578 + ], + "score": 1.0, + "content": "degrade. To address this problem, the framework of robust MDP was formulated in [Bagnell et al.,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 576, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 506, + 589 + ], + "score": 1.0, + "content": "2001, Nilim and El Ghaoui, 2004, Iyengar, 2005], where the transition kernel of the MDP is not fixed", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 587, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 506, + 599 + ], + "score": 1.0, + "content": "and lies in an uncertainty set, and the goal is to learn a policy that performs well under the worst-case", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 597, + 506, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 506, + 610 + ], + "score": 1.0, + "content": "MDP in the uncertainty set. In [Bagnell et al., 2001, Nilim and El Ghaoui, 2004, Iyengar, 2005], it", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 609, + 505, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 505, + 622 + ], + "score": 1.0, + "content": "was assumed that the uncertainty set is known beforehand, i.e., model-based approach, and dynamic", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 619, + 344, + 633 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 344, + 633 + ], + "score": 1.0, + "content": "programming can be used to find the optimal robust policy.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 107, + 636, + 505, + 713 + ], + "lines": [ + { + "bbox": [ + 105, + 636, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 505, + 648 + ], + "score": 1.0, + "content": "The model-based approach, however, requires a model of the uncertainty set known beforehand, and", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 648, + 506, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 506, + 660 + ], + "score": 1.0, + "content": "needs a large memory to store the model when the state and action spaces are large, which make it", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 658, + 505, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 505, + 671 + ], + "score": 1.0, + "content": "less applicable for many practical scenarios. This motivates the study in this paper, model-free robust", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 668, + 506, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 506, + 682 + ], + "score": 1.0, + "content": "RL with model uncertainty, which is to learn a robust policy using a single sample trajectory from a", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 680, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 505, + 693 + ], + "score": 1.0, + "content": "misspecified MDP, e.g., a simulator and a similar environment in which samples are easier to collect", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 691, + 505, + 703 + ], + "spans": [ + { + "bbox": [ + 105, + 691, + 505, + 703 + ], + "score": 1.0, + "content": "than in the target environment where the policy is going to be deployed. The major challenge lies in", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 701, + 505, + 714 + ], + "spans": [ + { + "bbox": [ + 105, + 701, + 505, + 714 + ], + "score": 1.0, + "content": "that the transition kernel of the misspecified MDP is not given beforehand, and thus, the uncertainty", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 43 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 731, + 385, + 741 + ], + "lines": [ + { + "bbox": [ + 105, + 730, + 387, + 744 + ], + "spans": [ + { + "bbox": [ + 105, + 730, + 387, + 744 + ], + "score": 1.0, + "content": "35th Conference on Neural Information Processing Systems (NeurIPS 2021).", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 115, + 97, + 497, + 137 + ], + "lines": [ + { + "bbox": [ + 115, + 97, + 497, + 118 + ], + "spans": [ + { + "bbox": [ + 115, + 97, + 497, + 118 + ], + "score": 1.0, + "content": "Online Robust Reinforcement Learning with Model", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 260, + 116, + 352, + 140 + ], + "spans": [ + { + "bbox": [ + 260, + 116, + 352, + 140 + ], + "score": 1.0, + "content": "Uncertainty", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "list", + "bbox": [ + 172, + 179, + 279, + 223 + ], + "lines": [ + { + "bbox": [ + 202, + 177, + 250, + 194 + ], + "spans": [ + { + "bbox": [ + 202, + 177, + 250, + 194 + ], + "score": 1.0, + "content": "Yue Wang", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 182, + 190, + 270, + 202 + ], + "spans": [ + { + "bbox": [ + 182, + 190, + 270, + 202 + ], + "score": 1.0, + "content": "University at Buffalo", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 185, + 201, + 265, + 212 + ], + "spans": [ + { + "bbox": [ + 185, + 201, + 265, + 212 + ], + "score": 1.0, + "content": "Buffalo, NY 14228", + "type": "text" + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 171, + 212, + 279, + 224 + ], + "spans": [ + { + "bbox": [ + 171, + 212, + 279, + 224 + ], + "score": 1.0, + "content": "ywang294@buffalo.edu", + "type": "text" + } + ], + "index": 8, + "is_list_start_line": true + } + ], + "index": 5.0, + "bbox_fs": [ + 171, + 177, + 279, + 224 + ] + }, + { + "type": "text", + "bbox": [ + 349, + 180, + 438, + 223 + ], + "lines": [ + { + "bbox": [ + 362, + 179, + 425, + 191 + ], + "spans": [ + { + "bbox": [ + 362, + 179, + 425, + 191 + ], + "score": 1.0, + "content": "Shaofeng Zou", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 350, + 190, + 437, + 202 + ], + "spans": [ + { + "bbox": [ + 350, + 190, + 437, + 202 + ], + "score": 1.0, + "content": "University at Buffalo", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 353, + 201, + 434, + 212 + ], + "spans": [ + { + "bbox": [ + 353, + 201, + 434, + 212 + ], + "score": 1.0, + "content": "Buffalo, NY 14228", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 348, + 212, + 439, + 223 + ], + "spans": [ + { + "bbox": [ + 348, + 212, + 439, + 223 + ], + "score": 1.0, + "content": "szou3@buffalo.edu", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6.0, + "bbox_fs": [ + 348, + 179, + 439, + 223 + ] + }, + { + "type": "title", + "bbox": [ + 283, + 252, + 328, + 265 + ], + "lines": [ + { + "bbox": [ + 281, + 251, + 330, + 267 + ], + "spans": [ + { + "bbox": [ + 281, + 251, + 330, + 267 + ], + "score": 1.0, + "content": "Abstract", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 143, + 277, + 469, + 463 + ], + "lines": [ + { + "bbox": [ + 141, + 277, + 471, + 291 + ], + "spans": [ + { + "bbox": [ + 141, + 277, + 471, + 291 + ], + "score": 1.0, + "content": "Robust reinforcement learning (RL) is to find a policy that optimizes the worst-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 288, + 470, + 302 + ], + "spans": [ + { + "bbox": [ + 141, + 288, + 470, + 302 + ], + "score": 1.0, + "content": "case performance over an uncertainty set of MDPs. In this paper, we focus on", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 299, + 470, + 312 + ], + "spans": [ + { + "bbox": [ + 141, + 299, + 470, + 312 + ], + "score": 1.0, + "content": "model-free robust RL, where the uncertainty set is defined to be centering at a", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 310, + 470, + 324 + ], + "spans": [ + { + "bbox": [ + 141, + 310, + 470, + 324 + ], + "score": 1.0, + "content": "misspecified MDP that generates a single sample trajectory sequentially, and is", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 321, + 469, + 333 + ], + "spans": [ + { + "bbox": [ + 141, + 321, + 469, + 333 + ], + "score": 1.0, + "content": "assumed to be unknown. We develop a sample-based approach to estimate the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 332, + 470, + 344 + ], + "spans": [ + { + "bbox": [ + 141, + 332, + 470, + 344 + ], + "score": 1.0, + "content": "unknown uncertainty set, and design robust Q-learning algorithm (tabular case) and", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 343, + 469, + 356 + ], + "spans": [ + { + "bbox": [ + 141, + 343, + 469, + 356 + ], + "score": 1.0, + "content": "robust TDC algorithm (function approximation setting), which can be implemented", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 141, + 353, + 469, + 366 + ], + "spans": [ + { + "bbox": [ + 141, + 353, + 469, + 366 + ], + "score": 1.0, + "content": "in an online and incremental fashion. For the robust Q-learning algorithm, we prove", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 365, + 470, + 378 + ], + "spans": [ + { + "bbox": [ + 141, + 365, + 470, + 378 + ], + "score": 1.0, + "content": "that it converges to the optimal robust Q function, and for the robust TDC algorithm,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 376, + 470, + 389 + ], + "spans": [ + { + "bbox": [ + 141, + 376, + 470, + 389 + ], + "score": 1.0, + "content": "we prove that it converges asymptotically to some stationary points. Unlike the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 387, + 469, + 399 + ], + "spans": [ + { + "bbox": [ + 141, + 387, + 469, + 399 + ], + "score": 1.0, + "content": "results in [Roy et al., 2017], our algorithms do not need any additional conditions", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 141, + 398, + 470, + 411 + ], + "spans": [ + { + "bbox": [ + 141, + 398, + 470, + 411 + ], + "score": 1.0, + "content": "on the discount factor to guarantee the convergence. We further characterize the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 141, + 408, + 470, + 421 + ], + "spans": [ + { + "bbox": [ + 141, + 408, + 470, + 421 + ], + "score": 1.0, + "content": "finite-time error bounds of the two algorithms, and show that both the robust Q-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 420, + 470, + 432 + ], + "spans": [ + { + "bbox": [ + 141, + 420, + 470, + 432 + ], + "score": 1.0, + "content": "learning and robust TDC algorithms converge as fast as their vanilla counterparts", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 141, + 430, + 469, + 442 + ], + "spans": [ + { + "bbox": [ + 141, + 430, + 469, + 442 + ], + "score": 1.0, + "content": "(within a constant factor). Our numerical experiments further demonstrate the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 141, + 441, + 469, + 454 + ], + "spans": [ + { + "bbox": [ + 141, + 441, + 469, + 454 + ], + "score": 1.0, + "content": "robustness of our algorithms. Our approach can be readily extended to robustify", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 141, + 453, + 423, + 465 + ], + "spans": [ + { + "bbox": [ + 141, + 453, + 423, + 465 + ], + "score": 1.0, + "content": "many other algorithms, e.g., TD, SARSA, and other GTD algorithms.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 19, + "bbox_fs": [ + 141, + 277, + 471, + 465 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 485, + 190, + 498 + ], + "lines": [ + { + "bbox": [ + 105, + 484, + 192, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 192, + 501 + ], + "score": 1.0, + "content": "1 Introduction", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 510, + 505, + 631 + ], + "lines": [ + { + "bbox": [ + 105, + 510, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 505, + 523 + ], + "score": 1.0, + "content": "Existing studies on Markov decision process (MDP) and reinforcement learning (RL) [Sutton and", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 520, + 505, + 535 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 505, + 535 + ], + "score": 1.0, + "content": "Barto, 2018] mostly rely on the crucial assumption that the environment on which a learned policy", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 533, + 505, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 505, + 545 + ], + "score": 1.0, + "content": "will be deployed is the same one that was used to generate the policy, which is often violated in", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 543, + 505, + 556 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 505, + 556 + ], + "score": 1.0, + "content": "practice – e.g., the simulator may be different from the true environment, and the MDP may evolve", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 554, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 554, + 505, + 567 + ], + "score": 1.0, + "content": "over time. Due to such model deviation, the actual performance of the learned policy can significantly", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 565, + 506, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 506, + 578 + ], + "score": 1.0, + "content": "degrade. To address this problem, the framework of robust MDP was formulated in [Bagnell et al.,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 576, + 506, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 506, + 589 + ], + "score": 1.0, + "content": "2001, Nilim and El Ghaoui, 2004, Iyengar, 2005], where the transition kernel of the MDP is not fixed", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 587, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 506, + 599 + ], + "score": 1.0, + "content": "and lies in an uncertainty set, and the goal is to learn a policy that performs well under the worst-case", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 597, + 506, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 506, + 610 + ], + "score": 1.0, + "content": "MDP in the uncertainty set. In [Bagnell et al., 2001, Nilim and El Ghaoui, 2004, Iyengar, 2005], it", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 609, + 505, + 622 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 505, + 622 + ], + "score": 1.0, + "content": "was assumed that the uncertainty set is known beforehand, i.e., model-based approach, and dynamic", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 619, + 344, + 633 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 344, + 633 + ], + "score": 1.0, + "content": "programming can be used to find the optimal robust policy.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 510, + 506, + 633 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 636, + 505, + 713 + ], + "lines": [ + { + "bbox": [ + 105, + 636, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 505, + 648 + ], + "score": 1.0, + "content": "The model-based approach, however, requires a model of the uncertainty set known beforehand, and", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 648, + 506, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 506, + 660 + ], + "score": 1.0, + "content": "needs a large memory to store the model when the state and action spaces are large, which make it", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 658, + 505, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 505, + 671 + ], + "score": 1.0, + "content": "less applicable for many practical scenarios. This motivates the study in this paper, model-free robust", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 668, + 506, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 506, + 682 + ], + "score": 1.0, + "content": "RL with model uncertainty, which is to learn a robust policy using a single sample trajectory from a", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 680, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 505, + 693 + ], + "score": 1.0, + "content": "misspecified MDP, e.g., a simulator and a similar environment in which samples are easier to collect", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 691, + 505, + 703 + ], + "spans": [ + { + "bbox": [ + 105, + 691, + 505, + 703 + ], + "score": 1.0, + "content": "than in the target environment where the policy is going to be deployed. The major challenge lies in", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 701, + 505, + 714 + ], + "spans": [ + { + "bbox": [ + 105, + 701, + 505, + 714 + ], + "score": 1.0, + "content": "that the transition kernel of the misspecified MDP is not given beforehand, and thus, the uncertainty", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 73, + 505, + 84 + ], + "spans": [ + { + "bbox": [ + 105, + 73, + 505, + 84 + ], + "score": 1.0, + "content": "set and the optimal robust policy need to be learned simultaneously using sequentially observed data", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 83, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 505, + 96 + ], + "score": 1.0, + "content": "from the misspecified MDP. Moreover, robust RL learns the value function of the worst-case MDP", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 94, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 506, + 107 + ], + "score": 1.0, + "content": "in the uncertainty set which is different from the misspecified MDP that generates samples. This is", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 106, + 506, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 106, + 506, + 118 + ], + "score": 1.0, + "content": "similar to the off-policy learning, which we refer to as the \"off-transition-kernel\" setting. Therefore,", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 115, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 506, + 129 + ], + "score": 1.0, + "content": "the learning may be unstable and could diverge especially when function approximation is used", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 127, + 166, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 166, + 139 + ], + "score": 1.0, + "content": "[Baird, 1995].", + "type": "text", + "cross_page": true + } + ], + "index": 5 + } + ], + "index": 43, + "bbox_fs": [ + 105, + 636, + 506, + 714 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 73, + 505, + 138 + ], + "lines": [ + { + "bbox": [ + 105, + 73, + 505, + 84 + ], + "spans": [ + { + "bbox": [ + 105, + 73, + 505, + 84 + ], + "score": 1.0, + "content": "set and the optimal robust policy need to be learned simultaneously using sequentially observed data", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 83, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 505, + 96 + ], + "score": 1.0, + "content": "from the misspecified MDP. Moreover, robust RL learns the value function of the worst-case MDP", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 94, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 506, + 107 + ], + "score": 1.0, + "content": "in the uncertainty set which is different from the misspecified MDP that generates samples. This is", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 106, + 506, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 106, + 506, + 118 + ], + "score": 1.0, + "content": "similar to the off-policy learning, which we refer to as the \"off-transition-kernel\" setting. Therefore,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 115, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 506, + 129 + ], + "score": 1.0, + "content": "the learning may be unstable and could diverge especially when function approximation is used", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 127, + 166, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 166, + 139 + ], + "score": 1.0, + "content": "[Baird, 1995].", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 144, + 504, + 165 + ], + "lines": [ + { + "bbox": [ + 105, + 142, + 506, + 157 + ], + "spans": [ + { + "bbox": [ + 105, + 142, + 506, + 157 + ], + "score": 1.0, + "content": "In this paper, we develop a model-free approach for robust RL with model uncertainty. Our major", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 155, + 325, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 155, + 325, + 167 + ], + "score": 1.0, + "content": "contributions in this paper are summarized as follows.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 106, + 175, + 505, + 263 + ], + "lines": [ + { + "bbox": [ + 105, + 174, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 506, + 189 + ], + "score": 1.0, + "content": "• Motivated by empirical studies of adversarial training in RL [Huang et al., 2017, Kos and Song,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 113, + 186, + 505, + 198 + ], + "spans": [ + { + "bbox": [ + 113, + 186, + 409, + 198 + ], + "score": 1.0, + "content": "2017, Lin et al., 2017, Pattanaik et al., 2018, Mandlekar et al., 2017] and the", + "type": "text" + }, + { + "bbox": [ + 410, + 187, + 418, + 196 + ], + "score": 0.81, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 186, + 505, + 198 + ], + "score": 1.0, + "content": "-contamination model", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 114, + 198, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 114, + 198, + 220, + 210 + ], + "score": 1.0, + "content": "in robust detection (called", + "type": "text" + }, + { + "bbox": [ + 220, + 199, + 225, + 207 + ], + "score": 0.65, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 198, + 505, + 210 + ], + "score": 1.0, + "content": "-contamination model in [Huber, 1965]), we design the uncertainty set", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 114, + 208, + 505, + 221 + ], + "spans": [ + { + "bbox": [ + 114, + 208, + 153, + 221 + ], + "score": 1.0, + "content": "using the", + "type": "text" + }, + { + "bbox": [ + 153, + 209, + 162, + 218 + ], + "score": 0.75, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 208, + 505, + 221 + ], + "score": 1.0, + "content": "-contamination model (see (4) for the details). We then develop an approach to estimate", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 114, + 219, + 505, + 231 + ], + "spans": [ + { + "bbox": [ + 114, + 219, + 505, + 231 + ], + "score": 1.0, + "content": "the unknown uncertainty set using only the current sample, which does not incur any additional", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 114, + 230, + 505, + 242 + ], + "spans": [ + { + "bbox": [ + 114, + 230, + 505, + 242 + ], + "score": 1.0, + "content": "memory cost. Unlike the approach in [Roy et al., 2017], where the uncertainty set is relaxed to one", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 114, + 241, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 114, + 241, + 505, + 253 + ], + "score": 1.0, + "content": "not depending on the misspecified MDP that generates samples so that an online algorithm can be", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 114, + 252, + 389, + 264 + ], + "spans": [ + { + "bbox": [ + 114, + 252, + 389, + 264 + ], + "score": 1.0, + "content": "constructed, our approach does not need to relax the uncertainty set.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 107, + 267, + 505, + 366 + ], + "lines": [ + { + "bbox": [ + 105, + 266, + 505, + 280 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 505, + 280 + ], + "score": 1.0, + "content": "• We develop a robust Q-learning algorithm for the tabular case, which can be implemented in an", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 114, + 279, + 506, + 290 + ], + "spans": [ + { + "bbox": [ + 114, + 279, + 506, + 290 + ], + "score": 1.0, + "content": "online and incremental fashion, and has the same memory cost as the vanilla Q-learning algorithm.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 113, + 289, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 113, + 289, + 505, + 302 + ], + "score": 1.0, + "content": "We show that our robust Q-learning algorithm converges asymptotically, and further characterize", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 114, + 300, + 505, + 313 + ], + "spans": [ + { + "bbox": [ + 114, + 300, + 505, + 313 + ], + "score": 1.0, + "content": "its finite-time error bound. Unlike the results in [Roy et al., 2017] where a stringent condition", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 113, + 310, + 506, + 324 + ], + "spans": [ + { + "bbox": [ + 113, + 310, + 506, + 324 + ], + "score": 1.0, + "content": "on the discount factor (which is due to the relaxation of the uncertainty set, and prevents the use", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 113, + 321, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 113, + 321, + 505, + 335 + ], + "score": 1.0, + "content": "of a discount factor close to 1 in practice) is needed to guarantee the convergence, our algorithm", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 334, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 115, + 334, + 505, + 345 + ], + "score": 1.0, + "content": "converges without the need of such condition. Furthermore, our robust Q-learning algorithm", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 114, + 343, + 507, + 357 + ], + "spans": [ + { + "bbox": [ + 114, + 343, + 507, + 357 + ], + "score": 1.0, + "content": "converges as fast as the vanilla Q-learning algorithm [Li et al., 2020] (within a constant factor),", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 113, + 354, + 278, + 369 + ], + "spans": [ + { + "bbox": [ + 113, + 354, + 278, + 369 + ], + "score": 1.0, + "content": "while being robust to model uncertainty.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 371, + 505, + 480 + ], + "lines": [ + { + "bbox": [ + 105, + 370, + 506, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 506, + 384 + ], + "score": 1.0, + "content": "• We generalize our approach to the case with function approximation (for large state/action space).", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 113, + 382, + 506, + 394 + ], + "spans": [ + { + "bbox": [ + 113, + 382, + 506, + 394 + ], + "score": 1.0, + "content": "We investigate the robust policy evaluation problem, i.e., evaluate a given policy under the worst-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 114, + 393, + 506, + 405 + ], + "spans": [ + { + "bbox": [ + 114, + 393, + 506, + 405 + ], + "score": 1.0, + "content": "case MDP in the uncertainty set. As mentioned before, the robust RL problem is essentially", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 114, + 402, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 114, + 402, + 505, + 417 + ], + "score": 1.0, + "content": "\"off-transition-kernel\", and therefore non-robust methods with function approximation may diverge", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 114, + 414, + 506, + 427 + ], + "spans": [ + { + "bbox": [ + 114, + 414, + 506, + 427 + ], + "score": 1.0, + "content": "[Baird, 1995] (also see our experiments). We develop a novel extension of the gradient TD (GTD)", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 114, + 425, + 506, + 438 + ], + "spans": [ + { + "bbox": [ + 114, + 425, + 506, + 438 + ], + "score": 1.0, + "content": "method [Maei et al., 2010, Maei, 2011, Sutton et al., 2008] to robust RL. Our approach introduces a", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 114, + 437, + 506, + 448 + ], + "spans": [ + { + "bbox": [ + 114, + 437, + 506, + 448 + ], + "score": 1.0, + "content": "novel smoothed robust Bellman operator to construct the smoothed mean-squared projected robust", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 113, + 447, + 506, + 460 + ], + "spans": [ + { + "bbox": [ + 113, + 447, + 506, + 460 + ], + "score": 1.0, + "content": "Bellman error (MSPRBE). Using our uncertainty set design and online sample-based estimation,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 113, + 457, + 506, + 470 + ], + "spans": [ + { + "bbox": [ + 113, + 457, + 506, + 470 + ], + "score": 1.0, + "content": "we develop a two time-scale robust TDC algorithm. We further characterize its convergence and", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 114, + 469, + 210, + 480 + ], + "spans": [ + { + "bbox": [ + 114, + 469, + 210, + 480 + ], + "score": 1.0, + "content": "finite-time error bound.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 107, + 484, + 506, + 528 + ], + "lines": [ + { + "bbox": [ + 105, + 483, + 507, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 507, + 498 + ], + "score": 1.0, + "content": "• We conduct numerical experiments to validate the robustness of our approach. In our experiments,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 114, + 495, + 507, + 509 + ], + "spans": [ + { + "bbox": [ + 114, + 495, + 507, + 509 + ], + "score": 1.0, + "content": "our robust Q-learning algorithm achieves a much higher reward than the vanilla Q-learning algo-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 114, + 507, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 114, + 507, + 505, + 519 + ], + "score": 1.0, + "content": "rithm when being trained on a misspecified MDP; and our robust TDC algorithm converges much", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 113, + 516, + 469, + 531 + ], + "spans": [ + { + "bbox": [ + 113, + 516, + 469, + 531 + ], + "score": 1.0, + "content": "faster than the vanilla TDC algorithm, and the vanilla TDC algorithm may even diverge.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36.5 + }, + { + "type": "title", + "bbox": [ + 107, + 542, + 190, + 554 + ], + "lines": [ + { + "bbox": [ + 105, + 541, + 191, + 556 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 191, + 556 + ], + "score": 1.0, + "content": "1.1 Related Work", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 106, + 563, + 505, + 651 + ], + "lines": [ + { + "bbox": [ + 105, + 563, + 506, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 506, + 576 + ], + "score": 1.0, + "content": "Model-Based Robust MDP. The framework of robust MDP was investigated in [Iyengar, 2005,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 574, + 506, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 506, + 586 + ], + "score": 1.0, + "content": "Nilim and El Ghaoui, 2004, Bagnell et al., 2001, Satia and Lave Jr, 1973, Wiesemann et al., 2013],", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 585, + 505, + 597 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 505, + 597 + ], + "score": 1.0, + "content": "where the transition kernel is assumed to be in some uncertainty set, and the problem can be solved by", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 596, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 505, + 608 + ], + "score": 1.0, + "content": "dynamic programming. This approach was further extended to the case with function approximation", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 104, + 606, + 506, + 620 + ], + "spans": [ + { + "bbox": [ + 104, + 606, + 506, + 620 + ], + "score": 1.0, + "content": "in [Tamar et al., 2014]. However, these studies are model-based, which assume beforehand knowledge", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 617, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 505, + 630 + ], + "score": 1.0, + "content": "of the uncertainty set. In this paper, we investigate the model-free setting, where the uncertainty set is", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 104, + 628, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 104, + 628, + 505, + 641 + ], + "score": 1.0, + "content": "a set of MDPs centered around some unknown Markov transition kernel from which a single sample", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 640, + 267, + 652 + ], + "spans": [ + { + "bbox": [ + 106, + 640, + 267, + 652 + ], + "score": 1.0, + "content": "trajectory can be sequentially observed.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 43.5 + }, + { + "type": "text", + "bbox": [ + 107, + 656, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 655, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 505, + 668 + ], + "score": 1.0, + "content": "Adversarial Robust RL. It was shown in [Iyengar, 2005] that the robust MDP problem is equivalent", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 104, + 666, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 104, + 666, + 506, + 681 + ], + "score": 1.0, + "content": "to a zero-sum game between the agent and the nature. Motivated by this fact, the adversarial training", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "score": 1.0, + "content": "approach, where an adversary perturbs the state transition, was studied in [Vinitsky et al., 2020, Pinto", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 688, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 506, + 701 + ], + "score": 1.0, + "content": "et al., 2017, Abdullah et al., 2019, Hou et al., 2020, Rajeswaran et al., 2017, Atkeson and Morimoto,", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "score": 1.0, + "content": "2003, Morimoto and Doya, 2005]. This method relies on a simulator, where the state transition can", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "score": 1.0, + "content": "be modified in an arbitrary way. Another approach is to modify the current state through adversarial", + "type": "text" + } + ], + "index": 53 + } + ], + "index": 50.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 742, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 741, + 310, + 753 + ], + "spans": [ + { + "bbox": [ + 301, + 741, + 310, + 753 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 73, + 505, + 138 + ], + "lines": [], + "index": 2.5, + "bbox_fs": [ + 105, + 73, + 506, + 139 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 144, + 504, + 165 + ], + "lines": [ + { + "bbox": [ + 105, + 142, + 506, + 157 + ], + "spans": [ + { + "bbox": [ + 105, + 142, + 506, + 157 + ], + "score": 1.0, + "content": "In this paper, we develop a model-free approach for robust RL with model uncertainty. Our major", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 155, + 325, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 155, + 325, + 167 + ], + "score": 1.0, + "content": "contributions in this paper are summarized as follows.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5, + "bbox_fs": [ + 105, + 142, + 506, + 167 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 175, + 505, + 263 + ], + "lines": [ + { + "bbox": [ + 105, + 174, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 506, + 189 + ], + "score": 1.0, + "content": "• Motivated by empirical studies of adversarial training in RL [Huang et al., 2017, Kos and Song,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 113, + 186, + 505, + 198 + ], + "spans": [ + { + "bbox": [ + 113, + 186, + 409, + 198 + ], + "score": 1.0, + "content": "2017, Lin et al., 2017, Pattanaik et al., 2018, Mandlekar et al., 2017] and the", + "type": "text" + }, + { + "bbox": [ + 410, + 187, + 418, + 196 + ], + "score": 0.81, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 186, + 505, + 198 + ], + "score": 1.0, + "content": "-contamination model", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 114, + 198, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 114, + 198, + 220, + 210 + ], + "score": 1.0, + "content": "in robust detection (called", + "type": "text" + }, + { + "bbox": [ + 220, + 199, + 225, + 207 + ], + "score": 0.65, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 198, + 505, + 210 + ], + "score": 1.0, + "content": "-contamination model in [Huber, 1965]), we design the uncertainty set", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 114, + 208, + 505, + 221 + ], + "spans": [ + { + "bbox": [ + 114, + 208, + 153, + 221 + ], + "score": 1.0, + "content": "using the", + "type": "text" + }, + { + "bbox": [ + 153, + 209, + 162, + 218 + ], + "score": 0.75, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 208, + 505, + 221 + ], + "score": 1.0, + "content": "-contamination model (see (4) for the details). We then develop an approach to estimate", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 114, + 219, + 505, + 231 + ], + "spans": [ + { + "bbox": [ + 114, + 219, + 505, + 231 + ], + "score": 1.0, + "content": "the unknown uncertainty set using only the current sample, which does not incur any additional", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 114, + 230, + 505, + 242 + ], + "spans": [ + { + "bbox": [ + 114, + 230, + 505, + 242 + ], + "score": 1.0, + "content": "memory cost. Unlike the approach in [Roy et al., 2017], where the uncertainty set is relaxed to one", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 114, + 241, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 114, + 241, + 505, + 253 + ], + "score": 1.0, + "content": "not depending on the misspecified MDP that generates samples so that an online algorithm can be", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 114, + 252, + 389, + 264 + ], + "spans": [ + { + "bbox": [ + 114, + 252, + 389, + 264 + ], + "score": 1.0, + "content": "constructed, our approach does not need to relax the uncertainty set.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 174, + 506, + 264 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 267, + 505, + 366 + ], + "lines": [ + { + "bbox": [ + 105, + 266, + 505, + 280 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 505, + 280 + ], + "score": 1.0, + "content": "• We develop a robust Q-learning algorithm for the tabular case, which can be implemented in an", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 114, + 279, + 506, + 290 + ], + "spans": [ + { + "bbox": [ + 114, + 279, + 506, + 290 + ], + "score": 1.0, + "content": "online and incremental fashion, and has the same memory cost as the vanilla Q-learning algorithm.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 113, + 289, + 505, + 302 + ], + "spans": [ + { + "bbox": [ + 113, + 289, + 505, + 302 + ], + "score": 1.0, + "content": "We show that our robust Q-learning algorithm converges asymptotically, and further characterize", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 114, + 300, + 505, + 313 + ], + "spans": [ + { + "bbox": [ + 114, + 300, + 505, + 313 + ], + "score": 1.0, + "content": "its finite-time error bound. Unlike the results in [Roy et al., 2017] where a stringent condition", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 113, + 310, + 506, + 324 + ], + "spans": [ + { + "bbox": [ + 113, + 310, + 506, + 324 + ], + "score": 1.0, + "content": "on the discount factor (which is due to the relaxation of the uncertainty set, and prevents the use", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 113, + 321, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 113, + 321, + 505, + 335 + ], + "score": 1.0, + "content": "of a discount factor close to 1 in practice) is needed to guarantee the convergence, our algorithm", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 334, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 115, + 334, + 505, + 345 + ], + "score": 1.0, + "content": "converges without the need of such condition. Furthermore, our robust Q-learning algorithm", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 114, + 343, + 507, + 357 + ], + "spans": [ + { + "bbox": [ + 114, + 343, + 507, + 357 + ], + "score": 1.0, + "content": "converges as fast as the vanilla Q-learning algorithm [Li et al., 2020] (within a constant factor),", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 113, + 354, + 278, + 369 + ], + "spans": [ + { + "bbox": [ + 113, + 354, + 278, + 369 + ], + "score": 1.0, + "content": "while being robust to model uncertainty.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 266, + 507, + 369 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 371, + 505, + 480 + ], + "lines": [ + { + "bbox": [ + 105, + 370, + 506, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 506, + 384 + ], + "score": 1.0, + "content": "• We generalize our approach to the case with function approximation (for large state/action space).", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 113, + 382, + 506, + 394 + ], + "spans": [ + { + "bbox": [ + 113, + 382, + 506, + 394 + ], + "score": 1.0, + "content": "We investigate the robust policy evaluation problem, i.e., evaluate a given policy under the worst-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 114, + 393, + 506, + 405 + ], + "spans": [ + { + "bbox": [ + 114, + 393, + 506, + 405 + ], + "score": 1.0, + "content": "case MDP in the uncertainty set. As mentioned before, the robust RL problem is essentially", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 114, + 402, + 505, + 417 + ], + "spans": [ + { + "bbox": [ + 114, + 402, + 505, + 417 + ], + "score": 1.0, + "content": "\"off-transition-kernel\", and therefore non-robust methods with function approximation may diverge", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 114, + 414, + 506, + 427 + ], + "spans": [ + { + "bbox": [ + 114, + 414, + 506, + 427 + ], + "score": 1.0, + "content": "[Baird, 1995] (also see our experiments). We develop a novel extension of the gradient TD (GTD)", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 114, + 425, + 506, + 438 + ], + "spans": [ + { + "bbox": [ + 114, + 425, + 506, + 438 + ], + "score": 1.0, + "content": "method [Maei et al., 2010, Maei, 2011, Sutton et al., 2008] to robust RL. Our approach introduces a", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 114, + 437, + 506, + 448 + ], + "spans": [ + { + "bbox": [ + 114, + 437, + 506, + 448 + ], + "score": 1.0, + "content": "novel smoothed robust Bellman operator to construct the smoothed mean-squared projected robust", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 113, + 447, + 506, + 460 + ], + "spans": [ + { + "bbox": [ + 113, + 447, + 506, + 460 + ], + "score": 1.0, + "content": "Bellman error (MSPRBE). Using our uncertainty set design and online sample-based estimation,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 113, + 457, + 506, + 470 + ], + "spans": [ + { + "bbox": [ + 113, + 457, + 506, + 470 + ], + "score": 1.0, + "content": "we develop a two time-scale robust TDC algorithm. We further characterize its convergence and", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 114, + 469, + 210, + 480 + ], + "spans": [ + { + "bbox": [ + 114, + 469, + 210, + 480 + ], + "score": 1.0, + "content": "finite-time error bound.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 370, + 506, + 480 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 484, + 506, + 528 + ], + "lines": [ + { + "bbox": [ + 105, + 483, + 507, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 507, + 498 + ], + "score": 1.0, + "content": "• We conduct numerical experiments to validate the robustness of our approach. In our experiments,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 114, + 495, + 507, + 509 + ], + "spans": [ + { + "bbox": [ + 114, + 495, + 507, + 509 + ], + "score": 1.0, + "content": "our robust Q-learning algorithm achieves a much higher reward than the vanilla Q-learning algo-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 114, + 507, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 114, + 507, + 505, + 519 + ], + "score": 1.0, + "content": "rithm when being trained on a misspecified MDP; and our robust TDC algorithm converges much", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 113, + 516, + 469, + 531 + ], + "spans": [ + { + "bbox": [ + 113, + 516, + 469, + 531 + ], + "score": 1.0, + "content": "faster than the vanilla TDC algorithm, and the vanilla TDC algorithm may even diverge.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36.5, + "bbox_fs": [ + 105, + 483, + 507, + 531 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 542, + 190, + 554 + ], + "lines": [ + { + "bbox": [ + 105, + 541, + 191, + 556 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 191, + 556 + ], + "score": 1.0, + "content": "1.1 Related Work", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 106, + 563, + 505, + 651 + ], + "lines": [ + { + "bbox": [ + 105, + 563, + 506, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 506, + 576 + ], + "score": 1.0, + "content": "Model-Based Robust MDP. The framework of robust MDP was investigated in [Iyengar, 2005,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 574, + 506, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 506, + 586 + ], + "score": 1.0, + "content": "Nilim and El Ghaoui, 2004, Bagnell et al., 2001, Satia and Lave Jr, 1973, Wiesemann et al., 2013],", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 585, + 505, + 597 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 505, + 597 + ], + "score": 1.0, + "content": "where the transition kernel is assumed to be in some uncertainty set, and the problem can be solved by", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 596, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 505, + 608 + ], + "score": 1.0, + "content": "dynamic programming. This approach was further extended to the case with function approximation", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 104, + 606, + 506, + 620 + ], + "spans": [ + { + "bbox": [ + 104, + 606, + 506, + 620 + ], + "score": 1.0, + "content": "in [Tamar et al., 2014]. However, these studies are model-based, which assume beforehand knowledge", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 617, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 505, + 630 + ], + "score": 1.0, + "content": "of the uncertainty set. In this paper, we investigate the model-free setting, where the uncertainty set is", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 104, + 628, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 104, + 628, + 505, + 641 + ], + "score": 1.0, + "content": "a set of MDPs centered around some unknown Markov transition kernel from which a single sample", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 640, + 267, + 652 + ], + "spans": [ + { + "bbox": [ + 106, + 640, + 267, + 652 + ], + "score": 1.0, + "content": "trajectory can be sequentially observed.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 43.5, + "bbox_fs": [ + 104, + 563, + 506, + 652 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 656, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 655, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 505, + 668 + ], + "score": 1.0, + "content": "Adversarial Robust RL. It was shown in [Iyengar, 2005] that the robust MDP problem is equivalent", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 104, + 666, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 104, + 666, + 506, + 681 + ], + "score": 1.0, + "content": "to a zero-sum game between the agent and the nature. Motivated by this fact, the adversarial training", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "score": 1.0, + "content": "approach, where an adversary perturbs the state transition, was studied in [Vinitsky et al., 2020, Pinto", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 688, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 506, + 701 + ], + "score": 1.0, + "content": "et al., 2017, Abdullah et al., 2019, Hou et al., 2020, Rajeswaran et al., 2017, Atkeson and Morimoto,", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "score": 1.0, + "content": "2003, Morimoto and Doya, 2005]. This method relies on a simulator, where the state transition can", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "score": 1.0, + "content": "be modified in an arbitrary way. Another approach is to modify the current state through adversarial", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 105, + 72, + 507, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 507, + 86 + ], + "score": 1.0, + "content": "samples, which is more heuristic, e.g., [Huang et al., 2017, Kos and Song, 2017, Lin et al., 2017,", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 83, + 507, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 507, + 97 + ], + "score": 1.0, + "content": "Pattanaik et al., 2018, Mandlekar et al., 2017]. Despite the empirical success of these approaches,", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 94, + 506, + 108 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 506, + 108 + ], + "score": 1.0, + "content": "theoretical performance guarantees, e.g., convergence to the optimal robust policy and convergence", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 106, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 505, + 118 + ], + "score": 1.0, + "content": "rate, are yet to be established. The main difference lies in that during the training, our approach does", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 117, + 504, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 117, + 504, + 128 + ], + "score": 1.0, + "content": "not need to manipulate the state transition of the MDP. More importantly, we develop the asymptotic", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 128, + 505, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 128, + 505, + 139 + ], + "score": 1.0, + "content": "convergence to the optimal robust policy and further characterize the finite-time error bound. In [Lim", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 137, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 505, + 151 + ], + "score": 1.0, + "content": "et al., 2013], the scenario where some unknown parts of the state space can have arbitrary transitions", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 149, + 506, + 162 + ], + "spans": [ + { + "bbox": [ + 106, + 149, + 506, + 162 + ], + "score": 1.0, + "content": "while other parts are purely stochastic was studied. Adaptive algorithm to adversarial behavior was", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 159, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 159, + 506, + 173 + ], + "score": 1.0, + "content": "designed, and its regret bound is shown to be similar to the purely stochastic case. In [Zhang et al.,", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 170, + 492, + 183 + ], + "spans": [ + { + "bbox": [ + 105, + 170, + 492, + 183 + ], + "score": 1.0, + "content": "2020a], the robust adversarial RL problem for the special linear quadratic case was investigated.", + "type": "text", + "cross_page": true + } + ], + "index": 9 + } + ], + "index": 50.5, + "bbox_fs": [ + 104, + 655, + 506, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 72, + 505, + 182 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 507, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 507, + 86 + ], + "score": 1.0, + "content": "samples, which is more heuristic, e.g., [Huang et al., 2017, Kos and Song, 2017, Lin et al., 2017,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 83, + 507, + 97 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 507, + 97 + ], + "score": 1.0, + "content": "Pattanaik et al., 2018, Mandlekar et al., 2017]. Despite the empirical success of these approaches,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 94, + 506, + 108 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 506, + 108 + ], + "score": 1.0, + "content": "theoretical performance guarantees, e.g., convergence to the optimal robust policy and convergence", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 106, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 505, + 118 + ], + "score": 1.0, + "content": "rate, are yet to be established. The main difference lies in that during the training, our approach does", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 117, + 504, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 117, + 504, + 128 + ], + "score": 1.0, + "content": "not need to manipulate the state transition of the MDP. More importantly, we develop the asymptotic", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 128, + 505, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 128, + 505, + 139 + ], + "score": 1.0, + "content": "convergence to the optimal robust policy and further characterize the finite-time error bound. In [Lim", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 137, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 505, + 151 + ], + "score": 1.0, + "content": "et al., 2013], the scenario where some unknown parts of the state space can have arbitrary transitions", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 149, + 506, + 162 + ], + "spans": [ + { + "bbox": [ + 106, + 149, + 506, + 162 + ], + "score": 1.0, + "content": "while other parts are purely stochastic was studied. Adaptive algorithm to adversarial behavior was", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 159, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 159, + 506, + 173 + ], + "score": 1.0, + "content": "designed, and its regret bound is shown to be similar to the purely stochastic case. In [Zhang et al.,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 170, + 492, + 183 + ], + "spans": [ + { + "bbox": [ + 105, + 170, + 492, + 183 + ], + "score": 1.0, + "content": "2020a], the robust adversarial RL problem for the special linear quadratic case was investigated.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 106, + 187, + 505, + 329 + ], + "lines": [ + { + "bbox": [ + 106, + 187, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 106, + 187, + 505, + 199 + ], + "score": 1.0, + "content": "Model-free Robust RL. In [Roy et al., 2017, Badrinath and Kalathil, 2021] model-free RL with", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 199, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 505, + 210 + ], + "score": 1.0, + "content": "model uncertainty was studied, where in order to construct an algorithm that can be implemented in an", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 209, + 506, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 506, + 222 + ], + "score": 1.0, + "content": "online and incremental fashion, the uncertainty set was firstly relaxed by dropping the dependency on", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 220, + 505, + 233 + ], + "spans": [ + { + "bbox": [ + 106, + 220, + 505, + 233 + ], + "score": 1.0, + "content": "the misspecified MDP that generates the samples (centroid of the uncertainty set). Such a relaxation", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 231, + 506, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 506, + 244 + ], + "score": 1.0, + "content": "is pessimistic since the relaxed uncertainty set is not centered at the misspecified MDP anymore", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 241, + 506, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 506, + 255 + ], + "score": 1.0, + "content": "(which is usually similar to the target MDP), making the robustness to the relaxed uncertainty set", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 253, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 106, + 253, + 505, + 264 + ], + "score": 1.0, + "content": "not well-justified. Such a relaxation will further incur a stringent condition on the discounted factor", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 264, + 506, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 506, + 276 + ], + "score": 1.0, + "content": "to guarantee the convergence, which prevents the use of a discount factor close to 1 in practice.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 274, + 505, + 287 + ], + "spans": [ + { + "bbox": [ + 105, + 274, + 505, + 287 + ], + "score": 1.0, + "content": "Moreover, only asymptotic convergence was established in [Roy et al., 2017]. In this paper, we do", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 285, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 505, + 298 + ], + "score": 1.0, + "content": "not relax the uncertainty set, and instead propose an online approach to estimate it. Our algorithms", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 296, + 505, + 309 + ], + "spans": [ + { + "bbox": [ + 105, + 296, + 505, + 309 + ], + "score": 1.0, + "content": "converge without the need of the condition on the discount factor. We also provides finite-time error", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 307, + 505, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 505, + 320 + ], + "score": 1.0, + "content": "bounds for our algorithms. The multi-agent RL robust to reward uncertainty was investigated in", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 319, + 499, + 331 + ], + "spans": [ + { + "bbox": [ + 106, + 319, + 499, + 331 + ], + "score": 1.0, + "content": "[Zhang et al., 2020b], where the reward uncertainty set is known, but the transition kernel is fixed.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 334, + 506, + 487 + ], + "lines": [ + { + "bbox": [ + 106, + 334, + 505, + 346 + ], + "spans": [ + { + "bbox": [ + 106, + 334, + 505, + 346 + ], + "score": 1.0, + "content": "Finite-time Error Bound for RL Algorithms. For the tubular case, Q-learning has been studied", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 345, + 506, + 358 + ], + "spans": [ + { + "bbox": [ + 105, + 345, + 506, + 358 + ], + "score": 1.0, + "content": "intensively, e.g., in [Even-Dar et al., 2003, Beck and Srikant, 2012, Qu and Wierman, 2020, Li et al.,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 356, + 506, + 369 + ], + "spans": [ + { + "bbox": [ + 106, + 356, + 506, + 369 + ], + "score": 1.0, + "content": "2020, Wainwright, 2019, Li et al., 2021]. TD with function approximation were studied in [Dalal,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 366, + 506, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 506, + 380 + ], + "score": 1.0, + "content": "Gal and Szörényi, Balázs and Thoppe, Gugan and Mannor, Shie, 2018, Bhandari et al., 2018, Srikant", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 378, + 506, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 506, + 390 + ], + "score": 1.0, + "content": "and Ying, 2019, Cai et al., 2019, Sun et al., 2020]. Q-learning and SARSA with linear function", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 389, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 506, + 401 + ], + "score": 1.0, + "content": "approximation were investigated in [Zou et al., 2019, Chen et al., 2019]. The finite-time error bounds", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 400, + 506, + 412 + ], + "spans": [ + { + "bbox": [ + 106, + 400, + 506, + 412 + ], + "score": 1.0, + "content": "for the gradient TD algorithms [Maei et al., 2010, Sutton et al., 2009, Maei et al., 2010] were further", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 410, + 506, + 423 + ], + "spans": [ + { + "bbox": [ + 106, + 410, + 506, + 423 + ], + "score": 1.0, + "content": "developed recently in [Dalal et al., 2018, Liu et al., 2015, Gupta et al., 2019, Xu et al., 2019, Dalal", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 422, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 105, + 422, + 506, + 434 + ], + "score": 1.0, + "content": "et al., 2020, Kaledin et al., 2020, Ma et al., 2020, Wang and Zou, 2020, Ma et al., 2021, Doan, 2021].", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 430, + 507, + 447 + ], + "spans": [ + { + "bbox": [ + 104, + 430, + 507, + 447 + ], + "score": 1.0, + "content": "There are also finite-time error bounds on the policy gradient methods and actor critic methods, e.g.,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 442, + 506, + 457 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 506, + 457 + ], + "score": 1.0, + "content": "[Wang et al., 2020, Yang et al., 2019, Kumar et al., 2019, Qiu et al., 2019, Wu et al., 2020, Cen", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 454, + 505, + 467 + ], + "spans": [ + { + "bbox": [ + 106, + 454, + 505, + 467 + ], + "score": 1.0, + "content": "et al., 2020, Bhandari and Russo, 2019, Agarwal et al., 2021, Mei et al., 2020]. We note that these", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 465, + 506, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 506, + 479 + ], + "score": 1.0, + "content": "studies are for the non-robust RL algorithms, and in this paper, we design robust RL algorithms, and", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 477, + 277, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 477, + 277, + 488 + ], + "score": 1.0, + "content": "characterize their finite-time error bounds.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 29.5 + }, + { + "type": "title", + "bbox": [ + 107, + 502, + 195, + 516 + ], + "lines": [ + { + "bbox": [ + 104, + 501, + 196, + 518 + ], + "spans": [ + { + "bbox": [ + 104, + 501, + 196, + 518 + ], + "score": 1.0, + "content": "2 Preliminaries", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 107, + 527, + 505, + 655 + ], + "lines": [ + { + "bbox": [ + 105, + 526, + 504, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 526, + 384, + 540 + ], + "score": 1.0, + "content": "Markov Decision Process. An MDP can be characterized by a tuple", + "type": "text" + }, + { + "bbox": [ + 384, + 527, + 439, + 539 + ], + "score": 0.91, + "content": "( \\mathcal { S } , \\mathcal { A } , \\mathsf { P } , c , \\gamma )", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 526, + 495, + 540 + ], + "score": 1.0, + "content": ", where S and", + "type": "text" + }, + { + "bbox": [ + 495, + 528, + 504, + 537 + ], + "score": 0.76, + "content": "\\mathcal { A }", + "type": "inline_equation" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 538, + 506, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 230, + 552 + ], + "score": 1.0, + "content": "are the state and action spaces,", + "type": "text" + }, + { + "bbox": [ + 230, + 538, + 359, + 552 + ], + "score": 0.93, + "content": "\\mathsf { P } = \\{ p _ { s } ^ { a } \\in \\varDelta _ { | \\mathcal { S } | } , a \\in \\mathcal { A } , s \\in \\mathcal { S } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 538, + 456, + 552 + ], + "score": 1.0, + "content": "is the transition kernel1,", + "type": "text" + }, + { + "bbox": [ + 456, + 541, + 462, + 549 + ], + "score": 0.73, + "content": "c", + "type": "inline_equation" + }, + { + "bbox": [ + 463, + 538, + 506, + 552 + ], + "score": 1.0, + "content": "is the cost", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 551, + 506, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 163, + 564 + ], + "score": 1.0, + "content": "function, and", + "type": "text" + }, + { + "bbox": [ + 163, + 551, + 203, + 563 + ], + "score": 0.91, + "content": "\\gamma \\in [ 0 , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 551, + 348, + 564 + ], + "score": 1.0, + "content": "is the discount factor. Specifically,", + "type": "text" + }, + { + "bbox": [ + 348, + 552, + 360, + 563 + ], + "score": 0.88, + "content": "p _ { s } ^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 551, + 506, + 564 + ], + "score": 1.0, + "content": "denotes the distribution of the next", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 560, + 506, + 578 + ], + "spans": [ + { + "bbox": [ + 104, + 560, + 191, + 578 + ], + "score": 1.0, + "content": "state if taking action", + "type": "text" + }, + { + "bbox": [ + 191, + 565, + 198, + 572 + ], + "score": 0.74, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 560, + 230, + 578 + ], + "score": 1.0, + "content": "at state", + "type": "text" + }, + { + "bbox": [ + 230, + 564, + 236, + 572 + ], + "score": 0.7, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 560, + 256, + 578 + ], + "score": 1.0, + "content": ". Let", + "type": "text" + }, + { + "bbox": [ + 256, + 562, + 325, + 576 + ], + "score": 0.95, + "content": "p _ { s } ^ { a } = \\{ p _ { s , s ^ { \\prime } } ^ { a } \\} _ { s ^ { \\prime } \\in \\mathcal S }", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 560, + 356, + 578 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 357, + 563, + 376, + 576 + ], + "score": 0.91, + "content": "p _ { s , s ^ { \\prime } } ^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 560, + 506, + 578 + ], + "score": 1.0, + "content": "denotes the probability that the", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 574, + 505, + 586 + ], + "spans": [ + { + "bbox": [ + 106, + 575, + 223, + 586 + ], + "score": 1.0, + "content": "environment transits to state", + "type": "text" + }, + { + "bbox": [ + 223, + 574, + 232, + 585 + ], + "score": 0.86, + "content": "s ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 575, + 297, + 586 + ], + "score": 1.0, + "content": "if taking action", + "type": "text" + }, + { + "bbox": [ + 298, + 577, + 304, + 585 + ], + "score": 0.78, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 575, + 336, + 586 + ], + "score": 1.0, + "content": "at state", + "type": "text" + }, + { + "bbox": [ + 336, + 577, + 342, + 585 + ], + "score": 0.75, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 575, + 450, + 586 + ], + "score": 1.0, + "content": ". The cost of taking action", + "type": "text" + }, + { + "bbox": [ + 450, + 577, + 456, + 585 + ], + "score": 0.73, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 575, + 488, + 586 + ], + "score": 1.0, + "content": "at state", + "type": "text" + }, + { + "bbox": [ + 489, + 577, + 495, + 585 + ], + "score": 0.76, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 575, + 505, + 586 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 585, + 506, + 598 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 144, + 598 + ], + "score": 1.0, + "content": "given by", + "type": "text" + }, + { + "bbox": [ + 144, + 586, + 171, + 597 + ], + "score": 0.92, + "content": "c ( s , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 585, + 256, + 598 + ], + "score": 1.0, + "content": ". A stationary policy", + "type": "text" + }, + { + "bbox": [ + 257, + 587, + 264, + 595 + ], + "score": 0.79, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 585, + 437, + 598 + ], + "score": 1.0, + "content": "is a mapping from S to a distribution over", + "type": "text" + }, + { + "bbox": [ + 437, + 586, + 446, + 595 + ], + "score": 0.69, + "content": "\\mathcal A", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 585, + 506, + 598 + ], + "score": 1.0, + "content": ". At each time", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 107, + 597, + 506, + 609 + ], + "spans": [ + { + "bbox": [ + 107, + 598, + 111, + 607 + ], + "score": 0.73, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 112, + 597, + 215, + 609 + ], + "score": 1.0, + "content": ", an agent takes an action", + "type": "text" + }, + { + "bbox": [ + 216, + 597, + 248, + 608 + ], + "score": 0.91, + "content": "A _ { t } \\in \\mathcal A", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 597, + 280, + 609 + ], + "score": 1.0, + "content": "at state", + "type": "text" + }, + { + "bbox": [ + 281, + 597, + 309, + 608 + ], + "score": 0.91, + "content": "S _ { t } \\in \\mathcal S", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 597, + 506, + 609 + ], + "score": 1.0, + "content": ". The environment then transits to the next state", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 107, + 605, + 509, + 627 + ], + "spans": [ + { + "bbox": [ + 107, + 609, + 127, + 621 + ], + "score": 0.91, + "content": "S _ { t + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 128, + 605, + 194, + 627 + ], + "score": 1.0, + "content": "with probability", + "type": "text" + }, + { + "bbox": [ + 194, + 608, + 228, + 623 + ], + "score": 0.92, + "content": "p _ { S _ { t } , S _ { t + 1 } } ^ { A _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 605, + 375, + 627 + ], + "score": 1.0, + "content": ", and the agent receives cost given by", + "type": "text" + }, + { + "bbox": [ + 375, + 608, + 414, + 621 + ], + "score": 0.93, + "content": "c ( S _ { t } , A _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 605, + 509, + 627 + ], + "score": 1.0, + "content": ". The value function of", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 102, + 621, + 509, + 656 + ], + "spans": [ + { + "bbox": [ + 102, + 621, + 140, + 656 + ], + "score": 1.0, + "content": "a policy by follo", + "type": "text" + }, + { + "bbox": [ + 140, + 624, + 147, + 631 + ], + "score": 0.73, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 621, + 160, + 656 + ], + "score": 1.0, + "content": "staing", + "type": "text" + }, + { + "bbox": [ + 264, + 621, + 287, + 632 + ], + "score": 0.88, + "content": "s \\in \\mathcal { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 621, + 434, + 656 + ], + "score": 1.0, + "content": "fined as the expected accumulatand the goal is to find the policy", + "type": "text" + }, + { + "bbox": [ + 443, + 621, + 509, + 656 + ], + "score": 1.0, + "content": "discounted costthat minimizes", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 160, + 632, + 443, + 645 + ], + "spans": [ + { + "bbox": [ + 160, + 632, + 300, + 645 + ], + "score": 0.9, + "content": "\\pi \\colon \\bar { \\mathbb { E } } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\dot { \\gamma } ^ { t } c ( S _ { t } , A _ { t } ) | S _ { 0 } = s , \\pi \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 634, + 443, + 642 + ], + "score": 0.76, + "content": "\\pi", + "type": "inline_equation" + } + ], + "index": 47 + }, + { + "bbox": [ + 284, + 644, + 307, + 654 + ], + "spans": [ + { + "bbox": [ + 284, + 644, + 307, + 654 + ], + "score": 0.88, + "content": "s \\in \\mathcal { S }", + "type": "inline_equation" + } + ], + "index": 48 + } + ], + "index": 43 + }, + { + "type": "text", + "bbox": [ + 107, + 659, + 505, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 659, + 506, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 506, + 672 + ], + "score": 1.0, + "content": "Robust Markov Decision Process. In the robust case, the transition kernel is not fixed and lies in", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 670, + 506, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 670, + 342, + 684 + ], + "score": 1.0, + "content": "some uncertainty set. Denote the transition kernel at time", + "type": "text" + }, + { + "bbox": [ + 343, + 672, + 347, + 681 + ], + "score": 0.74, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 670, + 361, + 684 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 361, + 671, + 372, + 682 + ], + "score": 0.87, + "content": "\\mathsf { P } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 670, + 406, + 684 + ], + "score": 1.0, + "content": ", and let", + "type": "text" + }, + { + "bbox": [ + 407, + 671, + 473, + 682 + ], + "score": 0.92, + "content": "\\kappa = ( \\mathsf { P } _ { 0 } , \\mathsf { P } _ { 1 } , \\ldots )", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 670, + 506, + 684 + ], + "score": 1.0, + "content": ", where", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 107, + 681, + 506, + 694 + ], + "spans": [ + { + "bbox": [ + 107, + 682, + 169, + 693 + ], + "score": 0.88, + "content": "\\mathsf { P } _ { t } \\in \\mathbf { P } , \\forall t \\ge 0", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 681, + 190, + 694 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 190, + 682, + 199, + 691 + ], + "score": 0.74, + "content": "\\mathbf { P }", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 681, + 437, + 694 + ], + "score": 1.0, + "content": "is the uncertainty set of the transition kernel. The sequence", + "type": "text" + }, + { + "bbox": [ + 438, + 684, + 445, + 691 + ], + "score": 0.78, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 681, + 506, + 694 + ], + "score": 1.0, + "content": "can be viewed", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 692, + 505, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 692, + 505, + 705 + ], + "score": 1.0, + "content": "as the policy of the nature, and is adversarially chosen by the nature [Bagnell et al., 2001, Nilim and", + "type": "text" + } + ], + "index": 52 + } + ], + "index": 50.5 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 115, + 711, + 484, + 723 + ], + "lines": [ + { + "bbox": [ + 119, + 708, + 483, + 726 + ], + "spans": [ + { + "bbox": [ + 119, + 708, + 123, + 726 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 123, + 711, + 136, + 721 + ], + "score": 0.76, + "content": "\\varDelta _ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 137, + 708, + 181, + 726 + ], + "score": 1.0, + "content": "denotes the", + "type": "text" + }, + { + "bbox": [ + 181, + 712, + 210, + 722 + ], + "score": 0.88, + "content": "( n - 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 708, + 333, + 726 + ], + "score": 1.0, + "content": "-dimensional probability simplex:", + "type": "text" + }, + { + "bbox": [ + 334, + 711, + 483, + 723 + ], + "score": 0.86, + "content": "\\begin{array} { r } { \\{ ( p _ { 1 } , . . . , p _ { n } ) | 0 \\leq p _ { i } \\leq 1 , \\sum _ { i = 1 } ^ { n } p _ { i } = 1 \\} . } \\end{array}", + "type": "inline_equation" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 72, + 505, + 182 + ], + "lines": [], + "index": 4.5, + "bbox_fs": [ + 105, + 72, + 507, + 183 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 187, + 505, + 329 + ], + "lines": [ + { + "bbox": [ + 106, + 187, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 106, + 187, + 505, + 199 + ], + "score": 1.0, + "content": "Model-free Robust RL. In [Roy et al., 2017, Badrinath and Kalathil, 2021] model-free RL with", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 199, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 505, + 210 + ], + "score": 1.0, + "content": "model uncertainty was studied, where in order to construct an algorithm that can be implemented in an", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 209, + 506, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 506, + 222 + ], + "score": 1.0, + "content": "online and incremental fashion, the uncertainty set was firstly relaxed by dropping the dependency on", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 220, + 505, + 233 + ], + "spans": [ + { + "bbox": [ + 106, + 220, + 505, + 233 + ], + "score": 1.0, + "content": "the misspecified MDP that generates the samples (centroid of the uncertainty set). Such a relaxation", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 231, + 506, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 506, + 244 + ], + "score": 1.0, + "content": "is pessimistic since the relaxed uncertainty set is not centered at the misspecified MDP anymore", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 241, + 506, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 506, + 255 + ], + "score": 1.0, + "content": "(which is usually similar to the target MDP), making the robustness to the relaxed uncertainty set", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 253, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 106, + 253, + 505, + 264 + ], + "score": 1.0, + "content": "not well-justified. Such a relaxation will further incur a stringent condition on the discounted factor", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 264, + 506, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 506, + 276 + ], + "score": 1.0, + "content": "to guarantee the convergence, which prevents the use of a discount factor close to 1 in practice.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 274, + 505, + 287 + ], + "spans": [ + { + "bbox": [ + 105, + 274, + 505, + 287 + ], + "score": 1.0, + "content": "Moreover, only asymptotic convergence was established in [Roy et al., 2017]. In this paper, we do", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 285, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 505, + 298 + ], + "score": 1.0, + "content": "not relax the uncertainty set, and instead propose an online approach to estimate it. Our algorithms", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 296, + 505, + 309 + ], + "spans": [ + { + "bbox": [ + 105, + 296, + 505, + 309 + ], + "score": 1.0, + "content": "converge without the need of the condition on the discount factor. We also provides finite-time error", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 307, + 505, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 505, + 320 + ], + "score": 1.0, + "content": "bounds for our algorithms. The multi-agent RL robust to reward uncertainty was investigated in", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 319, + 499, + 331 + ], + "spans": [ + { + "bbox": [ + 106, + 319, + 499, + 331 + ], + "score": 1.0, + "content": "[Zhang et al., 2020b], where the reward uncertainty set is known, but the transition kernel is fixed.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 187, + 506, + 331 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 334, + 506, + 487 + ], + "lines": [ + { + "bbox": [ + 106, + 334, + 505, + 346 + ], + "spans": [ + { + "bbox": [ + 106, + 334, + 505, + 346 + ], + "score": 1.0, + "content": "Finite-time Error Bound for RL Algorithms. For the tubular case, Q-learning has been studied", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 345, + 506, + 358 + ], + "spans": [ + { + "bbox": [ + 105, + 345, + 506, + 358 + ], + "score": 1.0, + "content": "intensively, e.g., in [Even-Dar et al., 2003, Beck and Srikant, 2012, Qu and Wierman, 2020, Li et al.,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 356, + 506, + 369 + ], + "spans": [ + { + "bbox": [ + 106, + 356, + 506, + 369 + ], + "score": 1.0, + "content": "2020, Wainwright, 2019, Li et al., 2021]. TD with function approximation were studied in [Dalal,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 366, + 506, + 380 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 506, + 380 + ], + "score": 1.0, + "content": "Gal and Szörényi, Balázs and Thoppe, Gugan and Mannor, Shie, 2018, Bhandari et al., 2018, Srikant", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 378, + 506, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 506, + 390 + ], + "score": 1.0, + "content": "and Ying, 2019, Cai et al., 2019, Sun et al., 2020]. Q-learning and SARSA with linear function", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 389, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 506, + 401 + ], + "score": 1.0, + "content": "approximation were investigated in [Zou et al., 2019, Chen et al., 2019]. The finite-time error bounds", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 400, + 506, + 412 + ], + "spans": [ + { + "bbox": [ + 106, + 400, + 506, + 412 + ], + "score": 1.0, + "content": "for the gradient TD algorithms [Maei et al., 2010, Sutton et al., 2009, Maei et al., 2010] were further", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 410, + 506, + 423 + ], + "spans": [ + { + "bbox": [ + 106, + 410, + 506, + 423 + ], + "score": 1.0, + "content": "developed recently in [Dalal et al., 2018, Liu et al., 2015, Gupta et al., 2019, Xu et al., 2019, Dalal", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 422, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 105, + 422, + 506, + 434 + ], + "score": 1.0, + "content": "et al., 2020, Kaledin et al., 2020, Ma et al., 2020, Wang and Zou, 2020, Ma et al., 2021, Doan, 2021].", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 430, + 507, + 447 + ], + "spans": [ + { + "bbox": [ + 104, + 430, + 507, + 447 + ], + "score": 1.0, + "content": "There are also finite-time error bounds on the policy gradient methods and actor critic methods, e.g.,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 442, + 506, + 457 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 506, + 457 + ], + "score": 1.0, + "content": "[Wang et al., 2020, Yang et al., 2019, Kumar et al., 2019, Qiu et al., 2019, Wu et al., 2020, Cen", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 454, + 505, + 467 + ], + "spans": [ + { + "bbox": [ + 106, + 454, + 505, + 467 + ], + "score": 1.0, + "content": "et al., 2020, Bhandari and Russo, 2019, Agarwal et al., 2021, Mei et al., 2020]. We note that these", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 465, + 506, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 506, + 479 + ], + "score": 1.0, + "content": "studies are for the non-robust RL algorithms, and in this paper, we design robust RL algorithms, and", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 477, + 277, + 488 + ], + "spans": [ + { + "bbox": [ + 106, + 477, + 277, + 488 + ], + "score": 1.0, + "content": "characterize their finite-time error bounds.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 29.5, + "bbox_fs": [ + 104, + 334, + 507, + 488 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 502, + 195, + 516 + ], + "lines": [ + { + "bbox": [ + 104, + 501, + 196, + 518 + ], + "spans": [ + { + "bbox": [ + 104, + 501, + 196, + 518 + ], + "score": 1.0, + "content": "2 Preliminaries", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "list", + "bbox": [ + 107, + 527, + 505, + 655 + ], + "lines": [ + { + "bbox": [ + 105, + 526, + 504, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 526, + 384, + 540 + ], + "score": 1.0, + "content": "Markov Decision Process. An MDP can be characterized by a tuple", + "type": "text" + }, + { + "bbox": [ + 384, + 527, + 439, + 539 + ], + "score": 0.91, + "content": "( \\mathcal { S } , \\mathcal { A } , \\mathsf { P } , c , \\gamma )", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 526, + 495, + 540 + ], + "score": 1.0, + "content": ", where S and", + "type": "text" + }, + { + "bbox": [ + 495, + 528, + 504, + 537 + ], + "score": 0.76, + "content": "\\mathcal { A }", + "type": "inline_equation" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 538, + 506, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 230, + 552 + ], + "score": 1.0, + "content": "are the state and action spaces,", + "type": "text" + }, + { + "bbox": [ + 230, + 538, + 359, + 552 + ], + "score": 0.93, + "content": "\\mathsf { P } = \\{ p _ { s } ^ { a } \\in \\varDelta _ { | \\mathcal { S } | } , a \\in \\mathcal { A } , s \\in \\mathcal { S } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 538, + 456, + 552 + ], + "score": 1.0, + "content": "is the transition kernel1,", + "type": "text" + }, + { + "bbox": [ + 456, + 541, + 462, + 549 + ], + "score": 0.73, + "content": "c", + "type": "inline_equation" + }, + { + "bbox": [ + 463, + 538, + 506, + 552 + ], + "score": 1.0, + "content": "is the cost", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 551, + 506, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 163, + 564 + ], + "score": 1.0, + "content": "function, and", + "type": "text" + }, + { + "bbox": [ + 163, + 551, + 203, + 563 + ], + "score": 0.91, + "content": "\\gamma \\in [ 0 , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 551, + 348, + 564 + ], + "score": 1.0, + "content": "is the discount factor. Specifically,", + "type": "text" + }, + { + "bbox": [ + 348, + 552, + 360, + 563 + ], + "score": 0.88, + "content": "p _ { s } ^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 551, + 506, + 564 + ], + "score": 1.0, + "content": "denotes the distribution of the next", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 560, + 506, + 578 + ], + "spans": [ + { + "bbox": [ + 104, + 560, + 191, + 578 + ], + "score": 1.0, + "content": "state if taking action", + "type": "text" + }, + { + "bbox": [ + 191, + 565, + 198, + 572 + ], + "score": 0.74, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 560, + 230, + 578 + ], + "score": 1.0, + "content": "at state", + "type": "text" + }, + { + "bbox": [ + 230, + 564, + 236, + 572 + ], + "score": 0.7, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 560, + 256, + 578 + ], + "score": 1.0, + "content": ". Let", + "type": "text" + }, + { + "bbox": [ + 256, + 562, + 325, + 576 + ], + "score": 0.95, + "content": "p _ { s } ^ { a } = \\{ p _ { s , s ^ { \\prime } } ^ { a } \\} _ { s ^ { \\prime } \\in \\mathcal S }", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 560, + 356, + 578 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 357, + 563, + 376, + 576 + ], + "score": 0.91, + "content": "p _ { s , s ^ { \\prime } } ^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 560, + 506, + 578 + ], + "score": 1.0, + "content": "denotes the probability that the", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 574, + 505, + 586 + ], + "spans": [ + { + "bbox": [ + 106, + 575, + 223, + 586 + ], + "score": 1.0, + "content": "environment transits to state", + "type": "text" + }, + { + "bbox": [ + 223, + 574, + 232, + 585 + ], + "score": 0.86, + "content": "s ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 575, + 297, + 586 + ], + "score": 1.0, + "content": "if taking action", + "type": "text" + }, + { + "bbox": [ + 298, + 577, + 304, + 585 + ], + "score": 0.78, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 575, + 336, + 586 + ], + "score": 1.0, + "content": "at state", + "type": "text" + }, + { + "bbox": [ + 336, + 577, + 342, + 585 + ], + "score": 0.75, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 575, + 450, + 586 + ], + "score": 1.0, + "content": ". The cost of taking action", + "type": "text" + }, + { + "bbox": [ + 450, + 577, + 456, + 585 + ], + "score": 0.73, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 575, + 488, + 586 + ], + "score": 1.0, + "content": "at state", + "type": "text" + }, + { + "bbox": [ + 489, + 577, + 495, + 585 + ], + "score": 0.76, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 575, + 505, + 586 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 585, + 506, + 598 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 144, + 598 + ], + "score": 1.0, + "content": "given by", + "type": "text" + }, + { + "bbox": [ + 144, + 586, + 171, + 597 + ], + "score": 0.92, + "content": "c ( s , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 585, + 256, + 598 + ], + "score": 1.0, + "content": ". A stationary policy", + "type": "text" + }, + { + "bbox": [ + 257, + 587, + 264, + 595 + ], + "score": 0.79, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 585, + 437, + 598 + ], + "score": 1.0, + "content": "is a mapping from S to a distribution over", + "type": "text" + }, + { + "bbox": [ + 437, + 586, + 446, + 595 + ], + "score": 0.69, + "content": "\\mathcal A", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 585, + 506, + 598 + ], + "score": 1.0, + "content": ". At each time", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 107, + 597, + 506, + 609 + ], + "spans": [ + { + "bbox": [ + 107, + 598, + 111, + 607 + ], + "score": 0.73, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 112, + 597, + 215, + 609 + ], + "score": 1.0, + "content": ", an agent takes an action", + "type": "text" + }, + { + "bbox": [ + 216, + 597, + 248, + 608 + ], + "score": 0.91, + "content": "A _ { t } \\in \\mathcal A", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 597, + 280, + 609 + ], + "score": 1.0, + "content": "at state", + "type": "text" + }, + { + "bbox": [ + 281, + 597, + 309, + 608 + ], + "score": 0.91, + "content": "S _ { t } \\in \\mathcal S", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 597, + 506, + 609 + ], + "score": 1.0, + "content": ". The environment then transits to the next state", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 107, + 605, + 509, + 627 + ], + "spans": [ + { + "bbox": [ + 107, + 609, + 127, + 621 + ], + "score": 0.91, + "content": "S _ { t + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 128, + 605, + 194, + 627 + ], + "score": 1.0, + "content": "with probability", + "type": "text" + }, + { + "bbox": [ + 194, + 608, + 228, + 623 + ], + "score": 0.92, + "content": "p _ { S _ { t } , S _ { t + 1 } } ^ { A _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 605, + 375, + 627 + ], + "score": 1.0, + "content": ", and the agent receives cost given by", + "type": "text" + }, + { + "bbox": [ + 375, + 608, + 414, + 621 + ], + "score": 0.93, + "content": "c ( S _ { t } , A _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 605, + 509, + 627 + ], + "score": 1.0, + "content": ". The value function of", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 102, + 621, + 509, + 656 + ], + "spans": [ + { + "bbox": [ + 102, + 621, + 140, + 656 + ], + "score": 1.0, + "content": "a policy by follo", + "type": "text" + }, + { + "bbox": [ + 140, + 624, + 147, + 631 + ], + "score": 0.73, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 621, + 160, + 656 + ], + "score": 1.0, + "content": "staing", + "type": "text" + }, + { + "bbox": [ + 264, + 621, + 287, + 632 + ], + "score": 0.88, + "content": "s \\in \\mathcal { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 621, + 434, + 656 + ], + "score": 1.0, + "content": "fined as the expected accumulatand the goal is to find the policy", + "type": "text" + }, + { + "bbox": [ + 443, + 621, + 509, + 656 + ], + "score": 1.0, + "content": "discounted costthat minimizes", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 160, + 632, + 443, + 645 + ], + "spans": [ + { + "bbox": [ + 160, + 632, + 300, + 645 + ], + "score": 0.9, + "content": "\\pi \\colon \\bar { \\mathbb { E } } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\dot { \\gamma } ^ { t } c ( S _ { t } , A _ { t } ) | S _ { 0 } = s , \\pi \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 634, + 443, + 642 + ], + "score": 0.76, + "content": "\\pi", + "type": "inline_equation" + } + ], + "index": 47, + "is_list_end_line": true + }, + { + "bbox": [ + 284, + 644, + 307, + 654 + ], + "spans": [ + { + "bbox": [ + 284, + 644, + 307, + 654 + ], + "score": 0.88, + "content": "s \\in \\mathcal { S }", + "type": "inline_equation" + } + ], + "index": 48, + "is_list_start_line": true, + "is_list_end_line": true + } + ], + "index": 43, + "bbox_fs": [ + 102, + 526, + 509, + 656 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 659, + 505, + 704 + ], + "lines": [ + { + "bbox": [ + 105, + 659, + 506, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 506, + 672 + ], + "score": 1.0, + "content": "Robust Markov Decision Process. In the robust case, the transition kernel is not fixed and lies in", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 670, + 506, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 670, + 342, + 684 + ], + "score": 1.0, + "content": "some uncertainty set. Denote the transition kernel at time", + "type": "text" + }, + { + "bbox": [ + 343, + 672, + 347, + 681 + ], + "score": 0.74, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 670, + 361, + 684 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 361, + 671, + 372, + 682 + ], + "score": 0.87, + "content": "\\mathsf { P } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 670, + 406, + 684 + ], + "score": 1.0, + "content": ", and let", + "type": "text" + }, + { + "bbox": [ + 407, + 671, + 473, + 682 + ], + "score": 0.92, + "content": "\\kappa = ( \\mathsf { P } _ { 0 } , \\mathsf { P } _ { 1 } , \\ldots )", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 670, + 506, + 684 + ], + "score": 1.0, + "content": ", where", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 107, + 681, + 506, + 694 + ], + "spans": [ + { + "bbox": [ + 107, + 682, + 169, + 693 + ], + "score": 0.88, + "content": "\\mathsf { P } _ { t } \\in \\mathbf { P } , \\forall t \\ge 0", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 681, + 190, + 694 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 190, + 682, + 199, + 691 + ], + "score": 0.74, + "content": "\\mathbf { P }", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 681, + 437, + 694 + ], + "score": 1.0, + "content": "is the uncertainty set of the transition kernel. The sequence", + "type": "text" + }, + { + "bbox": [ + 438, + 684, + 445, + 691 + ], + "score": 0.78, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 681, + 506, + 694 + ], + "score": 1.0, + "content": "can be viewed", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 692, + 505, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 692, + 505, + 705 + ], + "score": 1.0, + "content": "as the policy of the nature, and is adversarially chosen by the nature [Bagnell et al., 2001, Nilim and", + "type": "text" + } + ], + "index": 52 + } + ], + "index": 50.5, + "bbox_fs": [ + 105, + 659, + 506, + 705 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 72, + 505, + 106 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 505, + 85 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 424, + 85 + ], + "score": 1.0, + "content": "El Ghaoui, 2004, Iyengar, 2005]. Define the robust value function of a policy", + "type": "text" + }, + { + "bbox": [ + 424, + 75, + 432, + 83 + ], + "score": 0.75, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 72, + 505, + 85 + ], + "score": 1.0, + "content": "as the worst-case", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 83, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 365, + 96 + ], + "score": 1.0, + "content": "expected accumulated discounted cost following a fixed policy", + "type": "text" + }, + { + "bbox": [ + 365, + 86, + 372, + 93 + ], + "score": 0.73, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 83, + 505, + 96 + ], + "score": 1.0, + "content": "over all transition kernels in the", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 95, + 171, + 108 + ], + "spans": [ + { + "bbox": [ + 106, + 95, + 171, + 108 + ], + "score": 1.0, + "content": "uncertainty set:", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "interline_equation", + "bbox": [ + 207, + 111, + 402, + 145 + ], + "lines": [ + { + "bbox": [ + 207, + 111, + 402, + 145 + ], + "spans": [ + { + "bbox": [ + 207, + 111, + 402, + 145 + ], + "score": 0.94, + "content": "V ^ { \\pi } ( s ) = \\operatorname* { m a x } _ { \\kappa } \\mathbb { E } _ { \\kappa } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } c ( S _ { t } , A _ { t } ) | S _ { 0 } = s , \\pi \\right] ,", + "type": "interline_equation", + "image_path": "5c8479616301ab2d49dcead2362db851758cb5ec733965b24b83759529ec5c2b.jpg" + } + ] + } + ], + "index": 3.5, + "virtual_lines": [ + { + "bbox": [ + 207, + 111, + 402, + 128.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 207, + 128.0, + 402, + 145.0 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 150, + 506, + 194 + ], + "lines": [ + { + "bbox": [ + 106, + 150, + 506, + 162 + ], + "spans": [ + { + "bbox": [ + 106, + 150, + 132, + 162 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 151, + 146, + 161 + ], + "score": 0.89, + "content": "\\mathbb { E } _ { \\kappa }", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 150, + 384, + 162 + ], + "score": 1.0, + "content": "denotes the expectation when the state transits according to", + "type": "text" + }, + { + "bbox": [ + 385, + 153, + 391, + 160 + ], + "score": 0.72, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 150, + 506, + 162 + ], + "score": 1.0, + "content": ". Similarly, define the robust", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 102, + 158, + 508, + 177 + ], + "spans": [ + { + "bbox": [ + 102, + 158, + 245, + 177 + ], + "score": 1.0, + "content": "action-value function for a policy", + "type": "text" + }, + { + "bbox": [ + 246, + 161, + 501, + 174 + ], + "score": 0.84, + "content": "\\begin{array} { r } { \\pi \\colon Q ^ { \\pi } ( s , a ) = \\operatorname* { m a x } _ { \\kappa } \\mathbb { E } _ { \\kappa } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } c ( S _ { t } , A _ { t } ) | \\dot { S _ { 0 } } = s , A _ { 0 } = a , \\pi \\right] } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 158, + 508, + 177 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 171, + 507, + 185 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 336, + 185 + ], + "score": 1.0, + "content": "The goal of robust RL is to find the optimal robust policy", + "type": "text" + }, + { + "bbox": [ + 336, + 173, + 347, + 182 + ], + "score": 0.85, + "content": "\\pi ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 171, + 507, + 185 + ], + "score": 1.0, + "content": "that minimizes the worst-case accumu-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 183, + 195, + 195 + ], + "spans": [ + { + "bbox": [ + 105, + 183, + 195, + 195 + ], + "score": 1.0, + "content": "lated discounted cost:", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5 + }, + { + "type": "interline_equation", + "bbox": [ + 245, + 200, + 366, + 218 + ], + "lines": [ + { + "bbox": [ + 245, + 200, + 366, + 218 + ], + "spans": [ + { + "bbox": [ + 245, + 200, + 366, + 218 + ], + "score": 0.92, + "content": "\\pi ^ { * } = \\arg \\operatorname* { m i n } _ { \\pi } V ^ { \\pi } ( s ) , \\forall s \\in \\mathcal { S } .", + "type": "interline_equation", + "image_path": "323b5dbc7086c2a15a4b350fd9583f048099d6491412754c182179252321aee0.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 245, + 200, + 366, + 218 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 224, + 469, + 237 + ], + "lines": [ + { + "bbox": [ + 104, + 222, + 470, + 240 + ], + "spans": [ + { + "bbox": [ + 104, + 222, + 169, + 240 + ], + "score": 1.0, + "content": "We also denote", + "type": "text" + }, + { + "bbox": [ + 169, + 224, + 187, + 236 + ], + "score": 0.92, + "content": "V ^ { \\pi ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 222, + 206, + 240 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 207, + 224, + 224, + 237 + ], + "score": 0.91, + "content": "Q ^ { \\pi ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 222, + 238, + 240 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 239, + 225, + 252, + 236 + ], + "score": 0.85, + "content": "V ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 222, + 270, + 240 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 271, + 226, + 284, + 237 + ], + "score": 0.9, + "content": "Q ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 222, + 357, + 240 + ], + "score": 1.0, + "content": ", respectively, and", + "type": "text" + }, + { + "bbox": [ + 357, + 225, + 466, + 238 + ], + "score": 0.92, + "content": "\\begin{array} { r } { V ^ { * } ( s ) = \\operatorname* { m i n } _ { a \\in \\mathcal { A } } Q ^ { * } ( s , a ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 466, + 222, + 470, + 240 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 107, + 241, + 506, + 310 + ], + "lines": [ + { + "bbox": [ + 105, + 240, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 240, + 506, + 254 + ], + "score": 1.0, + "content": "Note that a transition kernel is a collection of conditional distributions. Therefore, the uncertainty set", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 252, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 253, + 115, + 263 + ], + "score": 0.64, + "content": "\\mathbf { P }", + "type": "inline_equation" + }, + { + "bbox": [ + 116, + 252, + 383, + 266 + ], + "score": 1.0, + "content": "of the transition kernel can be equivalently written as a collection of", + "type": "text" + }, + { + "bbox": [ + 383, + 253, + 395, + 265 + ], + "score": 0.9, + "content": "\\mathcal { P } _ { s } ^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 252, + 422, + 266 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 423, + 253, + 475, + 264 + ], + "score": 0.86, + "content": "s \\in \\mathcal { S } , a \\in \\mathcal { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 252, + 505, + 266 + ], + "score": 1.0, + "content": ", where", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 263, + 505, + 279 + ], + "spans": [ + { + "bbox": [ + 106, + 265, + 119, + 277 + ], + "score": 0.87, + "content": "\\mathcal { P } _ { s } ^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 263, + 259, + 279 + ], + "score": 1.0, + "content": "is a set of conditional distributions", + "type": "text" + }, + { + "bbox": [ + 260, + 266, + 271, + 277 + ], + "score": 0.89, + "content": "p _ { s } ^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 263, + 407, + 279 + ], + "score": 1.0, + "content": "over the state space S. Denote by", + "type": "text" + }, + { + "bbox": [ + 407, + 264, + 505, + 277 + ], + "score": 0.91, + "content": "\\sigma _ { \\mathcal { P } } ( v ) \\triangleq \\operatorname* { m a x } _ { p \\in \\mathcal { P } } ( p ^ { \\top } v )", + "type": "inline_equation" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 275, + 506, + 290 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 234, + 290 + ], + "score": 1.0, + "content": "the support function of vector", + "type": "text" + }, + { + "bbox": [ + 235, + 279, + 241, + 286 + ], + "score": 0.74, + "content": "v", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 275, + 401, + 290 + ], + "score": 1.0, + "content": "over a set of probability distributions", + "type": "text" + }, + { + "bbox": [ + 402, + 277, + 410, + 286 + ], + "score": 0.78, + "content": "\\mathcal { P }", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 275, + 506, + 290 + ], + "score": 1.0, + "content": ". For robust MDP, the", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 286, + 506, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 286, + 506, + 299 + ], + "score": 1.0, + "content": "following robust analogue of the Bellman recursion was provided in [Nilim and El Ghaoui, 2004,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 298, + 170, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 170, + 311 + ], + "score": 1.0, + "content": "Iyengar, 2005].", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 106, + 312, + 505, + 325 + ], + "lines": [ + { + "bbox": [ + 106, + 311, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 106, + 311, + 479, + 326 + ], + "score": 1.0, + "content": "Theorem 1. [Nilim and El Ghaoui, 2004] The following perfect duality condition holds for all", + "type": "text" + }, + { + "bbox": [ + 479, + 313, + 502, + 323 + ], + "score": 0.88, + "content": "s \\in \\mathcal { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 311, + 506, + 326 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "interline_equation", + "bbox": [ + 120, + 329, + 478, + 363 + ], + "lines": [ + { + "bbox": [ + 120, + 329, + 478, + 363 + ], + "spans": [ + { + "bbox": [ + 120, + 329, + 478, + 363 + ], + "score": 0.94, + "content": "\\operatorname* { m i n } _ { \\pi } \\operatorname* { m a x } _ { \\kappa } \\mathbb { E } _ { \\kappa } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } c ( S _ { t } , A _ { t } ) \\big | \\pi , S _ { 0 } = s \\right] = \\operatorname* { m a x } _ { \\kappa } \\operatorname* { m i n } _ { \\pi } \\mathbb { E } _ { \\kappa } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } c ( S _ { t } , A _ { t } ) \\big | \\pi , S _ { 0 } = s \\right] .", + "type": "interline_equation", + "image_path": "e93045239e0b7e332eb486ed1d6c8b17f4379e21d3033a60baeafcd33b64fda6.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 120, + 329, + 478, + 340.3333333333333 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 120, + 340.3333333333333, + 478, + 351.66666666666663 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 120, + 351.66666666666663, + 478, + 362.99999999999994 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 368, + 505, + 394 + ], + "lines": [ + { + "bbox": [ + 105, + 367, + 505, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 249, + 383 + ], + "score": 1.0, + "content": "The optimal robust value function", + "type": "text" + }, + { + "bbox": [ + 249, + 369, + 263, + 379 + ], + "score": 0.88, + "content": "V ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 367, + 301, + 383 + ], + "score": 1.0, + "content": "satisfies", + "type": "text" + }, + { + "bbox": [ + 301, + 368, + 467, + 381 + ], + "score": 0.89, + "content": "\\begin{array} { r } { V ^ { * } ( s ) = \\operatorname* { m i n } _ { a \\in \\mathcal { A } } \\bigl ( c ( s , a ) + \\gamma \\sigma _ { \\mathscr { P } _ { s } ^ { a } } ( V ^ { * } ) \\bigr ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 467, + 367, + 505, + 383 + ], + "score": 1.0, + "content": ", and the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 379, + 438, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 379, + 254, + 396 + ], + "score": 1.0, + "content": "optimal robust action-value function", + "type": "text" + }, + { + "bbox": [ + 254, + 381, + 268, + 393 + ], + "score": 0.89, + "content": "Q ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 379, + 303, + 396 + ], + "score": 1.0, + "content": "satisfies", + "type": "text" + }, + { + "bbox": [ + 303, + 382, + 433, + 394 + ], + "score": 0.89, + "content": "Q ^ { * } ( s , a ) = c ( s , a ) + \\gamma \\sigma _ { \\mathscr P _ { s } ^ { a } } ( V ^ { * } )", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 379, + 438, + 396 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 107, + 402, + 505, + 448 + ], + "lines": [ + { + "bbox": [ + 105, + 402, + 506, + 416 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 255, + 416 + ], + "score": 1.0, + "content": "Define the robust Bellman operator", + "type": "text" + }, + { + "bbox": [ + 256, + 403, + 266, + 413 + ], + "score": 0.64, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 402, + 280, + 416 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 280, + 403, + 470, + 416 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\mathbf { T } Q ( s , a ) = c ( s , a ) + \\gamma \\sigma _ { \\mathcal { P } _ { s } ^ { a } } ( \\operatorname* { m i n } _ { a \\in \\mathcal { A } } Q ( s , a ) ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 402, + 506, + 416 + ], + "score": 1.0, + "content": ". It was", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 414, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 106, + 414, + 336, + 425 + ], + "score": 1.0, + "content": "shown in [Nilim and El Ghaoui, 2004, Iyengar, 2005] that", + "type": "text" + }, + { + "bbox": [ + 336, + 415, + 345, + 424 + ], + "score": 0.76, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 414, + 505, + 425 + ], + "score": 1.0, + "content": "is a contraction and its fixed point is the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 424, + 506, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 165, + 438 + ], + "score": 1.0, + "content": "optimal robust", + "type": "text" + }, + { + "bbox": [ + 166, + 425, + 179, + 436 + ], + "score": 0.88, + "content": "Q ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 424, + 353, + 438 + ], + "score": 1.0, + "content": ". When the uncertainty set is known, so that", + "type": "text" + }, + { + "bbox": [ + 353, + 426, + 370, + 437 + ], + "score": 0.87, + "content": "\\sigma _ { \\mathcal P _ { s } ^ { a } }", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 424, + 473, + 438 + ], + "score": 1.0, + "content": "can be computed exactly,", + "type": "text" + }, + { + "bbox": [ + 474, + 425, + 487, + 434 + ], + "score": 0.86, + "content": "V ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 424, + 506, + 438 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 107, + 435, + 465, + 449 + ], + "spans": [ + { + "bbox": [ + 107, + 436, + 120, + 447 + ], + "score": 0.88, + "content": "Q ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 120, + 435, + 465, + 449 + ], + "score": 1.0, + "content": "can be solved by dynamic programming [Iyengar, 2005, Nilim and El Ghaoui, 2004].", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24.5 + }, + { + "type": "title", + "bbox": [ + 107, + 462, + 426, + 477 + ], + "lines": [ + { + "bbox": [ + 104, + 461, + 427, + 479 + ], + "spans": [ + { + "bbox": [ + 104, + 461, + 427, + 479 + ], + "score": 1.0, + "content": "3 R-Contamination Model For Uncertainty Set Construction", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 106, + 487, + 434, + 499 + ], + "lines": [ + { + "bbox": [ + 105, + 486, + 435, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 335, + 500 + ], + "score": 1.0, + "content": "In this section, we construct the uncertainty set using the", + "type": "text" + }, + { + "bbox": [ + 335, + 488, + 344, + 497 + ], + "score": 0.81, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 486, + 435, + 500 + ], + "score": 1.0, + "content": "-contamination model.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 503, + 506, + 560 + ], + "lines": [ + { + "bbox": [ + 105, + 504, + 506, + 517 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 123, + 517 + ], + "score": 1.0, + "content": "Let", + "type": "text" + }, + { + "bbox": [ + 123, + 504, + 226, + 516 + ], + "score": 0.9, + "content": "\\mathsf { P } = \\{ p _ { s } ^ { a } , s \\in \\mathscr { S } , a \\in \\mathscr { A } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 504, + 506, + 517 + ], + "score": 1.0, + "content": "be the centroid of the uncertainty set, i.e., the transition kernel that", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 515, + 506, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 256, + 527 + ], + "score": 1.0, + "content": "generates the sample trajectory, and", + "type": "text" + }, + { + "bbox": [ + 256, + 515, + 264, + 525 + ], + "score": 0.43, + "content": "\\mathsf { P }", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 515, + 506, + 527 + ], + "score": 1.0, + "content": "is unknown. For example, P can be the simulator at hand,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 525, + 506, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 265, + 538 + ], + "score": 1.0, + "content": "which may not be exactly accurate; and", + "type": "text" + }, + { + "bbox": [ + 265, + 526, + 273, + 536 + ], + "score": 0.48, + "content": "\\mathsf { P }", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 525, + 506, + 538 + ], + "score": 1.0, + "content": "can be the transition kernel of environment 1, from which", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 537, + 505, + 549 + ], + "spans": [ + { + "bbox": [ + 105, + 537, + 505, + 549 + ], + "score": 1.0, + "content": "we can take samples to learn a policy that will be deployed in a similar environment 2. The goal is to", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 547, + 501, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 547, + 243, + 560 + ], + "score": 1.0, + "content": "learn a policy using samples from", + "type": "text" + }, + { + "bbox": [ + 244, + 548, + 252, + 558 + ], + "score": 0.3, + "content": "\\mathsf { P }", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 547, + 501, + 560 + ], + "score": 1.0, + "content": "that performs well when applied to a perturbed MDP from P.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 106, + 564, + 506, + 597 + ], + "lines": [ + { + "bbox": [ + 106, + 564, + 506, + 576 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 506, + 576 + ], + "score": 1.0, + "content": "Motivated by empirical studies of adversarial training in RL [Huang et al., 2017, Kos and Song, 2017,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 575, + 505, + 586 + ], + "spans": [ + { + "bbox": [ + 106, + 575, + 395, + 586 + ], + "score": 1.0, + "content": "Lin et al., 2017, Pattanaik et al., 2018, Mandlekar et al., 2017] and the", + "type": "text" + }, + { + "bbox": [ + 395, + 576, + 404, + 585 + ], + "score": 0.8, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 575, + 505, + 586 + ], + "score": 1.0, + "content": "-contamination model in", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 586, + 492, + 598 + ], + "spans": [ + { + "bbox": [ + 106, + 586, + 279, + 598 + ], + "score": 1.0, + "content": "robust detection [Huber, 1965], we use the", + "type": "text" + }, + { + "bbox": [ + 279, + 586, + 288, + 596 + ], + "score": 0.83, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 586, + 492, + 598 + ], + "score": 1.0, + "content": "-contamination model to define the uncertainty set:", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35 + }, + { + "type": "interline_equation", + "bbox": [ + 159, + 602, + 452, + 617 + ], + "lines": [ + { + "bbox": [ + 159, + 602, + 452, + 617 + ], + "spans": [ + { + "bbox": [ + 159, + 602, + 452, + 617 + ], + "score": 0.83, + "content": "\\begin{array} { r } { \\mathfrak { P } _ { s } ^ { a } = \\big \\{ ( 1 - R ) p _ { s } ^ { a } + R q | q \\in \\mathcal { A } _ { | \\mathcal { S } | } \\big \\} , s \\in \\mathcal { S } , a \\in \\mathcal { A } , \\mathrm { f o r ~ s o m e ~ } 0 \\le R \\le 1 . } \\end{array}", + "type": "interline_equation", + "image_path": "65c42db17c4a64629886dd7a5cba73fa31dcb0cb47e4877794a019ea1569cb7c.jpg" + } + ] + } + ], + "index": 37, + "virtual_lines": [ + { + "bbox": [ + 159, + 602, + 452, + 617 + ], + "spans": [], + "index": 37 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 623, + 505, + 658 + ], + "lines": [ + { + "bbox": [ + 105, + 622, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 131, + 637 + ], + "score": 1.0, + "content": "Here,", + "type": "text" + }, + { + "bbox": [ + 131, + 624, + 142, + 635 + ], + "score": 0.89, + "content": "p _ { s } ^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 622, + 293, + 637 + ], + "score": 1.0, + "content": "is the centroid of the uncertainty set", + "type": "text" + }, + { + "bbox": [ + 293, + 624, + 306, + 635 + ], + "score": 0.9, + "content": "\\mathcal { P } _ { s } ^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 622, + 318, + 637 + ], + "score": 1.0, + "content": "at", + "type": "text" + }, + { + "bbox": [ + 318, + 623, + 340, + 636 + ], + "score": 0.9, + "content": "( s , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 622, + 441, + 637 + ], + "score": 1.0, + "content": ", which is unknown, and", + "type": "text" + }, + { + "bbox": [ + 441, + 624, + 450, + 633 + ], + "score": 0.83, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 622, + 505, + 637 + ], + "score": 1.0, + "content": "is the design", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 634, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 106, + 634, + 505, + 646 + ], + "score": 1.0, + "content": "parameter of the uncertainty set, which measures the size of the uncertainty set, and is assumed to be", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 102, + 640, + 339, + 663 + ], + "spans": [ + { + "bbox": [ + 102, + 640, + 253, + 663 + ], + "score": 1.0, + "content": "known in the algorithm. We then let", + "type": "text" + }, + { + "bbox": [ + 253, + 645, + 332, + 659 + ], + "score": 0.93, + "content": "\\mathbf { P } = \\bigotimes _ { s \\in \\mathcal { S } , a \\in \\mathcal { A } } \\mathcal { P } _ { s } ^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 640, + 339, + 663 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 106, + 661, + 506, + 725 + ], + "lines": [ + { + "bbox": [ + 105, + 661, + 506, + 674 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 156, + 674 + ], + "score": 1.0, + "content": "Remark 1.", + "type": "text" + }, + { + "bbox": [ + 156, + 662, + 165, + 672 + ], + "score": 0.73, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 661, + 506, + 674 + ], + "score": 1.0, + "content": "-contamination model is closely related to other uncertainty set models like total varia-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 673, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 106, + 673, + 142, + 684 + ], + "score": 1.0, + "content": "tion and", + "type": "text" + }, + { + "bbox": [ + 142, + 673, + 156, + 682 + ], + "score": 0.32, + "content": "K L", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 673, + 291, + 684 + ], + "score": 1.0, + "content": "-divergence. It can be shown that", + "type": "text" + }, + { + "bbox": [ + 291, + 673, + 299, + 682 + ], + "score": 0.78, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 673, + 422, + 684 + ], + "score": 1.0, + "content": "-contamination set certered at", + "type": "text" + }, + { + "bbox": [ + 422, + 675, + 428, + 684 + ], + "score": 0.5, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 673, + 505, + 684 + ], + "score": 1.0, + "content": "is a subset of total", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 682, + 507, + 698 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 172, + 698 + ], + "score": 1.0, + "content": "variation ball :", + "type": "text" + }, + { + "bbox": [ + 172, + 683, + 407, + 696 + ], + "score": 0.85, + "content": "\\{ ( 1 - R ) p + R q | q \\in \\varDelta _ { | \\mathcal { S } | } \\} \\subset \\{ q \\in \\varDelta _ { | \\mathcal { S } | } | d _ { T V } ( p , q ) \\leq R \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 682, + 507, + 698 + ], + "score": 1.0, + "content": ". Hence the total varia-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 694, + 505, + 707 + ], + "spans": [ + { + "bbox": [ + 105, + 694, + 305, + 707 + ], + "score": 1.0, + "content": "tion uncertainty set is less conservative than our", + "type": "text" + }, + { + "bbox": [ + 306, + 696, + 314, + 704 + ], + "score": 0.78, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 694, + 505, + 707 + ], + "score": 1.0, + "content": "-contamination uncertainty set. KL-divergence", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 104, + 705, + 506, + 726 + ], + "spans": [ + { + "bbox": [ + 104, + 708, + 400, + 725 + ], + "score": 1.0, + "content": "moreover can be related to total variation using Pinsker’s inequality, i.e.,", + "type": "text" + }, + { + "bbox": [ + 400, + 705, + 502, + 726 + ], + "score": 0.94, + "content": "\\begin{array} { r } { d _ { T V } ( p , q ) \\leq \\sqrt { \\frac 1 2 d _ { K L } ( p , q ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 708, + 506, + 725 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 43 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 310, + 752 + ], + "score": 1.0, + "content": "4", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 72, + 505, + 106 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 505, + 85 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 424, + 85 + ], + "score": 1.0, + "content": "El Ghaoui, 2004, Iyengar, 2005]. Define the robust value function of a policy", + "type": "text" + }, + { + "bbox": [ + 424, + 75, + 432, + 83 + ], + "score": 0.75, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 72, + 505, + 85 + ], + "score": 1.0, + "content": "as the worst-case", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 83, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 365, + 96 + ], + "score": 1.0, + "content": "expected accumulated discounted cost following a fixed policy", + "type": "text" + }, + { + "bbox": [ + 365, + 86, + 372, + 93 + ], + "score": 0.73, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 83, + 505, + 96 + ], + "score": 1.0, + "content": "over all transition kernels in the", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 95, + 171, + 108 + ], + "spans": [ + { + "bbox": [ + 106, + 95, + 171, + 108 + ], + "score": 1.0, + "content": "uncertainty set:", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 72, + 505, + 108 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 207, + 111, + 402, + 145 + ], + "lines": [ + { + "bbox": [ + 207, + 111, + 402, + 145 + ], + "spans": [ + { + "bbox": [ + 207, + 111, + 402, + 145 + ], + "score": 0.94, + "content": "V ^ { \\pi } ( s ) = \\operatorname* { m a x } _ { \\kappa } \\mathbb { E } _ { \\kappa } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } c ( S _ { t } , A _ { t } ) | S _ { 0 } = s , \\pi \\right] ,", + "type": "interline_equation", + "image_path": "5c8479616301ab2d49dcead2362db851758cb5ec733965b24b83759529ec5c2b.jpg" + } + ] + } + ], + "index": 3.5, + "virtual_lines": [ + { + "bbox": [ + 207, + 111, + 402, + 128.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 207, + 128.0, + 402, + 145.0 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 150, + 506, + 194 + ], + "lines": [ + { + "bbox": [ + 106, + 150, + 506, + 162 + ], + "spans": [ + { + "bbox": [ + 106, + 150, + 132, + 162 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 151, + 146, + 161 + ], + "score": 0.89, + "content": "\\mathbb { E } _ { \\kappa }", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 150, + 384, + 162 + ], + "score": 1.0, + "content": "denotes the expectation when the state transits according to", + "type": "text" + }, + { + "bbox": [ + 385, + 153, + 391, + 160 + ], + "score": 0.72, + "content": "\\kappa", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 150, + 506, + 162 + ], + "score": 1.0, + "content": ". Similarly, define the robust", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 102, + 158, + 508, + 177 + ], + "spans": [ + { + "bbox": [ + 102, + 158, + 245, + 177 + ], + "score": 1.0, + "content": "action-value function for a policy", + "type": "text" + }, + { + "bbox": [ + 246, + 161, + 501, + 174 + ], + "score": 0.84, + "content": "\\begin{array} { r } { \\pi \\colon Q ^ { \\pi } ( s , a ) = \\operatorname* { m a x } _ { \\kappa } \\mathbb { E } _ { \\kappa } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } c ( S _ { t } , A _ { t } ) | \\dot { S _ { 0 } } = s , A _ { 0 } = a , \\pi \\right] } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 158, + 508, + 177 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 171, + 507, + 185 + ], + "spans": [ + { + "bbox": [ + 105, + 171, + 336, + 185 + ], + "score": 1.0, + "content": "The goal of robust RL is to find the optimal robust policy", + "type": "text" + }, + { + "bbox": [ + 336, + 173, + 347, + 182 + ], + "score": 0.85, + "content": "\\pi ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 171, + 507, + 185 + ], + "score": 1.0, + "content": "that minimizes the worst-case accumu-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 183, + 195, + 195 + ], + "spans": [ + { + "bbox": [ + 105, + 183, + 195, + 195 + ], + "score": 1.0, + "content": "lated discounted cost:", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6.5, + "bbox_fs": [ + 102, + 150, + 508, + 195 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 245, + 200, + 366, + 218 + ], + "lines": [ + { + "bbox": [ + 245, + 200, + 366, + 218 + ], + "spans": [ + { + "bbox": [ + 245, + 200, + 366, + 218 + ], + "score": 0.92, + "content": "\\pi ^ { * } = \\arg \\operatorname* { m i n } _ { \\pi } V ^ { \\pi } ( s ) , \\forall s \\in \\mathcal { S } .", + "type": "interline_equation", + "image_path": "323b5dbc7086c2a15a4b350fd9583f048099d6491412754c182179252321aee0.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 245, + 200, + 366, + 218 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 224, + 469, + 237 + ], + "lines": [ + { + "bbox": [ + 104, + 222, + 470, + 240 + ], + "spans": [ + { + "bbox": [ + 104, + 222, + 169, + 240 + ], + "score": 1.0, + "content": "We also denote", + "type": "text" + }, + { + "bbox": [ + 169, + 224, + 187, + 236 + ], + "score": 0.92, + "content": "V ^ { \\pi ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 222, + 206, + 240 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 207, + 224, + 224, + 237 + ], + "score": 0.91, + "content": "Q ^ { \\pi ^ { * } }", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 222, + 238, + 240 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 239, + 225, + 252, + 236 + ], + "score": 0.85, + "content": "V ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 222, + 270, + 240 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 271, + 226, + 284, + 237 + ], + "score": 0.9, + "content": "Q ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 222, + 357, + 240 + ], + "score": 1.0, + "content": ", respectively, and", + "type": "text" + }, + { + "bbox": [ + 357, + 225, + 466, + 238 + ], + "score": 0.92, + "content": "\\begin{array} { r } { V ^ { * } ( s ) = \\operatorname* { m i n } _ { a \\in \\mathcal { A } } Q ^ { * } ( s , a ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 466, + 222, + 470, + 240 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10, + "bbox_fs": [ + 104, + 222, + 470, + 240 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 241, + 506, + 310 + ], + "lines": [ + { + "bbox": [ + 105, + 240, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 240, + 506, + 254 + ], + "score": 1.0, + "content": "Note that a transition kernel is a collection of conditional distributions. Therefore, the uncertainty set", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 252, + 505, + 266 + ], + "spans": [ + { + "bbox": [ + 106, + 253, + 115, + 263 + ], + "score": 0.64, + "content": "\\mathbf { P }", + "type": "inline_equation" + }, + { + "bbox": [ + 116, + 252, + 383, + 266 + ], + "score": 1.0, + "content": "of the transition kernel can be equivalently written as a collection of", + "type": "text" + }, + { + "bbox": [ + 383, + 253, + 395, + 265 + ], + "score": 0.9, + "content": "\\mathcal { P } _ { s } ^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 252, + 422, + 266 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 423, + 253, + 475, + 264 + ], + "score": 0.86, + "content": "s \\in \\mathcal { S } , a \\in \\mathcal { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 252, + 505, + 266 + ], + "score": 1.0, + "content": ", where", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 263, + 505, + 279 + ], + "spans": [ + { + "bbox": [ + 106, + 265, + 119, + 277 + ], + "score": 0.87, + "content": "\\mathcal { P } _ { s } ^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 263, + 259, + 279 + ], + "score": 1.0, + "content": "is a set of conditional distributions", + "type": "text" + }, + { + "bbox": [ + 260, + 266, + 271, + 277 + ], + "score": 0.89, + "content": "p _ { s } ^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 263, + 407, + 279 + ], + "score": 1.0, + "content": "over the state space S. Denote by", + "type": "text" + }, + { + "bbox": [ + 407, + 264, + 505, + 277 + ], + "score": 0.91, + "content": "\\sigma _ { \\mathcal { P } } ( v ) \\triangleq \\operatorname* { m a x } _ { p \\in \\mathcal { P } } ( p ^ { \\top } v )", + "type": "inline_equation" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 275, + 506, + 290 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 234, + 290 + ], + "score": 1.0, + "content": "the support function of vector", + "type": "text" + }, + { + "bbox": [ + 235, + 279, + 241, + 286 + ], + "score": 0.74, + "content": "v", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 275, + 401, + 290 + ], + "score": 1.0, + "content": "over a set of probability distributions", + "type": "text" + }, + { + "bbox": [ + 402, + 277, + 410, + 286 + ], + "score": 0.78, + "content": "\\mathcal { P }", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 275, + 506, + 290 + ], + "score": 1.0, + "content": ". For robust MDP, the", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 286, + 506, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 286, + 506, + 299 + ], + "score": 1.0, + "content": "following robust analogue of the Bellman recursion was provided in [Nilim and El Ghaoui, 2004,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 298, + 170, + 311 + ], + "spans": [ + { + "bbox": [ + 105, + 298, + 170, + 311 + ], + "score": 1.0, + "content": "Iyengar, 2005].", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 13.5, + "bbox_fs": [ + 105, + 240, + 506, + 311 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 312, + 505, + 325 + ], + "lines": [ + { + "bbox": [ + 106, + 311, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 106, + 311, + 479, + 326 + ], + "score": 1.0, + "content": "Theorem 1. [Nilim and El Ghaoui, 2004] The following perfect duality condition holds for all", + "type": "text" + }, + { + "bbox": [ + 479, + 313, + 502, + 323 + ], + "score": 0.88, + "content": "s \\in \\mathcal { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 311, + 506, + 326 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17, + "bbox_fs": [ + 106, + 311, + 506, + 326 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 120, + 329, + 478, + 363 + ], + "lines": [ + { + "bbox": [ + 120, + 329, + 478, + 363 + ], + "spans": [ + { + "bbox": [ + 120, + 329, + 478, + 363 + ], + "score": 0.94, + "content": "\\operatorname* { m i n } _ { \\pi } \\operatorname* { m a x } _ { \\kappa } \\mathbb { E } _ { \\kappa } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } c ( S _ { t } , A _ { t } ) \\big | \\pi , S _ { 0 } = s \\right] = \\operatorname* { m a x } _ { \\kappa } \\operatorname* { m i n } _ { \\pi } \\mathbb { E } _ { \\kappa } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } c ( S _ { t } , A _ { t } ) \\big | \\pi , S _ { 0 } = s \\right] .", + "type": "interline_equation", + "image_path": "e93045239e0b7e332eb486ed1d6c8b17f4379e21d3033a60baeafcd33b64fda6.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 120, + 329, + 478, + 340.3333333333333 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 120, + 340.3333333333333, + 478, + 351.66666666666663 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 120, + 351.66666666666663, + 478, + 362.99999999999994 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 368, + 505, + 394 + ], + "lines": [ + { + "bbox": [ + 105, + 367, + 505, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 249, + 383 + ], + "score": 1.0, + "content": "The optimal robust value function", + "type": "text" + }, + { + "bbox": [ + 249, + 369, + 263, + 379 + ], + "score": 0.88, + "content": "V ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 367, + 301, + 383 + ], + "score": 1.0, + "content": "satisfies", + "type": "text" + }, + { + "bbox": [ + 301, + 368, + 467, + 381 + ], + "score": 0.89, + "content": "\\begin{array} { r } { V ^ { * } ( s ) = \\operatorname* { m i n } _ { a \\in \\mathcal { A } } \\bigl ( c ( s , a ) + \\gamma \\sigma _ { \\mathscr { P } _ { s } ^ { a } } ( V ^ { * } ) \\bigr ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 467, + 367, + 505, + 383 + ], + "score": 1.0, + "content": ", and the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 379, + 438, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 379, + 254, + 396 + ], + "score": 1.0, + "content": "optimal robust action-value function", + "type": "text" + }, + { + "bbox": [ + 254, + 381, + 268, + 393 + ], + "score": 0.89, + "content": "Q ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 379, + 303, + 396 + ], + "score": 1.0, + "content": "satisfies", + "type": "text" + }, + { + "bbox": [ + 303, + 382, + 433, + 394 + ], + "score": 0.89, + "content": "Q ^ { * } ( s , a ) = c ( s , a ) + \\gamma \\sigma _ { \\mathscr P _ { s } ^ { a } } ( V ^ { * } )", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 379, + 438, + 396 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 367, + 505, + 396 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 402, + 505, + 448 + ], + "lines": [ + { + "bbox": [ + 105, + 402, + 506, + 416 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 255, + 416 + ], + "score": 1.0, + "content": "Define the robust Bellman operator", + "type": "text" + }, + { + "bbox": [ + 256, + 403, + 266, + 413 + ], + "score": 0.64, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 402, + 280, + 416 + ], + "score": 1.0, + "content": "by", + "type": "text" + }, + { + "bbox": [ + 280, + 403, + 470, + 416 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\mathbf { T } Q ( s , a ) = c ( s , a ) + \\gamma \\sigma _ { \\mathcal { P } _ { s } ^ { a } } ( \\operatorname* { m i n } _ { a \\in \\mathcal { A } } Q ( s , a ) ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 402, + 506, + 416 + ], + "score": 1.0, + "content": ". It was", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 414, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 106, + 414, + 336, + 425 + ], + "score": 1.0, + "content": "shown in [Nilim and El Ghaoui, 2004, Iyengar, 2005] that", + "type": "text" + }, + { + "bbox": [ + 336, + 415, + 345, + 424 + ], + "score": 0.76, + "content": "\\mathbf { T }", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 414, + 505, + 425 + ], + "score": 1.0, + "content": "is a contraction and its fixed point is the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 424, + 506, + 438 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 165, + 438 + ], + "score": 1.0, + "content": "optimal robust", + "type": "text" + }, + { + "bbox": [ + 166, + 425, + 179, + 436 + ], + "score": 0.88, + "content": "Q ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 424, + 353, + 438 + ], + "score": 1.0, + "content": ". When the uncertainty set is known, so that", + "type": "text" + }, + { + "bbox": [ + 353, + 426, + 370, + 437 + ], + "score": 0.87, + "content": "\\sigma _ { \\mathcal P _ { s } ^ { a } }", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 424, + 473, + 438 + ], + "score": 1.0, + "content": "can be computed exactly,", + "type": "text" + }, + { + "bbox": [ + 474, + 425, + 487, + 434 + ], + "score": 0.86, + "content": "V ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 424, + 506, + 438 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 107, + 435, + 465, + 449 + ], + "spans": [ + { + "bbox": [ + 107, + 436, + 120, + 447 + ], + "score": 0.88, + "content": "Q ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 120, + 435, + 465, + 449 + ], + "score": 1.0, + "content": "can be solved by dynamic programming [Iyengar, 2005, Nilim and El Ghaoui, 2004].", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 402, + 506, + 449 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 462, + 426, + 477 + ], + "lines": [ + { + "bbox": [ + 104, + 461, + 427, + 479 + ], + "spans": [ + { + "bbox": [ + 104, + 461, + 427, + 479 + ], + "score": 1.0, + "content": "3 R-Contamination Model For Uncertainty Set Construction", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 106, + 487, + 434, + 499 + ], + "lines": [ + { + "bbox": [ + 105, + 486, + 435, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 335, + 500 + ], + "score": 1.0, + "content": "In this section, we construct the uncertainty set using the", + "type": "text" + }, + { + "bbox": [ + 335, + 488, + 344, + 497 + ], + "score": 0.81, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 486, + 435, + 500 + ], + "score": 1.0, + "content": "-contamination model.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 486, + 435, + 500 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 503, + 506, + 560 + ], + "lines": [ + { + "bbox": [ + 105, + 504, + 506, + 517 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 123, + 517 + ], + "score": 1.0, + "content": "Let", + "type": "text" + }, + { + "bbox": [ + 123, + 504, + 226, + 516 + ], + "score": 0.9, + "content": "\\mathsf { P } = \\{ p _ { s } ^ { a } , s \\in \\mathscr { S } , a \\in \\mathscr { A } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 504, + 506, + 517 + ], + "score": 1.0, + "content": "be the centroid of the uncertainty set, i.e., the transition kernel that", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 515, + 506, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 256, + 527 + ], + "score": 1.0, + "content": "generates the sample trajectory, and", + "type": "text" + }, + { + "bbox": [ + 256, + 515, + 264, + 525 + ], + "score": 0.43, + "content": "\\mathsf { P }", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 515, + 506, + 527 + ], + "score": 1.0, + "content": "is unknown. For example, P can be the simulator at hand,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 525, + 506, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 265, + 538 + ], + "score": 1.0, + "content": "which may not be exactly accurate; and", + "type": "text" + }, + { + "bbox": [ + 265, + 526, + 273, + 536 + ], + "score": 0.48, + "content": "\\mathsf { P }", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 525, + 506, + 538 + ], + "score": 1.0, + "content": "can be the transition kernel of environment 1, from which", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 537, + 505, + 549 + ], + "spans": [ + { + "bbox": [ + 105, + 537, + 505, + 549 + ], + "score": 1.0, + "content": "we can take samples to learn a policy that will be deployed in a similar environment 2. The goal is to", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 547, + 501, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 547, + 243, + 560 + ], + "score": 1.0, + "content": "learn a policy using samples from", + "type": "text" + }, + { + "bbox": [ + 244, + 548, + 252, + 558 + ], + "score": 0.3, + "content": "\\mathsf { P }", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 547, + 501, + 560 + ], + "score": 1.0, + "content": "that performs well when applied to a perturbed MDP from P.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 504, + 506, + 560 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 564, + 506, + 597 + ], + "lines": [ + { + "bbox": [ + 106, + 564, + 506, + 576 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 506, + 576 + ], + "score": 1.0, + "content": "Motivated by empirical studies of adversarial training in RL [Huang et al., 2017, Kos and Song, 2017,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 575, + 505, + 586 + ], + "spans": [ + { + "bbox": [ + 106, + 575, + 395, + 586 + ], + "score": 1.0, + "content": "Lin et al., 2017, Pattanaik et al., 2018, Mandlekar et al., 2017] and the", + "type": "text" + }, + { + "bbox": [ + 395, + 576, + 404, + 585 + ], + "score": 0.8, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 575, + 505, + 586 + ], + "score": 1.0, + "content": "-contamination model in", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 586, + 492, + 598 + ], + "spans": [ + { + "bbox": [ + 106, + 586, + 279, + 598 + ], + "score": 1.0, + "content": "robust detection [Huber, 1965], we use the", + "type": "text" + }, + { + "bbox": [ + 279, + 586, + 288, + 596 + ], + "score": 0.83, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 288, + 586, + 492, + 598 + ], + "score": 1.0, + "content": "-contamination model to define the uncertainty set:", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35, + "bbox_fs": [ + 106, + 564, + 506, + 598 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 159, + 602, + 452, + 617 + ], + "lines": [ + { + "bbox": [ + 159, + 602, + 452, + 617 + ], + "spans": [ + { + "bbox": [ + 159, + 602, + 452, + 617 + ], + "score": 0.83, + "content": "\\begin{array} { r } { \\mathfrak { P } _ { s } ^ { a } = \\big \\{ ( 1 - R ) p _ { s } ^ { a } + R q | q \\in \\mathcal { A } _ { | \\mathcal { S } | } \\big \\} , s \\in \\mathcal { S } , a \\in \\mathcal { A } , \\mathrm { f o r ~ s o m e ~ } 0 \\le R \\le 1 . } \\end{array}", + "type": "interline_equation", + "image_path": "65c42db17c4a64629886dd7a5cba73fa31dcb0cb47e4877794a019ea1569cb7c.jpg" + } + ] + } + ], + "index": 37, + "virtual_lines": [ + { + "bbox": [ + 159, + 602, + 452, + 617 + ], + "spans": [], + "index": 37 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 623, + 505, + 658 + ], + "lines": [ + { + "bbox": [ + 105, + 622, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 131, + 637 + ], + "score": 1.0, + "content": "Here,", + "type": "text" + }, + { + "bbox": [ + 131, + 624, + 142, + 635 + ], + "score": 0.89, + "content": "p _ { s } ^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 622, + 293, + 637 + ], + "score": 1.0, + "content": "is the centroid of the uncertainty set", + "type": "text" + }, + { + "bbox": [ + 293, + 624, + 306, + 635 + ], + "score": 0.9, + "content": "\\mathcal { P } _ { s } ^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 622, + 318, + 637 + ], + "score": 1.0, + "content": "at", + "type": "text" + }, + { + "bbox": [ + 318, + 623, + 340, + 636 + ], + "score": 0.9, + "content": "( s , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 622, + 441, + 637 + ], + "score": 1.0, + "content": ", which is unknown, and", + "type": "text" + }, + { + "bbox": [ + 441, + 624, + 450, + 633 + ], + "score": 0.83, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 622, + 505, + 637 + ], + "score": 1.0, + "content": "is the design", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 634, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 106, + 634, + 505, + 646 + ], + "score": 1.0, + "content": "parameter of the uncertainty set, which measures the size of the uncertainty set, and is assumed to be", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 102, + 640, + 339, + 663 + ], + "spans": [ + { + "bbox": [ + 102, + 640, + 253, + 663 + ], + "score": 1.0, + "content": "known in the algorithm. We then let", + "type": "text" + }, + { + "bbox": [ + 253, + 645, + 332, + 659 + ], + "score": 0.93, + "content": "\\mathbf { P } = \\bigotimes _ { s \\in \\mathcal { S } , a \\in \\mathcal { A } } \\mathcal { P } _ { s } ^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 640, + 339, + 663 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39, + "bbox_fs": [ + 102, + 622, + 505, + 663 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 661, + 506, + 725 + ], + "lines": [ + { + "bbox": [ + 105, + 661, + 506, + 674 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 156, + 674 + ], + "score": 1.0, + "content": "Remark 1.", + "type": "text" + }, + { + "bbox": [ + 156, + 662, + 165, + 672 + ], + "score": 0.73, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 661, + 506, + 674 + ], + "score": 1.0, + "content": "-contamination model is closely related to other uncertainty set models like total varia-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 673, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 106, + 673, + 142, + 684 + ], + "score": 1.0, + "content": "tion and", + "type": "text" + }, + { + "bbox": [ + 142, + 673, + 156, + 682 + ], + "score": 0.32, + "content": "K L", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 673, + 291, + 684 + ], + "score": 1.0, + "content": "-divergence. It can be shown that", + "type": "text" + }, + { + "bbox": [ + 291, + 673, + 299, + 682 + ], + "score": 0.78, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 673, + 422, + 684 + ], + "score": 1.0, + "content": "-contamination set certered at", + "type": "text" + }, + { + "bbox": [ + 422, + 675, + 428, + 684 + ], + "score": 0.5, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 673, + 505, + 684 + ], + "score": 1.0, + "content": "is a subset of total", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 682, + 507, + 698 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 172, + 698 + ], + "score": 1.0, + "content": "variation ball :", + "type": "text" + }, + { + "bbox": [ + 172, + 683, + 407, + 696 + ], + "score": 0.85, + "content": "\\{ ( 1 - R ) p + R q | q \\in \\varDelta _ { | \\mathcal { S } | } \\} \\subset \\{ q \\in \\varDelta _ { | \\mathcal { S } | } | d _ { T V } ( p , q ) \\leq R \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 682, + 507, + 698 + ], + "score": 1.0, + "content": ". Hence the total varia-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 694, + 505, + 707 + ], + "spans": [ + { + "bbox": [ + 105, + 694, + 305, + 707 + ], + "score": 1.0, + "content": "tion uncertainty set is less conservative than our", + "type": "text" + }, + { + "bbox": [ + 306, + 696, + 314, + 704 + ], + "score": 0.78, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 694, + 505, + 707 + ], + "score": 1.0, + "content": "-contamination uncertainty set. KL-divergence", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 104, + 705, + 506, + 726 + ], + "spans": [ + { + "bbox": [ + 104, + 708, + 400, + 725 + ], + "score": 1.0, + "content": "moreover can be related to total variation using Pinsker’s inequality, i.e.,", + "type": "text" + }, + { + "bbox": [ + 400, + 705, + 502, + 726 + ], + "score": 0.94, + "content": "\\begin{array} { r } { d _ { T V } ( p , q ) \\leq \\sqrt { \\frac 1 2 d _ { K L } ( p , q ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 708, + 506, + 725 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 43, + "bbox_fs": [ + 104, + 661, + 507, + 726 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 70, + 302, + 85 + ], + "lines": [ + { + "bbox": [ + 104, + 68, + 303, + 88 + ], + "spans": [ + { + "bbox": [ + 104, + 68, + 303, + 88 + ], + "score": 1.0, + "content": "4 Tabular Case: Robust Q-Learning", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 96, + 505, + 141 + ], + "lines": [ + { + "bbox": [ + 106, + 97, + 504, + 108 + ], + "spans": [ + { + "bbox": [ + 106, + 97, + 504, + 108 + ], + "score": 1.0, + "content": "In this section, we focus on the tabular case with finite state and action spaces. We focus on the", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 107, + 505, + 120 + ], + "spans": [ + { + "bbox": [ + 105, + 107, + 505, + 120 + ], + "score": 1.0, + "content": "asynchronous setting where a single sample trajectory is available with Markovian noise. We will", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 118, + 506, + 132 + ], + "spans": [ + { + "bbox": [ + 106, + 118, + 395, + 132 + ], + "score": 1.0, + "content": "develop an efficient approach to estimate the unknown uncertainty set", + "type": "text" + }, + { + "bbox": [ + 395, + 119, + 404, + 129 + ], + "score": 0.51, + "content": "\\mathbf { P }", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 118, + 506, + 132 + ], + "score": 1.0, + "content": ", and further the support", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 129, + 371, + 142 + ], + "spans": [ + { + "bbox": [ + 105, + 129, + 142, + 142 + ], + "score": 1.0, + "content": "function", + "type": "text" + }, + { + "bbox": [ + 142, + 129, + 170, + 142 + ], + "score": 0.94, + "content": "\\sigma _ { \\mathcal { P } _ { s } ^ { a } } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 129, + 371, + 142 + ], + "score": 1.0, + "content": ", and then design our robust Q-learning algorithm.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 106, + 145, + 505, + 244 + ], + "lines": [ + { + "bbox": [ + 106, + 146, + 504, + 158 + ], + "spans": [ + { + "bbox": [ + 106, + 146, + 403, + 157 + ], + "score": 1.0, + "content": "We propose an efficient and data-driven approach to estimate the unknown", + "type": "text" + }, + { + "bbox": [ + 403, + 146, + 414, + 158 + ], + "score": 0.89, + "content": "p _ { s } ^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 146, + 504, + 157 + ], + "score": 1.0, + "content": "and thus the unknown", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 104, + 155, + 505, + 171 + ], + "spans": [ + { + "bbox": [ + 104, + 155, + 169, + 171 + ], + "score": 1.0, + "content": "uncertainty set", + "type": "text" + }, + { + "bbox": [ + 169, + 157, + 182, + 169 + ], + "score": 0.89, + "content": "\\mathcal { P } _ { s } ^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 155, + 216, + 171 + ], + "score": 1.0, + "content": "for any", + "type": "text" + }, + { + "bbox": [ + 216, + 157, + 241, + 167 + ], + "score": 0.9, + "content": "s \\in \\mathcal { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 155, + 260, + 171 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 261, + 157, + 288, + 167 + ], + "score": 0.9, + "content": "a \\in { \\mathcal { A } }", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 155, + 434, + 171 + ], + "score": 1.0, + "content": ". Specifically, denote the sample at", + "type": "text" + }, + { + "bbox": [ + 434, + 158, + 439, + 167 + ], + "score": 0.76, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 155, + 505, + 171 + ], + "score": 1.0, + "content": "-th time step by", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 166, + 504, + 186 + ], + "spans": [ + { + "bbox": [ + 106, + 169, + 184, + 181 + ], + "score": 0.92, + "content": "O _ { t } = ( s _ { t } , a _ { t } , s _ { t + 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 166, + 239, + 186 + ], + "score": 1.0, + "content": ". We then use", + "type": "text" + }, + { + "bbox": [ + 240, + 169, + 252, + 180 + ], + "score": 0.87, + "content": "O _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 166, + 459, + 186 + ], + "score": 1.0, + "content": "to obtain the maximum likelihood estimate (MLE)", + "type": "text" + }, + { + "bbox": [ + 460, + 168, + 504, + 182 + ], + "score": 0.83, + "content": "\\hat { p } _ { t } \\triangleq \\mathbb { 1 } _ { s _ { t + 1 } }", + "type": "inline_equation" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 178, + 504, + 195 + ], + "spans": [ + { + "bbox": [ + 104, + 178, + 203, + 195 + ], + "score": 1.0, + "content": "of the transition kernel", + "type": "text" + }, + { + "bbox": [ + 203, + 181, + 217, + 193 + ], + "score": 0.9, + "content": "p _ { s _ { t } } ^ { a _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 178, + 250, + 195 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 251, + 181, + 273, + 192 + ], + "score": 0.91, + "content": "\\mathbb { 1 } _ { s _ { t + 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 178, + 485, + 195 + ], + "score": 1.0, + "content": "is a probability distribution taking probability 1 at", + "type": "text" + }, + { + "bbox": [ + 485, + 183, + 504, + 192 + ], + "score": 0.88, + "content": "s _ { t + 1 }", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 191, + 506, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 420, + 204 + ], + "score": 1.0, + "content": "and 0 at other states. This is an unbiased estimate of the transition kernel", + "type": "text" + }, + { + "bbox": [ + 421, + 192, + 435, + 204 + ], + "score": 0.87, + "content": "p _ { s _ { t } } ^ { a _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 191, + 506, + 204 + ], + "score": 1.0, + "content": "conditioning on", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 202, + 504, + 220 + ], + "spans": [ + { + "bbox": [ + 107, + 205, + 139, + 217 + ], + "score": 0.92, + "content": "S _ { t } = s _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 202, + 158, + 220 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 158, + 205, + 192, + 217 + ], + "score": 0.92, + "content": "A _ { t } = a _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 202, + 363, + 220 + ], + "score": 1.0, + "content": ". We then design a sample-based estimate", + "type": "text" + }, + { + "bbox": [ + 363, + 203, + 504, + 218 + ], + "score": 0.92, + "content": "\\hat { \\mathcal { P } } _ { t } \\triangleq \\left\\{ ( 1 - R ) \\hat { p } _ { t } + R q | q \\in \\Delta _ { | \\mathcal { S } | } \\right\\}", + "type": "inline_equation" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 217, + 505, + 233 + ], + "spans": [ + { + "bbox": [ + 105, + 217, + 195, + 233 + ], + "score": 1.0, + "content": "of the uncertainty set", + "type": "text" + }, + { + "bbox": [ + 195, + 219, + 210, + 232 + ], + "score": 0.9, + "content": "\\mathcal { P } _ { s _ { t } } ^ { a _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 217, + 378, + 233 + ], + "score": 1.0, + "content": ". Using the sample-based uncertainty set", + "type": "text" + }, + { + "bbox": [ + 378, + 218, + 389, + 230 + ], + "score": 0.88, + "content": "\\hat { \\mathcal { P } } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 217, + 505, + 233 + ], + "score": 1.0, + "content": ", we construct the following", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 229, + 505, + 246 + ], + "spans": [ + { + "bbox": [ + 104, + 229, + 317, + 246 + ], + "score": 1.0, + "content": "robust Q-learning algorithm in Algorithm 1. For any", + "type": "text" + }, + { + "bbox": [ + 317, + 232, + 322, + 241 + ], + "score": 0.59, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 229, + 326, + 246 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 326, + 231, + 360, + 245 + ], + "score": 0.92, + "content": "\\sigma _ { \\hat { \\mathcal { P } } _ { t } } ( V _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 229, + 460, + 246 + ], + "score": 1.0, + "content": "can be easily computed:", + "type": "text" + }, + { + "bbox": [ + 460, + 231, + 505, + 245 + ], + "score": 0.94, + "content": "\\sigma _ { \\hat { \\Phi } _ { t } } ( V _ { t } ) =", + "type": "inline_equation" + } + ], + "index": 12 + } + ], + "index": 8.5 + }, + { + "type": "title", + "bbox": [ + 107, + 258, + 239, + 270 + ], + "lines": [ + { + "bbox": [ + 105, + 256, + 240, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 256, + 240, + 273 + ], + "score": 1.0, + "content": "Algorithm 1 Robust Q-Learning", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 272, + 448, + 285 + ], + "lines": [ + { + "bbox": [ + 105, + 270, + 448, + 287 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 167, + 287 + ], + "score": 1.0, + "content": "Initialization:", + "type": "text" + }, + { + "bbox": [ + 167, + 273, + 176, + 282 + ], + "score": 0.58, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 270, + 179, + 287 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 180, + 272, + 215, + 285 + ], + "score": 0.91, + "content": "Q _ { 0 } ( s , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 270, + 243, + 287 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 244, + 272, + 303, + 285 + ], + "score": 0.91, + "content": "( s , a ) \\in \\mathcal { S } \\times \\mathcal { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 270, + 371, + 287 + ], + "score": 1.0, + "content": ", behavior policy", + "type": "text" + }, + { + "bbox": [ + 372, + 275, + 382, + 284 + ], + "score": 0.69, + "content": "\\pi _ { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 270, + 386, + 287 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 387, + 275, + 396, + 284 + ], + "score": 0.61, + "content": "s _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 270, + 437, + 287 + ], + "score": 1.0, + "content": ", step size", + "type": "text" + }, + { + "bbox": [ + 437, + 275, + 448, + 284 + ], + "score": 0.7, + "content": "\\alpha _ { t }", + "type": "inline_equation" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 110, + 286, + 365, + 359 + ], + "lines": [ + { + "bbox": [ + 110, + 285, + 236, + 297 + ], + "spans": [ + { + "bbox": [ + 110, + 285, + 138, + 297 + ], + "score": 1.0, + "content": "1: for", + "type": "text" + }, + { + "bbox": [ + 138, + 286, + 222, + 297 + ], + "score": 0.85, + "content": "t = 0 , 1 , 2 , . . . , T - 1", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 285, + 236, + 297 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 111, + 296, + 261, + 309 + ], + "spans": [ + { + "bbox": [ + 111, + 297, + 121, + 308 + ], + "score": 1.0, + "content": "2:", + "type": "text" + }, + { + "bbox": [ + 132, + 296, + 165, + 309 + ], + "score": 1.0, + "content": "Choose", + "type": "text" + }, + { + "bbox": [ + 165, + 298, + 176, + 307 + ], + "score": 0.85, + "content": "a _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 296, + 229, + 309 + ], + "score": 1.0, + "content": "according to", + "type": "text" + }, + { + "bbox": [ + 229, + 296, + 261, + 308 + ], + "score": 0.9, + "content": "\\pi _ { b } ( \\cdot | s _ { t } )", + "type": "inline_equation" + } + ], + "index": 16 + }, + { + "bbox": [ + 111, + 307, + 216, + 320 + ], + "spans": [ + { + "bbox": [ + 111, + 307, + 121, + 319 + ], + "score": 1.0, + "content": "3:", + "type": "text" + }, + { + "bbox": [ + 133, + 307, + 168, + 320 + ], + "score": 1.0, + "content": "Observe", + "type": "text" + }, + { + "bbox": [ + 169, + 309, + 189, + 319 + ], + "score": 0.87, + "content": "s _ { t + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 307, + 206, + 320 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 207, + 309, + 216, + 318 + ], + "score": 0.81, + "content": "c _ { t }", + "type": "inline_equation" + } + ], + "index": 17 + }, + { + "bbox": [ + 111, + 318, + 270, + 330 + ], + "spans": [ + { + "bbox": [ + 111, + 318, + 121, + 329 + ], + "score": 1.0, + "content": "4:", + "type": "text" + }, + { + "bbox": [ + 133, + 319, + 270, + 330 + ], + "score": 0.74, + "content": "\\begin{array} { r } { V _ { t } ( s ) \\operatorname* { m i n } _ { a \\in \\mathcal { A } } Q _ { t } ( s , a ) , \\forall s \\in \\mathfrak { s } } \\end{array}", + "type": "inline_equation" + } + ], + "index": 18 + }, + { + "bbox": [ + 111, + 329, + 362, + 342 + ], + "spans": [ + { + "bbox": [ + 111, + 329, + 121, + 341 + ], + "score": 1.0, + "content": "5:", + "type": "text" + }, + { + "bbox": [ + 133, + 329, + 161, + 342 + ], + "score": 1.0, + "content": "Qt+1(", + "type": "text" + }, + { + "bbox": [ + 162, + 329, + 362, + 342 + ], + "score": 0.28, + "content": "s _ { t } , a _ { t } ) \\gets ( 1 - \\alpha _ { t } ) Q _ { t } ( s _ { t } , a _ { t } ) + \\alpha _ { t } ( c _ { t } + \\gamma \\sigma _ { \\hat { \\mathcal P } _ { t } } ( V _ { t } ) )", + "type": "inline_equation" + } + ], + "index": 19 + }, + { + "bbox": [ + 111, + 341, + 308, + 355 + ], + "spans": [ + { + "bbox": [ + 111, + 342, + 121, + 353 + ], + "score": 1.0, + "content": "6:", + "type": "text" + }, + { + "bbox": [ + 134, + 342, + 227, + 354 + ], + "score": 0.91, + "content": "Q _ { t + 1 } ( s , a ) \\gets Q _ { t } ( s , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 341, + 243, + 355 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 243, + 342, + 308, + 354 + ], + "score": 0.83, + "content": "( s , a ) \\neq ( s _ { t } , a _ { t } )", + "type": "inline_equation" + } + ], + "index": 20 + }, + { + "bbox": [ + 110, + 352, + 157, + 365 + ], + "spans": [ + { + "bbox": [ + 110, + 352, + 157, + 365 + ], + "score": 1.0, + "content": "7: end for", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 111, + 354, + 156, + 363 + ], + "lines": [ + { + "bbox": [ + 111, + 354, + 156, + 363 + ], + "spans": [], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 366, + 158, + 378 + ], + "lines": [ + { + "bbox": [ + 105, + 363, + 159, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 143, + 381 + ], + "score": 1.0, + "content": "Output:", + "type": "text" + }, + { + "bbox": [ + 144, + 366, + 159, + 378 + ], + "score": 0.86, + "content": "Q _ { T }", + "type": "inline_equation" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 104, + 391, + 491, + 404 + ], + "lines": [ + { + "bbox": [ + 106, + 391, + 492, + 405 + ], + "spans": [ + { + "bbox": [ + 106, + 392, + 254, + 405 + ], + "score": 0.89, + "content": "\\begin{array} { r } { R \\operatorname* { m a x } _ { s \\in \\mathcal { S } } V _ { t } ( s ) + ( 1 - R ) V _ { t } ( s _ { t + 1 } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 391, + 492, + 405 + ], + "score": 1.0, + "content": ". Hence the update in Algorithm 1 (line 5) can be written as", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "interline_equation", + "bbox": [ + 135, + 410, + 475, + 429 + ], + "lines": [ + { + "bbox": [ + 135, + 410, + 475, + 429 + ], + "spans": [ + { + "bbox": [ + 135, + 410, + 475, + 429 + ], + "score": 0.9, + "content": "Q _ { t + 1 } ( s _ { t } , a _ { t } ) \\gets ( 1 - \\alpha _ { t } ) Q _ { t } ( s _ { t } , a _ { t } ) + \\alpha _ { t } ( c _ { t } + \\gamma R \\operatorname* { m a x } _ { s \\in \\mathcal { S } } V _ { t } ( s ) + \\gamma ( 1 - R ) V _ { t } ( s _ { t + 1 } ) ) .", + "type": "interline_equation", + "image_path": "ebc655b9749ff56b32e6e65d9da89184ff1f6cd737704bf092f1717d68c980c8.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 135, + 410, + 475, + 429 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 440, + 505, + 529 + ], + "lines": [ + { + "bbox": [ + 106, + 440, + 506, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 440, + 506, + 453 + ], + "score": 1.0, + "content": "Compared to the model-based approach, our approach is model-free. It does not require the prior", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 451, + 506, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 344, + 465 + ], + "score": 1.0, + "content": "knowledge of the uncertainty set, i.e., the knowledge of", + "type": "text" + }, + { + "bbox": [ + 344, + 452, + 424, + 464 + ], + "score": 0.93, + "content": "p _ { s } ^ { a } , \\forall s \\in \\mathcal { S } , a \\in \\mathcal { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 451, + 506, + 465 + ], + "score": 1.0, + "content": ". Furthermore, the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 462, + 506, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 463, + 270, + 475 + ], + "score": 1.0, + "content": "memory requirement of our algorithm is", + "type": "text" + }, + { + "bbox": [ + 270, + 462, + 307, + 474 + ], + "score": 0.92, + "content": "| \\mathcal { S } | \\times | \\mathcal { A } |", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 463, + 506, + 475 + ], + "score": 1.0, + "content": "(used to store the Q-table), and unlike the model-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 473, + 506, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 298, + 487 + ], + "score": 1.0, + "content": "based approach it does not need a table of size", + "type": "text" + }, + { + "bbox": [ + 299, + 474, + 328, + 486 + ], + "score": 0.93, + "content": "| \\dot { \\mathcal { S } } | ^ { 2 } | \\mathcal { A } |", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 473, + 362, + 487 + ], + "score": 1.0, + "content": "to store", + "type": "text" + }, + { + "bbox": [ + 363, + 474, + 436, + 486 + ], + "score": 0.9, + "content": "p _ { s } ^ { a } , \\forall s \\in \\mathcal { S } , a \\in \\mathcal { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 473, + 506, + 487 + ], + "score": 1.0, + "content": ", which could be", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 485, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 505, + 497 + ], + "score": 1.0, + "content": "problematic if the state space is large. Moreover, our algorithm does not involve a relaxation of the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 496, + 505, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 505, + 508 + ], + "score": 1.0, + "content": "uncertainty set like the one in [Roy et al., 2017], which will incur a stringent condition on the discount", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 506, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 505, + 519 + ], + "score": 1.0, + "content": "factor to guarantee the convergence. As will be shown below, the convergence of our Algorithm 1", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 518, + 321, + 530 + ], + "spans": [ + { + "bbox": [ + 106, + 518, + 321, + 530 + ], + "score": 1.0, + "content": "does not require any condition on the discount factor.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 107, + 533, + 504, + 556 + ], + "lines": [ + { + "bbox": [ + 105, + 531, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 505, + 547 + ], + "score": 1.0, + "content": "We show in the following theorem that the robust Q-learning algorithm converges asymptotically to", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 545, + 285, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 268, + 558 + ], + "score": 1.0, + "content": "the optimal robust action-value function", + "type": "text" + }, + { + "bbox": [ + 268, + 545, + 282, + 556 + ], + "score": 0.89, + "content": "Q ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 545, + 285, + 558 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 108, + 559, + 502, + 583 + ], + "lines": [ + { + "bbox": [ + 103, + 555, + 503, + 578 + ], + "spans": [ + { + "bbox": [ + 103, + 555, + 308, + 578 + ], + "score": 1.0, + "content": "Theorem 2. (Asymptotic Convergence) If step sizes", + "type": "text" + }, + { + "bbox": [ + 309, + 562, + 320, + 571 + ], + "score": 0.84, + "content": "\\alpha _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 555, + 365, + 578 + ], + "score": 1.0, + "content": "satisfy that", + "type": "text" + }, + { + "bbox": [ + 365, + 560, + 425, + 573 + ], + "score": 0.89, + "content": "\\textstyle \\sum _ { t = 0 } ^ { \\infty } \\alpha _ { t } = \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 555, + 442, + 578 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 442, + 559, + 503, + 573 + ], + "score": 0.9, + "content": "\\textstyle \\sum _ { t = 0 } ^ { \\infty } \\alpha _ { t } ^ { 2 } < \\infty", + "type": "inline_equation" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 570, + 286, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 126, + 583 + ], + "score": 1.0, + "content": "then", + "type": "text" + }, + { + "bbox": [ + 126, + 571, + 166, + 582 + ], + "score": 0.92, + "content": "Q _ { t } \\to Q ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 570, + 179, + 583 + ], + "score": 1.0, + "content": "as", + "type": "text" + }, + { + "bbox": [ + 210, + 570, + 277, + 583 + ], + "score": 1.0, + "content": "with probability", + "type": "text" + }, + { + "bbox": [ + 283, + 570, + 286, + 583 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 108, + 592, + 503, + 615 + ], + "lines": [ + { + "bbox": [ + 106, + 592, + 504, + 605 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 504, + 605 + ], + "score": 1.0, + "content": "To further establish the finite-time error bound for our robust Q-learning algorithm in Algorithm 1,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 602, + 488, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 602, + 488, + 617 + ], + "score": 1.0, + "content": "we make the following assumption that is commonly used in the analysis of vanilla Q-learning.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 106, + 618, + 504, + 641 + ], + "lines": [ + { + "bbox": [ + 105, + 617, + 506, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 386, + 632 + ], + "score": 1.0, + "content": "Assumption 1. The Markov chain induced by the behavior policy", + "type": "text" + }, + { + "bbox": [ + 386, + 621, + 397, + 630 + ], + "score": 0.85, + "content": "\\pi _ { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 617, + 506, + 632 + ], + "score": 1.0, + "content": "and the transition kernel", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 107, + 629, + 266, + 644 + ], + "spans": [ + { + "bbox": [ + 107, + 630, + 179, + 642 + ], + "score": 0.9, + "content": "p _ { s } ^ { a } , \\forall s \\in \\mathcal { S } , a \\in \\mathcal { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 629, + 266, + 644 + ], + "score": 1.0, + "content": "is uniformly ergodic.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5 + }, + { + "type": "text", + "bbox": [ + 107, + 650, + 505, + 705 + ], + "lines": [ + { + "bbox": [ + 105, + 649, + 506, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 123, + 664 + ], + "score": 1.0, + "content": "Let", + "type": "text" + }, + { + "bbox": [ + 123, + 652, + 138, + 662 + ], + "score": 0.87, + "content": "\\mu _ { \\pi _ { b } }", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 649, + 299, + 664 + ], + "score": 1.0, + "content": "denote the stationary distribution over", + "type": "text" + }, + { + "bbox": [ + 300, + 651, + 327, + 661 + ], + "score": 0.89, + "content": "\\mathcal { S } \\times \\mathcal { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 649, + 376, + 664 + ], + "score": 1.0, + "content": "induced by", + "type": "text" + }, + { + "bbox": [ + 377, + 653, + 388, + 662 + ], + "score": 0.85, + "content": "\\pi _ { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 649, + 407, + 664 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 407, + 651, + 483, + 662 + ], + "score": 0.9, + "content": "p _ { s } ^ { a } , \\forall s \\in \\mathcal { S } , a \\in \\mathcal { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 649, + 506, + 664 + ], + "score": 1.0, + "content": ". We", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 660, + 507, + 677 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 183, + 677 + ], + "score": 1.0, + "content": "then further define", + "type": "text" + }, + { + "bbox": [ + 183, + 662, + 312, + 675 + ], + "score": 0.89, + "content": "\\begin{array} { r } { \\mu _ { \\mathrm { m i n } } = \\operatorname* { m i n } _ { ( s , a ) \\in \\mathcal { S } \\times \\mathcal { A } } \\mu _ { \\pi _ { b } } ( s , a ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 660, + 507, + 677 + ], + "score": 1.0, + "content": ". This quantity characterizes how many samples", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 673, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 673, + 505, + 684 + ], + "score": 1.0, + "content": "are needed to visit every state-action pair sufficiently often. Define the following mixing time of", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 681, + 504, + 698 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 220, + 698 + ], + "score": 1.0, + "content": "the induced Markov chain:", + "type": "text" + }, + { + "bbox": [ + 220, + 683, + 455, + 697 + ], + "score": 0.85, + "content": "\\begin{array} { r } { t _ { \\mathrm { m i x } } = \\operatorname* { m i n } \\left. t : \\operatorname* { m a x } _ { s \\in \\mathcal { S } } \\mathsf { \\bar { d } _ { \\mathrm { T V } } } ( \\mu _ { \\pi } , P ( s _ { t } = \\cdot | s _ { 0 } = s ) ) \\breve { \\leq } \\frac { 1 } { 4 } \\right. } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 681, + 487, + 698 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 488, + 684, + 504, + 695 + ], + "score": 0.87, + "content": "d _ { \\mathrm { T V } }", + "type": "inline_equation" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 693, + 225, + 706 + ], + "spans": [ + { + "bbox": [ + 105, + 693, + 225, + 706 + ], + "score": 1.0, + "content": "is the total variation distance.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 44 + }, + { + "type": "text", + "bbox": [ + 103, + 710, + 495, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 710, + 496, + 724 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 496, + 724 + ], + "score": 1.0, + "content": "The following theorem establishes the finite-time error bound of our robust Q-learning algorithm.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 47 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 740, + 309, + 753 + ], + "spans": [ + { + "bbox": [ + 302, + 740, + 309, + 753 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 70, + 302, + 85 + ], + "lines": [ + { + "bbox": [ + 104, + 68, + 303, + 88 + ], + "spans": [ + { + "bbox": [ + 104, + 68, + 303, + 88 + ], + "score": 1.0, + "content": "4 Tabular Case: Robust Q-Learning", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 96, + 505, + 141 + ], + "lines": [ + { + "bbox": [ + 106, + 97, + 504, + 108 + ], + "spans": [ + { + "bbox": [ + 106, + 97, + 504, + 108 + ], + "score": 1.0, + "content": "In this section, we focus on the tabular case with finite state and action spaces. We focus on the", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 107, + 505, + 120 + ], + "spans": [ + { + "bbox": [ + 105, + 107, + 505, + 120 + ], + "score": 1.0, + "content": "asynchronous setting where a single sample trajectory is available with Markovian noise. We will", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 118, + 506, + 132 + ], + "spans": [ + { + "bbox": [ + 106, + 118, + 395, + 132 + ], + "score": 1.0, + "content": "develop an efficient approach to estimate the unknown uncertainty set", + "type": "text" + }, + { + "bbox": [ + 395, + 119, + 404, + 129 + ], + "score": 0.51, + "content": "\\mathbf { P }", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 118, + 506, + 132 + ], + "score": 1.0, + "content": ", and further the support", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 129, + 371, + 142 + ], + "spans": [ + { + "bbox": [ + 105, + 129, + 142, + 142 + ], + "score": 1.0, + "content": "function", + "type": "text" + }, + { + "bbox": [ + 142, + 129, + 170, + 142 + ], + "score": 0.94, + "content": "\\sigma _ { \\mathcal { P } _ { s } ^ { a } } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 170, + 129, + 371, + 142 + ], + "score": 1.0, + "content": ", and then design our robust Q-learning algorithm.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2.5, + "bbox_fs": [ + 105, + 97, + 506, + 142 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 145, + 505, + 244 + ], + "lines": [ + { + "bbox": [ + 106, + 146, + 504, + 158 + ], + "spans": [ + { + "bbox": [ + 106, + 146, + 403, + 157 + ], + "score": 1.0, + "content": "We propose an efficient and data-driven approach to estimate the unknown", + "type": "text" + }, + { + "bbox": [ + 403, + 146, + 414, + 158 + ], + "score": 0.89, + "content": "p _ { s } ^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 146, + 504, + 157 + ], + "score": 1.0, + "content": "and thus the unknown", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 104, + 155, + 505, + 171 + ], + "spans": [ + { + "bbox": [ + 104, + 155, + 169, + 171 + ], + "score": 1.0, + "content": "uncertainty set", + "type": "text" + }, + { + "bbox": [ + 169, + 157, + 182, + 169 + ], + "score": 0.89, + "content": "\\mathcal { P } _ { s } ^ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 155, + 216, + 171 + ], + "score": 1.0, + "content": "for any", + "type": "text" + }, + { + "bbox": [ + 216, + 157, + 241, + 167 + ], + "score": 0.9, + "content": "s \\in \\mathcal { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 155, + 260, + 171 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 261, + 157, + 288, + 167 + ], + "score": 0.9, + "content": "a \\in { \\mathcal { A } }", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 155, + 434, + 171 + ], + "score": 1.0, + "content": ". Specifically, denote the sample at", + "type": "text" + }, + { + "bbox": [ + 434, + 158, + 439, + 167 + ], + "score": 0.76, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 440, + 155, + 505, + 171 + ], + "score": 1.0, + "content": "-th time step by", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 166, + 504, + 186 + ], + "spans": [ + { + "bbox": [ + 106, + 169, + 184, + 181 + ], + "score": 0.92, + "content": "O _ { t } = ( s _ { t } , a _ { t } , s _ { t + 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 166, + 239, + 186 + ], + "score": 1.0, + "content": ". We then use", + "type": "text" + }, + { + "bbox": [ + 240, + 169, + 252, + 180 + ], + "score": 0.87, + "content": "O _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 166, + 459, + 186 + ], + "score": 1.0, + "content": "to obtain the maximum likelihood estimate (MLE)", + "type": "text" + }, + { + "bbox": [ + 460, + 168, + 504, + 182 + ], + "score": 0.83, + "content": "\\hat { p } _ { t } \\triangleq \\mathbb { 1 } _ { s _ { t + 1 } }", + "type": "inline_equation" + } + ], + "index": 7 + }, + { + "bbox": [ + 104, + 178, + 504, + 195 + ], + "spans": [ + { + "bbox": [ + 104, + 178, + 203, + 195 + ], + "score": 1.0, + "content": "of the transition kernel", + "type": "text" + }, + { + "bbox": [ + 203, + 181, + 217, + 193 + ], + "score": 0.9, + "content": "p _ { s _ { t } } ^ { a _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 178, + 250, + 195 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 251, + 181, + 273, + 192 + ], + "score": 0.91, + "content": "\\mathbb { 1 } _ { s _ { t + 1 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 178, + 485, + 195 + ], + "score": 1.0, + "content": "is a probability distribution taking probability 1 at", + "type": "text" + }, + { + "bbox": [ + 485, + 183, + 504, + 192 + ], + "score": 0.88, + "content": "s _ { t + 1 }", + "type": "inline_equation" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 191, + 506, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 420, + 204 + ], + "score": 1.0, + "content": "and 0 at other states. This is an unbiased estimate of the transition kernel", + "type": "text" + }, + { + "bbox": [ + 421, + 192, + 435, + 204 + ], + "score": 0.87, + "content": "p _ { s _ { t } } ^ { a _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 435, + 191, + 506, + 204 + ], + "score": 1.0, + "content": "conditioning on", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 202, + 504, + 220 + ], + "spans": [ + { + "bbox": [ + 107, + 205, + 139, + 217 + ], + "score": 0.92, + "content": "S _ { t } = s _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 202, + 158, + 220 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 158, + 205, + 192, + 217 + ], + "score": 0.92, + "content": "A _ { t } = a _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 202, + 363, + 220 + ], + "score": 1.0, + "content": ". We then design a sample-based estimate", + "type": "text" + }, + { + "bbox": [ + 363, + 203, + 504, + 218 + ], + "score": 0.92, + "content": "\\hat { \\mathcal { P } } _ { t } \\triangleq \\left\\{ ( 1 - R ) \\hat { p } _ { t } + R q | q \\in \\Delta _ { | \\mathcal { S } | } \\right\\}", + "type": "inline_equation" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 217, + 505, + 233 + ], + "spans": [ + { + "bbox": [ + 105, + 217, + 195, + 233 + ], + "score": 1.0, + "content": "of the uncertainty set", + "type": "text" + }, + { + "bbox": [ + 195, + 219, + 210, + 232 + ], + "score": 0.9, + "content": "\\mathcal { P } _ { s _ { t } } ^ { a _ { t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 217, + 378, + 233 + ], + "score": 1.0, + "content": ". Using the sample-based uncertainty set", + "type": "text" + }, + { + "bbox": [ + 378, + 218, + 389, + 230 + ], + "score": 0.88, + "content": "\\hat { \\mathcal { P } } _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 217, + 505, + 233 + ], + "score": 1.0, + "content": ", we construct the following", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 229, + 505, + 246 + ], + "spans": [ + { + "bbox": [ + 104, + 229, + 317, + 246 + ], + "score": 1.0, + "content": "robust Q-learning algorithm in Algorithm 1. For any", + "type": "text" + }, + { + "bbox": [ + 317, + 232, + 322, + 241 + ], + "score": 0.59, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 229, + 326, + 246 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 326, + 231, + 360, + 245 + ], + "score": 0.92, + "content": "\\sigma _ { \\hat { \\mathcal { P } } _ { t } } ( V _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 229, + 460, + 246 + ], + "score": 1.0, + "content": "can be easily computed:", + "type": "text" + }, + { + "bbox": [ + 460, + 231, + 505, + 245 + ], + "score": 0.94, + "content": "\\sigma _ { \\hat { \\Phi } _ { t } } ( V _ { t } ) =", + "type": "inline_equation" + } + ], + "index": 12 + } + ], + "index": 8.5, + "bbox_fs": [ + 104, + 146, + 506, + 246 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 258, + 239, + 270 + ], + "lines": [ + { + "bbox": [ + 105, + 256, + 240, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 256, + 240, + 273 + ], + "score": 1.0, + "content": "Algorithm 1 Robust Q-Learning", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 272, + 448, + 285 + ], + "lines": [ + { + "bbox": [ + 105, + 270, + 448, + 287 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 167, + 287 + ], + "score": 1.0, + "content": "Initialization:", + "type": "text" + }, + { + "bbox": [ + 167, + 273, + 176, + 282 + ], + "score": 0.58, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 270, + 179, + 287 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 180, + 272, + 215, + 285 + ], + "score": 0.91, + "content": "Q _ { 0 } ( s , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 270, + 243, + 287 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 244, + 272, + 303, + 285 + ], + "score": 0.91, + "content": "( s , a ) \\in \\mathcal { S } \\times \\mathcal { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 270, + 371, + 287 + ], + "score": 1.0, + "content": ", behavior policy", + "type": "text" + }, + { + "bbox": [ + 372, + 275, + 382, + 284 + ], + "score": 0.69, + "content": "\\pi _ { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 270, + 386, + 287 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 387, + 275, + 396, + 284 + ], + "score": 0.61, + "content": "s _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 270, + 437, + 287 + ], + "score": 1.0, + "content": ", step size", + "type": "text" + }, + { + "bbox": [ + 437, + 275, + 448, + 284 + ], + "score": 0.7, + "content": "\\alpha _ { t }", + "type": "inline_equation" + } + ], + "index": 14 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 270, + 448, + 287 + ] + }, + { + "type": "index", + "bbox": [ + 110, + 286, + 365, + 359 + ], + "lines": [ + { + "bbox": [ + 110, + 285, + 236, + 297 + ], + "spans": [ + { + "bbox": [ + 110, + 285, + 138, + 297 + ], + "score": 1.0, + "content": "1: for", + "type": "text" + }, + { + "bbox": [ + 138, + 286, + 222, + 297 + ], + "score": 0.85, + "content": "t = 0 , 1 , 2 , . . . , T - 1", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 285, + 236, + 297 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 111, + 296, + 261, + 309 + ], + "spans": [ + { + "bbox": [ + 111, + 297, + 121, + 308 + ], + "score": 1.0, + "content": "2:", + "type": "text" + }, + { + "bbox": [ + 132, + 296, + 165, + 309 + ], + "score": 1.0, + "content": "Choose", + "type": "text" + }, + { + "bbox": [ + 165, + 298, + 176, + 307 + ], + "score": 0.85, + "content": "a _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 296, + 229, + 309 + ], + "score": 1.0, + "content": "according to", + "type": "text" + }, + { + "bbox": [ + 229, + 296, + 261, + 308 + ], + "score": 0.9, + "content": "\\pi _ { b } ( \\cdot | s _ { t } )", + "type": "inline_equation" + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 111, + 307, + 216, + 320 + ], + "spans": [ + { + "bbox": [ + 111, + 307, + 121, + 319 + ], + "score": 1.0, + "content": "3:", + "type": "text" + }, + { + "bbox": [ + 133, + 307, + 168, + 320 + ], + "score": 1.0, + "content": "Observe", + "type": "text" + }, + { + "bbox": [ + 169, + 309, + 189, + 319 + ], + "score": 0.87, + "content": "s _ { t + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 307, + 206, + 320 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 207, + 309, + 216, + 318 + ], + "score": 0.81, + "content": "c _ { t }", + "type": "inline_equation" + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 111, + 318, + 270, + 330 + ], + "spans": [ + { + "bbox": [ + 111, + 318, + 121, + 329 + ], + "score": 1.0, + "content": "4:", + "type": "text" + }, + { + "bbox": [ + 133, + 319, + 270, + 330 + ], + "score": 0.74, + "content": "\\begin{array} { r } { V _ { t } ( s ) \\operatorname* { m i n } _ { a \\in \\mathcal { A } } Q _ { t } ( s , a ) , \\forall s \\in \\mathfrak { s } } \\end{array}", + "type": "inline_equation" + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 111, + 329, + 362, + 342 + ], + "spans": [ + { + "bbox": [ + 111, + 329, + 121, + 341 + ], + "score": 1.0, + "content": "5:", + "type": "text" + }, + { + "bbox": [ + 133, + 329, + 161, + 342 + ], + "score": 1.0, + "content": "Qt+1(", + "type": "text" + }, + { + "bbox": [ + 162, + 329, + 362, + 342 + ], + "score": 0.28, + "content": "s _ { t } , a _ { t } ) \\gets ( 1 - \\alpha _ { t } ) Q _ { t } ( s _ { t } , a _ { t } ) + \\alpha _ { t } ( c _ { t } + \\gamma \\sigma _ { \\hat { \\mathcal P } _ { t } } ( V _ { t } ) )", + "type": "inline_equation" + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 111, + 341, + 308, + 355 + ], + "spans": [ + { + "bbox": [ + 111, + 342, + 121, + 353 + ], + "score": 1.0, + "content": "6:", + "type": "text" + }, + { + "bbox": [ + 134, + 342, + 227, + 354 + ], + "score": 0.91, + "content": "Q _ { t + 1 } ( s , a ) \\gets Q _ { t } ( s , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 341, + 243, + 355 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 243, + 342, + 308, + 354 + ], + "score": 0.83, + "content": "( s , a ) \\neq ( s _ { t } , a _ { t } )", + "type": "inline_equation" + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 352, + 157, + 365 + ], + "spans": [ + { + "bbox": [ + 110, + 352, + 157, + 365 + ], + "score": 1.0, + "content": "7: end for", + "type": "text" + } + ], + "index": 22, + "is_list_start_line": true + } + ], + "index": 18, + "bbox_fs": [ + 110, + 285, + 362, + 365 + ] + }, + { + "type": "text", + "bbox": [ + 111, + 354, + 156, + 363 + ], + "lines": [ + { + "bbox": [ + 111, + 354, + 156, + 363 + ], + "spans": [], + "index": 21 + } + ], + "index": 21, + "bbox_fs": [ + 111, + 354, + 156, + 363 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 366, + 158, + 378 + ], + "lines": [ + { + "bbox": [ + 105, + 363, + 159, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 143, + 381 + ], + "score": 1.0, + "content": "Output:", + "type": "text" + }, + { + "bbox": [ + 144, + 366, + 159, + 378 + ], + "score": 0.86, + "content": "Q _ { T }", + "type": "inline_equation" + } + ], + "index": 23 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 363, + 159, + 381 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 391, + 491, + 404 + ], + "lines": [ + { + "bbox": [ + 106, + 391, + 492, + 405 + ], + "spans": [ + { + "bbox": [ + 106, + 392, + 254, + 405 + ], + "score": 0.89, + "content": "\\begin{array} { r } { R \\operatorname* { m a x } _ { s \\in \\mathcal { S } } V _ { t } ( s ) + ( 1 - R ) V _ { t } ( s _ { t + 1 } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 391, + 492, + 405 + ], + "score": 1.0, + "content": ". Hence the update in Algorithm 1 (line 5) can be written as", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24, + "bbox_fs": [ + 106, + 391, + 492, + 405 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 135, + 410, + 475, + 429 + ], + "lines": [ + { + "bbox": [ + 135, + 410, + 475, + 429 + ], + "spans": [ + { + "bbox": [ + 135, + 410, + 475, + 429 + ], + "score": 0.9, + "content": "Q _ { t + 1 } ( s _ { t } , a _ { t } ) \\gets ( 1 - \\alpha _ { t } ) Q _ { t } ( s _ { t } , a _ { t } ) + \\alpha _ { t } ( c _ { t } + \\gamma R \\operatorname* { m a x } _ { s \\in \\mathcal { S } } V _ { t } ( s ) + \\gamma ( 1 - R ) V _ { t } ( s _ { t + 1 } ) ) .", + "type": "interline_equation", + "image_path": "ebc655b9749ff56b32e6e65d9da89184ff1f6cd737704bf092f1717d68c980c8.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 135, + 410, + 475, + 429 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 440, + 505, + 529 + ], + "lines": [ + { + "bbox": [ + 106, + 440, + 506, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 440, + 506, + 453 + ], + "score": 1.0, + "content": "Compared to the model-based approach, our approach is model-free. It does not require the prior", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 451, + 506, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 344, + 465 + ], + "score": 1.0, + "content": "knowledge of the uncertainty set, i.e., the knowledge of", + "type": "text" + }, + { + "bbox": [ + 344, + 452, + 424, + 464 + ], + "score": 0.93, + "content": "p _ { s } ^ { a } , \\forall s \\in \\mathcal { S } , a \\in \\mathcal { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 451, + 506, + 465 + ], + "score": 1.0, + "content": ". Furthermore, the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 462, + 506, + 475 + ], + "spans": [ + { + "bbox": [ + 105, + 463, + 270, + 475 + ], + "score": 1.0, + "content": "memory requirement of our algorithm is", + "type": "text" + }, + { + "bbox": [ + 270, + 462, + 307, + 474 + ], + "score": 0.92, + "content": "| \\mathcal { S } | \\times | \\mathcal { A } |", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 463, + 506, + 475 + ], + "score": 1.0, + "content": "(used to store the Q-table), and unlike the model-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 473, + 506, + 487 + ], + "spans": [ + { + "bbox": [ + 105, + 473, + 298, + 487 + ], + "score": 1.0, + "content": "based approach it does not need a table of size", + "type": "text" + }, + { + "bbox": [ + 299, + 474, + 328, + 486 + ], + "score": 0.93, + "content": "| \\dot { \\mathcal { S } } | ^ { 2 } | \\mathcal { A } |", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 473, + 362, + 487 + ], + "score": 1.0, + "content": "to store", + "type": "text" + }, + { + "bbox": [ + 363, + 474, + 436, + 486 + ], + "score": 0.9, + "content": "p _ { s } ^ { a } , \\forall s \\in \\mathcal { S } , a \\in \\mathcal { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 473, + 506, + 487 + ], + "score": 1.0, + "content": ", which could be", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 485, + 505, + 497 + ], + "spans": [ + { + "bbox": [ + 105, + 485, + 505, + 497 + ], + "score": 1.0, + "content": "problematic if the state space is large. Moreover, our algorithm does not involve a relaxation of the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 496, + 505, + 508 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 505, + 508 + ], + "score": 1.0, + "content": "uncertainty set like the one in [Roy et al., 2017], which will incur a stringent condition on the discount", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 506, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 505, + 519 + ], + "score": 1.0, + "content": "factor to guarantee the convergence. As will be shown below, the convergence of our Algorithm 1", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 518, + 321, + 530 + ], + "spans": [ + { + "bbox": [ + 106, + 518, + 321, + 530 + ], + "score": 1.0, + "content": "does not require any condition on the discount factor.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 440, + 506, + 530 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 533, + 504, + 556 + ], + "lines": [ + { + "bbox": [ + 105, + 531, + 505, + 547 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 505, + 547 + ], + "score": 1.0, + "content": "We show in the following theorem that the robust Q-learning algorithm converges asymptotically to", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 545, + 285, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 268, + 558 + ], + "score": 1.0, + "content": "the optimal robust action-value function", + "type": "text" + }, + { + "bbox": [ + 268, + 545, + 282, + 556 + ], + "score": 0.89, + "content": "Q ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 545, + 285, + 558 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5, + "bbox_fs": [ + 105, + 531, + 505, + 558 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 559, + 502, + 583 + ], + "lines": [ + { + "bbox": [ + 103, + 555, + 503, + 578 + ], + "spans": [ + { + "bbox": [ + 103, + 555, + 308, + 578 + ], + "score": 1.0, + "content": "Theorem 2. (Asymptotic Convergence) If step sizes", + "type": "text" + }, + { + "bbox": [ + 309, + 562, + 320, + 571 + ], + "score": 0.84, + "content": "\\alpha _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 555, + 365, + 578 + ], + "score": 1.0, + "content": "satisfy that", + "type": "text" + }, + { + "bbox": [ + 365, + 560, + 425, + 573 + ], + "score": 0.89, + "content": "\\textstyle \\sum _ { t = 0 } ^ { \\infty } \\alpha _ { t } = \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 555, + 442, + 578 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 442, + 559, + 503, + 573 + ], + "score": 0.9, + "content": "\\textstyle \\sum _ { t = 0 } ^ { \\infty } \\alpha _ { t } ^ { 2 } < \\infty", + "type": "inline_equation" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 570, + 286, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 126, + 583 + ], + "score": 1.0, + "content": "then", + "type": "text" + }, + { + "bbox": [ + 126, + 571, + 166, + 582 + ], + "score": 0.92, + "content": "Q _ { t } \\to Q ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 570, + 179, + 583 + ], + "score": 1.0, + "content": "as", + "type": "text" + }, + { + "bbox": [ + 210, + 570, + 277, + 583 + ], + "score": 1.0, + "content": "with probability", + "type": "text" + }, + { + "bbox": [ + 283, + 570, + 286, + 583 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5, + "bbox_fs": [ + 103, + 555, + 503, + 583 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 592, + 503, + 615 + ], + "lines": [ + { + "bbox": [ + 106, + 592, + 504, + 605 + ], + "spans": [ + { + "bbox": [ + 106, + 592, + 504, + 605 + ], + "score": 1.0, + "content": "To further establish the finite-time error bound for our robust Q-learning algorithm in Algorithm 1,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 602, + 488, + 617 + ], + "spans": [ + { + "bbox": [ + 105, + 602, + 488, + 617 + ], + "score": 1.0, + "content": "we make the following assumption that is commonly used in the analysis of vanilla Q-learning.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 105, + 592, + 504, + 617 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 618, + 504, + 641 + ], + "lines": [ + { + "bbox": [ + 105, + 617, + 506, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 386, + 632 + ], + "score": 1.0, + "content": "Assumption 1. The Markov chain induced by the behavior policy", + "type": "text" + }, + { + "bbox": [ + 386, + 621, + 397, + 630 + ], + "score": 0.85, + "content": "\\pi _ { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 617, + 506, + 632 + ], + "score": 1.0, + "content": "and the transition kernel", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 107, + 629, + 266, + 644 + ], + "spans": [ + { + "bbox": [ + 107, + 630, + 179, + 642 + ], + "score": 0.9, + "content": "p _ { s } ^ { a } , \\forall s \\in \\mathcal { S } , a \\in \\mathcal { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 629, + 266, + 644 + ], + "score": 1.0, + "content": "is uniformly ergodic.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5, + "bbox_fs": [ + 105, + 617, + 506, + 644 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 650, + 505, + 705 + ], + "lines": [ + { + "bbox": [ + 105, + 649, + 506, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 123, + 664 + ], + "score": 1.0, + "content": "Let", + "type": "text" + }, + { + "bbox": [ + 123, + 652, + 138, + 662 + ], + "score": 0.87, + "content": "\\mu _ { \\pi _ { b } }", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 649, + 299, + 664 + ], + "score": 1.0, + "content": "denote the stationary distribution over", + "type": "text" + }, + { + "bbox": [ + 300, + 651, + 327, + 661 + ], + "score": 0.89, + "content": "\\mathcal { S } \\times \\mathcal { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 649, + 376, + 664 + ], + "score": 1.0, + "content": "induced by", + "type": "text" + }, + { + "bbox": [ + 377, + 653, + 388, + 662 + ], + "score": 0.85, + "content": "\\pi _ { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 649, + 407, + 664 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 407, + 651, + 483, + 662 + ], + "score": 0.9, + "content": "p _ { s } ^ { a } , \\forall s \\in \\mathcal { S } , a \\in \\mathcal { A }", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 649, + 506, + 664 + ], + "score": 1.0, + "content": ". We", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 660, + 507, + 677 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 183, + 677 + ], + "score": 1.0, + "content": "then further define", + "type": "text" + }, + { + "bbox": [ + 183, + 662, + 312, + 675 + ], + "score": 0.89, + "content": "\\begin{array} { r } { \\mu _ { \\mathrm { m i n } } = \\operatorname* { m i n } _ { ( s , a ) \\in \\mathcal { S } \\times \\mathcal { A } } \\mu _ { \\pi _ { b } } ( s , a ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 660, + 507, + 677 + ], + "score": 1.0, + "content": ". This quantity characterizes how many samples", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 673, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 673, + 505, + 684 + ], + "score": 1.0, + "content": "are needed to visit every state-action pair sufficiently often. Define the following mixing time of", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 681, + 504, + 698 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 220, + 698 + ], + "score": 1.0, + "content": "the induced Markov chain:", + "type": "text" + }, + { + "bbox": [ + 220, + 683, + 455, + 697 + ], + "score": 0.85, + "content": "\\begin{array} { r } { t _ { \\mathrm { m i x } } = \\operatorname* { m i n } \\left. t : \\operatorname* { m a x } _ { s \\in \\mathcal { S } } \\mathsf { \\bar { d } _ { \\mathrm { T V } } } ( \\mu _ { \\pi } , P ( s _ { t } = \\cdot | s _ { 0 } = s ) ) \\breve { \\leq } \\frac { 1 } { 4 } \\right. } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 681, + 487, + 698 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 488, + 684, + 504, + 695 + ], + "score": 0.87, + "content": "d _ { \\mathrm { T V } }", + "type": "inline_equation" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 693, + 225, + 706 + ], + "spans": [ + { + "bbox": [ + 105, + 693, + 225, + 706 + ], + "score": 1.0, + "content": "is the total variation distance.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 44, + "bbox_fs": [ + 105, + 649, + 507, + 706 + ] + }, + { + "type": "text", + "bbox": [ + 103, + 710, + 495, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 710, + 496, + 724 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 496, + 724 + ], + "score": 1.0, + "content": "The following theorem establishes the finite-time error bound of our robust Q-learning algorithm.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 47, + "bbox_fs": [ + 106, + 710, + 496, + 724 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 104, + 72, + 505, + 97 + ], + "lines": [ + { + "bbox": [ + 106, + 72, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 106, + 72, + 412, + 86 + ], + "score": 1.0, + "content": "Theorem 3. (Finite-Time Error Bound) There exist some positive constants", + "type": "text" + }, + { + "bbox": [ + 412, + 75, + 422, + 84 + ], + "score": 0.85, + "content": "c _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 72, + 441, + 86 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 441, + 74, + 451, + 84 + ], + "score": 0.86, + "content": "c _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 72, + 505, + 86 + ], + "score": 1.0, + "content": "such that for", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 82, + 275, + 98 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 123, + 98 + ], + "score": 1.0, + "content": "any", + "type": "text" + }, + { + "bbox": [ + 124, + 84, + 147, + 94 + ], + "score": 0.87, + "content": "\\delta < 1", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 82, + 168, + 98 + ], + "score": 1.0, + "content": ", any", + "type": "text" + }, + { + "bbox": [ + 169, + 83, + 204, + 98 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\epsilon < \\frac { 1 } { 1 - \\gamma } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 82, + 225, + 98 + ], + "score": 1.0, + "content": ", any", + "type": "text" + }, + { + "bbox": [ + 225, + 84, + 233, + 93 + ], + "score": 0.78, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 82, + 275, + 98 + ], + "score": 1.0, + "content": "satisfying", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "interline_equation", + "bbox": [ + 147, + 104, + 462, + 132 + ], + "lines": [ + { + "bbox": [ + 147, + 104, + 462, + 132 + ], + "spans": [ + { + "bbox": [ + 147, + 104, + 462, + 132 + ], + "score": 0.92, + "content": "T \\geq c _ { 0 } \\left( \\frac { 1 } { \\mu _ { m i n } ( 1 - \\gamma ) ^ { 5 } \\epsilon ^ { 2 } } + \\frac { t _ { m i x } } { \\mu _ { m i n } ( 1 - \\gamma ) } \\right) \\log { \\left( \\frac { T | \\mathcal { S } | | \\mathcal { A } | } { \\delta } \\right) } \\log { \\left( \\frac { 1 } { \\epsilon ( 1 - \\gamma ) ^ { 2 } } \\right) } ,", + "type": "interline_equation", + "image_path": "9ec28bd95e4de19b60114b1e15ae076fa40d0282d19b6b1f9240d27b88488068.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 147, + 104, + 462, + 132 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 137, + 506, + 172 + ], + "lines": [ + { + "bbox": [ + 105, + 136, + 506, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 141, + 185, + 155 + ], + "score": 1.0, + "content": "and step size αt =", + "type": "text" + }, + { + "bbox": [ + 159, + 138, + 347, + 160 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\alpha _ { t } = \\frac { c _ { 1 } } { \\log \\left( \\frac { T | \\mathcal { S } | | \\mathcal { A } | } { \\delta } \\right) } \\operatorname* { m i n } \\left( \\frac { 1 } { t _ { m i x } } , \\frac { \\epsilon ^ { 2 } ( 1 - \\gamma ) ^ { 4 } } { \\gamma ^ { 2 } } \\right) , \\forall t \\geq 0 } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 136, + 477, + 159 + ], + "score": 1.0, + "content": "we have with probability at least", + "type": "text" + }, + { + "bbox": [ + 477, + 142, + 503, + 153 + ], + "score": 0.66, + "content": "1 - 6 \\delta", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 136, + 506, + 159 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 107, + 159, + 191, + 172 + ], + "spans": [ + { + "bbox": [ + 107, + 159, + 187, + 172 + ], + "score": 0.89, + "content": "\\| Q _ { T } - Q ^ { * } \\| _ { \\infty } \\leq 3 \\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 159, + 191, + 172 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 106, + 179, + 506, + 254 + ], + "lines": [ + { + "bbox": [ + 106, + 180, + 506, + 197 + ], + "spans": [ + { + "bbox": [ + 106, + 181, + 303, + 195 + ], + "score": 1.0, + "content": "From the theorem, we can see that to guarantee an", + "type": "text" + }, + { + "bbox": [ + 304, + 185, + 309, + 192 + ], + "score": 0.68, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 181, + 437, + 195 + ], + "score": 1.0, + "content": "-accurate estimate, a sample size", + "type": "text" + }, + { + "bbox": [ + 437, + 180, + 506, + 197 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\tilde { \\mathcal { O } } ( \\frac { 1 } { \\mu _ { \\mathrm { m i n } } ( 1 - \\gamma ) ^ { 5 } \\epsilon ^ { 2 } } + } \\end{array}", + "type": "inline_equation" + } + ], + "index": 5 + }, + { + "bbox": [ + 102, + 191, + 509, + 214 + ], + "spans": [ + { + "bbox": [ + 102, + 191, + 509, + 214 + ], + "score": 1.0, + "content": "tmixµmin(1−γ) ) (up to some logarithmic terms) is needed. This complexity matches with the one for the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 104, + 208, + 506, + 222 + ], + "spans": [ + { + "bbox": [ + 104, + 208, + 506, + 222 + ], + "score": 1.0, + "content": "vanilla Q-learning in [Li et al., 2020] (within a constant factor), while our algorithm also guarantees", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 220, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 105, + 220, + 505, + 232 + ], + "score": 1.0, + "content": "robustness to MDP model uncertainty. Our algorithm design and analysis can be readily extended", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 231, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 505, + 243 + ], + "score": 1.0, + "content": "to robustify TD and SARSA. The variance-reduction technique [Wainwright, 2019] can also be", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 241, + 486, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 451, + 255 + ], + "score": 1.0, + "content": "combined with our robust Q-learning algorithm to further improve the dependency on", + "type": "text" + }, + { + "bbox": [ + 451, + 242, + 482, + 254 + ], + "score": 0.92, + "content": "( 1 - \\gamma )", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 241, + 486, + 255 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 7.5 + }, + { + "type": "title", + "bbox": [ + 107, + 270, + 325, + 284 + ], + "lines": [ + { + "bbox": [ + 104, + 269, + 326, + 285 + ], + "spans": [ + { + "bbox": [ + 104, + 269, + 326, + 285 + ], + "score": 1.0, + "content": "5 Function Approximation: Robust TDC", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 106, + 295, + 506, + 362 + ], + "lines": [ + { + "bbox": [ + 105, + 295, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 505, + 308 + ], + "score": 1.0, + "content": "In this section, we investigate the case where the state and action spaces can be large or even", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 307, + 506, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 506, + 320 + ], + "score": 1.0, + "content": "continuous. A popular approach is to approximate the value function using a parameterized function,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 317, + 505, + 330 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 505, + 330 + ], + "score": 1.0, + "content": "e.g., linear function and neural network. In this section, we focus on the case with linear function", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 328, + 506, + 341 + ], + "spans": [ + { + "bbox": [ + 104, + 328, + 506, + 341 + ], + "score": 1.0, + "content": "approximation to illustrate the main idea of designing robust RL algorithms. Our approach can be", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 339, + 507, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 507, + 352 + ], + "score": 1.0, + "content": "extended to non-linear (smooth) function approximation using techniques in, e.g., [Cai et al., 2019,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 350, + 345, + 362 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 345, + 362 + ], + "score": 1.0, + "content": "Bhatnagar et al., 2009, Wai et al., 2019, Wang et al., 2021].", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14.5 + }, + { + "type": "text", + "bbox": [ + 106, + 366, + 506, + 455 + ], + "lines": [ + { + "bbox": [ + 104, + 366, + 504, + 379 + ], + "spans": [ + { + "bbox": [ + 104, + 366, + 489, + 379 + ], + "score": 1.0, + "content": "We focus on the problem of robust policy evaluation, i.e., estimate the robust value function", + "type": "text" + }, + { + "bbox": [ + 489, + 367, + 504, + 377 + ], + "score": 0.81, + "content": "V ^ { \\pi }", + "type": "inline_equation" + } + ], + "index": 18 + }, + { + "bbox": [ + 104, + 376, + 507, + 391 + ], + "spans": [ + { + "bbox": [ + 104, + 376, + 234, + 391 + ], + "score": 1.0, + "content": "defined in (1) for a given policy", + "type": "text" + }, + { + "bbox": [ + 235, + 380, + 242, + 388 + ], + "score": 0.72, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 376, + 507, + 391 + ], + "score": 1.0, + "content": "under the worst-case MDP transition kernel in the uncertainty set.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 104, + 387, + 506, + 402 + ], + "spans": [ + { + "bbox": [ + 104, + 387, + 506, + 402 + ], + "score": 1.0, + "content": "Note that for robust RL with model uncertainty, any policy evaluation problem can be viewed as", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 399, + 506, + 412 + ], + "spans": [ + { + "bbox": [ + 106, + 399, + 506, + 412 + ], + "score": 1.0, + "content": "\"off-transition-kernel\", as it is to evaluate the value function under the worst-case MDP using samples", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 410, + 506, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 506, + 423 + ], + "score": 1.0, + "content": "from a different MDP. Since the TD algorithm with function approximation may diverge under", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 420, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 506, + 434 + ], + "score": 1.0, + "content": "off-policy training [Baird, 1995] and importance sampling cannot be applied here due to unknown", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 432, + 505, + 444 + ], + "spans": [ + { + "bbox": [ + 106, + 432, + 505, + 444 + ], + "score": 1.0, + "content": "transition kernel, in this paper we generalize the GTD method [Maei et al., 2010, Maei, 2011] to the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 442, + 165, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 165, + 456 + ], + "score": 1.0, + "content": "robust setting.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 106, + 459, + 505, + 500 + ], + "lines": [ + { + "bbox": [ + 105, + 458, + 506, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 458, + 124, + 474 + ], + "score": 1.0, + "content": "Let", + "type": "text" + }, + { + "bbox": [ + 124, + 459, + 248, + 474 + ], + "score": 0.9, + "content": "\\left\\{ \\phi ^ { ( i ) } : \\mathbb { S } \\to \\mathbb { R } , i = 1 , \\ldots , N \\right\\}", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 458, + 299, + 474 + ], + "score": 1.0, + "content": "be a set of", + "type": "text" + }, + { + "bbox": [ + 299, + 461, + 309, + 470 + ], + "score": 0.81, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 310, + 458, + 431, + 474 + ], + "score": 1.0, + "content": "fixed base functions, where", + "type": "text" + }, + { + "bbox": [ + 431, + 460, + 486, + 473 + ], + "score": 0.92, + "content": "N \\ll | \\mathcal S | | \\mathcal A |", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 458, + 506, + 474 + ], + "score": 1.0, + "content": ". In", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 471, + 506, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 441, + 488 + ], + "score": 1.0, + "content": "particular, we approximate the robust value function using a linear combination of", + "type": "text" + }, + { + "bbox": [ + 106, + 485, + 188, + 501 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\sum _ { i = 1 } ^ { N } \\theta ^ { i } \\phi _ { s } ^ { ( i ) } = \\phi _ { s } ^ { \\top } \\theta } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 478, + 218, + 507 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 219, + 487, + 252, + 498 + ], + "score": 0.91, + "content": "\\boldsymbol { \\theta } \\in \\mathbb { R } ^ { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 478, + 340, + 507 + ], + "score": 1.0, + "content": "is the weight vector.", + "type": "text" + }, + { + "bbox": [ + 442, + 473, + 457, + 486 + ], + "score": 0.89, + "content": "\\phi ^ { ( i ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 458, + 471, + 469, + 488 + ], + "score": 1.0, + "content": "’s:", + "type": "text" + }, + { + "bbox": [ + 469, + 473, + 506, + 486 + ], + "score": 0.89, + "content": "V _ { \\theta } ( s ) =", + "type": "inline_equation" + } + ], + "index": 27 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 106, + 504, + 374, + 516 + ], + "lines": [ + { + "bbox": [ + 105, + 502, + 375, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 363, + 519 + ], + "score": 1.0, + "content": "Define the following robust Bellman operator for a given policy", + "type": "text" + }, + { + "bbox": [ + 363, + 507, + 370, + 514 + ], + "score": 0.72, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 502, + 375, + 519 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "interline_equation", + "bbox": [ + 146, + 521, + 465, + 574 + ], + "lines": [ + { + "bbox": [ + 146, + 521, + 465, + 574 + ], + "spans": [ + { + "bbox": [ + 146, + 521, + 465, + 574 + ], + "score": 0.92, + "content": "\\begin{array} { r l } & { \\mathrm { \\bf T } _ { \\pi } V ( s ) \\triangleq \\mathbb E _ { A \\sim \\pi ( \\cdot | s ) } [ c ( s , A ) + \\gamma \\sigma _ { \\mathcal { P } _ { s } ^ { A } } ( V ) ] } \\\\ & { \\quad \\quad \\quad = \\mathbb E _ { A \\sim \\pi ( \\cdot | s ) } \\left[ c ( s , A ) + \\gamma ( 1 - R ) \\displaystyle \\sum _ { s ^ { \\prime } \\in \\mathcal { S } } p _ { s , s ^ { \\prime } } ^ { A } V ( s ^ { \\prime } ) + \\gamma R \\operatorname* { m a x } _ { s ^ { \\prime } \\in \\mathcal { S } } V ( s ^ { \\prime } ) \\right] . } \\end{array}", + "type": "interline_equation", + "image_path": "9852dcd21269061985da10dcabeac8e4caafd9fd4a4cbfc50d42023c0a2ae526.jpg" + } + ] + } + ], + "index": 30, + "virtual_lines": [ + { + "bbox": [ + 146, + 521, + 465, + 538.6666666666666 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 146, + 538.6666666666666, + 465, + 556.3333333333333 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 146, + 556.3333333333333, + 465, + 573.9999999999999 + ], + "spans": [], + "index": 31 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 579, + 424, + 591 + ], + "lines": [ + { + "bbox": [ + 105, + 578, + 425, + 593 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 425, + 593 + ], + "score": 1.0, + "content": "We then define the mean squared projected robust Bellman error (MSPRBE) as", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "interline_equation", + "bbox": [ + 231, + 597, + 379, + 614 + ], + "lines": [ + { + "bbox": [ + 231, + 597, + 379, + 614 + ], + "spans": [ + { + "bbox": [ + 231, + 597, + 379, + 614 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\mathrm { M S P R B E } ( \\theta ) = \\| \\mathbf { I I T } _ { \\pi } V _ { \\theta } - V _ { \\theta } \\| _ { \\mu _ { \\pi } } ^ { 2 } , } \\end{array}", + "type": "interline_equation", + "image_path": "7fdf581c44a14004534015be69003a637c4f9373f971dff5a0e1cb387e623e39.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 231, + 597, + 379, + 614 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 620, + 505, + 681 + ], + "lines": [ + { + "bbox": [ + 105, + 621, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 133, + 635 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 621, + 231, + 635 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\| v \\| _ { \\mu _ { \\pi } } ^ { 2 } = \\int v ^ { 2 } ( s ) \\mu _ { \\pi } ( d s ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 621, + 235, + 635 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 235, + 623, + 248, + 633 + ], + "score": 0.78, + "content": "\\mu _ { \\pi }", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 621, + 407, + 635 + ], + "score": 1.0, + "content": "is the stationary distribution induced by", + "type": "text" + }, + { + "bbox": [ + 408, + 624, + 415, + 632 + ], + "score": 0.76, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 621, + 435, + 635 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 435, + 622, + 446, + 632 + ], + "score": 0.76, + "content": "\\mathbf { I I }", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 621, + 505, + 635 + ], + "score": 1.0, + "content": "is a projection", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 633, + 506, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 245, + 647 + ], + "score": 1.0, + "content": "onto the linear function space w.r.t.", + "type": "text" + }, + { + "bbox": [ + 246, + 634, + 272, + 646 + ], + "score": 0.91, + "content": "\\| \\cdot \\| _ { \\mu _ { \\pi } }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 633, + 506, + 647 + ], + "score": 1.0, + "content": ". We will develop a two time-scale gradient-based approach", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 645, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 645, + 329, + 657 + ], + "score": 1.0, + "content": "to minimize the MSPRBE. However, it can be seen that", + "type": "text" + }, + { + "bbox": [ + 329, + 645, + 377, + 657 + ], + "score": 0.83, + "content": "\\operatorname* { m a x } _ { s } V _ { \\theta } ( s )", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 645, + 469, + 657 + ], + "score": 1.0, + "content": "in (7) is not smooth in", + "type": "text" + }, + { + "bbox": [ + 469, + 645, + 474, + 655 + ], + "score": 0.79, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 645, + 505, + 657 + ], + "score": 1.0, + "content": ", which", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 655, + 506, + 669 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 506, + 669 + ], + "score": 1.0, + "content": "is troublesome in both algorithm design and analysis. To solve this issue, we introduce the following", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 667, + 461, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 247, + 682 + ], + "score": 1.0, + "content": "smoothed robust Bellman operator", + "type": "text" + }, + { + "bbox": [ + 247, + 667, + 261, + 680 + ], + "score": 0.91, + "content": "\\hat { \\mathbf { T } } _ { \\pi }", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 668, + 461, + 682 + ], + "score": 1.0, + "content": "by smoothing the max with a LSE(LogSumExp):", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36 + }, + { + "type": "interline_equation", + "bbox": [ + 149, + 686, + 461, + 721 + ], + "lines": [ + { + "bbox": [ + 149, + 686, + 461, + 721 + ], + "spans": [ + { + "bbox": [ + 149, + 686, + 461, + 721 + ], + "score": 0.93, + "content": "\\hat { \\mathbf { T } } _ { \\pi } V ( s ) = \\mathbb { E } _ { A \\sim \\pi ( \\cdot | s ) } \\left[ c ( s , A ) + \\gamma ( 1 - R ) \\sum _ { s ^ { \\prime } \\in \\mathcal { S } } p _ { s , s ^ { \\prime } } ^ { A } V ( s ^ { \\prime } ) + \\gamma R \\cdot \\mathrm { L S E } ( V ) \\right] ,", + "type": "interline_equation", + "image_path": "7543aac30207ffdd163f7a3ab81924732889d4c9ebbac9f5e3be94660e8a76c2.jpg" + } + ] + } + ], + "index": 40, + "virtual_lines": [ + { + "bbox": [ + 149, + 686, + 461, + 697.6666666666666 + ], + "spans": [], + "index": 39 + }, + { + "bbox": [ + 149, + 697.6666666666666, + 461, + 709.3333333333333 + ], + "spans": [], + "index": 40 + }, + { + "bbox": [ + 149, + 709.3333333333333, + 461, + 720.9999999999999 + ], + "spans": [], + "index": 41 + } + ] + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 104, + 72, + 505, + 97 + ], + "lines": [ + { + "bbox": [ + 106, + 72, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 106, + 72, + 412, + 86 + ], + "score": 1.0, + "content": "Theorem 3. (Finite-Time Error Bound) There exist some positive constants", + "type": "text" + }, + { + "bbox": [ + 412, + 75, + 422, + 84 + ], + "score": 0.85, + "content": "c _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 72, + 441, + 86 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 441, + 74, + 451, + 84 + ], + "score": 0.86, + "content": "c _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 72, + 505, + 86 + ], + "score": 1.0, + "content": "such that for", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 82, + 275, + 98 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 123, + 98 + ], + "score": 1.0, + "content": "any", + "type": "text" + }, + { + "bbox": [ + 124, + 84, + 147, + 94 + ], + "score": 0.87, + "content": "\\delta < 1", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 82, + 168, + 98 + ], + "score": 1.0, + "content": ", any", + "type": "text" + }, + { + "bbox": [ + 169, + 83, + 204, + 98 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\epsilon < \\frac { 1 } { 1 - \\gamma } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 82, + 225, + 98 + ], + "score": 1.0, + "content": ", any", + "type": "text" + }, + { + "bbox": [ + 225, + 84, + 233, + 93 + ], + "score": 0.78, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 82, + 275, + 98 + ], + "score": 1.0, + "content": "satisfying", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 105, + 72, + 505, + 98 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 147, + 104, + 462, + 132 + ], + "lines": [ + { + "bbox": [ + 147, + 104, + 462, + 132 + ], + "spans": [ + { + "bbox": [ + 147, + 104, + 462, + 132 + ], + "score": 0.92, + "content": "T \\geq c _ { 0 } \\left( \\frac { 1 } { \\mu _ { m i n } ( 1 - \\gamma ) ^ { 5 } \\epsilon ^ { 2 } } + \\frac { t _ { m i x } } { \\mu _ { m i n } ( 1 - \\gamma ) } \\right) \\log { \\left( \\frac { T | \\mathcal { S } | | \\mathcal { A } | } { \\delta } \\right) } \\log { \\left( \\frac { 1 } { \\epsilon ( 1 - \\gamma ) ^ { 2 } } \\right) } ,", + "type": "interline_equation", + "image_path": "9ec28bd95e4de19b60114b1e15ae076fa40d0282d19b6b1f9240d27b88488068.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 147, + 104, + 462, + 132 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 137, + 506, + 172 + ], + "lines": [ + { + "bbox": [ + 105, + 136, + 506, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 141, + 185, + 155 + ], + "score": 1.0, + "content": "and step size αt =", + "type": "text" + }, + { + "bbox": [ + 159, + 138, + 347, + 160 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\alpha _ { t } = \\frac { c _ { 1 } } { \\log \\left( \\frac { T | \\mathcal { S } | | \\mathcal { A } | } { \\delta } \\right) } \\operatorname* { m i n } \\left( \\frac { 1 } { t _ { m i x } } , \\frac { \\epsilon ^ { 2 } ( 1 - \\gamma ) ^ { 4 } } { \\gamma ^ { 2 } } \\right) , \\forall t \\geq 0 } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 136, + 477, + 159 + ], + "score": 1.0, + "content": "we have with probability at least", + "type": "text" + }, + { + "bbox": [ + 477, + 142, + 503, + 153 + ], + "score": 0.66, + "content": "1 - 6 \\delta", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 136, + 506, + 159 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 107, + 159, + 191, + 172 + ], + "spans": [ + { + "bbox": [ + 107, + 159, + 187, + 172 + ], + "score": 0.89, + "content": "\\| Q _ { T } - Q ^ { * } \\| _ { \\infty } \\leq 3 \\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 159, + 191, + 172 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 105, + 136, + 506, + 172 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 179, + 506, + 254 + ], + "lines": [ + { + "bbox": [ + 106, + 180, + 506, + 197 + ], + "spans": [ + { + "bbox": [ + 106, + 181, + 303, + 195 + ], + "score": 1.0, + "content": "From the theorem, we can see that to guarantee an", + "type": "text" + }, + { + "bbox": [ + 304, + 185, + 309, + 192 + ], + "score": 0.68, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 181, + 437, + 195 + ], + "score": 1.0, + "content": "-accurate estimate, a sample size", + "type": "text" + }, + { + "bbox": [ + 437, + 180, + 506, + 197 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\tilde { \\mathcal { O } } ( \\frac { 1 } { \\mu _ { \\mathrm { m i n } } ( 1 - \\gamma ) ^ { 5 } \\epsilon ^ { 2 } } + } \\end{array}", + "type": "inline_equation" + } + ], + "index": 5 + }, + { + "bbox": [ + 102, + 191, + 509, + 214 + ], + "spans": [ + { + "bbox": [ + 102, + 191, + 509, + 214 + ], + "score": 1.0, + "content": "tmixµmin(1−γ) ) (up to some logarithmic terms) is needed. This complexity matches with the one for the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 104, + 208, + 506, + 222 + ], + "spans": [ + { + "bbox": [ + 104, + 208, + 506, + 222 + ], + "score": 1.0, + "content": "vanilla Q-learning in [Li et al., 2020] (within a constant factor), while our algorithm also guarantees", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 220, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 105, + 220, + 505, + 232 + ], + "score": 1.0, + "content": "robustness to MDP model uncertainty. Our algorithm design and analysis can be readily extended", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 231, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 505, + 243 + ], + "score": 1.0, + "content": "to robustify TD and SARSA. The variance-reduction technique [Wainwright, 2019] can also be", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 241, + 486, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 451, + 255 + ], + "score": 1.0, + "content": "combined with our robust Q-learning algorithm to further improve the dependency on", + "type": "text" + }, + { + "bbox": [ + 451, + 242, + 482, + 254 + ], + "score": 0.92, + "content": "( 1 - \\gamma )", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 241, + 486, + 255 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 7.5, + "bbox_fs": [ + 102, + 180, + 509, + 255 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 270, + 325, + 284 + ], + "lines": [ + { + "bbox": [ + 104, + 269, + 326, + 285 + ], + "spans": [ + { + "bbox": [ + 104, + 269, + 326, + 285 + ], + "score": 1.0, + "content": "5 Function Approximation: Robust TDC", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 106, + 295, + 506, + 362 + ], + "lines": [ + { + "bbox": [ + 105, + 295, + 505, + 308 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 505, + 308 + ], + "score": 1.0, + "content": "In this section, we investigate the case where the state and action spaces can be large or even", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 307, + 506, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 506, + 320 + ], + "score": 1.0, + "content": "continuous. A popular approach is to approximate the value function using a parameterized function,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 317, + 505, + 330 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 505, + 330 + ], + "score": 1.0, + "content": "e.g., linear function and neural network. In this section, we focus on the case with linear function", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 328, + 506, + 341 + ], + "spans": [ + { + "bbox": [ + 104, + 328, + 506, + 341 + ], + "score": 1.0, + "content": "approximation to illustrate the main idea of designing robust RL algorithms. Our approach can be", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 339, + 507, + 352 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 507, + 352 + ], + "score": 1.0, + "content": "extended to non-linear (smooth) function approximation using techniques in, e.g., [Cai et al., 2019,", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 350, + 345, + 362 + ], + "spans": [ + { + "bbox": [ + 105, + 350, + 345, + 362 + ], + "score": 1.0, + "content": "Bhatnagar et al., 2009, Wai et al., 2019, Wang et al., 2021].", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14.5, + "bbox_fs": [ + 104, + 295, + 507, + 362 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 366, + 506, + 455 + ], + "lines": [ + { + "bbox": [ + 104, + 366, + 504, + 379 + ], + "spans": [ + { + "bbox": [ + 104, + 366, + 489, + 379 + ], + "score": 1.0, + "content": "We focus on the problem of robust policy evaluation, i.e., estimate the robust value function", + "type": "text" + }, + { + "bbox": [ + 489, + 367, + 504, + 377 + ], + "score": 0.81, + "content": "V ^ { \\pi }", + "type": "inline_equation" + } + ], + "index": 18 + }, + { + "bbox": [ + 104, + 376, + 507, + 391 + ], + "spans": [ + { + "bbox": [ + 104, + 376, + 234, + 391 + ], + "score": 1.0, + "content": "defined in (1) for a given policy", + "type": "text" + }, + { + "bbox": [ + 235, + 380, + 242, + 388 + ], + "score": 0.72, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 376, + 507, + 391 + ], + "score": 1.0, + "content": "under the worst-case MDP transition kernel in the uncertainty set.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 104, + 387, + 506, + 402 + ], + "spans": [ + { + "bbox": [ + 104, + 387, + 506, + 402 + ], + "score": 1.0, + "content": "Note that for robust RL with model uncertainty, any policy evaluation problem can be viewed as", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 399, + 506, + 412 + ], + "spans": [ + { + "bbox": [ + 106, + 399, + 506, + 412 + ], + "score": 1.0, + "content": "\"off-transition-kernel\", as it is to evaluate the value function under the worst-case MDP using samples", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 410, + 506, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 506, + 423 + ], + "score": 1.0, + "content": "from a different MDP. Since the TD algorithm with function approximation may diverge under", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 420, + 506, + 434 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 506, + 434 + ], + "score": 1.0, + "content": "off-policy training [Baird, 1995] and importance sampling cannot be applied here due to unknown", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 432, + 505, + 444 + ], + "spans": [ + { + "bbox": [ + 106, + 432, + 505, + 444 + ], + "score": 1.0, + "content": "transition kernel, in this paper we generalize the GTD method [Maei et al., 2010, Maei, 2011] to the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 442, + 165, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 165, + 456 + ], + "score": 1.0, + "content": "robust setting.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 21.5, + "bbox_fs": [ + 104, + 366, + 507, + 456 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 459, + 505, + 500 + ], + "lines": [ + { + "bbox": [ + 105, + 458, + 506, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 458, + 124, + 474 + ], + "score": 1.0, + "content": "Let", + "type": "text" + }, + { + "bbox": [ + 124, + 459, + 248, + 474 + ], + "score": 0.9, + "content": "\\left\\{ \\phi ^ { ( i ) } : \\mathbb { S } \\to \\mathbb { R } , i = 1 , \\ldots , N \\right\\}", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 458, + 299, + 474 + ], + "score": 1.0, + "content": "be a set of", + "type": "text" + }, + { + "bbox": [ + 299, + 461, + 309, + 470 + ], + "score": 0.81, + "content": "N", + "type": "inline_equation" + }, + { + "bbox": [ + 310, + 458, + 431, + 474 + ], + "score": 1.0, + "content": "fixed base functions, where", + "type": "text" + }, + { + "bbox": [ + 431, + 460, + 486, + 473 + ], + "score": 0.92, + "content": "N \\ll | \\mathcal S | | \\mathcal A |", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 458, + 506, + 474 + ], + "score": 1.0, + "content": ". In", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 471, + 506, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 441, + 488 + ], + "score": 1.0, + "content": "particular, we approximate the robust value function using a linear combination of", + "type": "text" + }, + { + "bbox": [ + 106, + 485, + 188, + 501 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\sum _ { i = 1 } ^ { N } \\theta ^ { i } \\phi _ { s } ^ { ( i ) } = \\phi _ { s } ^ { \\top } \\theta } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 478, + 218, + 507 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 219, + 487, + 252, + 498 + ], + "score": 0.91, + "content": "\\boldsymbol { \\theta } \\in \\mathbb { R } ^ { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 478, + 340, + 507 + ], + "score": 1.0, + "content": "is the weight vector.", + "type": "text" + }, + { + "bbox": [ + 442, + 473, + 457, + 486 + ], + "score": 0.89, + "content": "\\phi ^ { ( i ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 458, + 471, + 469, + 488 + ], + "score": 1.0, + "content": "’s:", + "type": "text" + }, + { + "bbox": [ + 469, + 473, + 506, + 486 + ], + "score": 0.89, + "content": "V _ { \\theta } ( s ) =", + "type": "inline_equation" + } + ], + "index": 27 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 458, + 506, + 507 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 504, + 374, + 516 + ], + "lines": [ + { + "bbox": [ + 105, + 502, + 375, + 519 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 363, + 519 + ], + "score": 1.0, + "content": "Define the following robust Bellman operator for a given policy", + "type": "text" + }, + { + "bbox": [ + 363, + 507, + 370, + 514 + ], + "score": 0.72, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 502, + 375, + 519 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 502, + 375, + 519 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 146, + 521, + 465, + 574 + ], + "lines": [ + { + "bbox": [ + 146, + 521, + 465, + 574 + ], + "spans": [ + { + "bbox": [ + 146, + 521, + 465, + 574 + ], + "score": 0.92, + "content": "\\begin{array} { r l } & { \\mathrm { \\bf T } _ { \\pi } V ( s ) \\triangleq \\mathbb E _ { A \\sim \\pi ( \\cdot | s ) } [ c ( s , A ) + \\gamma \\sigma _ { \\mathcal { P } _ { s } ^ { A } } ( V ) ] } \\\\ & { \\quad \\quad \\quad = \\mathbb E _ { A \\sim \\pi ( \\cdot | s ) } \\left[ c ( s , A ) + \\gamma ( 1 - R ) \\displaystyle \\sum _ { s ^ { \\prime } \\in \\mathcal { S } } p _ { s , s ^ { \\prime } } ^ { A } V ( s ^ { \\prime } ) + \\gamma R \\operatorname* { m a x } _ { s ^ { \\prime } \\in \\mathcal { S } } V ( s ^ { \\prime } ) \\right] . } \\end{array}", + "type": "interline_equation", + "image_path": "9852dcd21269061985da10dcabeac8e4caafd9fd4a4cbfc50d42023c0a2ae526.jpg" + } + ] + } + ], + "index": 30, + "virtual_lines": [ + { + "bbox": [ + 146, + 521, + 465, + 538.6666666666666 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 146, + 538.6666666666666, + 465, + 556.3333333333333 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 146, + 556.3333333333333, + 465, + 573.9999999999999 + ], + "spans": [], + "index": 31 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 579, + 424, + 591 + ], + "lines": [ + { + "bbox": [ + 105, + 578, + 425, + 593 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 425, + 593 + ], + "score": 1.0, + "content": "We then define the mean squared projected robust Bellman error (MSPRBE) as", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32, + "bbox_fs": [ + 105, + 578, + 425, + 593 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 231, + 597, + 379, + 614 + ], + "lines": [ + { + "bbox": [ + 231, + 597, + 379, + 614 + ], + "spans": [ + { + "bbox": [ + 231, + 597, + 379, + 614 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\mathrm { M S P R B E } ( \\theta ) = \\| \\mathbf { I I T } _ { \\pi } V _ { \\theta } - V _ { \\theta } \\| _ { \\mu _ { \\pi } } ^ { 2 } , } \\end{array}", + "type": "interline_equation", + "image_path": "7fdf581c44a14004534015be69003a637c4f9373f971dff5a0e1cb387e623e39.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 231, + 597, + 379, + 614 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 620, + 505, + 681 + ], + "lines": [ + { + "bbox": [ + 105, + 621, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 133, + 635 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 621, + 231, + 635 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\| v \\| _ { \\mu _ { \\pi } } ^ { 2 } = \\int v ^ { 2 } ( s ) \\mu _ { \\pi } ( d s ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 231, + 621, + 235, + 635 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 235, + 623, + 248, + 633 + ], + "score": 0.78, + "content": "\\mu _ { \\pi }", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 621, + 407, + 635 + ], + "score": 1.0, + "content": "is the stationary distribution induced by", + "type": "text" + }, + { + "bbox": [ + 408, + 624, + 415, + 632 + ], + "score": 0.76, + "content": "\\pi", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 621, + 435, + 635 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 435, + 622, + 446, + 632 + ], + "score": 0.76, + "content": "\\mathbf { I I }", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 621, + 505, + 635 + ], + "score": 1.0, + "content": "is a projection", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 633, + 506, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 245, + 647 + ], + "score": 1.0, + "content": "onto the linear function space w.r.t.", + "type": "text" + }, + { + "bbox": [ + 246, + 634, + 272, + 646 + ], + "score": 0.91, + "content": "\\| \\cdot \\| _ { \\mu _ { \\pi } }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 633, + 506, + 647 + ], + "score": 1.0, + "content": ". We will develop a two time-scale gradient-based approach", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 645, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 645, + 329, + 657 + ], + "score": 1.0, + "content": "to minimize the MSPRBE. However, it can be seen that", + "type": "text" + }, + { + "bbox": [ + 329, + 645, + 377, + 657 + ], + "score": 0.83, + "content": "\\operatorname* { m a x } _ { s } V _ { \\theta } ( s )", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 645, + 469, + 657 + ], + "score": 1.0, + "content": "in (7) is not smooth in", + "type": "text" + }, + { + "bbox": [ + 469, + 645, + 474, + 655 + ], + "score": 0.79, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 645, + 505, + 657 + ], + "score": 1.0, + "content": ", which", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 655, + 506, + 669 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 506, + 669 + ], + "score": 1.0, + "content": "is troublesome in both algorithm design and analysis. To solve this issue, we introduce the following", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 667, + 461, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 247, + 682 + ], + "score": 1.0, + "content": "smoothed robust Bellman operator", + "type": "text" + }, + { + "bbox": [ + 247, + 667, + 261, + 680 + ], + "score": 0.91, + "content": "\\hat { \\mathbf { T } } _ { \\pi }", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 668, + 461, + 682 + ], + "score": 1.0, + "content": "by smoothing the max with a LSE(LogSumExp):", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36, + "bbox_fs": [ + 105, + 621, + 506, + 682 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 149, + 686, + 461, + 721 + ], + "lines": [ + { + "bbox": [ + 149, + 686, + 461, + 721 + ], + "spans": [ + { + "bbox": [ + 149, + 686, + 461, + 721 + ], + "score": 0.93, + "content": "\\hat { \\mathbf { T } } _ { \\pi } V ( s ) = \\mathbb { E } _ { A \\sim \\pi ( \\cdot | s ) } \\left[ c ( s , A ) + \\gamma ( 1 - R ) \\sum _ { s ^ { \\prime } \\in \\mathcal { S } } p _ { s , s ^ { \\prime } } ^ { A } V ( s ^ { \\prime } ) + \\gamma R \\cdot \\mathrm { L S E } ( V ) \\right] ,", + "type": "interline_equation", + "image_path": "7543aac30207ffdd163f7a3ab81924732889d4c9ebbac9f5e3be94660e8a76c2.jpg" + } + ] + } + ], + "index": 40, + "virtual_lines": [ + { + "bbox": [ + 149, + 686, + 461, + 697.6666666666666 + ], + "spans": [], + "index": 39 + }, + { + "bbox": [ + 149, + 697.6666666666666, + 461, + 709.3333333333333 + ], + "spans": [], + "index": 40 + }, + { + "bbox": [ + 149, + 709.3333333333333, + 461, + 720.9999999999999 + ], + "spans": [], + "index": 41 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 69, + 506, + 137 + ], + "lines": [ + { + "bbox": [ + 105, + 70, + 506, + 90 + ], + "spans": [ + { + "bbox": [ + 105, + 70, + 132, + 90 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 70, + 236, + 90 + ], + "score": 0.94, + "content": "\\begin{array} { r } { \\mathrm { L S E } ( V ) = \\frac { \\log \\left( \\sum _ { s } e ^ { \\varrho V ( s ) } \\right) } { \\varrho } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 70, + 335, + 90 + ], + "score": 1.0, + "content": "is the LogSumExp w.r.t.", + "type": "text" + }, + { + "bbox": [ + 335, + 76, + 345, + 86 + ], + "score": 0.77, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 70, + 414, + 90 + ], + "score": 1.0, + "content": "with a parameter", + "type": "text" + }, + { + "bbox": [ + 414, + 76, + 438, + 87 + ], + "score": 0.9, + "content": "\\varrho > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 70, + 506, + 90 + ], + "score": 1.0, + "content": ". Note that when", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 107, + 89, + 505, + 104 + ], + "spans": [ + { + "bbox": [ + 107, + 93, + 138, + 102 + ], + "score": 0.89, + "content": "\\varrho \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 90, + 299, + 104 + ], + "score": 1.0, + "content": ", the smoothed robust Bellman operator", + "type": "text" + }, + { + "bbox": [ + 299, + 89, + 342, + 102 + ], + "score": 0.92, + "content": "\\hat { \\mathbf { T } } _ { \\pi } \\to \\mathbf { T } _ { \\pi }", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 90, + 505, + 104 + ], + "score": 1.0, + "content": ". The LSE operator can also be replaced", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 102, + 505, + 114 + ], + "spans": [ + { + "bbox": [ + 105, + 102, + 505, + 114 + ], + "score": 1.0, + "content": "by some other operator that approximates the max operator and is smooth, e.g., mellow-max [Asadi", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 113, + 505, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 405, + 127 + ], + "score": 1.0, + "content": "and Littman, 2017]. In the following, we first show that the fixed point of", + "type": "text" + }, + { + "bbox": [ + 406, + 113, + 420, + 126 + ], + "score": 0.9, + "content": "\\hat { \\mathbf { T } } _ { \\pi }", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 114, + 478, + 127 + ], + "score": 1.0, + "content": "exists for any", + "type": "text" + }, + { + "bbox": [ + 478, + 117, + 484, + 126 + ], + "score": 0.74, + "content": "\\varrho", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 114, + 505, + 127 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 124, + 324, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 124, + 261, + 139 + ], + "score": 1.0, + "content": "the fixed points converge to the one of", + "type": "text" + }, + { + "bbox": [ + 261, + 126, + 275, + 136 + ], + "score": 0.9, + "content": "\\mathbf { T } _ { \\pi }", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 124, + 313, + 139 + ], + "score": 1.0, + "content": "for large", + "type": "text" + }, + { + "bbox": [ + 314, + 128, + 320, + 137 + ], + "score": 0.76, + "content": "\\varrho", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 124, + 324, + 139 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 106, + 141, + 305, + 154 + ], + "lines": [ + { + "bbox": [ + 105, + 140, + 306, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 140, + 210, + 156 + ], + "score": 1.0, + "content": "Theorem 4. (1). For any", + "type": "text" + }, + { + "bbox": [ + 210, + 145, + 217, + 154 + ], + "score": 0.56, + "content": "\\varrho", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 140, + 219, + 156 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 220, + 140, + 235, + 154 + ], + "score": 0.83, + "content": "\\hat { \\mathbf { T } } _ { \\pi }", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 140, + 306, + 156 + ], + "score": 1.0, + "content": "has a fixed point.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 105, + 158, + 393, + 172 + ], + "lines": [ + { + "bbox": [ + 105, + 158, + 393, + 174 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 138, + 174 + ], + "score": 1.0, + "content": "(2). Let", + "type": "text" + }, + { + "bbox": [ + 138, + 160, + 150, + 171 + ], + "score": 0.87, + "content": "V _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 158, + 168, + 174 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 169, + 160, + 180, + 171 + ], + "score": 0.87, + "content": "V _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 158, + 266, + 174 + ], + "score": 1.0, + "content": "be the fixed points of", + "type": "text" + }, + { + "bbox": [ + 266, + 158, + 281, + 172 + ], + "score": 0.9, + "content": "\\hat { \\mathbf { T } } _ { \\pi }", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 158, + 299, + 174 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 299, + 160, + 313, + 172 + ], + "score": 0.89, + "content": "\\mathbf { T } _ { \\pi }", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 158, + 393, + 174 + ], + "score": 1.0, + "content": ", respectively. Then", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "interline_equation", + "bbox": [ + 209, + 176, + 402, + 204 + ], + "lines": [ + { + "bbox": [ + 209, + 176, + 402, + 204 + ], + "spans": [ + { + "bbox": [ + 209, + 176, + 402, + 204 + ], + "score": 0.94, + "content": "\\| V _ { 1 } - V _ { 2 } \\| _ { \\infty } \\leq \\frac { \\gamma R } { 1 - \\gamma } \\frac { \\log | \\mathcal { S } | } { \\varrho } 0 , a s \\varrho \\infty .", + "type": "interline_equation", + "image_path": "ad911689bf0003f0de2b6c5368d62b6e5ae5d9dd29ce3c7f51bc819eed442bbf.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 209, + 176, + 402, + 204 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 213, + 457, + 226 + ], + "lines": [ + { + "bbox": [ + 105, + 213, + 458, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 183, + 228 + ], + "score": 1.0, + "content": "We then denote by", + "type": "text" + }, + { + "bbox": [ + 183, + 214, + 203, + 226 + ], + "score": 0.92, + "content": "J ( \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 213, + 458, + 228 + ], + "score": 1.0, + "content": "the smoothed MSPRBE with the LSE operator, and the goal is:", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "interline_equation", + "bbox": [ + 228, + 231, + 381, + 257 + ], + "lines": [ + { + "bbox": [ + 228, + 231, + 381, + 257 + ], + "spans": [ + { + "bbox": [ + 228, + 231, + 381, + 257 + ], + "score": 0.94, + "content": "\\operatorname* { m i n } _ { \\theta } J ( \\theta ) = \\operatorname* { m i n } _ { \\theta } \\Big \\| \\mathbf { H } \\hat { \\mathbf { T } } _ { \\pi } V _ { \\theta } - V _ { \\theta } \\Big \\| _ { \\mu _ { \\pi } } ^ { 2 } .", + "type": "interline_equation", + "image_path": "fa3206f9b868d359d6943578e15679723de365442f3536e6f3308bb5498eeceb.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 228, + 231, + 381, + 257 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "title", + "bbox": [ + 107, + 267, + 233, + 280 + ], + "lines": [ + { + "bbox": [ + 105, + 266, + 234, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 234, + 282 + ], + "score": 1.0, + "content": "5.1 Algorithm Development", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 105, + 287, + 506, + 390 + ], + "lines": [ + { + "bbox": [ + 105, + 287, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 287, + 505, + 300 + ], + "score": 1.0, + "content": "In the following, we develop the robust TDC algorithm to solve the problem in (11). We will", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 299, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 106, + 299, + 333, + 310 + ], + "score": 1.0, + "content": "first derive the gradient of the smoothed MSPRBE,", + "type": "text" + }, + { + "bbox": [ + 333, + 299, + 353, + 311 + ], + "score": 0.91, + "content": "J ( \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 299, + 505, + 310 + ], + "score": 1.0, + "content": ", and then design a two time-scale", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 310, + 507, + 322 + ], + "spans": [ + { + "bbox": [ + 106, + 310, + 507, + 322 + ], + "score": 1.0, + "content": "update rule using the weight doubling trick in [Sutton et al., 2009] to solve the double sam-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 321, + 506, + 335 + ], + "spans": [ + { + "bbox": [ + 106, + 321, + 207, + 335 + ], + "score": 1.0, + "content": "pling problem. Define", + "type": "text" + }, + { + "bbox": [ + 207, + 321, + 472, + 334 + ], + "score": 0.88, + "content": "\\delta _ { s , a , s ^ { \\prime } } ( \\theta ) \\ \\triangleq \\ c ( s , a ) + \\gamma ( 1 - R ) V _ { \\theta } ( s ^ { \\prime } ) + \\gamma R \\mathrm { L S E } ( V _ { \\theta } ) - V _ { \\theta } ( s )", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 321, + 506, + 335 + ], + "score": 1.0, + "content": ", where", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 333, + 507, + 348 + ], + "spans": [ + { + "bbox": [ + 106, + 334, + 144, + 347 + ], + "score": 0.82, + "content": "\\operatorname { L S E } ( V _ { \\theta } )", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 333, + 292, + 348 + ], + "score": 1.0, + "content": "is the LogSumExp function w.r.t.", + "type": "text" + }, + { + "bbox": [ + 292, + 334, + 339, + 347 + ], + "score": 0.92, + "content": "V _ { \\theta } = \\theta ^ { \\top } \\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 333, + 394, + 348 + ], + "score": 1.0, + "content": ". Denote by", + "type": "text" + }, + { + "bbox": [ + 395, + 334, + 471, + 348 + ], + "score": 0.94, + "content": "C \\triangleq \\mathbb { E } _ { \\mu _ { \\pi } } \\left[ \\phi _ { S } ^ { \\intercal } \\phi _ { S } \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 333, + 507, + 348 + ], + "score": 1.0, + "content": ". Then,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 347, + 507, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 347, + 282, + 366 + ], + "score": 0.85, + "content": "\\mathbb { E } _ { \\mu } [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } ] \\ = \\ \\varPhi ^ { \\top } D \\left( \\hat { \\mathbf { T } } _ { \\pi } V _ { \\theta } - V _ { \\theta } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 348, + 317, + 365 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 317, + 351, + 484, + 364 + ], + "score": 0.89, + "content": "\\textit { D } = \\ \\mathrm { d i a g } ( \\mu _ { \\pi } ( s _ { 1 } ) , \\mu _ { \\pi } ( s _ { 2 } ) , . . . , \\mu _ { \\pi } ( s _ { | \\mathcal { S } | } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 348, + 507, + 365 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 364, + 507, + 381 + ], + "spans": [ + { + "bbox": [ + 106, + 366, + 254, + 381 + ], + "score": 0.86, + "content": "\\boldsymbol { \\varPhi } = ( \\phi _ { s _ { 1 } } , \\phi _ { s _ { 2 } } , . . . , \\phi _ { s _ { | \\mathcal { S } | } } ) ^ { \\top } \\in \\dot { \\mathbb { R } } ^ { | \\mathcal { S } | \\times N }", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 364, + 316, + 381 + ], + "score": 1.0, + "content": ". We know that", + "type": "text" + }, + { + "bbox": [ + 316, + 366, + 454, + 379 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\pmb { \\Pi } ^ { \\top } D \\pmb { \\Pi } = D ^ { \\top } \\pmb { \\varPhi } ( \\pmb { \\varPhi } ^ { \\top } D \\pmb { \\varPhi } ) ^ { - 1 } \\pmb { \\varPhi } ^ { \\top } D } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 454, + 364, + 507, + 381 + ], + "score": 1.0, + "content": "from [Maei,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 377, + 200, + 390 + ], + "spans": [ + { + "bbox": [ + 105, + 377, + 200, + 390 + ], + "score": 1.0, + "content": "2011]. Hence we have", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 14.5 + }, + { + "type": "interline_equation", + "bbox": [ + 154, + 394, + 457, + 420 + ], + "lines": [ + { + "bbox": [ + 154, + 394, + 457, + 420 + ], + "spans": [ + { + "bbox": [ + 154, + 394, + 457, + 420 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\boldsymbol J ( \\theta ) = \\Big \\| \\mathbf { \\hat { I } } \\mathbf { \\hat { T } } _ { \\boldsymbol \\pi } V _ { \\theta } - V _ { \\theta } \\Big \\| _ { \\mu _ { \\boldsymbol \\pi } } ^ { 2 } = \\mathbb { E } _ { \\mu _ { \\boldsymbol \\pi } } [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } ] ^ { \\top } \\boldsymbol C ^ { - 1 } \\mathbb { E } _ { \\mu _ { \\boldsymbol \\pi } } [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } ] . } \\end{array}", + "type": "interline_equation", + "image_path": "65951d26b16d119c3bba00a86bde2fc36e1073e08e65ed93c7a1b72420890a6a.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 154, + 394, + 457, + 420 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 424, + 250, + 436 + ], + "lines": [ + { + "bbox": [ + 106, + 423, + 251, + 437 + ], + "spans": [ + { + "bbox": [ + 106, + 423, + 251, + 437 + ], + "score": 1.0, + "content": "Then, its gradient can be written as:", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "interline_equation", + "bbox": [ + 153, + 440, + 457, + 493 + ], + "lines": [ + { + "bbox": [ + 153, + 440, + 457, + 493 + ], + "spans": [ + { + "bbox": [ + 153, + 440, + 457, + 493 + ], + "score": 0.92, + "content": "\\begin{array} { l } { \\displaystyle - \\frac { 1 } { 2 } \\nabla J ( \\theta ) = - \\mathbb { E } _ { \\mu _ { \\pi } } \\big [ ( \\nabla \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) ) \\phi _ { S } \\big ] ^ { \\top } C ^ { - 1 } \\mathbb { E } _ { \\mu _ { \\pi } } \\big [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } \\big ] } \\\\ { = \\mathbb { E } _ { \\mu _ { \\pi } } \\big [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } \\big ] - \\gamma \\mathbb { E } _ { \\mu _ { \\pi } } \\bigg [ \\bigg ( ( 1 - R ) \\phi _ { S ^ { \\prime } } + R \\cdot \\nabla \\mathrm { L S E } ( V _ { \\theta } ) \\bigg ) \\phi _ { S } ^ { \\top } \\bigg ] \\omega ( \\theta ) , } \\end{array}", + "type": "interline_equation", + "image_path": "f48da04f35475f740a3388ea84655013bc612fe39b6277c8a6300ce42221782b.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 153, + 440, + 457, + 457.6666666666667 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 153, + 457.6666666666667, + 457, + 475.33333333333337 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 153, + 475.33333333333337, + 457, + 493.00000000000006 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 497, + 506, + 589 + ], + "lines": [ + { + "bbox": [ + 105, + 497, + 507, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 133, + 512 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 498, + 261, + 511 + ], + "score": 0.92, + "content": "\\omega ( \\theta ) = C ^ { - 1 } \\mathbb { E } _ { \\mu _ { \\pi } } [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 497, + 474, + 512 + ], + "score": 1.0, + "content": ". It can be seen that to obtain an unbiased estimate of", + "type": "text" + }, + { + "bbox": [ + 474, + 498, + 502, + 511 + ], + "score": 0.91, + "content": "\\nabla J ( \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 497, + 507, + 512 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 510, + 506, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 506, + 522 + ], + "score": 1.0, + "content": "two independent samples are needed as there exists a multiplication of two expectations, which is", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 520, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 505, + 532 + ], + "score": 1.0, + "content": "not applicable when there is only one sample trajectory. We then utilize the weight doubling trick in", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 532, + 506, + 544 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 506, + 544 + ], + "score": 1.0, + "content": "[Sutton et al., 2009], and design the robust TDC algorithm in Algorithm 2. Specifically, we introduce", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 541, + 506, + 556 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 217, + 556 + ], + "score": 1.0, + "content": "a fast time scale to estimate", + "type": "text" + }, + { + "bbox": [ + 218, + 542, + 237, + 554 + ], + "score": 0.91, + "content": "\\omega ( \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 541, + 372, + 556 + ], + "score": 1.0, + "content": ", and a slow time scale to estimate", + "type": "text" + }, + { + "bbox": [ + 373, + 542, + 401, + 554 + ], + "score": 0.91, + "content": "\\nabla J ( \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 541, + 506, + 556 + ], + "score": 1.0, + "content": ". Denote the projection by", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 553, + 507, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 554, + 244, + 568 + ], + "score": 0.91, + "content": "\\Pi _ { K } ( x ) \\triangleq \\arg \\operatorname* { m i n } _ { \\| y \\| \\leq K } \\| y - x \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 553, + 277, + 569 + ], + "score": 1.0, + "content": "for any", + "type": "text" + }, + { + "bbox": [ + 277, + 554, + 311, + 565 + ], + "score": 0.89, + "content": "\\boldsymbol { x } \\in \\mathbb { R } ^ { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 553, + 507, + 569 + ], + "score": 1.0, + "content": ". Our robust TDC algorithm in Algorithm 2 can", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 565, + 506, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 506, + 579 + ], + "score": 1.0, + "content": "be implemented in an online and incremental fashion. If the uncertainty set becomes a singleton, i.e.,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 107, + 577, + 362, + 589 + ], + "spans": [ + { + "bbox": [ + 107, + 577, + 133, + 587 + ], + "score": 0.87, + "content": "R = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 577, + 362, + 589 + ], + "score": 1.0, + "content": ", then Algorithm 2 reduces to the vanilla TDC algorithm.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 27.5 + }, + { + "type": "title", + "bbox": [ + 106, + 600, + 306, + 613 + ], + "lines": [ + { + "bbox": [ + 106, + 601, + 307, + 614 + ], + "spans": [ + { + "bbox": [ + 106, + 601, + 307, + 614 + ], + "score": 1.0, + "content": "5.2 Finite-Time Error Bound of Robust TDC", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 106, + 621, + 506, + 665 + ], + "lines": [ + { + "bbox": [ + 106, + 621, + 506, + 634 + ], + "spans": [ + { + "bbox": [ + 106, + 621, + 252, + 634 + ], + "score": 1.0, + "content": "Unlike the vanilla TDC algorithm,", + "type": "text" + }, + { + "bbox": [ + 252, + 621, + 272, + 633 + ], + "score": 0.91, + "content": "J ( \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 621, + 506, + 634 + ], + "score": 1.0, + "content": "here is non-convex. Therefore, we are interested in the", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 632, + 506, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 312, + 645 + ], + "score": 1.0, + "content": "convergence to stationary points, i.e., the rate of", + "type": "text" + }, + { + "bbox": [ + 312, + 632, + 375, + 644 + ], + "score": 0.93, + "content": "\\| \\nabla J ( \\theta ) \\| 0", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 632, + 506, + 645 + ], + "score": 1.0, + "content": ". We first make some standard", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 643, + 506, + 655 + ], + "spans": [ + { + "bbox": [ + 104, + 643, + 506, + 655 + ], + "score": 1.0, + "content": "assumptions which are commonly used in RL algorithm analysis, e.g., [Wang and Zou, 2020, Kaledin", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 654, + 409, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 409, + 667 + ], + "score": 1.0, + "content": "et al., 2020, Xu et al., 2019, Srikant and Ying, 2019, Bhandari et al., 2018].", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 105, + 668, + 328, + 681 + ], + "lines": [ + { + "bbox": [ + 105, + 667, + 327, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 247, + 682 + ], + "score": 1.0, + "content": "Assumption 2 (Bounded feature).", + "type": "text" + }, + { + "bbox": [ + 248, + 668, + 324, + 681 + ], + "score": 0.9, + "content": "\\| \\phi _ { s } \\| _ { 2 } \\leq 1 , \\forall s \\in \\mathcal { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 667, + 327, + 682 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 105, + 683, + 417, + 696 + ], + "lines": [ + { + "bbox": [ + 105, + 682, + 418, + 697 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 271, + 697 + ], + "score": 1.0, + "content": "Assumption 3 (Bounded cost function).", + "type": "text" + }, + { + "bbox": [ + 271, + 684, + 369, + 696 + ], + "score": 0.83, + "content": "| c ( s , a ) | \\leq c _ { \\mathrm { m a x } } , \\forall s \\in \\mathcal { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 682, + 388, + 697 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 388, + 684, + 414, + 694 + ], + "score": 0.87, + "content": "a \\in { \\mathcal { A } }", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 682, + 418, + 697 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 38 + }, + { + "type": "text", + "bbox": [ + 106, + 699, + 505, + 723 + ], + "lines": [ + { + "bbox": [ + 105, + 697, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 303, + 714 + ], + "score": 1.0, + "content": "Assumption 4 (Problem solvability). The matrix", + "type": "text" + }, + { + "bbox": [ + 303, + 699, + 372, + 712 + ], + "score": 0.93, + "content": "C = \\mathbb { E } _ { \\mu _ { \\pi } } [ \\phi _ { S } \\phi _ { S } ^ { \\top } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 697, + 454, + 714 + ], + "score": 1.0, + "content": "is non-singular with", + "type": "text" + }, + { + "bbox": [ + 454, + 700, + 480, + 711 + ], + "score": 0.87, + "content": "\\lambda > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 697, + 506, + 714 + ], + "score": 1.0, + "content": "being", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 711, + 201, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 201, + 723 + ], + "score": 1.0, + "content": "its smallest eigenvalue.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 69, + 506, + 137 + ], + "lines": [ + { + "bbox": [ + 105, + 70, + 506, + 90 + ], + "spans": [ + { + "bbox": [ + 105, + 70, + 132, + 90 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 132, + 70, + 236, + 90 + ], + "score": 0.94, + "content": "\\begin{array} { r } { \\mathrm { L S E } ( V ) = \\frac { \\log \\left( \\sum _ { s } e ^ { \\varrho V ( s ) } \\right) } { \\varrho } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 70, + 335, + 90 + ], + "score": 1.0, + "content": "is the LogSumExp w.r.t.", + "type": "text" + }, + { + "bbox": [ + 335, + 76, + 345, + 86 + ], + "score": 0.77, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 70, + 414, + 90 + ], + "score": 1.0, + "content": "with a parameter", + "type": "text" + }, + { + "bbox": [ + 414, + 76, + 438, + 87 + ], + "score": 0.9, + "content": "\\varrho > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 70, + 506, + 90 + ], + "score": 1.0, + "content": ". Note that when", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 107, + 89, + 505, + 104 + ], + "spans": [ + { + "bbox": [ + 107, + 93, + 138, + 102 + ], + "score": 0.89, + "content": "\\varrho \\to \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 90, + 299, + 104 + ], + "score": 1.0, + "content": ", the smoothed robust Bellman operator", + "type": "text" + }, + { + "bbox": [ + 299, + 89, + 342, + 102 + ], + "score": 0.92, + "content": "\\hat { \\mathbf { T } } _ { \\pi } \\to \\mathbf { T } _ { \\pi }", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 90, + 505, + 104 + ], + "score": 1.0, + "content": ". The LSE operator can also be replaced", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 102, + 505, + 114 + ], + "spans": [ + { + "bbox": [ + 105, + 102, + 505, + 114 + ], + "score": 1.0, + "content": "by some other operator that approximates the max operator and is smooth, e.g., mellow-max [Asadi", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 113, + 505, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 405, + 127 + ], + "score": 1.0, + "content": "and Littman, 2017]. In the following, we first show that the fixed point of", + "type": "text" + }, + { + "bbox": [ + 406, + 113, + 420, + 126 + ], + "score": 0.9, + "content": "\\hat { \\mathbf { T } } _ { \\pi }", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 114, + 478, + 127 + ], + "score": 1.0, + "content": "exists for any", + "type": "text" + }, + { + "bbox": [ + 478, + 117, + 484, + 126 + ], + "score": 0.74, + "content": "\\varrho", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 114, + 505, + 127 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 124, + 324, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 124, + 261, + 139 + ], + "score": 1.0, + "content": "the fixed points converge to the one of", + "type": "text" + }, + { + "bbox": [ + 261, + 126, + 275, + 136 + ], + "score": 0.9, + "content": "\\mathbf { T } _ { \\pi }", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 124, + 313, + 139 + ], + "score": 1.0, + "content": "for large", + "type": "text" + }, + { + "bbox": [ + 314, + 128, + 320, + 137 + ], + "score": 0.76, + "content": "\\varrho", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 124, + 324, + 139 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 70, + 506, + 139 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 141, + 305, + 154 + ], + "lines": [ + { + "bbox": [ + 105, + 140, + 306, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 140, + 210, + 156 + ], + "score": 1.0, + "content": "Theorem 4. (1). For any", + "type": "text" + }, + { + "bbox": [ + 210, + 145, + 217, + 154 + ], + "score": 0.56, + "content": "\\varrho", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 140, + 219, + 156 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 220, + 140, + 235, + 154 + ], + "score": 0.83, + "content": "\\hat { \\mathbf { T } } _ { \\pi }", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 140, + 306, + 156 + ], + "score": 1.0, + "content": "has a fixed point.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 140, + 306, + 156 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 158, + 393, + 172 + ], + "lines": [ + { + "bbox": [ + 105, + 158, + 393, + 174 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 138, + 174 + ], + "score": 1.0, + "content": "(2). Let", + "type": "text" + }, + { + "bbox": [ + 138, + 160, + 150, + 171 + ], + "score": 0.87, + "content": "V _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 158, + 168, + 174 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 169, + 160, + 180, + 171 + ], + "score": 0.87, + "content": "V _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 158, + 266, + 174 + ], + "score": 1.0, + "content": "be the fixed points of", + "type": "text" + }, + { + "bbox": [ + 266, + 158, + 281, + 172 + ], + "score": 0.9, + "content": "\\hat { \\mathbf { T } } _ { \\pi }", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 158, + 299, + 174 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 299, + 160, + 313, + 172 + ], + "score": 0.89, + "content": "\\mathbf { T } _ { \\pi }", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 158, + 393, + 174 + ], + "score": 1.0, + "content": ", respectively. Then", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6, + "bbox_fs": [ + 105, + 158, + 393, + 174 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 209, + 176, + 402, + 204 + ], + "lines": [ + { + "bbox": [ + 209, + 176, + 402, + 204 + ], + "spans": [ + { + "bbox": [ + 209, + 176, + 402, + 204 + ], + "score": 0.94, + "content": "\\| V _ { 1 } - V _ { 2 } \\| _ { \\infty } \\leq \\frac { \\gamma R } { 1 - \\gamma } \\frac { \\log | \\mathcal { S } | } { \\varrho } 0 , a s \\varrho \\infty .", + "type": "interline_equation", + "image_path": "ad911689bf0003f0de2b6c5368d62b6e5ae5d9dd29ce3c7f51bc819eed442bbf.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 209, + 176, + 402, + 204 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 213, + 457, + 226 + ], + "lines": [ + { + "bbox": [ + 105, + 213, + 458, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 183, + 228 + ], + "score": 1.0, + "content": "We then denote by", + "type": "text" + }, + { + "bbox": [ + 183, + 214, + 203, + 226 + ], + "score": 0.92, + "content": "J ( \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 213, + 458, + 228 + ], + "score": 1.0, + "content": "the smoothed MSPRBE with the LSE operator, and the goal is:", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 105, + 213, + 458, + 228 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 228, + 231, + 381, + 257 + ], + "lines": [ + { + "bbox": [ + 228, + 231, + 381, + 257 + ], + "spans": [ + { + "bbox": [ + 228, + 231, + 381, + 257 + ], + "score": 0.94, + "content": "\\operatorname* { m i n } _ { \\theta } J ( \\theta ) = \\operatorname* { m i n } _ { \\theta } \\Big \\| \\mathbf { H } \\hat { \\mathbf { T } } _ { \\pi } V _ { \\theta } - V _ { \\theta } \\Big \\| _ { \\mu _ { \\pi } } ^ { 2 } .", + "type": "interline_equation", + "image_path": "fa3206f9b868d359d6943578e15679723de365442f3536e6f3308bb5498eeceb.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 228, + 231, + 381, + 257 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "title", + "bbox": [ + 107, + 267, + 233, + 280 + ], + "lines": [ + { + "bbox": [ + 105, + 266, + 234, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 266, + 234, + 282 + ], + "score": 1.0, + "content": "5.1 Algorithm Development", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 105, + 287, + 506, + 390 + ], + "lines": [ + { + "bbox": [ + 105, + 287, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 287, + 505, + 300 + ], + "score": 1.0, + "content": "In the following, we develop the robust TDC algorithm to solve the problem in (11). We will", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 299, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 106, + 299, + 333, + 310 + ], + "score": 1.0, + "content": "first derive the gradient of the smoothed MSPRBE,", + "type": "text" + }, + { + "bbox": [ + 333, + 299, + 353, + 311 + ], + "score": 0.91, + "content": "J ( \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 299, + 505, + 310 + ], + "score": 1.0, + "content": ", and then design a two time-scale", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 310, + 507, + 322 + ], + "spans": [ + { + "bbox": [ + 106, + 310, + 507, + 322 + ], + "score": 1.0, + "content": "update rule using the weight doubling trick in [Sutton et al., 2009] to solve the double sam-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 321, + 506, + 335 + ], + "spans": [ + { + "bbox": [ + 106, + 321, + 207, + 335 + ], + "score": 1.0, + "content": "pling problem. Define", + "type": "text" + }, + { + "bbox": [ + 207, + 321, + 472, + 334 + ], + "score": 0.88, + "content": "\\delta _ { s , a , s ^ { \\prime } } ( \\theta ) \\ \\triangleq \\ c ( s , a ) + \\gamma ( 1 - R ) V _ { \\theta } ( s ^ { \\prime } ) + \\gamma R \\mathrm { L S E } ( V _ { \\theta } ) - V _ { \\theta } ( s )", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 321, + 506, + 335 + ], + "score": 1.0, + "content": ", where", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 333, + 507, + 348 + ], + "spans": [ + { + "bbox": [ + 106, + 334, + 144, + 347 + ], + "score": 0.82, + "content": "\\operatorname { L S E } ( V _ { \\theta } )", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 333, + 292, + 348 + ], + "score": 1.0, + "content": "is the LogSumExp function w.r.t.", + "type": "text" + }, + { + "bbox": [ + 292, + 334, + 339, + 347 + ], + "score": 0.92, + "content": "V _ { \\theta } = \\theta ^ { \\top } \\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 333, + 394, + 348 + ], + "score": 1.0, + "content": ". Denote by", + "type": "text" + }, + { + "bbox": [ + 395, + 334, + 471, + 348 + ], + "score": 0.94, + "content": "C \\triangleq \\mathbb { E } _ { \\mu _ { \\pi } } \\left[ \\phi _ { S } ^ { \\intercal } \\phi _ { S } \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 333, + 507, + 348 + ], + "score": 1.0, + "content": ". Then,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 347, + 507, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 347, + 282, + 366 + ], + "score": 0.85, + "content": "\\mathbb { E } _ { \\mu } [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } ] \\ = \\ \\varPhi ^ { \\top } D \\left( \\hat { \\mathbf { T } } _ { \\pi } V _ { \\theta } - V _ { \\theta } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 348, + 317, + 365 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 317, + 351, + 484, + 364 + ], + "score": 0.89, + "content": "\\textit { D } = \\ \\mathrm { d i a g } ( \\mu _ { \\pi } ( s _ { 1 } ) , \\mu _ { \\pi } ( s _ { 2 } ) , . . . , \\mu _ { \\pi } ( s _ { | \\mathcal { S } | } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 348, + 507, + 365 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 364, + 507, + 381 + ], + "spans": [ + { + "bbox": [ + 106, + 366, + 254, + 381 + ], + "score": 0.86, + "content": "\\boldsymbol { \\varPhi } = ( \\phi _ { s _ { 1 } } , \\phi _ { s _ { 2 } } , . . . , \\phi _ { s _ { | \\mathcal { S } | } } ) ^ { \\top } \\in \\dot { \\mathbb { R } } ^ { | \\mathcal { S } | \\times N }", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 364, + 316, + 381 + ], + "score": 1.0, + "content": ". We know that", + "type": "text" + }, + { + "bbox": [ + 316, + 366, + 454, + 379 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\pmb { \\Pi } ^ { \\top } D \\pmb { \\Pi } = D ^ { \\top } \\pmb { \\varPhi } ( \\pmb { \\varPhi } ^ { \\top } D \\pmb { \\varPhi } ) ^ { - 1 } \\pmb { \\varPhi } ^ { \\top } D } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 454, + 364, + 507, + 381 + ], + "score": 1.0, + "content": "from [Maei,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 377, + 200, + 390 + ], + "spans": [ + { + "bbox": [ + 105, + 377, + 200, + 390 + ], + "score": 1.0, + "content": "2011]. Hence we have", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 287, + 507, + 390 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 154, + 394, + 457, + 420 + ], + "lines": [ + { + "bbox": [ + 154, + 394, + 457, + 420 + ], + "spans": [ + { + "bbox": [ + 154, + 394, + 457, + 420 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\boldsymbol J ( \\theta ) = \\Big \\| \\mathbf { \\hat { I } } \\mathbf { \\hat { T } } _ { \\boldsymbol \\pi } V _ { \\theta } - V _ { \\theta } \\Big \\| _ { \\mu _ { \\boldsymbol \\pi } } ^ { 2 } = \\mathbb { E } _ { \\mu _ { \\boldsymbol \\pi } } [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } ] ^ { \\top } \\boldsymbol C ^ { - 1 } \\mathbb { E } _ { \\mu _ { \\boldsymbol \\pi } } [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } ] . } \\end{array}", + "type": "interline_equation", + "image_path": "65951d26b16d119c3bba00a86bde2fc36e1073e08e65ed93c7a1b72420890a6a.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 154, + 394, + 457, + 420 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 424, + 250, + 436 + ], + "lines": [ + { + "bbox": [ + 106, + 423, + 251, + 437 + ], + "spans": [ + { + "bbox": [ + 106, + 423, + 251, + 437 + ], + "score": 1.0, + "content": "Then, its gradient can be written as:", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20, + "bbox_fs": [ + 106, + 423, + 251, + 437 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 153, + 440, + 457, + 493 + ], + "lines": [ + { + "bbox": [ + 153, + 440, + 457, + 493 + ], + "spans": [ + { + "bbox": [ + 153, + 440, + 457, + 493 + ], + "score": 0.92, + "content": "\\begin{array} { l } { \\displaystyle - \\frac { 1 } { 2 } \\nabla J ( \\theta ) = - \\mathbb { E } _ { \\mu _ { \\pi } } \\big [ ( \\nabla \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) ) \\phi _ { S } \\big ] ^ { \\top } C ^ { - 1 } \\mathbb { E } _ { \\mu _ { \\pi } } \\big [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } \\big ] } \\\\ { = \\mathbb { E } _ { \\mu _ { \\pi } } \\big [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } \\big ] - \\gamma \\mathbb { E } _ { \\mu _ { \\pi } } \\bigg [ \\bigg ( ( 1 - R ) \\phi _ { S ^ { \\prime } } + R \\cdot \\nabla \\mathrm { L S E } ( V _ { \\theta } ) \\bigg ) \\phi _ { S } ^ { \\top } \\bigg ] \\omega ( \\theta ) , } \\end{array}", + "type": "interline_equation", + "image_path": "f48da04f35475f740a3388ea84655013bc612fe39b6277c8a6300ce42221782b.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 153, + 440, + 457, + 457.6666666666667 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 153, + 457.6666666666667, + 457, + 475.33333333333337 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 153, + 475.33333333333337, + 457, + 493.00000000000006 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 497, + 506, + 589 + ], + "lines": [ + { + "bbox": [ + 105, + 497, + 507, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 133, + 512 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 498, + 261, + 511 + ], + "score": 0.92, + "content": "\\omega ( \\theta ) = C ^ { - 1 } \\mathbb { E } _ { \\mu _ { \\pi } } [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 497, + 474, + 512 + ], + "score": 1.0, + "content": ". It can be seen that to obtain an unbiased estimate of", + "type": "text" + }, + { + "bbox": [ + 474, + 498, + 502, + 511 + ], + "score": 0.91, + "content": "\\nabla J ( \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 497, + 507, + 512 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 510, + 506, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 506, + 522 + ], + "score": 1.0, + "content": "two independent samples are needed as there exists a multiplication of two expectations, which is", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 520, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 505, + 532 + ], + "score": 1.0, + "content": "not applicable when there is only one sample trajectory. We then utilize the weight doubling trick in", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 532, + 506, + 544 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 506, + 544 + ], + "score": 1.0, + "content": "[Sutton et al., 2009], and design the robust TDC algorithm in Algorithm 2. Specifically, we introduce", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 541, + 506, + 556 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 217, + 556 + ], + "score": 1.0, + "content": "a fast time scale to estimate", + "type": "text" + }, + { + "bbox": [ + 218, + 542, + 237, + 554 + ], + "score": 0.91, + "content": "\\omega ( \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 541, + 372, + 556 + ], + "score": 1.0, + "content": ", and a slow time scale to estimate", + "type": "text" + }, + { + "bbox": [ + 373, + 542, + 401, + 554 + ], + "score": 0.91, + "content": "\\nabla J ( \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 541, + 506, + 556 + ], + "score": 1.0, + "content": ". Denote the projection by", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 553, + 507, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 554, + 244, + 568 + ], + "score": 0.91, + "content": "\\Pi _ { K } ( x ) \\triangleq \\arg \\operatorname* { m i n } _ { \\| y \\| \\leq K } \\| y - x \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 553, + 277, + 569 + ], + "score": 1.0, + "content": "for any", + "type": "text" + }, + { + "bbox": [ + 277, + 554, + 311, + 565 + ], + "score": 0.89, + "content": "\\boldsymbol { x } \\in \\mathbb { R } ^ { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 553, + 507, + 569 + ], + "score": 1.0, + "content": ". Our robust TDC algorithm in Algorithm 2 can", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 565, + 506, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 506, + 579 + ], + "score": 1.0, + "content": "be implemented in an online and incremental fashion. If the uncertainty set becomes a singleton, i.e.,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 107, + 577, + 362, + 589 + ], + "spans": [ + { + "bbox": [ + 107, + 577, + 133, + 587 + ], + "score": 0.87, + "content": "R = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 577, + 362, + 589 + ], + "score": 1.0, + "content": ", then Algorithm 2 reduces to the vanilla TDC algorithm.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 27.5, + "bbox_fs": [ + 105, + 497, + 507, + 589 + ] + }, + { + "type": "title", + "bbox": [ + 106, + 600, + 306, + 613 + ], + "lines": [ + { + "bbox": [ + 106, + 601, + 307, + 614 + ], + "spans": [ + { + "bbox": [ + 106, + 601, + 307, + 614 + ], + "score": 1.0, + "content": "5.2 Finite-Time Error Bound of Robust TDC", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 106, + 621, + 506, + 665 + ], + "lines": [ + { + "bbox": [ + 106, + 621, + 506, + 634 + ], + "spans": [ + { + "bbox": [ + 106, + 621, + 252, + 634 + ], + "score": 1.0, + "content": "Unlike the vanilla TDC algorithm,", + "type": "text" + }, + { + "bbox": [ + 252, + 621, + 272, + 633 + ], + "score": 0.91, + "content": "J ( \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 621, + 506, + 634 + ], + "score": 1.0, + "content": "here is non-convex. Therefore, we are interested in the", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 632, + 506, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 312, + 645 + ], + "score": 1.0, + "content": "convergence to stationary points, i.e., the rate of", + "type": "text" + }, + { + "bbox": [ + 312, + 632, + 375, + 644 + ], + "score": 0.93, + "content": "\\| \\nabla J ( \\theta ) \\| 0", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 632, + 506, + 645 + ], + "score": 1.0, + "content": ". We first make some standard", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 643, + 506, + 655 + ], + "spans": [ + { + "bbox": [ + 104, + 643, + 506, + 655 + ], + "score": 1.0, + "content": "assumptions which are commonly used in RL algorithm analysis, e.g., [Wang and Zou, 2020, Kaledin", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 654, + 409, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 409, + 667 + ], + "score": 1.0, + "content": "et al., 2020, Xu et al., 2019, Srikant and Ying, 2019, Bhandari et al., 2018].", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5, + "bbox_fs": [ + 104, + 621, + 506, + 667 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 668, + 328, + 681 + ], + "lines": [ + { + "bbox": [ + 105, + 667, + 327, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 247, + 682 + ], + "score": 1.0, + "content": "Assumption 2 (Bounded feature).", + "type": "text" + }, + { + "bbox": [ + 248, + 668, + 324, + 681 + ], + "score": 0.9, + "content": "\\| \\phi _ { s } \\| _ { 2 } \\leq 1 , \\forall s \\in \\mathcal { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 667, + 327, + 682 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37, + "bbox_fs": [ + 105, + 667, + 327, + 682 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 683, + 417, + 696 + ], + "lines": [ + { + "bbox": [ + 105, + 682, + 418, + 697 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 271, + 697 + ], + "score": 1.0, + "content": "Assumption 3 (Bounded cost function).", + "type": "text" + }, + { + "bbox": [ + 271, + 684, + 369, + 696 + ], + "score": 0.83, + "content": "| c ( s , a ) | \\leq c _ { \\mathrm { m a x } } , \\forall s \\in \\mathcal { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 682, + 388, + 697 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 388, + 684, + 414, + 694 + ], + "score": 0.87, + "content": "a \\in { \\mathcal { A } }", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 682, + 418, + 697 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 38, + "bbox_fs": [ + 105, + 682, + 418, + 697 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 699, + 505, + 723 + ], + "lines": [ + { + "bbox": [ + 105, + 697, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 105, + 697, + 303, + 714 + ], + "score": 1.0, + "content": "Assumption 4 (Problem solvability). The matrix", + "type": "text" + }, + { + "bbox": [ + 303, + 699, + 372, + 712 + ], + "score": 0.93, + "content": "C = \\mathbb { E } _ { \\mu _ { \\pi } } [ \\phi _ { S } \\phi _ { S } ^ { \\top } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 697, + 454, + 714 + ], + "score": 1.0, + "content": "is non-singular with", + "type": "text" + }, + { + "bbox": [ + 454, + 700, + 480, + 711 + ], + "score": 0.87, + "content": "\\lambda > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 697, + 506, + 714 + ], + "score": 1.0, + "content": "being", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 711, + 201, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 201, + 723 + ], + "score": 1.0, + "content": "its smallest eigenvalue.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5, + "bbox_fs": [ + 105, + 697, + 506, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 86, + 490, + 214 + ], + "lines": [ + { + "bbox": [ + 106, + 84, + 338, + 98 + ], + "spans": [ + { + "bbox": [ + 106, + 84, + 137, + 98 + ], + "score": 1.0, + "content": "Input:", + "type": "text" + }, + { + "bbox": [ + 137, + 86, + 189, + 97 + ], + "score": 0.27, + "content": "T , \\alpha , \\beta , \\varrho , \\phi _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 84, + 204, + 98 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 205, + 86, + 253, + 97 + ], + "score": 0.89, + "content": "i = 1 , . . . , N", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 84, + 327, + 98 + ], + "score": 1.0, + "content": ", projection radius", + "type": "text" + }, + { + "bbox": [ + 327, + 86, + 338, + 96 + ], + "score": 0.7, + "content": "K", + "type": "inline_equation" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 96, + 207, + 108 + ], + "spans": [ + { + "bbox": [ + 106, + 96, + 207, + 108 + ], + "score": 1.0, + "content": "Initialization: θ0,w0, s0", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 109, + 107, + 280, + 122 + ], + "spans": [ + { + "bbox": [ + 109, + 107, + 155, + 122 + ], + "score": 1.0, + "content": "1: Choose", + "type": "text" + }, + { + "bbox": [ + 156, + 109, + 280, + 121 + ], + "score": 0.82, + "content": "W \\sim \\mathrm { U n i f o r m } ( 0 , 1 , . . . , T - 1 )", + "type": "inline_equation" + } + ], + "index": 2 + }, + { + "bbox": [ + 110, + 119, + 241, + 132 + ], + "spans": [ + { + "bbox": [ + 110, + 119, + 138, + 132 + ], + "score": 1.0, + "content": "2: for", + "type": "text" + }, + { + "bbox": [ + 138, + 121, + 225, + 132 + ], + "score": 0.88, + "content": "t = 0 , 1 , 2 , . . . , W - 1", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 119, + 241, + 132 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 108, + 129, + 361, + 145 + ], + "spans": [ + { + "bbox": [ + 108, + 129, + 234, + 145 + ], + "score": 1.0, + "content": "3: Take action according to", + "type": "text" + }, + { + "bbox": [ + 234, + 131, + 262, + 144 + ], + "score": 0.92, + "content": "\\pi ( \\cdot | s _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 129, + 314, + 145 + ], + "score": 1.0, + "content": "and observe", + "type": "text" + }, + { + "bbox": [ + 314, + 133, + 333, + 143 + ], + "score": 0.9, + "content": "s _ { t + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 129, + 351, + 145 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 352, + 133, + 361, + 142 + ], + "score": 0.77, + "content": "c _ { t }", + "type": "inline_equation" + } + ], + "index": 4 + }, + { + "bbox": [ + 110, + 141, + 173, + 157 + ], + "spans": [ + { + "bbox": [ + 110, + 141, + 133, + 157 + ], + "score": 1.0, + "content": "4:", + "type": "text" + }, + { + "bbox": [ + 133, + 142, + 173, + 155 + ], + "score": 0.74, + "content": "\\phi _ { t } \\phi _ { s _ { t } }", + "type": "inline_equation" + } + ], + "index": 5 + }, + { + "bbox": [ + 108, + 148, + 488, + 214 + ], + "spans": [ + { + "bbox": [ + 108, + 198, + 131, + 214 + ], + "score": 1.0, + "content": "7:", + "type": "text" + }, + { + "bbox": [ + 110, + 148, + 131, + 202 + ], + "score": 1.0, + "content": "5: 6:", + "type": "text" + }, + { + "bbox": [ + 132, + 152, + 488, + 214 + ], + "score": 0.91, + "content": "\\begin{array} { r l } & { \\ddot { \\delta _ { t } } ( \\theta _ { t } ) \\gets c _ { t } + \\gamma ( 1 - R ) V _ { \\theta _ { t } } ( s _ { t + 1 } ) + \\gamma R \\frac { \\log ( \\sum _ { s } e ^ { \\varrho \\theta ^ { \\top } \\phi _ { s } } ) } { \\varrho } - V _ { \\theta _ { t } } ( s _ { t } ) } \\\\ & { \\theta _ { t + 1 } \\gets \\Pi _ { K } \\left( \\theta _ { t } + \\alpha \\left( \\delta _ { t } ( \\theta _ { t } ) \\phi _ { t } - \\gamma \\bigg ( ( 1 - R ) \\phi _ { t + 1 } + R \\sum _ { s \\in \\mathcal { S } } \\left( \\frac { e ^ { \\sigma V _ { \\theta } ( s ) } \\phi _ { s } } { \\sum _ { j \\in \\mathcal { S } } e ^ { \\sigma V _ { \\theta } ( j ) } } \\right) \\bigg ) \\phi _ { t } ^ { \\top } \\omega _ { t } \\right) \\right) } \\\\ & { \\omega _ { t + 1 } \\gets \\Pi _ { K } ( \\omega _ { t } + \\beta ( \\delta _ { t } ( \\theta _ { t } ) - \\phi _ { t } ^ { \\top } \\omega _ { t } ) \\phi _ { t } ) } \\end{array}", + "type": "inline_equation", + "image_path": "8504c8f66aa9dae624c071a98e50fb92a72dc33f61233228fc326889e3085dd1.jpg" + } + ], + "index": 6 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 111, + 212, + 156, + 222 + ], + "lines": [ + { + "bbox": [ + 110, + 210, + 157, + 223 + ], + "spans": [ + { + "bbox": [ + 110, + 210, + 157, + 223 + ], + "score": 1.0, + "content": "8: end for", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 107, + 224, + 158, + 235 + ], + "lines": [ + { + "bbox": [ + 105, + 221, + 158, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 221, + 143, + 239 + ], + "score": 1.0, + "content": "Output:", + "type": "text" + }, + { + "bbox": [ + 143, + 223, + 158, + 235 + ], + "score": 0.73, + "content": "\\theta _ { W }", + "type": "inline_equation" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 106, + 258, + 503, + 282 + ], + "lines": [ + { + "bbox": [ + 105, + 256, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 256, + 415, + 272 + ], + "score": 1.0, + "content": "Assumption 5 (Geometric uniform ergodicity). There exist some constants", + "type": "text" + }, + { + "bbox": [ + 416, + 259, + 444, + 269 + ], + "score": 0.89, + "content": "m > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 256, + 464, + 272 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 464, + 258, + 505, + 270 + ], + "score": 0.92, + "content": "\\rho \\in ( 0 , 1 )", + "type": "inline_equation" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 268, + 364, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 176, + 282 + ], + "score": 1.0, + "content": "such that for any", + "type": "text" + }, + { + "bbox": [ + 176, + 270, + 199, + 280 + ], + "score": 0.83, + "content": "t > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 268, + 201, + 282 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 202, + 269, + 359, + 281 + ], + "score": 0.83, + "content": "\\begin{array} { r } { \\operatorname* { m a x } _ { s \\in \\mathcal { S } } d _ { T V } ( \\bar { \\mathbb { P } } ( s _ { t } | s _ { 0 } = s ) , \\mu _ { \\pi } ) \\leq m \\rho ^ { t } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 268, + 364, + 282 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 107, + 289, + 505, + 322 + ], + "lines": [ + { + "bbox": [ + 105, + 289, + 505, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 505, + 301 + ], + "score": 1.0, + "content": "In the following theorem, we characterize the finite-time error bound for the convergence of our", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 300, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 440, + 312 + ], + "score": 1.0, + "content": "robust TDC algorithm. Here we only provide the order of the bounds in terms of", + "type": "text" + }, + { + "bbox": [ + 440, + 301, + 448, + 310 + ], + "score": 0.81, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 300, + 505, + 312 + ], + "score": 1.0, + "content": ". The explicit", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 310, + 298, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 310, + 298, + 324 + ], + "score": 1.0, + "content": "bounds can be found in (129) in Appendix D.3.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 325, + 505, + 348 + ], + "lines": [ + { + "bbox": [ + 105, + 324, + 505, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 302, + 341 + ], + "score": 1.0, + "content": "Theorem 5. Consider the following step-sizes:", + "type": "text" + }, + { + "bbox": [ + 302, + 325, + 353, + 339 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\beta = \\mathcal { O } \\left( \\frac { 1 } { T ^ { b } } \\right) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 324, + 375, + 341 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 376, + 325, + 427, + 339 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\alpha = \\mathcal { O } \\left( \\frac { 1 } { T ^ { a } } \\right) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 324, + 459, + 341 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 460, + 325, + 505, + 339 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\frac { 1 } { 2 } < a \\leq 1 } \\end{array}", + "type": "inline_equation" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 336, + 246, + 349 + ], + "spans": [ + { + "bbox": [ + 106, + 336, + 124, + 349 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 337, + 166, + 348 + ], + "score": 0.91, + "content": "0 < b \\leq a", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 336, + 246, + 349 + ], + "score": 1.0, + "content": ". Then we have that", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5 + }, + { + "type": "interline_equation", + "bbox": [ + 172, + 353, + 437, + 380 + ], + "lines": [ + { + "bbox": [ + 172, + 353, + 437, + 380 + ], + "spans": [ + { + "bbox": [ + 172, + 353, + 437, + 380 + ], + "score": 0.92, + "content": "\\mathbb { E } [ \\| \\nabla J ( \\theta _ { W } ) \\| ^ { 2 } ] = \\odot \\left( \\frac { 1 } { T \\alpha } + \\alpha \\log ( 1 / \\alpha ) + \\frac { 1 } { T \\beta } + \\beta \\log ( 1 / \\beta ) \\right) ,", + "type": "interline_equation", + "image_path": "a654cd452a7e1f7e9a634b553d55226b856e45fcf1332c2009f8149f8fab6ff9.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 172, + 353, + 437, + 380 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 385, + 367, + 406 + ], + "lines": [ + { + "bbox": [ + 101, + 381, + 365, + 409 + ], + "spans": [ + { + "bbox": [ + 101, + 381, + 171, + 409 + ], + "score": 1.0, + "content": "If we further let", + "type": "text" + }, + { + "bbox": [ + 172, + 389, + 221, + 401 + ], + "score": 0.82, + "content": "a = b = 0 . 5", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 381, + 245, + 409 + ], + "score": 1.0, + "content": ", then", + "type": "text" + }, + { + "bbox": [ + 245, + 385, + 365, + 406 + ], + "score": 0.8, + "content": "\\begin{array} { r } { \\mathbb { E } [ \\| \\nabla J ( \\theta _ { W } ) \\| ^ { 2 } ] = \\odot \\left( \\frac { \\log T } { \\sqrt { T } } \\right) . } \\end{array}", + "type": "inline_equation" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 106, + 413, + 506, + 457 + ], + "lines": [ + { + "bbox": [ + 105, + 412, + 507, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 507, + 426 + ], + "score": 1.0, + "content": "The robust TDC has a matching complexity with the vanilla TDC with non-linear function approxi-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 424, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 106, + 424, + 505, + 435 + ], + "score": 1.0, + "content": "mation [Wang et al., 2021], but provides the additional robustness to model uncertainty. It does not", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 435, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 106, + 435, + 506, + 447 + ], + "score": 1.0, + "content": "need to relax the uncertainty set like in [Roy et al., 2017], and our convergence results do not need a", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 446, + 237, + 458 + ], + "spans": [ + { + "bbox": [ + 106, + 446, + 237, + 458 + ], + "score": 1.0, + "content": "condition on the discount factor.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19.5 + }, + { + "type": "title", + "bbox": [ + 107, + 473, + 191, + 486 + ], + "lines": [ + { + "bbox": [ + 104, + 471, + 193, + 489 + ], + "spans": [ + { + "bbox": [ + 104, + 471, + 193, + 489 + ], + "score": 1.0, + "content": "6 Experiments", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "title", + "bbox": [ + 107, + 497, + 214, + 509 + ], + "lines": [ + { + "bbox": [ + 105, + 495, + 215, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 215, + 512 + ], + "score": 1.0, + "content": "6.1 Robust Q-Learning", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 106, + 517, + 506, + 662 + ], + "lines": [ + { + "bbox": [ + 105, + 516, + 506, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 506, + 531 + ], + "score": 1.0, + "content": "In this section, we compare our robust Q-learning with the vanilla non-robust Q-learning. We use", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 529, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 106, + 529, + 505, + 540 + ], + "score": 1.0, + "content": "OpenAI gym framework [Brockman et al., 2016], and consider two different problems: Frozen lake", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 539, + 506, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 506, + 552 + ], + "score": 1.0, + "content": "and Cart-Pole. One more example of the taxi problem is given in the appendix. To demonstrate the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 550, + 506, + 562 + ], + "spans": [ + { + "bbox": [ + 106, + 550, + 506, + 562 + ], + "score": 1.0, + "content": "robustness, the policy is learned in a perturbed MDP, and is then tested on the true unperturbed MDP.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 561, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 316, + 574 + ], + "score": 1.0, + "content": "Specifically, during the training, we set a probability", + "type": "text" + }, + { + "bbox": [ + 317, + 563, + 323, + 573 + ], + "score": 0.8, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 561, + 505, + 574 + ], + "score": 1.0, + "content": "such that after the agent takes an action, with", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 572, + 506, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 153, + 585 + ], + "score": 1.0, + "content": "probability", + "type": "text" + }, + { + "bbox": [ + 153, + 574, + 159, + 583 + ], + "score": 0.74, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 160, + 572, + 404, + 585 + ], + "score": 1.0, + "content": ", the state transition is uniformly over S, and with probability", + "type": "text" + }, + { + "bbox": [ + 405, + 573, + 428, + 583 + ], + "score": 0.89, + "content": "1 - p", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 572, + 506, + 585 + ], + "score": 1.0, + "content": "the state transition", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "is according to the true unperturbed transition kernel. The behavior policy for all the experiments", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 592, + 506, + 608 + ], + "spans": [ + { + "bbox": [ + 104, + 592, + 432, + 608 + ], + "score": 1.0, + "content": "below is set to be a uniform distribution over the action space given any state, i.e.,", + "type": "text" + }, + { + "bbox": [ + 432, + 593, + 490, + 608 + ], + "score": 0.94, + "content": "\\begin{array} { r } { \\pi _ { b } ( a | s ) ^ { \\bullet } = \\frac { 1 } { | \\mathcal { A } | } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 490, + 594, + 506, + 606 + ], + "score": 1.0, + "content": "for", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 606, + 506, + 619 + ], + "spans": [ + { + "bbox": [ + 105, + 606, + 123, + 619 + ], + "score": 1.0, + "content": "any", + "type": "text" + }, + { + "bbox": [ + 124, + 607, + 147, + 618 + ], + "score": 0.89, + "content": "s \\in \\mathcal { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 606, + 165, + 619 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 165, + 607, + 192, + 617 + ], + "score": 0.89, + "content": "a \\in { \\mathcal { A } }", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 606, + 506, + 619 + ], + "score": 1.0, + "content": ". We then evaluate the performance of the obtained policy in the unperturbed", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 617, + 506, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 506, + 630 + ], + "score": 1.0, + "content": "environment. At each time t, the policy we evaluate is the greedy-policy w.r.t. the current estimate of", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 629, + 506, + 642 + ], + "spans": [ + { + "bbox": [ + 106, + 629, + 186, + 642 + ], + "score": 1.0, + "content": "the Q-function, i.e.,", + "type": "text" + }, + { + "bbox": [ + 186, + 629, + 296, + 641 + ], + "score": 0.92, + "content": "\\pi _ { t } ( s ) = \\arg \\operatorname* { m a x } _ { a } Q _ { t } ( s , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 629, + 506, + 642 + ], + "score": 1.0, + "content": ". A Monte-Carlo method with horizon 100 is used to", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 639, + 506, + 653 + ], + "spans": [ + { + "bbox": [ + 106, + 639, + 506, + 653 + ], + "score": 1.0, + "content": "evaluate the accumulated discounted reward of the learned policy on the unperturbed MDP. We take", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 650, + 408, + 664 + ], + "spans": [ + { + "bbox": [ + 106, + 650, + 408, + 664 + ], + "score": 1.0, + "content": "the average over 30 trajectories. More details are provided in the appendix.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 107, + 667, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 667, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 505, + 680 + ], + "score": 1.0, + "content": "In Figure 1 and Figure 2, we plot the accumulated discounted reward of both algorithms under", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 677, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 144, + 691 + ], + "score": 1.0, + "content": "different", + "type": "text" + }, + { + "bbox": [ + 144, + 680, + 151, + 689 + ], + "score": 0.81, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 677, + 169, + 691 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 169, + 678, + 178, + 688 + ], + "score": 0.81, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 677, + 505, + 691 + ], + "score": 1.0, + "content": "for both problems. The upper and lower envelopes of the curves correspond to", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 687, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 506, + 702 + ], + "score": 1.0, + "content": "the 95 and 5 percentiles of the 30 trajectories, respectively. It can be seen that overall our robust", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "score": 1.0, + "content": "Q-learning algorithm achieves a much higher reward than the vanilla Q-learning. This demonstrates", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 710, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 441, + 724 + ], + "score": 1.0, + "content": "the robustness of our robust Q-learning algorithm to model uncertainty. Moreover, as", + "type": "text" + }, + { + "bbox": [ + 441, + 713, + 448, + 722 + ], + "score": 0.79, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 710, + 465, + 724 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 465, + 711, + 474, + 721 + ], + "score": 0.78, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 710, + 505, + 724 + ], + "score": 1.0, + "content": "getting", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 39 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 742, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 9 + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 106, + 72, + 363, + 84 + ], + "lines": [ + { + "bbox": [ + 105, + 70, + 364, + 87 + ], + "spans": [ + { + "bbox": [ + 105, + 70, + 364, + 87 + ], + "score": 1.0, + "content": "Algorithm 2 Robust TDC with Linear Function Approximation", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 106, + 86, + 490, + 214 + ], + "lines": [ + { + "bbox": [ + 106, + 84, + 338, + 98 + ], + "spans": [ + { + "bbox": [ + 106, + 84, + 137, + 98 + ], + "score": 1.0, + "content": "Input:", + "type": "text" + }, + { + "bbox": [ + 137, + 86, + 189, + 97 + ], + "score": 0.27, + "content": "T , \\alpha , \\beta , \\varrho , \\phi _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 84, + 204, + 98 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 205, + 86, + 253, + 97 + ], + "score": 0.89, + "content": "i = 1 , . . . , N", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 84, + 327, + 98 + ], + "score": 1.0, + "content": ", projection radius", + "type": "text" + }, + { + "bbox": [ + 327, + 86, + 338, + 96 + ], + "score": 0.7, + "content": "K", + "type": "inline_equation" + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 106, + 96, + 207, + 108 + ], + "spans": [ + { + "bbox": [ + 106, + 96, + 207, + 108 + ], + "score": 1.0, + "content": "Initialization: θ0,w0, s0", + "type": "text" + } + ], + "index": 1, + "is_list_start_line": true + }, + { + "bbox": [ + 109, + 107, + 280, + 122 + ], + "spans": [ + { + "bbox": [ + 109, + 107, + 155, + 122 + ], + "score": 1.0, + "content": "1: Choose", + "type": "text" + }, + { + "bbox": [ + 156, + 109, + 280, + 121 + ], + "score": 0.82, + "content": "W \\sim \\mathrm { U n i f o r m } ( 0 , 1 , . . . , T - 1 )", + "type": "inline_equation" + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 119, + 241, + 132 + ], + "spans": [ + { + "bbox": [ + 110, + 119, + 138, + 132 + ], + "score": 1.0, + "content": "2: for", + "type": "text" + }, + { + "bbox": [ + 138, + 121, + 225, + 132 + ], + "score": 0.88, + "content": "t = 0 , 1 , 2 , . . . , W - 1", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 119, + 241, + 132 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 108, + 129, + 361, + 145 + ], + "spans": [ + { + "bbox": [ + 108, + 129, + 234, + 145 + ], + "score": 1.0, + "content": "3: Take action according to", + "type": "text" + }, + { + "bbox": [ + 234, + 131, + 262, + 144 + ], + "score": 0.92, + "content": "\\pi ( \\cdot | s _ { t } )", + "type": "inline_equation" + }, + { + "bbox": [ + 263, + 129, + 314, + 145 + ], + "score": 1.0, + "content": "and observe", + "type": "text" + }, + { + "bbox": [ + 314, + 133, + 333, + 143 + ], + "score": 0.9, + "content": "s _ { t + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 129, + 351, + 145 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 352, + 133, + 361, + 142 + ], + "score": 0.77, + "content": "c _ { t }", + "type": "inline_equation" + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 141, + 173, + 157 + ], + "spans": [ + { + "bbox": [ + 110, + 141, + 133, + 157 + ], + "score": 1.0, + "content": "4:", + "type": "text" + }, + { + "bbox": [ + 133, + 142, + 173, + 155 + ], + "score": 0.74, + "content": "\\phi _ { t } \\phi _ { s _ { t } }", + "type": "inline_equation" + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 108, + 148, + 488, + 214 + ], + "spans": [ + { + "bbox": [ + 108, + 198, + 131, + 214 + ], + "score": 1.0, + "content": "7:", + "type": "text" + }, + { + "bbox": [ + 110, + 148, + 131, + 202 + ], + "score": 1.0, + "content": "5: 6:", + "type": "text" + }, + { + "bbox": [ + 132, + 152, + 488, + 214 + ], + "score": 0.91, + "content": "\\begin{array} { r l } & { \\ddot { \\delta _ { t } } ( \\theta _ { t } ) \\gets c _ { t } + \\gamma ( 1 - R ) V _ { \\theta _ { t } } ( s _ { t + 1 } ) + \\gamma R \\frac { \\log ( \\sum _ { s } e ^ { \\varrho \\theta ^ { \\top } \\phi _ { s } } ) } { \\varrho } - V _ { \\theta _ { t } } ( s _ { t } ) } \\\\ & { \\theta _ { t + 1 } \\gets \\Pi _ { K } \\left( \\theta _ { t } + \\alpha \\left( \\delta _ { t } ( \\theta _ { t } ) \\phi _ { t } - \\gamma \\bigg ( ( 1 - R ) \\phi _ { t + 1 } + R \\sum _ { s \\in \\mathcal { S } } \\left( \\frac { e ^ { \\sigma V _ { \\theta } ( s ) } \\phi _ { s } } { \\sum _ { j \\in \\mathcal { S } } e ^ { \\sigma V _ { \\theta } ( j ) } } \\right) \\bigg ) \\phi _ { t } ^ { \\top } \\omega _ { t } \\right) \\right) } \\\\ & { \\omega _ { t + 1 } \\gets \\Pi _ { K } ( \\omega _ { t } + \\beta ( \\delta _ { t } ( \\theta _ { t } ) - \\phi _ { t } ^ { \\top } \\omega _ { t } ) \\phi _ { t } ) } \\end{array}", + "type": "inline_equation", + "image_path": "8504c8f66aa9dae624c071a98e50fb92a72dc33f61233228fc326889e3085dd1.jpg" + } + ], + "index": 6, + "is_list_start_line": true + } + ], + "index": 3, + "bbox_fs": [ + 106, + 84, + 488, + 214 + ] + }, + { + "type": "text", + "bbox": [ + 111, + 212, + 156, + 222 + ], + "lines": [ + { + "bbox": [ + 110, + 210, + 157, + 223 + ], + "spans": [ + { + "bbox": [ + 110, + 210, + 157, + 223 + ], + "score": 1.0, + "content": "8: end for", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7, + "bbox_fs": [ + 110, + 210, + 157, + 223 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 224, + 158, + 235 + ], + "lines": [ + { + "bbox": [ + 105, + 221, + 158, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 221, + 143, + 239 + ], + "score": 1.0, + "content": "Output:", + "type": "text" + }, + { + "bbox": [ + 143, + 223, + 158, + 235 + ], + "score": 0.73, + "content": "\\theta _ { W }", + "type": "inline_equation" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 105, + 221, + 158, + 239 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 258, + 503, + 282 + ], + "lines": [ + { + "bbox": [ + 105, + 256, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 256, + 415, + 272 + ], + "score": 1.0, + "content": "Assumption 5 (Geometric uniform ergodicity). There exist some constants", + "type": "text" + }, + { + "bbox": [ + 416, + 259, + 444, + 269 + ], + "score": 0.89, + "content": "m > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 256, + 464, + 272 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 464, + 258, + 505, + 270 + ], + "score": 0.92, + "content": "\\rho \\in ( 0 , 1 )", + "type": "inline_equation" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 268, + 364, + 282 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 176, + 282 + ], + "score": 1.0, + "content": "such that for any", + "type": "text" + }, + { + "bbox": [ + 176, + 270, + 199, + 280 + ], + "score": 0.83, + "content": "t > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 268, + 201, + 282 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 202, + 269, + 359, + 281 + ], + "score": 0.83, + "content": "\\begin{array} { r } { \\operatorname* { m a x } _ { s \\in \\mathcal { S } } d _ { T V } ( \\bar { \\mathbb { P } } ( s _ { t } | s _ { 0 } = s ) , \\mu _ { \\pi } ) \\leq m \\rho ^ { t } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 268, + 364, + 282 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 256, + 505, + 282 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 289, + 505, + 322 + ], + "lines": [ + { + "bbox": [ + 105, + 289, + 505, + 301 + ], + "spans": [ + { + "bbox": [ + 105, + 289, + 505, + 301 + ], + "score": 1.0, + "content": "In the following theorem, we characterize the finite-time error bound for the convergence of our", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 300, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 440, + 312 + ], + "score": 1.0, + "content": "robust TDC algorithm. Here we only provide the order of the bounds in terms of", + "type": "text" + }, + { + "bbox": [ + 440, + 301, + 448, + 310 + ], + "score": 0.81, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 300, + 505, + 312 + ], + "score": 1.0, + "content": ". The explicit", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 310, + 298, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 310, + 298, + 324 + ], + "score": 1.0, + "content": "bounds can be found in (129) in Appendix D.3.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12, + "bbox_fs": [ + 105, + 289, + 505, + 324 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 325, + 505, + 348 + ], + "lines": [ + { + "bbox": [ + 105, + 324, + 505, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 302, + 341 + ], + "score": 1.0, + "content": "Theorem 5. Consider the following step-sizes:", + "type": "text" + }, + { + "bbox": [ + 302, + 325, + 353, + 339 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\beta = \\mathcal { O } \\left( \\frac { 1 } { T ^ { b } } \\right) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 324, + 375, + 341 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 376, + 325, + 427, + 339 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\alpha = \\mathcal { O } \\left( \\frac { 1 } { T ^ { a } } \\right) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 427, + 324, + 459, + 341 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 460, + 325, + 505, + 339 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\frac { 1 } { 2 } < a \\leq 1 } \\end{array}", + "type": "inline_equation" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 336, + 246, + 349 + ], + "spans": [ + { + "bbox": [ + 106, + 336, + 124, + 349 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 337, + 166, + 348 + ], + "score": 0.91, + "content": "0 < b \\leq a", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 336, + 246, + 349 + ], + "score": 1.0, + "content": ". Then we have that", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 324, + 505, + 349 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 172, + 353, + 437, + 380 + ], + "lines": [ + { + "bbox": [ + 172, + 353, + 437, + 380 + ], + "spans": [ + { + "bbox": [ + 172, + 353, + 437, + 380 + ], + "score": 0.92, + "content": "\\mathbb { E } [ \\| \\nabla J ( \\theta _ { W } ) \\| ^ { 2 } ] = \\odot \\left( \\frac { 1 } { T \\alpha } + \\alpha \\log ( 1 / \\alpha ) + \\frac { 1 } { T \\beta } + \\beta \\log ( 1 / \\beta ) \\right) ,", + "type": "interline_equation", + "image_path": "a654cd452a7e1f7e9a634b553d55226b856e45fcf1332c2009f8149f8fab6ff9.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 172, + 353, + 437, + 380 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 385, + 367, + 406 + ], + "lines": [ + { + "bbox": [ + 101, + 381, + 365, + 409 + ], + "spans": [ + { + "bbox": [ + 101, + 381, + 171, + 409 + ], + "score": 1.0, + "content": "If we further let", + "type": "text" + }, + { + "bbox": [ + 172, + 389, + 221, + 401 + ], + "score": 0.82, + "content": "a = b = 0 . 5", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 381, + 245, + 409 + ], + "score": 1.0, + "content": ", then", + "type": "text" + }, + { + "bbox": [ + 245, + 385, + 365, + 406 + ], + "score": 0.8, + "content": "\\begin{array} { r } { \\mathbb { E } [ \\| \\nabla J ( \\theta _ { W } ) \\| ^ { 2 } ] = \\odot \\left( \\frac { \\log T } { \\sqrt { T } } \\right) . } \\end{array}", + "type": "inline_equation" + } + ], + "index": 17 + } + ], + "index": 17, + "bbox_fs": [ + 101, + 381, + 365, + 409 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 413, + 506, + 457 + ], + "lines": [ + { + "bbox": [ + 105, + 412, + 507, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 507, + 426 + ], + "score": 1.0, + "content": "The robust TDC has a matching complexity with the vanilla TDC with non-linear function approxi-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 424, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 106, + 424, + 505, + 435 + ], + "score": 1.0, + "content": "mation [Wang et al., 2021], but provides the additional robustness to model uncertainty. It does not", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 435, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 106, + 435, + 506, + 447 + ], + "score": 1.0, + "content": "need to relax the uncertainty set like in [Roy et al., 2017], and our convergence results do not need a", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 446, + 237, + 458 + ], + "spans": [ + { + "bbox": [ + 106, + 446, + 237, + 458 + ], + "score": 1.0, + "content": "condition on the discount factor.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19.5, + "bbox_fs": [ + 105, + 412, + 507, + 458 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 473, + 191, + 486 + ], + "lines": [ + { + "bbox": [ + 104, + 471, + 193, + 489 + ], + "spans": [ + { + "bbox": [ + 104, + 471, + 193, + 489 + ], + "score": 1.0, + "content": "6 Experiments", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "title", + "bbox": [ + 107, + 497, + 214, + 509 + ], + "lines": [ + { + "bbox": [ + 105, + 495, + 215, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 215, + 512 + ], + "score": 1.0, + "content": "6.1 Robust Q-Learning", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 106, + 517, + 506, + 662 + ], + "lines": [ + { + "bbox": [ + 105, + 516, + 506, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 506, + 531 + ], + "score": 1.0, + "content": "In this section, we compare our robust Q-learning with the vanilla non-robust Q-learning. We use", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 529, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 106, + 529, + 505, + 540 + ], + "score": 1.0, + "content": "OpenAI gym framework [Brockman et al., 2016], and consider two different problems: Frozen lake", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 539, + 506, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 506, + 552 + ], + "score": 1.0, + "content": "and Cart-Pole. One more example of the taxi problem is given in the appendix. To demonstrate the", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 550, + 506, + 562 + ], + "spans": [ + { + "bbox": [ + 106, + 550, + 506, + 562 + ], + "score": 1.0, + "content": "robustness, the policy is learned in a perturbed MDP, and is then tested on the true unperturbed MDP.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 561, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 316, + 574 + ], + "score": 1.0, + "content": "Specifically, during the training, we set a probability", + "type": "text" + }, + { + "bbox": [ + 317, + 563, + 323, + 573 + ], + "score": 0.8, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 561, + 505, + 574 + ], + "score": 1.0, + "content": "such that after the agent takes an action, with", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 572, + 506, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 153, + 585 + ], + "score": 1.0, + "content": "probability", + "type": "text" + }, + { + "bbox": [ + 153, + 574, + 159, + 583 + ], + "score": 0.74, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 160, + 572, + 404, + 585 + ], + "score": 1.0, + "content": ", the state transition is uniformly over S, and with probability", + "type": "text" + }, + { + "bbox": [ + 405, + 573, + 428, + 583 + ], + "score": 0.89, + "content": "1 - p", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 572, + 506, + 585 + ], + "score": 1.0, + "content": "the state transition", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "is according to the true unperturbed transition kernel. The behavior policy for all the experiments", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 592, + 506, + 608 + ], + "spans": [ + { + "bbox": [ + 104, + 592, + 432, + 608 + ], + "score": 1.0, + "content": "below is set to be a uniform distribution over the action space given any state, i.e.,", + "type": "text" + }, + { + "bbox": [ + 432, + 593, + 490, + 608 + ], + "score": 0.94, + "content": "\\begin{array} { r } { \\pi _ { b } ( a | s ) ^ { \\bullet } = \\frac { 1 } { | \\mathcal { A } | } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 490, + 594, + 506, + 606 + ], + "score": 1.0, + "content": "for", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 606, + 506, + 619 + ], + "spans": [ + { + "bbox": [ + 105, + 606, + 123, + 619 + ], + "score": 1.0, + "content": "any", + "type": "text" + }, + { + "bbox": [ + 124, + 607, + 147, + 618 + ], + "score": 0.89, + "content": "s \\in \\mathcal { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 606, + 165, + 619 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 165, + 607, + 192, + 617 + ], + "score": 0.89, + "content": "a \\in { \\mathcal { A } }", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 606, + 506, + 619 + ], + "score": 1.0, + "content": ". We then evaluate the performance of the obtained policy in the unperturbed", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 617, + 506, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 506, + 630 + ], + "score": 1.0, + "content": "environment. At each time t, the policy we evaluate is the greedy-policy w.r.t. the current estimate of", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 629, + 506, + 642 + ], + "spans": [ + { + "bbox": [ + 106, + 629, + 186, + 642 + ], + "score": 1.0, + "content": "the Q-function, i.e.,", + "type": "text" + }, + { + "bbox": [ + 186, + 629, + 296, + 641 + ], + "score": 0.92, + "content": "\\pi _ { t } ( s ) = \\arg \\operatorname* { m a x } _ { a } Q _ { t } ( s , a )", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 629, + 506, + 642 + ], + "score": 1.0, + "content": ". A Monte-Carlo method with horizon 100 is used to", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 639, + 506, + 653 + ], + "spans": [ + { + "bbox": [ + 106, + 639, + 506, + 653 + ], + "score": 1.0, + "content": "evaluate the accumulated discounted reward of the learned policy on the unperturbed MDP. We take", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 650, + 408, + 664 + ], + "spans": [ + { + "bbox": [ + 106, + 650, + 408, + 664 + ], + "score": 1.0, + "content": "the average over 30 trajectories. More details are provided in the appendix.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 30, + "bbox_fs": [ + 104, + 516, + 506, + 664 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 667, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 667, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 505, + 680 + ], + "score": 1.0, + "content": "In Figure 1 and Figure 2, we plot the accumulated discounted reward of both algorithms under", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 677, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 144, + 691 + ], + "score": 1.0, + "content": "different", + "type": "text" + }, + { + "bbox": [ + 144, + 680, + 151, + 689 + ], + "score": 0.81, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 677, + 169, + 691 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 169, + 678, + 178, + 688 + ], + "score": 0.81, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 677, + 505, + 691 + ], + "score": 1.0, + "content": "for both problems. The upper and lower envelopes of the curves correspond to", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 687, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 506, + 702 + ], + "score": 1.0, + "content": "the 95 and 5 percentiles of the 30 trajectories, respectively. It can be seen that overall our robust", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "score": 1.0, + "content": "Q-learning algorithm achieves a much higher reward than the vanilla Q-learning. This demonstrates", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 710, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 441, + 724 + ], + "score": 1.0, + "content": "the robustness of our robust Q-learning algorithm to model uncertainty. Moreover, as", + "type": "text" + }, + { + "bbox": [ + 441, + 713, + 448, + 722 + ], + "score": 0.79, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 710, + 465, + 724 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 465, + 711, + 474, + 721 + ], + "score": 0.78, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 710, + 505, + 724 + ], + "score": 1.0, + "content": "getting", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 370, + 505, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 505, + 384 + ], + "score": 1.0, + "content": "larger, i.e., as the MDP that we learn the policy deviates from the MDP we test the policy, the", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 381, + 505, + 395 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 505, + 395 + ], + "score": 1.0, + "content": "advantage of our robust Q-learning algorithm is getting more significant compared to the vanilla", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 393, + 196, + 406 + ], + "spans": [ + { + "bbox": [ + 106, + 393, + 196, + 406 + ], + "score": 1.0, + "content": "Q-learning algorithm.", + "type": "text", + "cross_page": true + } + ], + "index": 10 + } + ], + "index": 39, + "bbox_fs": [ + 105, + 667, + 506, + 724 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 118, + 85, + 492, + 183 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 118, + 85, + 492, + 183 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 118, + 85, + 492, + 183 + ], + "spans": [ + { + "bbox": [ + 118, + 85, + 492, + 183 + ], + "score": 0.968, + "type": "image", + "image_path": "f6e7e9e2735f3d2f3cb6a78235a2c13acfd85270e0532cc9f312127eab2de53d.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 118, + 85, + 492, + 117.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 118, + 117.66666666666666, + 492, + 150.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 118, + 150.33333333333331, + 492, + 182.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 158, + 192, + 451, + 204 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 158, + 190, + 452, + 207 + ], + "spans": [ + { + "bbox": [ + 158, + 190, + 452, + 207 + ], + "score": 1.0, + "content": "Figure 1: FrozenLake-v0: robust Q-learning v.s. non-robust Q-learning.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "image", + "bbox": [ + 116, + 229, + 491, + 328 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 116, + 229, + 491, + 328 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 116, + 229, + 491, + 328 + ], + "spans": [ + { + "bbox": [ + 116, + 229, + 491, + 328 + ], + "score": 0.969, + "type": "image", + "image_path": "a43d2928aff120beef997e93a1a33000db2a33693b2a692ee8c4a435bdabda59.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 116, + 229, + 491, + 262.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 116, + 262.0, + 491, + 295.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 116, + 295.0, + 491, + 328.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 165, + 338, + 444, + 349 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 164, + 334, + 446, + 353 + ], + "spans": [ + { + "bbox": [ + 164, + 334, + 446, + 353 + ], + "score": 1.0, + "content": "Figure 2: CartPole-v0: robust Q-learning v.s. non-robust Q-learning.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + } + ], + "index": 6.0 + }, + { + "type": "text", + "bbox": [ + 107, + 370, + 505, + 404 + ], + "lines": [ + { + "bbox": [ + 105, + 370, + 505, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 505, + 384 + ], + "score": 1.0, + "content": "larger, i.e., as the MDP that we learn the policy deviates from the MDP we test the policy, the", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 381, + 505, + 395 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 505, + 395 + ], + "score": 1.0, + "content": "advantage of our robust Q-learning algorithm is getting more significant compared to the vanilla", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 393, + 196, + 406 + ], + "spans": [ + { + "bbox": [ + 106, + 393, + 196, + 406 + ], + "score": 1.0, + "content": "Q-learning algorithm.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + }, + { + "type": "title", + "bbox": [ + 106, + 417, + 345, + 429 + ], + "lines": [ + { + "bbox": [ + 105, + 415, + 346, + 432 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 346, + 432 + ], + "score": 1.0, + "content": "6.2 Robust TDC with Linear Function Approximation", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 108, + 437, + 504, + 470 + ], + "lines": [ + { + "bbox": [ + 105, + 437, + 505, + 449 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 505, + 449 + ], + "score": 1.0, + "content": "In this section we compare our robust TDC with the vanilla non-robust TDC with linear function", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 448, + 506, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 196, + 460 + ], + "score": 1.0, + "content": "approximation on the", + "type": "text" + }, + { + "bbox": [ + 197, + 448, + 221, + 459 + ], + "score": 0.89, + "content": "4 \\times 4", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 448, + 506, + 460 + ], + "score": 1.0, + "content": "Frozen Lake problem. The problem setting is the same as the one in", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 459, + 438, + 471 + ], + "spans": [ + { + "bbox": [ + 106, + 459, + 438, + 471 + ], + "score": 1.0, + "content": "Section 6.1. More details about the experiment setup are provided in the appendix.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 106, + 475, + 505, + 574 + ], + "lines": [ + { + "bbox": [ + 105, + 475, + 506, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 506, + 488 + ], + "score": 1.0, + "content": "We implement the two algorithms using samples from the perturbed MDP both for 30 times, and", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 104, + 484, + 507, + 501 + ], + "spans": [ + { + "bbox": [ + 104, + 484, + 206, + 501 + ], + "score": 1.0, + "content": "obtain 30 sequences of", + "type": "text" + }, + { + "bbox": [ + 207, + 486, + 239, + 498 + ], + "score": 0.88, + "content": "\\{ \\theta _ { t } ^ { i } \\} _ { t = 1 } ^ { \\infty }", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 484, + 244, + 501 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 245, + 487, + 308, + 498 + ], + "score": 0.83, + "content": "i = 1 , 2 , . . . , 3 0", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 484, + 507, + 501 + ], + "score": 1.0, + "content": ". We then compute the squared gradient norm", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 107, + 495, + 507, + 513 + ], + "spans": [ + { + "bbox": [ + 107, + 497, + 150, + 509 + ], + "score": 0.92, + "content": "\\| \\nabla J ( \\theta ) \\| ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 495, + 348, + 513 + ], + "score": 1.0, + "content": "on the true unperturbed MDP, and see whether", + "type": "text" + }, + { + "bbox": [ + 349, + 497, + 381, + 509 + ], + "score": 0.93, + "content": "\\{ \\theta _ { t } ^ { i } \\} _ { t = 1 } ^ { \\infty }", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 495, + 507, + 513 + ], + "score": 1.0, + "content": "converges to some stationary", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 104, + 507, + 504, + 521 + ], + "spans": [ + { + "bbox": [ + 104, + 507, + 461, + 521 + ], + "score": 1.0, + "content": "points on the true unperturbed MDP. In Fig. 3, we plot the average squared gradient norm", + "type": "text" + }, + { + "bbox": [ + 461, + 507, + 504, + 520 + ], + "score": 0.92, + "content": "\\| \\nabla J ( \\theta ) \\| ^ { 2 }", + "type": "inline_equation" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 519, + 506, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 156, + 532 + ], + "score": 1.0, + "content": "for different", + "type": "text" + }, + { + "bbox": [ + 157, + 521, + 164, + 531 + ], + "score": 0.8, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 519, + 181, + 532 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 181, + 520, + 190, + 529 + ], + "score": 0.72, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 519, + 506, + 532 + ], + "score": 1.0, + "content": ". The upper and lower envelops are the 95 and 5 percentiles of the 30 curves. It", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 529, + 506, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 506, + 542 + ], + "score": 1.0, + "content": "can be seen that our robust TDC converges much faster than vanilla TDC, and as the model mismatch", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 541, + 505, + 553 + ], + "spans": [ + { + "bbox": [ + 106, + 541, + 505, + 553 + ], + "score": 1.0, + "content": "between the training and test MDPs enlarges, the vanilla TDC may diverge (Fig. 3(c)), while our", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 551, + 506, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 506, + 565 + ], + "score": 1.0, + "content": "robust TDC still converges to some stationary point. Also, the robust TDC has a much smaller", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 561, + 406, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 406, + 576 + ], + "score": 1.0, + "content": "variance, which indicates a much stable behavior under model uncertainty.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 19 + }, + { + "type": "image", + "bbox": [ + 117, + 599, + 491, + 698 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 117, + 599, + 491, + 698 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 117, + 599, + 491, + 698 + ], + "spans": [ + { + "bbox": [ + 117, + 599, + 491, + 698 + ], + "score": 0.971, + "type": "image", + "image_path": "88a93ae259a06e7bc8fa9f4ad18dbea7ab033fd1a750f80210fcc36ef5a44b84.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 117, + 599, + 491, + 632.0 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 117, + 632.0, + 491, + 665.0 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 117, + 665.0, + 491, + 698.0 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 155, + 707, + 456, + 720 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 154, + 706, + 456, + 721 + ], + "spans": [ + { + "bbox": [ + 154, + 706, + 265, + 721 + ], + "score": 1.0, + "content": "Figure 3: FrozenLake-v0:", + "type": "text" + }, + { + "bbox": [ + 265, + 707, + 308, + 720 + ], + "score": 0.92, + "content": "\\| \\nabla J ( \\theta ) \\| ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 706, + 456, + 721 + ], + "score": 1.0, + "content": "of robust TDC and non-robust TDC.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + } + ], + "index": 26.0 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 118, + 85, + 492, + 183 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 118, + 85, + 492, + 183 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 118, + 85, + 492, + 183 + ], + "spans": [ + { + "bbox": [ + 118, + 85, + 492, + 183 + ], + "score": 0.968, + "type": "image", + "image_path": "f6e7e9e2735f3d2f3cb6a78235a2c13acfd85270e0532cc9f312127eab2de53d.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 118, + 85, + 492, + 117.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 118, + 117.66666666666666, + 492, + 150.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 118, + 150.33333333333331, + 492, + 182.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 158, + 192, + 451, + 204 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 158, + 190, + 452, + 207 + ], + "spans": [ + { + "bbox": [ + 158, + 190, + 452, + 207 + ], + "score": 1.0, + "content": "Figure 1: FrozenLake-v0: robust Q-learning v.s. non-robust Q-learning.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "image", + "bbox": [ + 116, + 229, + 491, + 328 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 116, + 229, + 491, + 328 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 116, + 229, + 491, + 328 + ], + "spans": [ + { + "bbox": [ + 116, + 229, + 491, + 328 + ], + "score": 0.969, + "type": "image", + "image_path": "a43d2928aff120beef997e93a1a33000db2a33693b2a692ee8c4a435bdabda59.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 116, + 229, + 491, + 262.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 116, + 262.0, + 491, + 295.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 116, + 295.0, + 491, + 328.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 165, + 338, + 444, + 349 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 164, + 334, + 446, + 353 + ], + "spans": [ + { + "bbox": [ + 164, + 334, + 446, + 353 + ], + "score": 1.0, + "content": "Figure 2: CartPole-v0: robust Q-learning v.s. non-robust Q-learning.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + } + ], + "index": 6.0 + }, + { + "type": "text", + "bbox": [ + 107, + 370, + 505, + 404 + ], + "lines": [], + "index": 9, + "bbox_fs": [ + 105, + 370, + 505, + 406 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 106, + 417, + 345, + 429 + ], + "lines": [ + { + "bbox": [ + 105, + 415, + 346, + 432 + ], + "spans": [ + { + "bbox": [ + 105, + 415, + 346, + 432 + ], + "score": 1.0, + "content": "6.2 Robust TDC with Linear Function Approximation", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 108, + 437, + 504, + 470 + ], + "lines": [ + { + "bbox": [ + 105, + 437, + 505, + 449 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 505, + 449 + ], + "score": 1.0, + "content": "In this section we compare our robust TDC with the vanilla non-robust TDC with linear function", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 448, + 506, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 196, + 460 + ], + "score": 1.0, + "content": "approximation on the", + "type": "text" + }, + { + "bbox": [ + 197, + 448, + 221, + 459 + ], + "score": 0.89, + "content": "4 \\times 4", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 448, + 506, + 460 + ], + "score": 1.0, + "content": "Frozen Lake problem. The problem setting is the same as the one in", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 459, + 438, + 471 + ], + "spans": [ + { + "bbox": [ + 106, + 459, + 438, + 471 + ], + "score": 1.0, + "content": "Section 6.1. More details about the experiment setup are provided in the appendix.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13, + "bbox_fs": [ + 105, + 437, + 506, + 471 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 475, + 505, + 574 + ], + "lines": [ + { + "bbox": [ + 105, + 475, + 506, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 506, + 488 + ], + "score": 1.0, + "content": "We implement the two algorithms using samples from the perturbed MDP both for 30 times, and", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 104, + 484, + 507, + 501 + ], + "spans": [ + { + "bbox": [ + 104, + 484, + 206, + 501 + ], + "score": 1.0, + "content": "obtain 30 sequences of", + "type": "text" + }, + { + "bbox": [ + 207, + 486, + 239, + 498 + ], + "score": 0.88, + "content": "\\{ \\theta _ { t } ^ { i } \\} _ { t = 1 } ^ { \\infty }", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 484, + 244, + 501 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 245, + 487, + 308, + 498 + ], + "score": 0.83, + "content": "i = 1 , 2 , . . . , 3 0", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 484, + 507, + 501 + ], + "score": 1.0, + "content": ". We then compute the squared gradient norm", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 107, + 495, + 507, + 513 + ], + "spans": [ + { + "bbox": [ + 107, + 497, + 150, + 509 + ], + "score": 0.92, + "content": "\\| \\nabla J ( \\theta ) \\| ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 495, + 348, + 513 + ], + "score": 1.0, + "content": "on the true unperturbed MDP, and see whether", + "type": "text" + }, + { + "bbox": [ + 349, + 497, + 381, + 509 + ], + "score": 0.93, + "content": "\\{ \\theta _ { t } ^ { i } \\} _ { t = 1 } ^ { \\infty }", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 495, + 507, + 513 + ], + "score": 1.0, + "content": "converges to some stationary", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 104, + 507, + 504, + 521 + ], + "spans": [ + { + "bbox": [ + 104, + 507, + 461, + 521 + ], + "score": 1.0, + "content": "points on the true unperturbed MDP. In Fig. 3, we plot the average squared gradient norm", + "type": "text" + }, + { + "bbox": [ + 461, + 507, + 504, + 520 + ], + "score": 0.92, + "content": "\\| \\nabla J ( \\theta ) \\| ^ { 2 }", + "type": "inline_equation" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 519, + 506, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 156, + 532 + ], + "score": 1.0, + "content": "for different", + "type": "text" + }, + { + "bbox": [ + 157, + 521, + 164, + 531 + ], + "score": 0.8, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 519, + 181, + 532 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 181, + 520, + 190, + 529 + ], + "score": 0.72, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 519, + 506, + 532 + ], + "score": 1.0, + "content": ". The upper and lower envelops are the 95 and 5 percentiles of the 30 curves. It", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 529, + 506, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 529, + 506, + 542 + ], + "score": 1.0, + "content": "can be seen that our robust TDC converges much faster than vanilla TDC, and as the model mismatch", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 541, + 505, + 553 + ], + "spans": [ + { + "bbox": [ + 106, + 541, + 505, + 553 + ], + "score": 1.0, + "content": "between the training and test MDPs enlarges, the vanilla TDC may diverge (Fig. 3(c)), while our", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 551, + 506, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 506, + 565 + ], + "score": 1.0, + "content": "robust TDC still converges to some stationary point. Also, the robust TDC has a much smaller", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 561, + 406, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 406, + 576 + ], + "score": 1.0, + "content": "variance, which indicates a much stable behavior under model uncertainty.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 19, + "bbox_fs": [ + 104, + 475, + 507, + 576 + ] + }, + { + "type": "image", + "bbox": [ + 117, + 599, + 491, + 698 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 117, + 599, + 491, + 698 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 117, + 599, + 491, + 698 + ], + "spans": [ + { + "bbox": [ + 117, + 599, + 491, + 698 + ], + "score": 0.971, + "type": "image", + "image_path": "88a93ae259a06e7bc8fa9f4ad18dbea7ab033fd1a750f80210fcc36ef5a44b84.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 117, + 599, + 491, + 632.0 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 117, + 632.0, + 491, + 665.0 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 117, + 665.0, + 491, + 698.0 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 155, + 707, + 456, + 720 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 154, + 706, + 456, + 721 + ], + "spans": [ + { + "bbox": [ + 154, + 706, + 265, + 721 + ], + "score": 1.0, + "content": "Figure 3: FrozenLake-v0:", + "type": "text" + }, + { + "bbox": [ + 265, + 707, + 308, + 720 + ], + "score": 0.92, + "content": "\\| \\nabla J ( \\theta ) \\| ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 706, + 456, + 721 + ], + "score": 1.0, + "content": "of robust TDC and non-robust TDC.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + } + ], + "index": 26.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 106, + 72, + 360, + 84 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 362, + 87 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 362, + 87 + ], + "score": 1.0, + "content": "6.3 Comparison with The Adversarial Training Approach", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 92, + 505, + 182 + ], + "lines": [ + { + "bbox": [ + 105, + 91, + 506, + 105 + ], + "spans": [ + { + "bbox": [ + 105, + 91, + 506, + 105 + ], + "score": 1.0, + "content": "We also compare our robust Q-learning with Robust Adversarial Reinforcement Learning (RARL)", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 103, + 506, + 116 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 506, + 116 + ], + "score": 1.0, + "content": "in [Pinto et al., 2017]. To apply their algorithm to our problem setting, we model the nature as an", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "score": 1.0, + "content": "adversarial player, and its goal is to minimize the reward that the agent receives. The action space", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 126, + 505, + 140 + ], + "spans": [ + { + "bbox": [ + 106, + 127, + 124, + 138 + ], + "score": 0.9, + "content": "\\mathcal { A } _ { a d }", + "type": "inline_equation" + }, + { + "bbox": [ + 125, + 126, + 283, + 140 + ], + "score": 1.0, + "content": "of the nature is set to be the state space", + "type": "text" + }, + { + "bbox": [ + 283, + 126, + 320, + 138 + ], + "score": 0.94, + "content": "\\mathcal { A } _ { a d } \\triangleq \\mathcal { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 126, + 505, + 140 + ], + "score": 1.0, + "content": ". Then the perturbed training environment can", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 136, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 136, + 436, + 151 + ], + "score": 1.0, + "content": "be viewed as an adversarial model: both the agent and the adversary take actions", + "type": "text" + }, + { + "bbox": [ + 437, + 139, + 466, + 149 + ], + "score": 0.88, + "content": "a _ { a } , a _ { a d }", + "type": "inline_equation" + }, + { + "bbox": [ + 466, + 136, + 505, + 151 + ], + "score": 1.0, + "content": ", then the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 149, + 505, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 149, + 235, + 160 + ], + "score": 1.0, + "content": "environment will transit to state", + "type": "text" + }, + { + "bbox": [ + 236, + 150, + 252, + 160 + ], + "score": 0.87, + "content": "a _ { a d }", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 149, + 319, + 160 + ], + "score": 1.0, + "content": "with probability", + "type": "text" + }, + { + "bbox": [ + 320, + 149, + 329, + 159 + ], + "score": 0.79, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 149, + 505, + 160 + ], + "score": 1.0, + "content": "and transit following the unperturbed MDP", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 159, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 106, + 160, + 121, + 172 + ], + "score": 0.88, + "content": "p _ { s } ^ { a _ { a } }", + "type": "inline_equation" + }, + { + "bbox": [ + 122, + 159, + 189, + 173 + ], + "score": 1.0, + "content": "with probability", + "type": "text" + }, + { + "bbox": [ + 189, + 160, + 216, + 170 + ], + "score": 0.88, + "content": "1 - R", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 159, + 506, + 173 + ], + "score": 1.0, + "content": ". The goal of the maximize its accumulated reward, while the goal of the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 170, + 207, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 170, + 207, + 182 + ], + "score": 1.0, + "content": "natural is to minimize it.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 107, + 187, + 505, + 220 + ], + "lines": [ + { + "bbox": [ + 105, + 186, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 505, + 199 + ], + "score": 1.0, + "content": "Following the RARL algorithm [Pinto et al., 2017], in each iteration of the training, we first fix the", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 197, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 197, + 465, + 210 + ], + "score": 1.0, + "content": "adversarial policy and use Q-learning to optimize the agent’s policy and obtain the Q-table", + "type": "text" + }, + { + "bbox": [ + 465, + 198, + 478, + 209 + ], + "score": 0.88, + "content": "Q _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 197, + 505, + 210 + ], + "score": 1.0, + "content": ". Then", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 208, + 351, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 208, + 351, + 222 + ], + "score": 1.0, + "content": "we fix the agent’s policy and optimize the adversarial policy.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 225, + 506, + 313 + ], + "lines": [ + { + "bbox": [ + 105, + 225, + 506, + 238 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 506, + 238 + ], + "score": 1.0, + "content": "After each training iteration, we test the performance of the greedy policies w.r.t. Q-tables obtained", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 236, + 506, + 248 + ], + "spans": [ + { + "bbox": [ + 106, + 236, + 506, + 248 + ], + "score": 1.0, + "content": "from robust Q-learning and RARL. The testing environment is set to be the worst-case, i.e., after", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 246, + 506, + 259 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 506, + 259 + ], + "score": 1.0, + "content": "the agent takes an action, the environment transits to the state which has the minimal value function", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 107, + 258, + 507, + 270 + ], + "spans": [ + { + "bbox": [ + 107, + 258, + 183, + 270 + ], + "score": 0.62, + "content": "( \\mathrm { a r g m i n } _ { s \\in \\mathcal { S } } V _ { t } ( s ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 258, + 254, + 270 + ], + "score": 1.0, + "content": "with probability", + "type": "text" + }, + { + "bbox": [ + 254, + 260, + 260, + 270 + ], + "score": 0.74, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 258, + 507, + 270 + ], + "score": 1.0, + "content": ". We plot the accumulated discounted rewards of both algo-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 268, + 507, + 281 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 418, + 281 + ], + "score": 1.0, + "content": "rithms against number of training iterations under different parameters. We set", + "type": "text" + }, + { + "bbox": [ + 419, + 269, + 452, + 279 + ], + "score": 0.89, + "content": "\\alpha = 0 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 268, + 470, + 281 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 470, + 269, + 503, + 280 + ], + "score": 0.9, + "content": "\\gamma = 0 . 9", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 268, + 507, + 281 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 279, + 506, + 292 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 506, + 292 + ], + "score": 1.0, + "content": "It can be seen from Fig. 4 that our robust Q-learning achieves a higher accumulative reward, and thus", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 291, + 506, + 303 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 506, + 303 + ], + "score": 1.0, + "content": "is more robust that the RARL algorithm in [Pinto et al., 2017]. Also our robust Q-learning is more", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 302, + 307, + 314 + ], + "spans": [ + { + "bbox": [ + 106, + 302, + 307, + 314 + ], + "score": 1.0, + "content": "stable during training, i.e., the variance is smaller.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 15.5 + }, + { + "type": "image", + "bbox": [ + 121, + 337, + 484, + 433 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 121, + 337, + 484, + 433 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 121, + 337, + 484, + 433 + ], + "spans": [ + { + "bbox": [ + 121, + 337, + 484, + 433 + ], + "score": 0.972, + "type": "image", + "image_path": "05ae19de592c7a368e9a827c529316238af525b6bd10c6a6b56e811a7ad5cc4c.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 121, + 337, + 484, + 369.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 121, + 369.0, + 484, + 401.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 121, + 401.0, + 484, + 433.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 206, + 442, + 403, + 454 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 206, + 441, + 404, + 456 + ], + "spans": [ + { + "bbox": [ + 206, + 441, + 404, + 456 + ], + "score": 1.0, + "content": "Figure 4: Taxi-v3: robust Q-learning v.s. RARL.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + } + ], + "index": 22.0 + }, + { + "type": "title", + "bbox": [ + 107, + 482, + 183, + 495 + ], + "lines": [ + { + "bbox": [ + 104, + 479, + 185, + 498 + ], + "spans": [ + { + "bbox": [ + 104, + 479, + 185, + 498 + ], + "score": 1.0, + "content": "7 Conclusion", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 498, + 505, + 564 + ], + "lines": [ + { + "bbox": [ + 105, + 497, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 505, + 510 + ], + "score": 1.0, + "content": "In this paper, we develop a novel approach for solving model-free robust RL problems with model", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 509, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 505, + 522 + ], + "score": 1.0, + "content": "uncertainty. Our algorithms can be implemented in an online and incremental fashion, do not require", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 519, + 506, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 506, + 533 + ], + "score": 1.0, + "content": "additional memory than their non-robust counterparts. We theoretically proved the convergence", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 531, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 505, + 543 + ], + "score": 1.0, + "content": "of our algorithms under no additional assumption on the discount factor, and further characterized", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 541, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 541, + 505, + 554 + ], + "score": 1.0, + "content": "their finite-time error bounds, which match with their non-robust counterparts (within a constant", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 553, + 506, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 506, + 565 + ], + "score": 1.0, + "content": "factor). Our approach can be readily extended to robustify TD, SARSA and other GTD algorithms.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 107, + 564, + 506, + 608 + ], + "lines": [ + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "score": 1.0, + "content": "Limitations: It is also of future interest to investigate robustness to reward uncertainty, and other", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 574, + 506, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 506, + 587 + ], + "score": 1.0, + "content": "types of uncertainty sets, e.g., ones defined by KL divergence, Wasserstein distance and total", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 584, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 584, + 505, + 599 + ], + "score": 1.0, + "content": "variation. Negative societal impact: To the best of the authors’ knowledge, this study does not have", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 596, + 285, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 285, + 610 + ], + "score": 1.0, + "content": "any potential negative impact on the society.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32.5 + }, + { + "type": "title", + "bbox": [ + 107, + 623, + 214, + 636 + ], + "lines": [ + { + "bbox": [ + 105, + 621, + 217, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 217, + 639 + ], + "score": 1.0, + "content": "8 Acknowledgment", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 107, + 648, + 504, + 669 + ], + "lines": [ + { + "bbox": [ + 105, + 647, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 647, + 505, + 661 + ], + "score": 1.0, + "content": "The work of Y. Wang and S. Zou was supported by the National Science Foundation under Grants", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 658, + 247, + 670 + ], + "spans": [ + { + "bbox": [ + 106, + 658, + 247, + 670 + ], + "score": 1.0, + "content": "CCF-2106560 and CCF- 2007783.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 301, + 742, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 106, + 72, + 360, + 84 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 362, + 87 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 362, + 87 + ], + "score": 1.0, + "content": "6.3 Comparison with The Adversarial Training Approach", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 106, + 92, + 505, + 182 + ], + "lines": [ + { + "bbox": [ + 105, + 91, + 506, + 105 + ], + "spans": [ + { + "bbox": [ + 105, + 91, + 506, + 105 + ], + "score": 1.0, + "content": "We also compare our robust Q-learning with Robust Adversarial Reinforcement Learning (RARL)", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 103, + 506, + 116 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 506, + 116 + ], + "score": 1.0, + "content": "in [Pinto et al., 2017]. To apply their algorithm to our problem setting, we model the nature as an", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "score": 1.0, + "content": "adversarial player, and its goal is to minimize the reward that the agent receives. The action space", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 126, + 505, + 140 + ], + "spans": [ + { + "bbox": [ + 106, + 127, + 124, + 138 + ], + "score": 0.9, + "content": "\\mathcal { A } _ { a d }", + "type": "inline_equation" + }, + { + "bbox": [ + 125, + 126, + 283, + 140 + ], + "score": 1.0, + "content": "of the nature is set to be the state space", + "type": "text" + }, + { + "bbox": [ + 283, + 126, + 320, + 138 + ], + "score": 0.94, + "content": "\\mathcal { A } _ { a d } \\triangleq \\mathcal { S }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 126, + 505, + 140 + ], + "score": 1.0, + "content": ". Then the perturbed training environment can", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 136, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 136, + 436, + 151 + ], + "score": 1.0, + "content": "be viewed as an adversarial model: both the agent and the adversary take actions", + "type": "text" + }, + { + "bbox": [ + 437, + 139, + 466, + 149 + ], + "score": 0.88, + "content": "a _ { a } , a _ { a d }", + "type": "inline_equation" + }, + { + "bbox": [ + 466, + 136, + 505, + 151 + ], + "score": 1.0, + "content": ", then the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 149, + 505, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 149, + 235, + 160 + ], + "score": 1.0, + "content": "environment will transit to state", + "type": "text" + }, + { + "bbox": [ + 236, + 150, + 252, + 160 + ], + "score": 0.87, + "content": "a _ { a d }", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 149, + 319, + 160 + ], + "score": 1.0, + "content": "with probability", + "type": "text" + }, + { + "bbox": [ + 320, + 149, + 329, + 159 + ], + "score": 0.79, + "content": "R", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 149, + 505, + 160 + ], + "score": 1.0, + "content": "and transit following the unperturbed MDP", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 159, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 106, + 160, + 121, + 172 + ], + "score": 0.88, + "content": "p _ { s } ^ { a _ { a } }", + "type": "inline_equation" + }, + { + "bbox": [ + 122, + 159, + 189, + 173 + ], + "score": 1.0, + "content": "with probability", + "type": "text" + }, + { + "bbox": [ + 189, + 160, + 216, + 170 + ], + "score": 0.88, + "content": "1 - R", + "type": "inline_equation" + }, + { + "bbox": [ + 216, + 159, + 506, + 173 + ], + "score": 1.0, + "content": ". The goal of the maximize its accumulated reward, while the goal of the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 170, + 207, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 170, + 207, + 182 + ], + "score": 1.0, + "content": "natural is to minimize it.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 4.5, + "bbox_fs": [ + 105, + 91, + 506, + 182 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 187, + 505, + 220 + ], + "lines": [ + { + "bbox": [ + 105, + 186, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 505, + 199 + ], + "score": 1.0, + "content": "Following the RARL algorithm [Pinto et al., 2017], in each iteration of the training, we first fix the", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 197, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 197, + 465, + 210 + ], + "score": 1.0, + "content": "adversarial policy and use Q-learning to optimize the agent’s policy and obtain the Q-table", + "type": "text" + }, + { + "bbox": [ + 465, + 198, + 478, + 209 + ], + "score": 0.88, + "content": "Q _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 197, + 505, + 210 + ], + "score": 1.0, + "content": ". Then", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 208, + 351, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 208, + 351, + 222 + ], + "score": 1.0, + "content": "we fix the agent’s policy and optimize the adversarial policy.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 186, + 505, + 222 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 225, + 506, + 313 + ], + "lines": [ + { + "bbox": [ + 105, + 225, + 506, + 238 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 506, + 238 + ], + "score": 1.0, + "content": "After each training iteration, we test the performance of the greedy policies w.r.t. Q-tables obtained", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 236, + 506, + 248 + ], + "spans": [ + { + "bbox": [ + 106, + 236, + 506, + 248 + ], + "score": 1.0, + "content": "from robust Q-learning and RARL. The testing environment is set to be the worst-case, i.e., after", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 246, + 506, + 259 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 506, + 259 + ], + "score": 1.0, + "content": "the agent takes an action, the environment transits to the state which has the minimal value function", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 107, + 258, + 507, + 270 + ], + "spans": [ + { + "bbox": [ + 107, + 258, + 183, + 270 + ], + "score": 0.62, + "content": "( \\mathrm { a r g m i n } _ { s \\in \\mathcal { S } } V _ { t } ( s ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 258, + 254, + 270 + ], + "score": 1.0, + "content": "with probability", + "type": "text" + }, + { + "bbox": [ + 254, + 260, + 260, + 270 + ], + "score": 0.74, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 258, + 507, + 270 + ], + "score": 1.0, + "content": ". We plot the accumulated discounted rewards of both algo-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 268, + 507, + 281 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 418, + 281 + ], + "score": 1.0, + "content": "rithms against number of training iterations under different parameters. We set", + "type": "text" + }, + { + "bbox": [ + 419, + 269, + 452, + 279 + ], + "score": 0.89, + "content": "\\alpha = 0 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 268, + 470, + 281 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 470, + 269, + 503, + 280 + ], + "score": 0.9, + "content": "\\gamma = 0 . 9", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 268, + 507, + 281 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 279, + 506, + 292 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 506, + 292 + ], + "score": 1.0, + "content": "It can be seen from Fig. 4 that our robust Q-learning achieves a higher accumulative reward, and thus", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 291, + 506, + 303 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 506, + 303 + ], + "score": 1.0, + "content": "is more robust that the RARL algorithm in [Pinto et al., 2017]. Also our robust Q-learning is more", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 302, + 307, + 314 + ], + "spans": [ + { + "bbox": [ + 106, + 302, + 307, + 314 + ], + "score": 1.0, + "content": "stable during training, i.e., the variance is smaller.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 225, + 507, + 314 + ] + }, + { + "type": "image", + "bbox": [ + 121, + 337, + 484, + 433 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 121, + 337, + 484, + 433 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 121, + 337, + 484, + 433 + ], + "spans": [ + { + "bbox": [ + 121, + 337, + 484, + 433 + ], + "score": 0.972, + "type": "image", + "image_path": "05ae19de592c7a368e9a827c529316238af525b6bd10c6a6b56e811a7ad5cc4c.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 121, + 337, + 484, + 369.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 121, + 369.0, + 484, + 401.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 121, + 401.0, + 484, + 433.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 206, + 442, + 403, + 454 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 206, + 441, + 404, + 456 + ], + "spans": [ + { + "bbox": [ + 206, + 441, + 404, + 456 + ], + "score": 1.0, + "content": "Figure 4: Taxi-v3: robust Q-learning v.s. RARL.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + } + ], + "index": 22.0 + }, + { + "type": "title", + "bbox": [ + 107, + 482, + 183, + 495 + ], + "lines": [ + { + "bbox": [ + 104, + 479, + 185, + 498 + ], + "spans": [ + { + "bbox": [ + 104, + 479, + 185, + 498 + ], + "score": 1.0, + "content": "7 Conclusion", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 498, + 505, + 564 + ], + "lines": [ + { + "bbox": [ + 105, + 497, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 505, + 510 + ], + "score": 1.0, + "content": "In this paper, we develop a novel approach for solving model-free robust RL problems with model", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 509, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 505, + 522 + ], + "score": 1.0, + "content": "uncertainty. Our algorithms can be implemented in an online and incremental fashion, do not require", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 519, + 506, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 506, + 533 + ], + "score": 1.0, + "content": "additional memory than their non-robust counterparts. We theoretically proved the convergence", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 531, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 505, + 543 + ], + "score": 1.0, + "content": "of our algorithms under no additional assumption on the discount factor, and further characterized", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 541, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 541, + 505, + 554 + ], + "score": 1.0, + "content": "their finite-time error bounds, which match with their non-robust counterparts (within a constant", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 553, + 506, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 506, + 565 + ], + "score": 1.0, + "content": "factor). Our approach can be readily extended to robustify TD, SARSA and other GTD algorithms.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27.5, + "bbox_fs": [ + 105, + 497, + 506, + 565 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 564, + 506, + 608 + ], + "lines": [ + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "score": 1.0, + "content": "Limitations: It is also of future interest to investigate robustness to reward uncertainty, and other", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 574, + 506, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 506, + 587 + ], + "score": 1.0, + "content": "types of uncertainty sets, e.g., ones defined by KL divergence, Wasserstein distance and total", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 584, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 584, + 505, + 599 + ], + "score": 1.0, + "content": "variation. Negative societal impact: To the best of the authors’ knowledge, this study does not have", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 596, + 285, + 610 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 285, + 610 + ], + "score": 1.0, + "content": "any potential negative impact on the society.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32.5, + "bbox_fs": [ + 105, + 563, + 506, + 610 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 623, + 214, + 636 + ], + "lines": [ + { + "bbox": [ + 105, + 621, + 217, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 217, + 639 + ], + "score": 1.0, + "content": "8 Acknowledgment", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 107, + 648, + 504, + 669 + ], + "lines": [ + { + "bbox": [ + 105, + 647, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 647, + 505, + 661 + ], + "score": 1.0, + "content": "The work of Y. Wang and S. Zou was supported by the National Science Foundation under Grants", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 658, + 247, + 670 + ], + "spans": [ + { + "bbox": [ + 106, + 658, + 247, + 670 + ], + "score": 1.0, + "content": "CCF-2106560 and CCF- 2007783.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5, + "bbox_fs": [ + 105, + 647, + 505, + 670 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 73, + 155, + 84 + ], + "lines": [ + { + "bbox": [ + 106, + 72, + 155, + 85 + ], + "spans": [ + { + "bbox": [ + 106, + 72, + 155, + 85 + ], + "score": 1.0, + "content": "References", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 89, + 506, + 122 + ], + "lines": [ + { + "bbox": [ + 105, + 87, + 506, + 102 + ], + "spans": [ + { + "bbox": [ + 105, + 87, + 506, + 102 + ], + "score": 1.0, + "content": "Mohammed Amin Abdullah, Hang Ren, Haitham Bou Ammar, Vladimir Milenkovic, Rui Luo,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 100, + 505, + 113 + ], + "spans": [ + { + "bbox": [ + 115, + 100, + 505, + 113 + ], + "score": 1.0, + "content": "Mingtian Zhang, and Jun Wang. Wasserstein robust reinforcement learning. arXiv preprint", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 111, + 219, + 122 + ], + "spans": [ + { + "bbox": [ + 115, + 111, + 219, + 122 + ], + "score": 1.0, + "content": "arXiv:1907.13196, 2019.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 105, + 131, + 505, + 165 + ], + "lines": [ + { + "bbox": [ + 105, + 130, + 505, + 145 + ], + "spans": [ + { + "bbox": [ + 105, + 130, + 505, + 145 + ], + "score": 1.0, + "content": "Alekh Agarwal, Sham M Kakade, Jason D Lee, and Gaurav Mahajan. On the theory of policy", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 141, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 116, + 141, + 505, + 155 + ], + "score": 1.0, + "content": "gradient methods: Optimality, approximation, and distribution shift. Journal of Machine Learning", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 153, + 237, + 164 + ], + "spans": [ + { + "bbox": [ + 116, + 153, + 237, + 164 + ], + "score": 1.0, + "content": "Research, 22(98):1–76, 2021.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 173, + 505, + 207 + ], + "lines": [ + { + "bbox": [ + 106, + 173, + 505, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 173, + 505, + 185 + ], + "score": 1.0, + "content": "Kavosh Asadi and Michael L Littman. An alternative softmax operator for reinforcement learning. In", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 183, + 506, + 197 + ], + "spans": [ + { + "bbox": [ + 115, + 183, + 506, + 197 + ], + "score": 1.0, + "content": "Proc. International Conference on Machine Learning (ICML), volume 70, pages 243–252. JMLR,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 194, + 143, + 208 + ], + "spans": [ + { + "bbox": [ + 115, + 194, + 143, + 208 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 108, + 215, + 504, + 249 + ], + "lines": [ + { + "bbox": [ + 106, + 215, + 505, + 228 + ], + "spans": [ + { + "bbox": [ + 106, + 215, + 505, + 228 + ], + "score": 1.0, + "content": "Christopher G Atkeson and Jun Morimoto. Nonparametric representation of policies and value", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 226, + 506, + 240 + ], + "spans": [ + { + "bbox": [ + 115, + 226, + 506, + 240 + ], + "score": 1.0, + "content": "functions: A trajectory-based approach. In Proc. Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 237, + 283, + 249 + ], + "spans": [ + { + "bbox": [ + 115, + 237, + 283, + 249 + ], + "score": 1.0, + "content": "Systems (NIPS), pages 1643–1650, 2003.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 107, + 258, + 503, + 291 + ], + "lines": [ + { + "bbox": [ + 105, + 257, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 257, + 505, + 272 + ], + "score": 1.0, + "content": "Kishan Panaganti Badrinath and Dileep Kalathil. Robust reinforcement learning using least squares", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 269, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 115, + 269, + 505, + 282 + ], + "score": 1.0, + "content": "policy iteration with provable performance guarantees. In Proc. International Conference on", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 279, + 351, + 293 + ], + "spans": [ + { + "bbox": [ + 115, + 279, + 351, + 293 + ], + "score": 1.0, + "content": "Machine Learning (ICML), pages 511–520. PMLR, 2021.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 104, + 299, + 505, + 313 + ], + "lines": [ + { + "bbox": [ + 105, + 299, + 507, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 299, + 507, + 313 + ], + "score": 1.0, + "content": "J Andrew Bagnell, Andrew Y Ng, and Jeff G Schneider. Solving uncertain markov decision. 09 2001.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 104, + 320, + 504, + 343 + ], + "lines": [ + { + "bbox": [ + 105, + 320, + 505, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 505, + 333 + ], + "score": 1.0, + "content": "Leemon Baird. Residual algorithms: Reinforcement learning with function approximation. In", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 331, + 364, + 344 + ], + "spans": [ + { + "bbox": [ + 115, + 331, + 364, + 344 + ], + "score": 1.0, + "content": "Machine Learning Proceedings, pages 30–37. Elsevier, 1995.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 104, + 351, + 505, + 375 + ], + "lines": [ + { + "bbox": [ + 106, + 352, + 505, + 365 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 505, + 365 + ], + "score": 1.0, + "content": "Carolyn L Beck and Rayadurgam Srikant. Error bounds for constant step-size Q-learning. Systems &", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 362, + 281, + 375 + ], + "spans": [ + { + "bbox": [ + 115, + 362, + 281, + 375 + ], + "score": 1.0, + "content": "control letters, 61(12):1203–1208, 2012.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 106, + 383, + 504, + 406 + ], + "lines": [ + { + "bbox": [ + 106, + 383, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 505, + 396 + ], + "score": 1.0, + "content": "Jalaj Bhandari and Daniel Russo. Global optimality guarantees for policy gradient methods. arXiv", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 394, + 254, + 406 + ], + "spans": [ + { + "bbox": [ + 115, + 394, + 254, + 406 + ], + "score": 1.0, + "content": "preprint arXiv:1906.01786, 2019.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 106, + 414, + 504, + 448 + ], + "lines": [ + { + "bbox": [ + 105, + 414, + 506, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 506, + 428 + ], + "score": 1.0, + "content": "Jalaj Bhandari, Daniel Russo, and Raghav Singal. A finite time analysis of temporal difference", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 425, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 115, + 425, + 506, + 439 + ], + "score": 1.0, + "content": "learning with linear function approximation. In Proc. Annual Conference on Learning Theory", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 437, + 282, + 449 + ], + "spans": [ + { + "bbox": [ + 115, + 437, + 282, + 449 + ], + "score": 1.0, + "content": "(CoLT), pages 1691–1692. PMLR, 2018.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 108, + 456, + 505, + 491 + ], + "lines": [ + { + "bbox": [ + 106, + 457, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 106, + 457, + 506, + 469 + ], + "score": 1.0, + "content": "Shalabh Bhatnagar, Doina Precup, David Silver, Richard S Sutton, Hamid Maei, and Csaba Szepesvári.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 468, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 115, + 468, + 506, + 480 + ], + "score": 1.0, + "content": "Convergent temporal-difference learning with arbitrary smooth function approximation. In Proc.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 114, + 478, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 114, + 478, + 506, + 492 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems (NIPS), volume 22, pages 1204–1212, 2009.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 105, + 498, + 504, + 522 + ], + "lines": [ + { + "bbox": [ + 106, + 499, + 505, + 512 + ], + "spans": [ + { + "bbox": [ + 106, + 499, + 505, + 512 + ], + "score": 1.0, + "content": "Vivek S Borkar and Sean P Meyn. The ODE method for convergence of stochastic approximation", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 116, + 509, + 497, + 523 + ], + "spans": [ + { + "bbox": [ + 116, + 509, + 497, + 523 + ], + "score": 1.0, + "content": "and reinforcement learning. SIAM Journal on Control and Optimization, 38(2):447–469, 2000.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 106, + 530, + 504, + 553 + ], + "lines": [ + { + "bbox": [ + 106, + 531, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 505, + 543 + ], + "score": 1.0, + "content": "Greg Brockman, Vicki Cheung, Ludwig Pettersson, Jonas Schneider, John Schulman, Jie Tang, and", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 541, + 420, + 554 + ], + "spans": [ + { + "bbox": [ + 115, + 541, + 420, + 554 + ], + "score": 1.0, + "content": "Wojciech Zaremba. OpenAI Gym. arXiv preprint arXiv:1606.01540, 2016.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 106, + 561, + 506, + 595 + ], + "lines": [ + { + "bbox": [ + 106, + 561, + 507, + 576 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 507, + 576 + ], + "score": 1.0, + "content": "Qi Cai, Zhuoran Yang, Jason D Lee, and Zhaoran Wang. Neural temporal-difference learning con-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 573, + 506, + 586 + ], + "spans": [ + { + "bbox": [ + 115, + 573, + 506, + 586 + ], + "score": 1.0, + "content": "verges to global optima. In Proc. Advances in Neural Information Processing Systems (NeurIPS),", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 585, + 227, + 596 + ], + "spans": [ + { + "bbox": [ + 115, + 585, + 227, + 596 + ], + "score": 1.0, + "content": "pages 11312–11322, 2019.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 106, + 604, + 505, + 637 + ], + "lines": [ + { + "bbox": [ + 106, + 604, + 506, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 604, + 506, + 617 + ], + "score": 1.0, + "content": "Shicong Cen, Chen Cheng, Yuxin Chen, Yuting Wei, and Yuejie Chi. Fast global convergence of", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 614, + 506, + 628 + ], + "spans": [ + { + "bbox": [ + 115, + 614, + 506, + 628 + ], + "score": 1.0, + "content": "natural policy gradient methods with entropy regularization. arXiv preprint arXiv:2007.06558,", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 115, + 624, + 142, + 639 + ], + "spans": [ + { + "bbox": [ + 115, + 624, + 142, + 639 + ], + "score": 1.0, + "content": "2020.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 105, + 646, + 505, + 680 + ], + "lines": [ + { + "bbox": [ + 106, + 645, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 106, + 645, + 505, + 659 + ], + "score": 1.0, + "content": "Zaiwei Chen, Sheng Zhang, Thinh T Doan, Siva Theja Maguluri, and John-Paul Clarke. Performance", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 116, + 657, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 116, + 657, + 505, + 669 + ], + "score": 1.0, + "content": "of Q-learning with linear function approuimation: Stability and finite-time analysis. arXiv preprint", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 668, + 219, + 680 + ], + "spans": [ + { + "bbox": [ + 115, + 668, + 219, + 680 + ], + "score": 1.0, + "content": "arXiv:1905.11425, 2019.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 107, + 689, + 506, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 687, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 506, + 702 + ], + "score": 1.0, + "content": "Gal Dalal, Balázs Szörényi, Gugan Thoppe, and Shie Mannor. Finite sample analysis of two-timescale", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 700, + 504, + 712 + ], + "spans": [ + { + "bbox": [ + 116, + 700, + 504, + 712 + ], + "score": 1.0, + "content": "stochastic approximation with applications to reinforcement learning. Proceedings of Machine", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 116, + 711, + 259, + 723 + ], + "spans": [ + { + "bbox": [ + 116, + 711, + 259, + 723 + ], + "score": 1.0, + "content": "Learning Research, 75:1–35, 2018.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 310, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 312, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 312, + 754 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 73, + 155, + 84 + ], + "lines": [ + { + "bbox": [ + 106, + 72, + 155, + 85 + ], + "spans": [ + { + "bbox": [ + 106, + 72, + 155, + 85 + ], + "score": 1.0, + "content": "References", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 89, + 506, + 122 + ], + "lines": [ + { + "bbox": [ + 105, + 87, + 506, + 102 + ], + "spans": [ + { + "bbox": [ + 105, + 87, + 506, + 102 + ], + "score": 1.0, + "content": "Mohammed Amin Abdullah, Hang Ren, Haitham Bou Ammar, Vladimir Milenkovic, Rui Luo,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 100, + 505, + 113 + ], + "spans": [ + { + "bbox": [ + 115, + 100, + 505, + 113 + ], + "score": 1.0, + "content": "Mingtian Zhang, and Jun Wang. Wasserstein robust reinforcement learning. arXiv preprint", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 111, + 219, + 122 + ], + "spans": [ + { + "bbox": [ + 115, + 111, + 219, + 122 + ], + "score": 1.0, + "content": "arXiv:1907.13196, 2019.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 87, + 506, + 122 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 131, + 505, + 165 + ], + "lines": [ + { + "bbox": [ + 105, + 130, + 505, + 145 + ], + "spans": [ + { + "bbox": [ + 105, + 130, + 505, + 145 + ], + "score": 1.0, + "content": "Alekh Agarwal, Sham M Kakade, Jason D Lee, and Gaurav Mahajan. On the theory of policy", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 141, + 505, + 155 + ], + "spans": [ + { + "bbox": [ + 116, + 141, + 505, + 155 + ], + "score": 1.0, + "content": "gradient methods: Optimality, approximation, and distribution shift. Journal of Machine Learning", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 153, + 237, + 164 + ], + "spans": [ + { + "bbox": [ + 116, + 153, + 237, + 164 + ], + "score": 1.0, + "content": "Research, 22(98):1–76, 2021.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 130, + 505, + 164 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 173, + 505, + 207 + ], + "lines": [ + { + "bbox": [ + 106, + 173, + 505, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 173, + 505, + 185 + ], + "score": 1.0, + "content": "Kavosh Asadi and Michael L Littman. An alternative softmax operator for reinforcement learning. In", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 183, + 506, + 197 + ], + "spans": [ + { + "bbox": [ + 115, + 183, + 506, + 197 + ], + "score": 1.0, + "content": "Proc. International Conference on Machine Learning (ICML), volume 70, pages 243–252. JMLR,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 194, + 143, + 208 + ], + "spans": [ + { + "bbox": [ + 115, + 194, + 143, + 208 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8, + "bbox_fs": [ + 106, + 173, + 506, + 208 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 215, + 504, + 249 + ], + "lines": [ + { + "bbox": [ + 106, + 215, + 505, + 228 + ], + "spans": [ + { + "bbox": [ + 106, + 215, + 505, + 228 + ], + "score": 1.0, + "content": "Christopher G Atkeson and Jun Morimoto. Nonparametric representation of policies and value", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 226, + 506, + 240 + ], + "spans": [ + { + "bbox": [ + 115, + 226, + 506, + 240 + ], + "score": 1.0, + "content": "functions: A trajectory-based approach. In Proc. Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 237, + 283, + 249 + ], + "spans": [ + { + "bbox": [ + 115, + 237, + 283, + 249 + ], + "score": 1.0, + "content": "Systems (NIPS), pages 1643–1650, 2003.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11, + "bbox_fs": [ + 106, + 215, + 506, + 249 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 258, + 503, + 291 + ], + "lines": [ + { + "bbox": [ + 105, + 257, + 505, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 257, + 505, + 272 + ], + "score": 1.0, + "content": "Kishan Panaganti Badrinath and Dileep Kalathil. Robust reinforcement learning using least squares", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 269, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 115, + 269, + 505, + 282 + ], + "score": 1.0, + "content": "policy iteration with provable performance guarantees. In Proc. International Conference on", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 279, + 351, + 293 + ], + "spans": [ + { + "bbox": [ + 115, + 279, + 351, + 293 + ], + "score": 1.0, + "content": "Machine Learning (ICML), pages 511–520. PMLR, 2021.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 257, + 505, + 293 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 299, + 505, + 313 + ], + "lines": [ + { + "bbox": [ + 105, + 299, + 507, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 299, + 507, + 313 + ], + "score": 1.0, + "content": "J Andrew Bagnell, Andrew Y Ng, and Jeff G Schneider. Solving uncertain markov decision. 09 2001.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 299, + 507, + 313 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 320, + 504, + 343 + ], + "lines": [ + { + "bbox": [ + 105, + 320, + 505, + 333 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 505, + 333 + ], + "score": 1.0, + "content": "Leemon Baird. Residual algorithms: Reinforcement learning with function approximation. In", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 331, + 364, + 344 + ], + "spans": [ + { + "bbox": [ + 115, + 331, + 364, + 344 + ], + "score": 1.0, + "content": "Machine Learning Proceedings, pages 30–37. Elsevier, 1995.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 105, + 320, + 505, + 344 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 351, + 505, + 375 + ], + "lines": [ + { + "bbox": [ + 106, + 352, + 505, + 365 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 505, + 365 + ], + "score": 1.0, + "content": "Carolyn L Beck and Rayadurgam Srikant. Error bounds for constant step-size Q-learning. Systems &", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 362, + 281, + 375 + ], + "spans": [ + { + "bbox": [ + 115, + 362, + 281, + 375 + ], + "score": 1.0, + "content": "control letters, 61(12):1203–1208, 2012.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 106, + 352, + 505, + 375 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 383, + 504, + 406 + ], + "lines": [ + { + "bbox": [ + 106, + 383, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 505, + 396 + ], + "score": 1.0, + "content": "Jalaj Bhandari and Daniel Russo. Global optimality guarantees for policy gradient methods. arXiv", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 394, + 254, + 406 + ], + "spans": [ + { + "bbox": [ + 115, + 394, + 254, + 406 + ], + "score": 1.0, + "content": "preprint arXiv:1906.01786, 2019.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5, + "bbox_fs": [ + 106, + 383, + 505, + 406 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 414, + 504, + 448 + ], + "lines": [ + { + "bbox": [ + 105, + 414, + 506, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 506, + 428 + ], + "score": 1.0, + "content": "Jalaj Bhandari, Daniel Russo, and Raghav Singal. A finite time analysis of temporal difference", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 425, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 115, + 425, + 506, + 439 + ], + "score": 1.0, + "content": "learning with linear function approximation. In Proc. Annual Conference on Learning Theory", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 437, + 282, + 449 + ], + "spans": [ + { + "bbox": [ + 115, + 437, + 282, + 449 + ], + "score": 1.0, + "content": "(CoLT), pages 1691–1692. PMLR, 2018.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 414, + 506, + 449 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 456, + 505, + 491 + ], + "lines": [ + { + "bbox": [ + 106, + 457, + 506, + 469 + ], + "spans": [ + { + "bbox": [ + 106, + 457, + 506, + 469 + ], + "score": 1.0, + "content": "Shalabh Bhatnagar, Doina Precup, David Silver, Richard S Sutton, Hamid Maei, and Csaba Szepesvári.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 468, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 115, + 468, + 506, + 480 + ], + "score": 1.0, + "content": "Convergent temporal-difference learning with arbitrary smooth function approximation. In Proc.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 114, + 478, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 114, + 478, + 506, + 492 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems (NIPS), volume 22, pages 1204–1212, 2009.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27, + "bbox_fs": [ + 106, + 457, + 506, + 492 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 498, + 504, + 522 + ], + "lines": [ + { + "bbox": [ + 106, + 499, + 505, + 512 + ], + "spans": [ + { + "bbox": [ + 106, + 499, + 505, + 512 + ], + "score": 1.0, + "content": "Vivek S Borkar and Sean P Meyn. The ODE method for convergence of stochastic approximation", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 116, + 509, + 497, + 523 + ], + "spans": [ + { + "bbox": [ + 116, + 509, + 497, + 523 + ], + "score": 1.0, + "content": "and reinforcement learning. SIAM Journal on Control and Optimization, 38(2):447–469, 2000.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5, + "bbox_fs": [ + 106, + 499, + 505, + 523 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 530, + 504, + 553 + ], + "lines": [ + { + "bbox": [ + 106, + 531, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 505, + 543 + ], + "score": 1.0, + "content": "Greg Brockman, Vicki Cheung, Ludwig Pettersson, Jonas Schneider, John Schulman, Jie Tang, and", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 541, + 420, + 554 + ], + "spans": [ + { + "bbox": [ + 115, + 541, + 420, + 554 + ], + "score": 1.0, + "content": "Wojciech Zaremba. OpenAI Gym. arXiv preprint arXiv:1606.01540, 2016.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5, + "bbox_fs": [ + 106, + 531, + 505, + 554 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 561, + 506, + 595 + ], + "lines": [ + { + "bbox": [ + 106, + 561, + 507, + 576 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 507, + 576 + ], + "score": 1.0, + "content": "Qi Cai, Zhuoran Yang, Jason D Lee, and Zhaoran Wang. Neural temporal-difference learning con-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 573, + 506, + 586 + ], + "spans": [ + { + "bbox": [ + 115, + 573, + 506, + 586 + ], + "score": 1.0, + "content": "verges to global optima. In Proc. Advances in Neural Information Processing Systems (NeurIPS),", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 585, + 227, + 596 + ], + "spans": [ + { + "bbox": [ + 115, + 585, + 227, + 596 + ], + "score": 1.0, + "content": "pages 11312–11322, 2019.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34, + "bbox_fs": [ + 106, + 561, + 507, + 596 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 604, + 505, + 637 + ], + "lines": [ + { + "bbox": [ + 106, + 604, + 506, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 604, + 506, + 617 + ], + "score": 1.0, + "content": "Shicong Cen, Chen Cheng, Yuxin Chen, Yuting Wei, and Yuejie Chi. Fast global convergence of", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 614, + 506, + 628 + ], + "spans": [ + { + "bbox": [ + 115, + 614, + 506, + 628 + ], + "score": 1.0, + "content": "natural policy gradient methods with entropy regularization. arXiv preprint arXiv:2007.06558,", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 115, + 624, + 142, + 639 + ], + "spans": [ + { + "bbox": [ + 115, + 624, + 142, + 639 + ], + "score": 1.0, + "content": "2020.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37, + "bbox_fs": [ + 106, + 604, + 506, + 639 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 646, + 505, + 680 + ], + "lines": [ + { + "bbox": [ + 106, + 645, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 106, + 645, + 505, + 659 + ], + "score": 1.0, + "content": "Zaiwei Chen, Sheng Zhang, Thinh T Doan, Siva Theja Maguluri, and John-Paul Clarke. Performance", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 116, + 657, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 116, + 657, + 505, + 669 + ], + "score": 1.0, + "content": "of Q-learning with linear function approuimation: Stability and finite-time analysis. arXiv preprint", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 668, + 219, + 680 + ], + "spans": [ + { + "bbox": [ + 115, + 668, + 219, + 680 + ], + "score": 1.0, + "content": "arXiv:1905.11425, 2019.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40, + "bbox_fs": [ + 106, + 645, + 505, + 680 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 689, + 506, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 687, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 506, + 702 + ], + "score": 1.0, + "content": "Gal Dalal, Balázs Szörényi, Gugan Thoppe, and Shie Mannor. Finite sample analysis of two-timescale", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 700, + 504, + 712 + ], + "spans": [ + { + "bbox": [ + 116, + 700, + 504, + 712 + ], + "score": 1.0, + "content": "stochastic approximation with applications to reinforcement learning. Proceedings of Machine", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 116, + 711, + 259, + 723 + ], + "spans": [ + { + "bbox": [ + 116, + 711, + 259, + 723 + ], + "score": 1.0, + "content": "Learning Research, 75:1–35, 2018.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43, + "bbox_fs": [ + 105, + 687, + 506, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 72, + 504, + 106 + ], + "lines": [ + { + "bbox": [ + 105, + 71, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 71, + 505, + 86 + ], + "score": 1.0, + "content": "Gal Dalal, Balázs Szörényi, and Gugan Thoppe. A tale of two-timescale reinforcement learning", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 83, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 115, + 83, + 505, + 96 + ], + "score": 1.0, + "content": "with the tightest finite-time bound. In Proc. Conference on Artificial Intelligence (AAAI), pages", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 95, + 191, + 105 + ], + "spans": [ + { + "bbox": [ + 116, + 95, + 191, + 105 + ], + "score": 1.0, + "content": "3701–3708, 2020.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 107, + 113, + 504, + 147 + ], + "lines": [ + { + "bbox": [ + 105, + 112, + 506, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 112, + 506, + 127 + ], + "score": 1.0, + "content": "Dalal, Gal and Szörényi, Balázs and Thoppe, Gugan and Mannor, Shie. Finite sample analyses for", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 123, + 506, + 137 + ], + "spans": [ + { + "bbox": [ + 115, + 123, + 506, + 137 + ], + "score": 1.0, + "content": "TD(0) with function approximation. In Proc. AAAI Conference on Artificial Intelligence (AAAI),", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 136, + 216, + 146 + ], + "spans": [ + { + "bbox": [ + 115, + 136, + 216, + 146 + ], + "score": 1.0, + "content": "pages 6144–6160, 2018.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 105, + 154, + 504, + 177 + ], + "lines": [ + { + "bbox": [ + 106, + 154, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 106, + 154, + 505, + 167 + ], + "score": 1.0, + "content": "Thinh T Doan. Finite-time analysis and restarting scheme for linear two-time-scale stochastic", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 165, + 459, + 177 + ], + "spans": [ + { + "bbox": [ + 115, + 165, + 459, + 177 + ], + "score": 1.0, + "content": "approximation. SIAM Journal on Control and Optimization, 59(4):2798–2819, 2021.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "text", + "bbox": [ + 106, + 183, + 504, + 207 + ], + "lines": [ + { + "bbox": [ + 105, + 182, + 506, + 198 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 506, + 198 + ], + "score": 1.0, + "content": "Eyal Even-Dar, Yishay Mansour, and Peter Bartlett. Learning rates for Q-learning. Journal of", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 116, + 195, + 276, + 207 + ], + "spans": [ + { + "bbox": [ + 116, + 195, + 276, + 207 + ], + "score": 1.0, + "content": "machine learning Research, 5(1), 2003.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 106, + 213, + 505, + 248 + ], + "lines": [ + { + "bbox": [ + 105, + 213, + 505, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 505, + 227 + ], + "score": 1.0, + "content": "Harsh Gupta, R Srikant, and Lei Ying. Finite-time performance bounds and adaptive learning rate", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 225, + 505, + 237 + ], + "spans": [ + { + "bbox": [ + 116, + 225, + 505, + 237 + ], + "score": 1.0, + "content": "selection for two time-scale reinforcement learning. In Proc. Advances in Neural Information", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 116, + 236, + 343, + 248 + ], + "spans": [ + { + "bbox": [ + 116, + 236, + 343, + 248 + ], + "score": 1.0, + "content": "Processing Systems (NeurIPS), pages 4706–4715, 2019.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 106, + 254, + 503, + 277 + ], + "lines": [ + { + "bbox": [ + 105, + 254, + 505, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 505, + 267 + ], + "score": 1.0, + "content": "Linfang Hou, Liang Pang, Xin Hong, Yanyan Lan, Zhiming Ma, and Dawei Yin. Robust reinforcement", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 264, + 430, + 277 + ], + "spans": [ + { + "bbox": [ + 115, + 264, + 430, + 277 + ], + "score": 1.0, + "content": "learning with Wasserstein constraint. arXiv preprint arXiv:2006.00945, 2020.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 106, + 284, + 506, + 317 + ], + "lines": [ + { + "bbox": [ + 106, + 284, + 505, + 297 + ], + "spans": [ + { + "bbox": [ + 106, + 284, + 505, + 297 + ], + "score": 1.0, + "content": "Sandy Huang, Nicolas Papernot, Ian Goodfellow, Yan Duan, and Pieter Abbeel. Adversarial attacks", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 296, + 506, + 308 + ], + "spans": [ + { + "bbox": [ + 115, + 296, + 506, + 308 + ], + "score": 1.0, + "content": "on neural network policies. In Proc. International Conference on Learning Representations (ICLR),", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 305, + 142, + 317 + ], + "spans": [ + { + "bbox": [ + 115, + 305, + 142, + 317 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 324, + 505, + 348 + ], + "lines": [ + { + "bbox": [ + 105, + 323, + 507, + 339 + ], + "spans": [ + { + "bbox": [ + 105, + 323, + 507, + 339 + ], + "score": 1.0, + "content": "Peter J Huber. A robust version of the probability ratio test. The Annals of Mathematical Statistics,", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 336, + 216, + 348 + ], + "spans": [ + { + "bbox": [ + 115, + 336, + 216, + 348 + ], + "score": 1.0, + "content": "pages 1753–1758, 1965.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 106, + 354, + 504, + 378 + ], + "lines": [ + { + "bbox": [ + 106, + 354, + 506, + 368 + ], + "spans": [ + { + "bbox": [ + 106, + 354, + 506, + 368 + ], + "score": 1.0, + "content": "Garud N Iyengar. Robust dynamic programming. Mathematics of Operations Research, 30(2):", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 116, + 365, + 182, + 378 + ], + "spans": [ + { + "bbox": [ + 116, + 365, + 182, + 378 + ], + "score": 1.0, + "content": "257–280, 2005.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 106, + 384, + 506, + 419 + ], + "lines": [ + { + "bbox": [ + 106, + 384, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 505, + 397 + ], + "score": 1.0, + "content": "Maxim Kaledin, Eric Moulines, Alexey Naumov, Vladislav Tadic, and Hoi-To Wai. Finite time", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 116, + 395, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 116, + 395, + 505, + 407 + ], + "score": 1.0, + "content": "analysis of linear two-timescale stochastic approximation with Markovian noise. In Proc. Annual", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 116, + 406, + 412, + 419 + ], + "spans": [ + { + "bbox": [ + 116, + 406, + 412, + 419 + ], + "score": 1.0, + "content": "Conference on Learning Theory (CoLT), pages 2144–2203. PMLR, 2020.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 106, + 425, + 504, + 448 + ], + "lines": [ + { + "bbox": [ + 106, + 425, + 506, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 425, + 506, + 438 + ], + "score": 1.0, + "content": "Jernej Kos and Dawn Song. Delving into adversarial attacks on deep policies. In Proc. International", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 436, + 341, + 449 + ], + "spans": [ + { + "bbox": [ + 115, + 436, + 341, + 449 + ], + "score": 1.0, + "content": "Conference on Learning Representations (ICLR), 2017.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 106, + 455, + 506, + 488 + ], + "lines": [ + { + "bbox": [ + 105, + 454, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 505, + 468 + ], + "score": 1.0, + "content": "Harshat Kumar, Alec Koppel, and Alejandro Ribeiro. On the sample complexity of actor-critic", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 466, + 506, + 479 + ], + "spans": [ + { + "bbox": [ + 115, + 466, + 506, + 479 + ], + "score": 1.0, + "content": "method for reinforcement learning with function approximation. arXiv preprint arXiv:1910.08412,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 475, + 143, + 490 + ], + "spans": [ + { + "bbox": [ + 115, + 475, + 143, + 490 + ], + "score": 1.0, + "content": "2019.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 107, + 495, + 504, + 529 + ], + "lines": [ + { + "bbox": [ + 105, + 495, + 505, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 505, + 509 + ], + "score": 1.0, + "content": "Gen Li, Yuting Wei, Yuejie Chi, Yuantao Gu, and Yuxin Chen. Sample complexity of asynchronous", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 507, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 116, + 507, + 505, + 519 + ], + "score": 1.0, + "content": "Q-learning: Sharper analysis and variance reduction. In Proc. Advances in Neural Information", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 116, + 517, + 267, + 531 + ], + "spans": [ + { + "bbox": [ + 116, + 517, + 267, + 531 + ], + "score": 1.0, + "content": "Processing Systems (NeurIPS), 2020.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 104, + 536, + 504, + 560 + ], + "lines": [ + { + "bbox": [ + 106, + 536, + 505, + 550 + ], + "spans": [ + { + "bbox": [ + 106, + 536, + 505, + 550 + ], + "score": 1.0, + "content": "Gen Li, Changxiao Cai, Yuxin Chen, Yuantao Gu, Yuting Wei, and Yuejie Chi. Is Q-learning minimax", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 548, + 464, + 560 + ], + "spans": [ + { + "bbox": [ + 115, + 548, + 464, + 560 + ], + "score": 1.0, + "content": "optimal? A tight sample complexity analysis. arXiv preprint arXiv:2102.06548, 2021.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 107, + 566, + 505, + 600 + ], + "lines": [ + { + "bbox": [ + 105, + 566, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 566, + 505, + 579 + ], + "score": 1.0, + "content": "Shiau Hong Lim, Huan Xu, and Shie Mannor. Reinforcement learning in robust Markov decision", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 578, + 506, + 590 + ], + "spans": [ + { + "bbox": [ + 115, + 578, + 506, + 590 + ], + "score": 1.0, + "content": "processes. In Proc. Advances in Neural Information Processing Systems (NIPS), pages 701–709,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 114, + 587, + 142, + 600 + ], + "spans": [ + { + "bbox": [ + 114, + 587, + 142, + 600 + ], + "score": 1.0, + "content": "2013.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 108, + 606, + 505, + 641 + ], + "lines": [ + { + "bbox": [ + 106, + 606, + 506, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 606, + 506, + 621 + ], + "score": 1.0, + "content": "Yen-Chen Lin, Zhang-Wei Hong, Yuan-Hong Liao, Meng-Li Shih, Ming-Yu Liu, and Min Sun.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 618, + 505, + 631 + ], + "spans": [ + { + "bbox": [ + 115, + 618, + 505, + 631 + ], + "score": 1.0, + "content": "Tactics of adversarial attack on deep reinforcement learning agents. In Proc. International Joint", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 117, + 629, + 403, + 642 + ], + "spans": [ + { + "bbox": [ + 117, + 629, + 403, + 642 + ], + "score": 1.0, + "content": "Conferences on Artificial Intelligence (IJCAI), pages 3756–3762, 2017.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 107, + 647, + 504, + 682 + ], + "lines": [ + { + "bbox": [ + 106, + 648, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 106, + 648, + 505, + 660 + ], + "score": 1.0, + "content": "Bo Liu, Ji Liu, Mohammad Ghavamzadeh, Sridhar Mahadevan, and Marek Petrik. Finite-sample", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 659, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 116, + 659, + 505, + 672 + ], + "score": 1.0, + "content": "analysis of proximal gradient TD algorithms. In Proc. International Conference on Uncertainty in", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 114, + 669, + 358, + 683 + ], + "spans": [ + { + "bbox": [ + 114, + 669, + 358, + 683 + ], + "score": 1.0, + "content": "Artificial Intelligence (UAI), pages 504–513. Citeseer, 2015.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42 + }, + { + "type": "text", + "bbox": [ + 107, + 688, + 506, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "Shaocong Ma, Yi Zhou, and Shaofeng Zou. Variance-reduced off-policy TDC learning: Non-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 700, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 115, + 700, + 505, + 713 + ], + "score": 1.0, + "content": "asymptotic convergence analysis. In Proc. Advances in Neural Information Processing Systems", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 115, + 711, + 318, + 723 + ], + "spans": [ + { + "bbox": [ + 115, + 711, + 318, + 723 + ], + "score": 1.0, + "content": "(NeurIPS), volume 33, pages 14796–14806, 2020.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 72, + 504, + 106 + ], + "lines": [ + { + "bbox": [ + 105, + 71, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 71, + 505, + 86 + ], + "score": 1.0, + "content": "Gal Dalal, Balázs Szörényi, and Gugan Thoppe. A tale of two-timescale reinforcement learning", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 83, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 115, + 83, + 505, + 96 + ], + "score": 1.0, + "content": "with the tightest finite-time bound. In Proc. Conference on Artificial Intelligence (AAAI), pages", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 95, + 191, + 105 + ], + "spans": [ + { + "bbox": [ + 116, + 95, + 191, + 105 + ], + "score": 1.0, + "content": "3701–3708, 2020.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 71, + 505, + 105 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 113, + 504, + 147 + ], + "lines": [ + { + "bbox": [ + 105, + 112, + 506, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 112, + 506, + 127 + ], + "score": 1.0, + "content": "Dalal, Gal and Szörényi, Balázs and Thoppe, Gugan and Mannor, Shie. Finite sample analyses for", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 123, + 506, + 137 + ], + "spans": [ + { + "bbox": [ + 115, + 123, + 506, + 137 + ], + "score": 1.0, + "content": "TD(0) with function approximation. In Proc. AAAI Conference on Artificial Intelligence (AAAI),", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 136, + 216, + 146 + ], + "spans": [ + { + "bbox": [ + 115, + 136, + 216, + 146 + ], + "score": 1.0, + "content": "pages 6144–6160, 2018.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4, + "bbox_fs": [ + 105, + 112, + 506, + 146 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 154, + 504, + 177 + ], + "lines": [ + { + "bbox": [ + 106, + 154, + 505, + 167 + ], + "spans": [ + { + "bbox": [ + 106, + 154, + 505, + 167 + ], + "score": 1.0, + "content": "Thinh T Doan. Finite-time analysis and restarting scheme for linear two-time-scale stochastic", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 165, + 459, + 177 + ], + "spans": [ + { + "bbox": [ + 115, + 165, + 459, + 177 + ], + "score": 1.0, + "content": "approximation. SIAM Journal on Control and Optimization, 59(4):2798–2819, 2021.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5, + "bbox_fs": [ + 106, + 154, + 505, + 177 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 183, + 504, + 207 + ], + "lines": [ + { + "bbox": [ + 105, + 182, + 506, + 198 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 506, + 198 + ], + "score": 1.0, + "content": "Eyal Even-Dar, Yishay Mansour, and Peter Bartlett. Learning rates for Q-learning. Journal of", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 116, + 195, + 276, + 207 + ], + "spans": [ + { + "bbox": [ + 116, + 195, + 276, + 207 + ], + "score": 1.0, + "content": "machine learning Research, 5(1), 2003.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 182, + 506, + 207 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 213, + 505, + 248 + ], + "lines": [ + { + "bbox": [ + 105, + 213, + 505, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 505, + 227 + ], + "score": 1.0, + "content": "Harsh Gupta, R Srikant, and Lei Ying. Finite-time performance bounds and adaptive learning rate", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 225, + 505, + 237 + ], + "spans": [ + { + "bbox": [ + 116, + 225, + 505, + 237 + ], + "score": 1.0, + "content": "selection for two time-scale reinforcement learning. In Proc. Advances in Neural Information", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 116, + 236, + 343, + 248 + ], + "spans": [ + { + "bbox": [ + 116, + 236, + 343, + 248 + ], + "score": 1.0, + "content": "Processing Systems (NeurIPS), pages 4706–4715, 2019.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11, + "bbox_fs": [ + 105, + 213, + 505, + 248 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 254, + 503, + 277 + ], + "lines": [ + { + "bbox": [ + 105, + 254, + 505, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 505, + 267 + ], + "score": 1.0, + "content": "Linfang Hou, Liang Pang, Xin Hong, Yanyan Lan, Zhiming Ma, and Dawei Yin. Robust reinforcement", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 264, + 430, + 277 + ], + "spans": [ + { + "bbox": [ + 115, + 264, + 430, + 277 + ], + "score": 1.0, + "content": "learning with Wasserstein constraint. arXiv preprint arXiv:2006.00945, 2020.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5, + "bbox_fs": [ + 105, + 254, + 505, + 277 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 284, + 506, + 317 + ], + "lines": [ + { + "bbox": [ + 106, + 284, + 505, + 297 + ], + "spans": [ + { + "bbox": [ + 106, + 284, + 505, + 297 + ], + "score": 1.0, + "content": "Sandy Huang, Nicolas Papernot, Ian Goodfellow, Yan Duan, and Pieter Abbeel. Adversarial attacks", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 296, + 506, + 308 + ], + "spans": [ + { + "bbox": [ + 115, + 296, + 506, + 308 + ], + "score": 1.0, + "content": "on neural network policies. In Proc. International Conference on Learning Representations (ICLR),", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 305, + 142, + 317 + ], + "spans": [ + { + "bbox": [ + 115, + 305, + 142, + 317 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16, + "bbox_fs": [ + 106, + 284, + 506, + 317 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 324, + 505, + 348 + ], + "lines": [ + { + "bbox": [ + 105, + 323, + 507, + 339 + ], + "spans": [ + { + "bbox": [ + 105, + 323, + 507, + 339 + ], + "score": 1.0, + "content": "Peter J Huber. A robust version of the probability ratio test. The Annals of Mathematical Statistics,", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 336, + 216, + 348 + ], + "spans": [ + { + "bbox": [ + 115, + 336, + 216, + 348 + ], + "score": 1.0, + "content": "pages 1753–1758, 1965.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 323, + 507, + 348 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 354, + 504, + 378 + ], + "lines": [ + { + "bbox": [ + 106, + 354, + 506, + 368 + ], + "spans": [ + { + "bbox": [ + 106, + 354, + 506, + 368 + ], + "score": 1.0, + "content": "Garud N Iyengar. Robust dynamic programming. Mathematics of Operations Research, 30(2):", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 116, + 365, + 182, + 378 + ], + "spans": [ + { + "bbox": [ + 116, + 365, + 182, + 378 + ], + "score": 1.0, + "content": "257–280, 2005.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5, + "bbox_fs": [ + 106, + 354, + 506, + 378 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 384, + 506, + 419 + ], + "lines": [ + { + "bbox": [ + 106, + 384, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 505, + 397 + ], + "score": 1.0, + "content": "Maxim Kaledin, Eric Moulines, Alexey Naumov, Vladislav Tadic, and Hoi-To Wai. Finite time", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 116, + 395, + 505, + 407 + ], + "spans": [ + { + "bbox": [ + 116, + 395, + 505, + 407 + ], + "score": 1.0, + "content": "analysis of linear two-timescale stochastic approximation with Markovian noise. In Proc. Annual", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 116, + 406, + 412, + 419 + ], + "spans": [ + { + "bbox": [ + 116, + 406, + 412, + 419 + ], + "score": 1.0, + "content": "Conference on Learning Theory (CoLT), pages 2144–2203. PMLR, 2020.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23, + "bbox_fs": [ + 106, + 384, + 505, + 419 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 425, + 504, + 448 + ], + "lines": [ + { + "bbox": [ + 106, + 425, + 506, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 425, + 506, + 438 + ], + "score": 1.0, + "content": "Jernej Kos and Dawn Song. Delving into adversarial attacks on deep policies. In Proc. International", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 436, + 341, + 449 + ], + "spans": [ + { + "bbox": [ + 115, + 436, + 341, + 449 + ], + "score": 1.0, + "content": "Conference on Learning Representations (ICLR), 2017.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 106, + 425, + 506, + 449 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 455, + 506, + 488 + ], + "lines": [ + { + "bbox": [ + 105, + 454, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 505, + 468 + ], + "score": 1.0, + "content": "Harshat Kumar, Alec Koppel, and Alejandro Ribeiro. On the sample complexity of actor-critic", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 466, + 506, + 479 + ], + "spans": [ + { + "bbox": [ + 115, + 466, + 506, + 479 + ], + "score": 1.0, + "content": "method for reinforcement learning with function approximation. arXiv preprint arXiv:1910.08412,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 475, + 143, + 490 + ], + "spans": [ + { + "bbox": [ + 115, + 475, + 143, + 490 + ], + "score": 1.0, + "content": "2019.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 454, + 506, + 490 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 495, + 504, + 529 + ], + "lines": [ + { + "bbox": [ + 105, + 495, + 505, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 495, + 505, + 509 + ], + "score": 1.0, + "content": "Gen Li, Yuting Wei, Yuejie Chi, Yuantao Gu, and Yuxin Chen. Sample complexity of asynchronous", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 116, + 507, + 505, + 519 + ], + "spans": [ + { + "bbox": [ + 116, + 507, + 505, + 519 + ], + "score": 1.0, + "content": "Q-learning: Sharper analysis and variance reduction. In Proc. Advances in Neural Information", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 116, + 517, + 267, + 531 + ], + "spans": [ + { + "bbox": [ + 116, + 517, + 267, + 531 + ], + "score": 1.0, + "content": "Processing Systems (NeurIPS), 2020.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 495, + 505, + 531 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 536, + 504, + 560 + ], + "lines": [ + { + "bbox": [ + 106, + 536, + 505, + 550 + ], + "spans": [ + { + "bbox": [ + 106, + 536, + 505, + 550 + ], + "score": 1.0, + "content": "Gen Li, Changxiao Cai, Yuxin Chen, Yuantao Gu, Yuting Wei, and Yuejie Chi. Is Q-learning minimax", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 548, + 464, + 560 + ], + "spans": [ + { + "bbox": [ + 115, + 548, + 464, + 560 + ], + "score": 1.0, + "content": "optimal? A tight sample complexity analysis. arXiv preprint arXiv:2102.06548, 2021.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5, + "bbox_fs": [ + 106, + 536, + 505, + 560 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 566, + 505, + 600 + ], + "lines": [ + { + "bbox": [ + 105, + 566, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 566, + 505, + 579 + ], + "score": 1.0, + "content": "Shiau Hong Lim, Huan Xu, and Shie Mannor. Reinforcement learning in robust Markov decision", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 578, + 506, + 590 + ], + "spans": [ + { + "bbox": [ + 115, + 578, + 506, + 590 + ], + "score": 1.0, + "content": "processes. In Proc. Advances in Neural Information Processing Systems (NIPS), pages 701–709,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 114, + 587, + 142, + 600 + ], + "spans": [ + { + "bbox": [ + 114, + 587, + 142, + 600 + ], + "score": 1.0, + "content": "2013.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36, + "bbox_fs": [ + 105, + 566, + 506, + 600 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 606, + 505, + 641 + ], + "lines": [ + { + "bbox": [ + 106, + 606, + 506, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 606, + 506, + 621 + ], + "score": 1.0, + "content": "Yen-Chen Lin, Zhang-Wei Hong, Yuan-Hong Liao, Meng-Li Shih, Ming-Yu Liu, and Min Sun.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 618, + 505, + 631 + ], + "spans": [ + { + "bbox": [ + 115, + 618, + 505, + 631 + ], + "score": 1.0, + "content": "Tactics of adversarial attack on deep reinforcement learning agents. In Proc. International Joint", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 117, + 629, + 403, + 642 + ], + "spans": [ + { + "bbox": [ + 117, + 629, + 403, + 642 + ], + "score": 1.0, + "content": "Conferences on Artificial Intelligence (IJCAI), pages 3756–3762, 2017.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39, + "bbox_fs": [ + 106, + 606, + 506, + 642 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 647, + 504, + 682 + ], + "lines": [ + { + "bbox": [ + 106, + 648, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 106, + 648, + 505, + 660 + ], + "score": 1.0, + "content": "Bo Liu, Ji Liu, Mohammad Ghavamzadeh, Sridhar Mahadevan, and Marek Petrik. Finite-sample", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 659, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 116, + 659, + 505, + 672 + ], + "score": 1.0, + "content": "analysis of proximal gradient TD algorithms. In Proc. International Conference on Uncertainty in", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 114, + 669, + 358, + 683 + ], + "spans": [ + { + "bbox": [ + 114, + 669, + 358, + 683 + ], + "score": 1.0, + "content": "Artificial Intelligence (UAI), pages 504–513. Citeseer, 2015.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42, + "bbox_fs": [ + 106, + 648, + 505, + 683 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 688, + 506, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "Shaocong Ma, Yi Zhou, and Shaofeng Zou. Variance-reduced off-policy TDC learning: Non-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 115, + 700, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 115, + 700, + 505, + 713 + ], + "score": 1.0, + "content": "asymptotic convergence analysis. In Proc. Advances in Neural Information Processing Systems", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 115, + 711, + 318, + 723 + ], + "spans": [ + { + "bbox": [ + 115, + 711, + 318, + 723 + ], + "score": 1.0, + "content": "(NeurIPS), volume 33, pages 14796–14806, 2020.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45, + "bbox_fs": [ + 106, + 688, + 506, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 72, + 505, + 106 + ], + "lines": [ + { + "bbox": [ + 106, + 73, + 505, + 84 + ], + "spans": [ + { + "bbox": [ + 106, + 73, + 505, + 84 + ], + "score": 1.0, + "content": "Shaocong Ma, Yi Zhou, and Shaofeng Zou. Greedy-GQ with variance reduction: Finite-time analysis", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 84, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 115, + 84, + 506, + 96 + ], + "score": 1.0, + "content": "and improved complexity. In Proc. International Conference on Learning Representations (ICLR),", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 94, + 142, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 94, + 142, + 106 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 106, + 114, + 504, + 137 + ], + "lines": [ + { + "bbox": [ + 105, + 112, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 112, + 506, + 128 + ], + "score": 1.0, + "content": "Hamid Reza Maei. Gradient temporal-difference learning algorithms. Thesis, University of Alberta,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 114, + 124, + 143, + 138 + ], + "spans": [ + { + "bbox": [ + 114, + 124, + 143, + 138 + ], + "score": 1.0, + "content": "2011.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 106, + 145, + 506, + 179 + ], + "lines": [ + { + "bbox": [ + 104, + 143, + 506, + 160 + ], + "spans": [ + { + "bbox": [ + 104, + 143, + 506, + 160 + ], + "score": 1.0, + "content": "Hamid Reza Maei, Csaba Szepesvári, Shalabh Bhatnagar, and Richard S Sutton. Toward off-policy", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 156, + 505, + 169 + ], + "spans": [ + { + "bbox": [ + 116, + 156, + 505, + 169 + ], + "score": 1.0, + "content": "learning control with function approximation. In Proc. International Conference on Machine", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 116, + 168, + 280, + 180 + ], + "spans": [ + { + "bbox": [ + 116, + 168, + 280, + 180 + ], + "score": 1.0, + "content": "Learning (ICML), pages 719–726, 2010.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 107, + 187, + 506, + 231 + ], + "lines": [ + { + "bbox": [ + 105, + 187, + 506, + 200 + ], + "spans": [ + { + "bbox": [ + 105, + 187, + 506, + 200 + ], + "score": 1.0, + "content": "Ajay Mandlekar, Yuke Zhu, Animesh Garg, Li Fei-Fei, and Silvio Savarese. Adversarially robust", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 199, + 506, + 211 + ], + "spans": [ + { + "bbox": [ + 115, + 199, + 506, + 211 + ], + "score": 1.0, + "content": "policy learning: Active construction of physically-plausible perturbations. In 2017 IEEE/RSJ", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 209, + 506, + 222 + ], + "spans": [ + { + "bbox": [ + 115, + 209, + 506, + 222 + ], + "score": 1.0, + "content": "International Conference on Intelligent Robots and Systems (IROS), pages 3932–3939. IEEE,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 114, + 219, + 143, + 232 + ], + "spans": [ + { + "bbox": [ + 114, + 219, + 143, + 232 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 107, + 239, + 504, + 273 + ], + "lines": [ + { + "bbox": [ + 105, + 239, + 506, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 506, + 253 + ], + "score": 1.0, + "content": "Jincheng Mei, Chenjun Xiao, Csaba Szepesvari, and Dale Schuurmans. On the global convergence", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 250, + 506, + 265 + ], + "spans": [ + { + "bbox": [ + 115, + 250, + 506, + 265 + ], + "score": 1.0, + "content": "rates of softmax policy gradient methods. In Proc. International Conference on Machine Learning", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 263, + 283, + 274 + ], + "spans": [ + { + "bbox": [ + 115, + 263, + 283, + 274 + ], + "score": 1.0, + "content": "(ICML), pages 6820–6829. PMLR, 2020.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 105, + 281, + 505, + 305 + ], + "lines": [ + { + "bbox": [ + 105, + 281, + 506, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 506, + 295 + ], + "score": 1.0, + "content": "Jun Morimoto and Kenji Doya. Robust reinforcement learning. Neural computation, 17(2):335–359,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 114, + 292, + 143, + 306 + ], + "spans": [ + { + "bbox": [ + 114, + 292, + 143, + 306 + ], + "score": 1.0, + "content": "2005.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 107, + 312, + 506, + 347 + ], + "lines": [ + { + "bbox": [ + 106, + 313, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 106, + 313, + 505, + 325 + ], + "score": 1.0, + "content": "Arnab Nilim and Laurent El Ghaoui. Robustness in Markov decision problems with uncertain", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 323, + 506, + 338 + ], + "spans": [ + { + "bbox": [ + 115, + 323, + 506, + 338 + ], + "score": 1.0, + "content": "transition matrices. In Proc. Advances in Neural Information Processing Systems (NIPS), pages", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 116, + 335, + 181, + 347 + ], + "spans": [ + { + "bbox": [ + 116, + 335, + 181, + 347 + ], + "score": 1.0, + "content": "839–846, 2004.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 354, + 505, + 389 + ], + "lines": [ + { + "bbox": [ + 106, + 355, + 505, + 368 + ], + "spans": [ + { + "bbox": [ + 106, + 355, + 505, + 368 + ], + "score": 1.0, + "content": "Anay Pattanaik, Zhenyi Tang, Shuijing Liu, Gautham Bommannan, and Girish Chowdhary. Robust", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 366, + 505, + 379 + ], + "spans": [ + { + "bbox": [ + 115, + 366, + 505, + 379 + ], + "score": 1.0, + "content": "deep reinforcement learning with adversarial attacks. In Proc. International Conference on", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 114, + 376, + 398, + 389 + ], + "spans": [ + { + "bbox": [ + 114, + 376, + 398, + 389 + ], + "score": 1.0, + "content": "Autonomous Agents and MultiAgent Systems, pages 2040–2042, 2018.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 396, + 506, + 430 + ], + "lines": [ + { + "bbox": [ + 105, + 396, + 506, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 506, + 410 + ], + "score": 1.0, + "content": "Lerrel Pinto, James Davidson, Rahul Sukthankar, and Abhinav Gupta. Robust adversarial reinforce-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 116, + 408, + 506, + 420 + ], + "spans": [ + { + "bbox": [ + 116, + 408, + 506, + 420 + ], + "score": 1.0, + "content": "ment learning. In Proc. International Conference on Machine Learning (ICML), pages 2817–2826.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 116, + 419, + 174, + 430 + ], + "spans": [ + { + "bbox": [ + 116, + 419, + 174, + 430 + ], + "score": 1.0, + "content": "PMLR, 2017.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 438, + 505, + 473 + ], + "lines": [ + { + "bbox": [ + 105, + 438, + 506, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 506, + 452 + ], + "score": 1.0, + "content": "Shuang Qiu, Zhuoran Yang, Jieping Ye, and Zhaoran Wang. On the finite-time convergence of", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 449, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 115, + 449, + 505, + 464 + ], + "score": 1.0, + "content": "actor-critic algorithm. In Proc. Optimization Foundations for Reinforcement Learning Workshop", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 460, + 410, + 474 + ], + "spans": [ + { + "bbox": [ + 115, + 460, + 410, + 474 + ], + "score": 1.0, + "content": "at Advances in Neural Information Processing Systems (NeurIPS), 2019.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 107, + 480, + 506, + 514 + ], + "lines": [ + { + "bbox": [ + 106, + 480, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 106, + 480, + 505, + 493 + ], + "score": 1.0, + "content": "Guannan Qu and Adam Wierman. Finite-time analysis of asynchronous stochastic approximation and", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 116, + 492, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 116, + 492, + 506, + 504 + ], + "score": 1.0, + "content": "Q-learning. In Proc. Annual Conference on Learning Theory (CoLT), pages 3185–3205. PMLR,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 501, + 142, + 515 + ], + "spans": [ + { + "bbox": [ + 115, + 501, + 142, + 515 + ], + "score": 1.0, + "content": "2020.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 106, + 522, + 506, + 556 + ], + "lines": [ + { + "bbox": [ + 105, + 520, + 506, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 506, + 537 + ], + "score": 1.0, + "content": "Aravind Rajeswaran, Sarvjeet Ghotra, Balaraman Ravindran, and Sergey Levine. Epopt: Learning", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 116, + 534, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 116, + 534, + 505, + 546 + ], + "score": 1.0, + "content": "robust neural network policies using model ensembles. In Proc. International Conference on", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 545, + 280, + 557 + ], + "spans": [ + { + "bbox": [ + 116, + 545, + 280, + 557 + ], + "score": 1.0, + "content": "Learning Representations (ICLR), 2017.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 106, + 564, + 505, + 588 + ], + "lines": [ + { + "bbox": [ + 107, + 565, + 504, + 577 + ], + "spans": [ + { + "bbox": [ + 107, + 565, + 504, + 577 + ], + "score": 1.0, + "content": "Aurko Roy, Huan Xu, and Sebastian Pokutta. Reinforcement learning under model mismatch. In", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 116, + 576, + 484, + 588 + ], + "spans": [ + { + "bbox": [ + 116, + 576, + 484, + 588 + ], + "score": 1.0, + "content": "Proc. Advances in Neural Information Processing Systems (NIPS), pages 3046–3055, 2017.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 104, + 595, + 505, + 618 + ], + "lines": [ + { + "bbox": [ + 106, + 595, + 506, + 608 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 506, + 608 + ], + "score": 1.0, + "content": "Jay K Satia and Roy E Lave Jr. Markovian decision processes with uncertain transition probabilities.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 607, + 294, + 618 + ], + "spans": [ + { + "bbox": [ + 116, + 607, + 294, + 618 + ], + "score": 1.0, + "content": "Operations Research, 21(3):728–740, 1973.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 104, + 626, + 506, + 650 + ], + "lines": [ + { + "bbox": [ + 105, + 625, + 507, + 641 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 507, + 641 + ], + "score": 1.0, + "content": "R. Srikant and Lei Ying. Finite-time error bounds for linear stochastic approximation and TD learning.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 637, + 447, + 650 + ], + "spans": [ + { + "bbox": [ + 115, + 637, + 447, + 650 + ], + "score": 1.0, + "content": "In Proc. Annual Conference on Learning Theory (CoLT), pages 2803–2830, 2019.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5 + }, + { + "type": "text", + "bbox": [ + 107, + 657, + 504, + 691 + ], + "lines": [ + { + "bbox": [ + 105, + 657, + 506, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 657, + 506, + 672 + ], + "score": 1.0, + "content": "Jun Sun, Gang Wang, Georgios B Giannakis, Qinmin Yang, and Zaiyue Yang. Finite-sample", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 669, + 506, + 682 + ], + "spans": [ + { + "bbox": [ + 116, + 669, + 506, + 682 + ], + "score": 1.0, + "content": "analysis of decentralized temporal-difference learning with linear function approximation. In Proc.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 680, + 444, + 692 + ], + "spans": [ + { + "bbox": [ + 116, + 680, + 444, + 692 + ], + "score": 1.0, + "content": "International Conference on Artifical Intelligence and Statistics (AISTATS), 2020.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42 + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 504, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 504, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 504, + 712 + ], + "score": 1.0, + "content": "Richard S. Sutton and Andrew G. Barto. Reinforcement Learning: An Introduction, Second Edition.", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 117, + 711, + 317, + 722 + ], + "spans": [ + { + "bbox": [ + 117, + 711, + 317, + 722 + ], + "score": 1.0, + "content": "The MIT Press, Cambridge, Massachusetts, 2018.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44.5 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "13", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 72, + 505, + 106 + ], + "lines": [ + { + "bbox": [ + 106, + 73, + 505, + 84 + ], + "spans": [ + { + "bbox": [ + 106, + 73, + 505, + 84 + ], + "score": 1.0, + "content": "Shaocong Ma, Yi Zhou, and Shaofeng Zou. Greedy-GQ with variance reduction: Finite-time analysis", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 84, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 115, + 84, + 506, + 96 + ], + "score": 1.0, + "content": "and improved complexity. In Proc. International Conference on Learning Representations (ICLR),", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 94, + 142, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 94, + 142, + 106 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 106, + 73, + 506, + 106 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 114, + 504, + 137 + ], + "lines": [ + { + "bbox": [ + 105, + 112, + 506, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 112, + 506, + 128 + ], + "score": 1.0, + "content": "Hamid Reza Maei. Gradient temporal-difference learning algorithms. Thesis, University of Alberta,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 114, + 124, + 143, + 138 + ], + "spans": [ + { + "bbox": [ + 114, + 124, + 143, + 138 + ], + "score": 1.0, + "content": "2011.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 105, + 112, + 506, + 138 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 145, + 506, + 179 + ], + "lines": [ + { + "bbox": [ + 104, + 143, + 506, + 160 + ], + "spans": [ + { + "bbox": [ + 104, + 143, + 506, + 160 + ], + "score": 1.0, + "content": "Hamid Reza Maei, Csaba Szepesvári, Shalabh Bhatnagar, and Richard S Sutton. Toward off-policy", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 156, + 505, + 169 + ], + "spans": [ + { + "bbox": [ + 116, + 156, + 505, + 169 + ], + "score": 1.0, + "content": "learning control with function approximation. In Proc. International Conference on Machine", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 116, + 168, + 280, + 180 + ], + "spans": [ + { + "bbox": [ + 116, + 168, + 280, + 180 + ], + "score": 1.0, + "content": "Learning (ICML), pages 719–726, 2010.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6, + "bbox_fs": [ + 104, + 143, + 506, + 180 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 187, + 506, + 231 + ], + "lines": [ + { + "bbox": [ + 105, + 187, + 506, + 200 + ], + "spans": [ + { + "bbox": [ + 105, + 187, + 506, + 200 + ], + "score": 1.0, + "content": "Ajay Mandlekar, Yuke Zhu, Animesh Garg, Li Fei-Fei, and Silvio Savarese. Adversarially robust", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 115, + 199, + 506, + 211 + ], + "spans": [ + { + "bbox": [ + 115, + 199, + 506, + 211 + ], + "score": 1.0, + "content": "policy learning: Active construction of physically-plausible perturbations. In 2017 IEEE/RSJ", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 209, + 506, + 222 + ], + "spans": [ + { + "bbox": [ + 115, + 209, + 506, + 222 + ], + "score": 1.0, + "content": "International Conference on Intelligent Robots and Systems (IROS), pages 3932–3939. IEEE,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 114, + 219, + 143, + 232 + ], + "spans": [ + { + "bbox": [ + 114, + 219, + 143, + 232 + ], + "score": 1.0, + "content": "2017.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 187, + 506, + 232 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 239, + 504, + 273 + ], + "lines": [ + { + "bbox": [ + 105, + 239, + 506, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 239, + 506, + 253 + ], + "score": 1.0, + "content": "Jincheng Mei, Chenjun Xiao, Csaba Szepesvari, and Dale Schuurmans. On the global convergence", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 250, + 506, + 265 + ], + "spans": [ + { + "bbox": [ + 115, + 250, + 506, + 265 + ], + "score": 1.0, + "content": "rates of softmax policy gradient methods. In Proc. International Conference on Machine Learning", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 263, + 283, + 274 + ], + "spans": [ + { + "bbox": [ + 115, + 263, + 283, + 274 + ], + "score": 1.0, + "content": "(ICML), pages 6820–6829. PMLR, 2020.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13, + "bbox_fs": [ + 105, + 239, + 506, + 274 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 281, + 505, + 305 + ], + "lines": [ + { + "bbox": [ + 105, + 281, + 506, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 506, + 295 + ], + "score": 1.0, + "content": "Jun Morimoto and Kenji Doya. Robust reinforcement learning. Neural computation, 17(2):335–359,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 114, + 292, + 143, + 306 + ], + "spans": [ + { + "bbox": [ + 114, + 292, + 143, + 306 + ], + "score": 1.0, + "content": "2005.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 281, + 506, + 306 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 312, + 506, + 347 + ], + "lines": [ + { + "bbox": [ + 106, + 313, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 106, + 313, + 505, + 325 + ], + "score": 1.0, + "content": "Arnab Nilim and Laurent El Ghaoui. Robustness in Markov decision problems with uncertain", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 323, + 506, + 338 + ], + "spans": [ + { + "bbox": [ + 115, + 323, + 506, + 338 + ], + "score": 1.0, + "content": "transition matrices. In Proc. Advances in Neural Information Processing Systems (NIPS), pages", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 116, + 335, + 181, + 347 + ], + "spans": [ + { + "bbox": [ + 116, + 335, + 181, + 347 + ], + "score": 1.0, + "content": "839–846, 2004.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18, + "bbox_fs": [ + 106, + 313, + 506, + 347 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 354, + 505, + 389 + ], + "lines": [ + { + "bbox": [ + 106, + 355, + 505, + 368 + ], + "spans": [ + { + "bbox": [ + 106, + 355, + 505, + 368 + ], + "score": 1.0, + "content": "Anay Pattanaik, Zhenyi Tang, Shuijing Liu, Gautham Bommannan, and Girish Chowdhary. Robust", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 366, + 505, + 379 + ], + "spans": [ + { + "bbox": [ + 115, + 366, + 505, + 379 + ], + "score": 1.0, + "content": "deep reinforcement learning with adversarial attacks. In Proc. International Conference on", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 114, + 376, + 398, + 389 + ], + "spans": [ + { + "bbox": [ + 114, + 376, + 398, + 389 + ], + "score": 1.0, + "content": "Autonomous Agents and MultiAgent Systems, pages 2040–2042, 2018.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21, + "bbox_fs": [ + 106, + 355, + 505, + 389 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 396, + 506, + 430 + ], + "lines": [ + { + "bbox": [ + 105, + 396, + 506, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 506, + 410 + ], + "score": 1.0, + "content": "Lerrel Pinto, James Davidson, Rahul Sukthankar, and Abhinav Gupta. Robust adversarial reinforce-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 116, + 408, + 506, + 420 + ], + "spans": [ + { + "bbox": [ + 116, + 408, + 506, + 420 + ], + "score": 1.0, + "content": "ment learning. In Proc. International Conference on Machine Learning (ICML), pages 2817–2826.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 116, + 419, + 174, + 430 + ], + "spans": [ + { + "bbox": [ + 116, + 419, + 174, + 430 + ], + "score": 1.0, + "content": "PMLR, 2017.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 396, + 506, + 430 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 438, + 505, + 473 + ], + "lines": [ + { + "bbox": [ + 105, + 438, + 506, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 506, + 452 + ], + "score": 1.0, + "content": "Shuang Qiu, Zhuoran Yang, Jieping Ye, and Zhaoran Wang. On the finite-time convergence of", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 449, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 115, + 449, + 505, + 464 + ], + "score": 1.0, + "content": "actor-critic algorithm. In Proc. Optimization Foundations for Reinforcement Learning Workshop", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 115, + 460, + 410, + 474 + ], + "spans": [ + { + "bbox": [ + 115, + 460, + 410, + 474 + ], + "score": 1.0, + "content": "at Advances in Neural Information Processing Systems (NeurIPS), 2019.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27, + "bbox_fs": [ + 105, + 438, + 506, + 474 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 480, + 506, + 514 + ], + "lines": [ + { + "bbox": [ + 106, + 480, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 106, + 480, + 505, + 493 + ], + "score": 1.0, + "content": "Guannan Qu and Adam Wierman. Finite-time analysis of asynchronous stochastic approximation and", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 116, + 492, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 116, + 492, + 506, + 504 + ], + "score": 1.0, + "content": "Q-learning. In Proc. Annual Conference on Learning Theory (CoLT), pages 3185–3205. PMLR,", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 501, + 142, + 515 + ], + "spans": [ + { + "bbox": [ + 115, + 501, + 142, + 515 + ], + "score": 1.0, + "content": "2020.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30, + "bbox_fs": [ + 106, + 480, + 506, + 515 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 522, + 506, + 556 + ], + "lines": [ + { + "bbox": [ + 105, + 520, + 506, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 506, + 537 + ], + "score": 1.0, + "content": "Aravind Rajeswaran, Sarvjeet Ghotra, Balaraman Ravindran, and Sergey Levine. Epopt: Learning", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 116, + 534, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 116, + 534, + 505, + 546 + ], + "score": 1.0, + "content": "robust neural network policies using model ensembles. In Proc. International Conference on", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 545, + 280, + 557 + ], + "spans": [ + { + "bbox": [ + 116, + 545, + 280, + 557 + ], + "score": 1.0, + "content": "Learning Representations (ICLR), 2017.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33, + "bbox_fs": [ + 105, + 520, + 506, + 557 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 564, + 505, + 588 + ], + "lines": [ + { + "bbox": [ + 107, + 565, + 504, + 577 + ], + "spans": [ + { + "bbox": [ + 107, + 565, + 504, + 577 + ], + "score": 1.0, + "content": "Aurko Roy, Huan Xu, and Sebastian Pokutta. Reinforcement learning under model mismatch. In", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 116, + 576, + 484, + 588 + ], + "spans": [ + { + "bbox": [ + 116, + 576, + 484, + 588 + ], + "score": 1.0, + "content": "Proc. Advances in Neural Information Processing Systems (NIPS), pages 3046–3055, 2017.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5, + "bbox_fs": [ + 107, + 565, + 504, + 588 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 595, + 505, + 618 + ], + "lines": [ + { + "bbox": [ + 106, + 595, + 506, + 608 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 506, + 608 + ], + "score": 1.0, + "content": "Jay K Satia and Roy E Lave Jr. Markovian decision processes with uncertain transition probabilities.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 607, + 294, + 618 + ], + "spans": [ + { + "bbox": [ + 116, + 607, + 294, + 618 + ], + "score": 1.0, + "content": "Operations Research, 21(3):728–740, 1973.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5, + "bbox_fs": [ + 106, + 595, + 506, + 618 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 626, + 506, + 650 + ], + "lines": [ + { + "bbox": [ + 105, + 625, + 507, + 641 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 507, + 641 + ], + "score": 1.0, + "content": "R. Srikant and Lei Ying. Finite-time error bounds for linear stochastic approximation and TD learning.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 637, + 447, + 650 + ], + "spans": [ + { + "bbox": [ + 115, + 637, + 447, + 650 + ], + "score": 1.0, + "content": "In Proc. Annual Conference on Learning Theory (CoLT), pages 2803–2830, 2019.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5, + "bbox_fs": [ + 105, + 625, + 507, + 650 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 657, + 504, + 691 + ], + "lines": [ + { + "bbox": [ + 105, + 657, + 506, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 657, + 506, + 672 + ], + "score": 1.0, + "content": "Jun Sun, Gang Wang, Georgios B Giannakis, Qinmin Yang, and Zaiyue Yang. Finite-sample", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 669, + 506, + 682 + ], + "spans": [ + { + "bbox": [ + 116, + 669, + 506, + 682 + ], + "score": 1.0, + "content": "analysis of decentralized temporal-difference learning with linear function approximation. In Proc.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 680, + 444, + 692 + ], + "spans": [ + { + "bbox": [ + 116, + 680, + 444, + 692 + ], + "score": 1.0, + "content": "International Conference on Artifical Intelligence and Statistics (AISTATS), 2020.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42, + "bbox_fs": [ + 105, + 657, + 506, + 692 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 504, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 504, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 504, + 712 + ], + "score": 1.0, + "content": "Richard S. Sutton and Andrew G. Barto. Reinforcement Learning: An Introduction, Second Edition.", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 117, + 711, + 317, + 722 + ], + "spans": [ + { + "bbox": [ + 117, + 711, + 317, + 722 + ], + "score": 1.0, + "content": "The MIT Press, Cambridge, Massachusetts, 2018.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44.5, + "bbox_fs": [ + 106, + 699, + 504, + 722 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 72, + 504, + 106 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 506, + 86 + ], + "score": 1.0, + "content": "Richard S Sutton, Csaba Szepesvári, and Hamid Reza Maei. A convergent O(n) algorithm for", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 84, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 116, + 84, + 505, + 96 + ], + "score": 1.0, + "content": "off-policy temporal-difference learning with linear function approximation. In Proc. Advances in", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 94, + 504, + 108 + ], + "spans": [ + { + "bbox": [ + 115, + 94, + 504, + 108 + ], + "score": 1.0, + "content": "Neural Information Processing Systems (NIPS), volume 21, pages 1609–1616. MIT Press, 2008.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 106, + 112, + 505, + 157 + ], + "lines": [ + { + "bbox": [ + 106, + 113, + 505, + 125 + ], + "spans": [ + { + "bbox": [ + 106, + 113, + 505, + 125 + ], + "score": 1.0, + "content": "Richard S Sutton, Hamid Reza Maei, Doina Precup, Shalabh Bhatnagar, David Silver, Csaba", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 116, + 124, + 506, + 136 + ], + "spans": [ + { + "bbox": [ + 116, + 124, + 506, + 136 + ], + "score": 1.0, + "content": "Szepesvári, and Eric Wiewiora. Fast gradient-descent methods for temporal-difference learn-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 135, + 505, + 148 + ], + "spans": [ + { + "bbox": [ + 116, + 135, + 505, + 148 + ], + "score": 1.0, + "content": "ing with linear function approximation. In Proc. International Conference on Machine Learning", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 146, + 246, + 159 + ], + "spans": [ + { + "bbox": [ + 115, + 146, + 246, + 159 + ], + "score": 1.0, + "content": "(ICML), pages 993–1000, 2009.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 107, + 164, + 503, + 188 + ], + "lines": [ + { + "bbox": [ + 106, + 164, + 505, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 164, + 505, + 177 + ], + "score": 1.0, + "content": "Aviv Tamar, Shie Mannor, and Huan Xu. Scaling up robust MDPs using function approximation. In", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 175, + 490, + 188 + ], + "spans": [ + { + "bbox": [ + 115, + 175, + 490, + 188 + ], + "score": 1.0, + "content": "Proc. International Conference on Machine Learning (ICML), pages 181–189. PMLR, 2014.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 106, + 194, + 505, + 227 + ], + "lines": [ + { + "bbox": [ + 105, + 193, + 506, + 208 + ], + "spans": [ + { + "bbox": [ + 105, + 193, + 506, + 208 + ], + "score": 1.0, + "content": "Eugene Vinitsky, Yuqing Du, Kanaad Parvate, Kathy Jang, Pieter Abbeel, and Alexandre Bayen.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 205, + 507, + 218 + ], + "spans": [ + { + "bbox": [ + 115, + 205, + 507, + 218 + ], + "score": 1.0, + "content": "Robust reinforcement learning using adversarial populations. arXiv preprint arXiv:2008.01825,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 214, + 142, + 229 + ], + "spans": [ + { + "bbox": [ + 115, + 214, + 142, + 229 + ], + "score": 1.0, + "content": "2020.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 234, + 506, + 268 + ], + "lines": [ + { + "bbox": [ + 105, + 234, + 505, + 248 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 505, + 248 + ], + "score": 1.0, + "content": "Hoi-To Wai, Mingyi Hong, Zhuoran Yang, Zhaoran Wang, and Kexin Tang. Variance reduced", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 246, + 505, + 259 + ], + "spans": [ + { + "bbox": [ + 115, + 246, + 505, + 259 + ], + "score": 1.0, + "content": "policy evaluation with smooth function approximation. In Proc. Advances in Neural Information", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 116, + 257, + 389, + 269 + ], + "spans": [ + { + "bbox": [ + 116, + 257, + 389, + 269 + ], + "score": 1.0, + "content": "Processing Systems (NeurIPS), volume 32, pages 5784–5795, 2019.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 106, + 275, + 505, + 298 + ], + "lines": [ + { + "bbox": [ + 105, + 274, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 105, + 274, + 505, + 288 + ], + "score": 1.0, + "content": "Martin J Wainwright. Variance-reduced Q-learning is minimax optimal. arXiv preprint", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 286, + 219, + 297 + ], + "spans": [ + { + "bbox": [ + 115, + 286, + 219, + 297 + ], + "score": 1.0, + "content": "arXiv:1906.04697, 2019.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 106, + 304, + 505, + 338 + ], + "lines": [ + { + "bbox": [ + 105, + 304, + 506, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 506, + 318 + ], + "score": 1.0, + "content": "Lingxiao Wang, Qi Cai, Zhuoran Yang, and Zhaoran Wang. Neural policy gradient methods:", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 315, + 505, + 330 + ], + "spans": [ + { + "bbox": [ + 115, + 315, + 505, + 330 + ], + "score": 1.0, + "content": "Global optimality and rates of convergence. In Proc. International Conference on Learning", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 116, + 327, + 241, + 339 + ], + "spans": [ + { + "bbox": [ + 116, + 327, + 241, + 339 + ], + "score": 1.0, + "content": "Representations (ICLR), 2020.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 345, + 506, + 379 + ], + "lines": [ + { + "bbox": [ + 106, + 345, + 506, + 359 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 506, + 359 + ], + "score": 1.0, + "content": "Yue Wang and Shaofeng Zou. Finite-sample analysis of Greedy-GQ with linear function approxi-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 356, + 506, + 370 + ], + "spans": [ + { + "bbox": [ + 115, + 356, + 506, + 370 + ], + "score": 1.0, + "content": "mation under Markovian noise. In Proc. International Conference on Uncertainty in Artificial", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 367, + 306, + 381 + ], + "spans": [ + { + "bbox": [ + 115, + 367, + 306, + 381 + ], + "score": 1.0, + "content": "Intelligence (UAI), pages 11–20. PMLR, 2020.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 385, + 505, + 419 + ], + "lines": [ + { + "bbox": [ + 107, + 386, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 107, + 386, + 505, + 398 + ], + "score": 1.0, + "content": "Yue Wang, Shaofeng Zou, and Yi Zhou. Non-asymptotic analysis for two time-scale TDC with", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 396, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 115, + 396, + 505, + 411 + ], + "score": 1.0, + "content": "general smooth function approximation. In Proc. Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 116, + 408, + 269, + 420 + ], + "spans": [ + { + "bbox": [ + 116, + 408, + 269, + 420 + ], + "score": 1.0, + "content": "Systems (NeurIPS), volume 34, 2021.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 426, + 504, + 450 + ], + "lines": [ + { + "bbox": [ + 106, + 425, + 505, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 425, + 505, + 439 + ], + "score": 1.0, + "content": "Wolfram Wiesemann, Daniel Kuhn, and Berç Rustem. Robust Markov decision processes. Mathe-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 438, + 333, + 449 + ], + "spans": [ + { + "bbox": [ + 116, + 438, + 333, + 449 + ], + "score": 1.0, + "content": "matics of Operations Research, 38(1):153–183, 2013.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 106, + 456, + 504, + 479 + ], + "lines": [ + { + "bbox": [ + 107, + 456, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 107, + 456, + 505, + 469 + ], + "score": 1.0, + "content": "Yue Wu, Weitong Zhang, Pan Xu, and Quanquan Gu. A finite time analysis of two time-scale actor", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 116, + 467, + 341, + 479 + ], + "spans": [ + { + "bbox": [ + 116, + 467, + 341, + 479 + ], + "score": 1.0, + "content": "critic methods. arXiv preprint arXiv:2005.01350, 2020.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 107, + 486, + 504, + 520 + ], + "lines": [ + { + "bbox": [ + 105, + 486, + 506, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 506, + 499 + ], + "score": 1.0, + "content": "Tengyu Xu and Yingbin Liang. Sample complexity bounds for two timescale value-based reinforce-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 497, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 115, + 497, + 505, + 510 + ], + "score": 1.0, + "content": "ment learning algorithms. In Proc. International Conference on Artifical Intelligence and Statistics", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 508, + 286, + 520 + ], + "spans": [ + { + "bbox": [ + 115, + 508, + 286, + 520 + ], + "score": 1.0, + "content": "(AISTATS), pages 811–819. PMLR, 2021.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 107, + 526, + 504, + 560 + ], + "lines": [ + { + "bbox": [ + 106, + 526, + 506, + 540 + ], + "spans": [ + { + "bbox": [ + 106, + 526, + 506, + 540 + ], + "score": 1.0, + "content": "Tengyu Xu, Shaofeng Zou, and Yingbin Liang. Two time-scale off-policy TD learning: Non-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 114, + 536, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 114, + 536, + 506, + 551 + ], + "score": 1.0, + "content": "asymptotic analysis over Markovian samples. In Proc. Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 548, + 307, + 560 + ], + "spans": [ + { + "bbox": [ + 115, + 548, + 307, + 560 + ], + "score": 1.0, + "content": "Systems (NeurIPS), pages 10633–10643, 2019.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 108, + 566, + 505, + 601 + ], + "lines": [ + { + "bbox": [ + 106, + 566, + 506, + 580 + ], + "spans": [ + { + "bbox": [ + 106, + 566, + 506, + 580 + ], + "score": 1.0, + "content": "Zhuoran Yang, Yongxin Chen, Mingyi Hong, and Zhaoran Wang. Provably global convergence of", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 578, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 115, + 578, + 505, + 591 + ], + "score": 1.0, + "content": "actor-critic: A case for linear quadratic regulator with ergodic cost. In Proc. Advances in Neural", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 115, + 588, + 393, + 602 + ], + "spans": [ + { + "bbox": [ + 115, + 588, + 393, + 602 + ], + "score": 1.0, + "content": "Information Processing Systems (NeurIPS), pages 8353–8365, 2019.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 107, + 607, + 504, + 641 + ], + "lines": [ + { + "bbox": [ + 106, + 607, + 506, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 607, + 506, + 621 + ], + "score": 1.0, + "content": "Kaiqing Zhang, Bin Hu, and Tamer Basar. On the stability and convergence of robust adversarial", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 619, + 505, + 631 + ], + "spans": [ + { + "bbox": [ + 115, + 619, + 505, + 631 + ], + "score": 1.0, + "content": "reinforcement learning: A case study on linear quadratic systems. In Proc. Advances in Neural", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 629, + 370, + 643 + ], + "spans": [ + { + "bbox": [ + 115, + 629, + 370, + 643 + ], + "score": 1.0, + "content": "Information Processing Systems (NeurIPS), volume 33, 2020a.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 107, + 648, + 504, + 682 + ], + "lines": [ + { + "bbox": [ + 105, + 648, + 506, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 506, + 660 + ], + "score": 1.0, + "content": "Kaiqing Zhang, Tao Sun, Yunzhe Tao, Sahika Genc, Sunil Mallya, and Tamer Basar. Robust multi-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 659, + 506, + 672 + ], + "spans": [ + { + "bbox": [ + 115, + 659, + 506, + 672 + ], + "score": 1.0, + "content": "agent reinforcement learning with model uncertainty. In Proc. Advances in Neural Information", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 669, + 319, + 683 + ], + "spans": [ + { + "bbox": [ + 115, + 669, + 319, + 683 + ], + "score": 1.0, + "content": "Processing Systems (NeurIPS), volume 33, 2020b.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43 + }, + { + "type": "text", + "bbox": [ + 107, + 688, + 506, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "Shaofeng Zou, Tengyu Xu, and Yingbin Liang. Finite-sample analysis for SARSA with linear", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 116, + 700, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 116, + 700, + 506, + 713 + ], + "score": 1.0, + "content": "function approximation. In Proc. Advances in Neural Information Processing Systems (NeurIPS),", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 115, + 711, + 217, + 722 + ], + "spans": [ + { + "bbox": [ + 115, + 711, + 217, + 722 + ], + "score": 1.0, + "content": "pages 8665–8675, 2019.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 46 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "14", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 72, + 504, + 106 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 506, + 86 + ], + "score": 1.0, + "content": "Richard S Sutton, Csaba Szepesvári, and Hamid Reza Maei. A convergent O(n) algorithm for", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 84, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 116, + 84, + 505, + 96 + ], + "score": 1.0, + "content": "off-policy temporal-difference learning with linear function approximation. In Proc. Advances in", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 94, + 504, + 108 + ], + "spans": [ + { + "bbox": [ + 115, + 94, + 504, + 108 + ], + "score": 1.0, + "content": "Neural Information Processing Systems (NIPS), volume 21, pages 1609–1616. MIT Press, 2008.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 72, + 506, + 108 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 112, + 505, + 157 + ], + "lines": [ + { + "bbox": [ + 106, + 113, + 505, + 125 + ], + "spans": [ + { + "bbox": [ + 106, + 113, + 505, + 125 + ], + "score": 1.0, + "content": "Richard S Sutton, Hamid Reza Maei, Doina Precup, Shalabh Bhatnagar, David Silver, Csaba", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 116, + 124, + 506, + 136 + ], + "spans": [ + { + "bbox": [ + 116, + 124, + 506, + 136 + ], + "score": 1.0, + "content": "Szepesvári, and Eric Wiewiora. Fast gradient-descent methods for temporal-difference learn-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 135, + 505, + 148 + ], + "spans": [ + { + "bbox": [ + 116, + 135, + 505, + 148 + ], + "score": 1.0, + "content": "ing with linear function approximation. In Proc. International Conference on Machine Learning", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 146, + 246, + 159 + ], + "spans": [ + { + "bbox": [ + 115, + 146, + 246, + 159 + ], + "score": 1.0, + "content": "(ICML), pages 993–1000, 2009.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5, + "bbox_fs": [ + 106, + 113, + 506, + 159 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 164, + 503, + 188 + ], + "lines": [ + { + "bbox": [ + 106, + 164, + 505, + 177 + ], + "spans": [ + { + "bbox": [ + 106, + 164, + 505, + 177 + ], + "score": 1.0, + "content": "Aviv Tamar, Shie Mannor, and Huan Xu. Scaling up robust MDPs using function approximation. In", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 175, + 490, + 188 + ], + "spans": [ + { + "bbox": [ + 115, + 175, + 490, + 188 + ], + "score": 1.0, + "content": "Proc. International Conference on Machine Learning (ICML), pages 181–189. PMLR, 2014.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 106, + 164, + 505, + 188 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 194, + 505, + 227 + ], + "lines": [ + { + "bbox": [ + 105, + 193, + 506, + 208 + ], + "spans": [ + { + "bbox": [ + 105, + 193, + 506, + 208 + ], + "score": 1.0, + "content": "Eugene Vinitsky, Yuqing Du, Kanaad Parvate, Kathy Jang, Pieter Abbeel, and Alexandre Bayen.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 205, + 507, + 218 + ], + "spans": [ + { + "bbox": [ + 115, + 205, + 507, + 218 + ], + "score": 1.0, + "content": "Robust reinforcement learning using adversarial populations. arXiv preprint arXiv:2008.01825,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 214, + 142, + 229 + ], + "spans": [ + { + "bbox": [ + 115, + 214, + 142, + 229 + ], + "score": 1.0, + "content": "2020.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 193, + 507, + 229 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 234, + 506, + 268 + ], + "lines": [ + { + "bbox": [ + 105, + 234, + 505, + 248 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 505, + 248 + ], + "score": 1.0, + "content": "Hoi-To Wai, Mingyi Hong, Zhuoran Yang, Zhaoran Wang, and Kexin Tang. Variance reduced", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 246, + 505, + 259 + ], + "spans": [ + { + "bbox": [ + 115, + 246, + 505, + 259 + ], + "score": 1.0, + "content": "policy evaluation with smooth function approximation. In Proc. Advances in Neural Information", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 116, + 257, + 389, + 269 + ], + "spans": [ + { + "bbox": [ + 116, + 257, + 389, + 269 + ], + "score": 1.0, + "content": "Processing Systems (NeurIPS), volume 32, pages 5784–5795, 2019.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13, + "bbox_fs": [ + 105, + 234, + 505, + 269 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 275, + 505, + 298 + ], + "lines": [ + { + "bbox": [ + 105, + 274, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 105, + 274, + 505, + 288 + ], + "score": 1.0, + "content": "Martin J Wainwright. Variance-reduced Q-learning is minimax optimal. arXiv preprint", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 286, + 219, + 297 + ], + "spans": [ + { + "bbox": [ + 115, + 286, + 219, + 297 + ], + "score": 1.0, + "content": "arXiv:1906.04697, 2019.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 274, + 505, + 297 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 304, + 505, + 338 + ], + "lines": [ + { + "bbox": [ + 105, + 304, + 506, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 506, + 318 + ], + "score": 1.0, + "content": "Lingxiao Wang, Qi Cai, Zhuoran Yang, and Zhaoran Wang. Neural policy gradient methods:", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 315, + 505, + 330 + ], + "spans": [ + { + "bbox": [ + 115, + 315, + 505, + 330 + ], + "score": 1.0, + "content": "Global optimality and rates of convergence. In Proc. International Conference on Learning", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 116, + 327, + 241, + 339 + ], + "spans": [ + { + "bbox": [ + 116, + 327, + 241, + 339 + ], + "score": 1.0, + "content": "Representations (ICLR), 2020.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 304, + 506, + 339 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 345, + 506, + 379 + ], + "lines": [ + { + "bbox": [ + 106, + 345, + 506, + 359 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 506, + 359 + ], + "score": 1.0, + "content": "Yue Wang and Shaofeng Zou. Finite-sample analysis of Greedy-GQ with linear function approxi-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 356, + 506, + 370 + ], + "spans": [ + { + "bbox": [ + 115, + 356, + 506, + 370 + ], + "score": 1.0, + "content": "mation under Markovian noise. In Proc. International Conference on Uncertainty in Artificial", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 367, + 306, + 381 + ], + "spans": [ + { + "bbox": [ + 115, + 367, + 306, + 381 + ], + "score": 1.0, + "content": "Intelligence (UAI), pages 11–20. PMLR, 2020.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21, + "bbox_fs": [ + 106, + 345, + 506, + 381 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 385, + 505, + 419 + ], + "lines": [ + { + "bbox": [ + 107, + 386, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 107, + 386, + 505, + 398 + ], + "score": 1.0, + "content": "Yue Wang, Shaofeng Zou, and Yi Zhou. Non-asymptotic analysis for two time-scale TDC with", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 396, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 115, + 396, + 505, + 411 + ], + "score": 1.0, + "content": "general smooth function approximation. In Proc. Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 116, + 408, + 269, + 420 + ], + "spans": [ + { + "bbox": [ + 116, + 408, + 269, + 420 + ], + "score": 1.0, + "content": "Systems (NeurIPS), volume 34, 2021.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24, + "bbox_fs": [ + 107, + 386, + 505, + 420 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 426, + 504, + 450 + ], + "lines": [ + { + "bbox": [ + 106, + 425, + 505, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 425, + 505, + 439 + ], + "score": 1.0, + "content": "Wolfram Wiesemann, Daniel Kuhn, and Berç Rustem. Robust Markov decision processes. Mathe-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 438, + 333, + 449 + ], + "spans": [ + { + "bbox": [ + 116, + 438, + 333, + 449 + ], + "score": 1.0, + "content": "matics of Operations Research, 38(1):153–183, 2013.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5, + "bbox_fs": [ + 106, + 425, + 505, + 449 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 456, + 504, + 479 + ], + "lines": [ + { + "bbox": [ + 107, + 456, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 107, + 456, + 505, + 469 + ], + "score": 1.0, + "content": "Yue Wu, Weitong Zhang, Pan Xu, and Quanquan Gu. A finite time analysis of two time-scale actor", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 116, + 467, + 341, + 479 + ], + "spans": [ + { + "bbox": [ + 116, + 467, + 341, + 479 + ], + "score": 1.0, + "content": "critic methods. arXiv preprint arXiv:2005.01350, 2020.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5, + "bbox_fs": [ + 107, + 456, + 505, + 479 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 486, + 504, + 520 + ], + "lines": [ + { + "bbox": [ + 105, + 486, + 506, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 506, + 499 + ], + "score": 1.0, + "content": "Tengyu Xu and Yingbin Liang. Sample complexity bounds for two timescale value-based reinforce-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 497, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 115, + 497, + 505, + 510 + ], + "score": 1.0, + "content": "ment learning algorithms. In Proc. International Conference on Artifical Intelligence and Statistics", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 508, + 286, + 520 + ], + "spans": [ + { + "bbox": [ + 115, + 508, + 286, + 520 + ], + "score": 1.0, + "content": "(AISTATS), pages 811–819. PMLR, 2021.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 486, + 506, + 520 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 526, + 504, + 560 + ], + "lines": [ + { + "bbox": [ + 106, + 526, + 506, + 540 + ], + "spans": [ + { + "bbox": [ + 106, + 526, + 506, + 540 + ], + "score": 1.0, + "content": "Tengyu Xu, Shaofeng Zou, and Yingbin Liang. Two time-scale off-policy TD learning: Non-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 114, + 536, + 506, + 551 + ], + "spans": [ + { + "bbox": [ + 114, + 536, + 506, + 551 + ], + "score": 1.0, + "content": "asymptotic analysis over Markovian samples. In Proc. Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 548, + 307, + 560 + ], + "spans": [ + { + "bbox": [ + 115, + 548, + 307, + 560 + ], + "score": 1.0, + "content": "Systems (NeurIPS), pages 10633–10643, 2019.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34, + "bbox_fs": [ + 106, + 526, + 506, + 560 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 566, + 505, + 601 + ], + "lines": [ + { + "bbox": [ + 106, + 566, + 506, + 580 + ], + "spans": [ + { + "bbox": [ + 106, + 566, + 506, + 580 + ], + "score": 1.0, + "content": "Zhuoran Yang, Yongxin Chen, Mingyi Hong, and Zhaoran Wang. Provably global convergence of", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 578, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 115, + 578, + 505, + 591 + ], + "score": 1.0, + "content": "actor-critic: A case for linear quadratic regulator with ergodic cost. In Proc. Advances in Neural", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 115, + 588, + 393, + 602 + ], + "spans": [ + { + "bbox": [ + 115, + 588, + 393, + 602 + ], + "score": 1.0, + "content": "Information Processing Systems (NeurIPS), pages 8353–8365, 2019.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37, + "bbox_fs": [ + 106, + 566, + 506, + 602 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 607, + 504, + 641 + ], + "lines": [ + { + "bbox": [ + 106, + 607, + 506, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 607, + 506, + 621 + ], + "score": 1.0, + "content": "Kaiqing Zhang, Bin Hu, and Tamer Basar. On the stability and convergence of robust adversarial", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 619, + 505, + 631 + ], + "spans": [ + { + "bbox": [ + 115, + 619, + 505, + 631 + ], + "score": 1.0, + "content": "reinforcement learning: A case study on linear quadratic systems. In Proc. Advances in Neural", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 629, + 370, + 643 + ], + "spans": [ + { + "bbox": [ + 115, + 629, + 370, + 643 + ], + "score": 1.0, + "content": "Information Processing Systems (NeurIPS), volume 33, 2020a.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40, + "bbox_fs": [ + 106, + 607, + 506, + 643 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 648, + 504, + 682 + ], + "lines": [ + { + "bbox": [ + 105, + 648, + 506, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 506, + 660 + ], + "score": 1.0, + "content": "Kaiqing Zhang, Tao Sun, Yunzhe Tao, Sahika Genc, Sunil Mallya, and Tamer Basar. Robust multi-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 659, + 506, + 672 + ], + "spans": [ + { + "bbox": [ + 115, + 659, + 506, + 672 + ], + "score": 1.0, + "content": "agent reinforcement learning with model uncertainty. In Proc. Advances in Neural Information", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 115, + 669, + 319, + 683 + ], + "spans": [ + { + "bbox": [ + 115, + 669, + 319, + 683 + ], + "score": 1.0, + "content": "Processing Systems (NeurIPS), volume 33, 2020b.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43, + "bbox_fs": [ + 105, + 648, + 506, + 683 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 688, + 506, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "Shaofeng Zou, Tengyu Xu, and Yingbin Liang. Finite-sample analysis for SARSA with linear", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 116, + 700, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 116, + 700, + 506, + 713 + ], + "score": 1.0, + "content": "function approximation. In Proc. Advances in Neural Information Processing Systems (NeurIPS),", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 115, + 711, + 217, + 722 + ], + "spans": [ + { + "bbox": [ + 115, + 711, + 217, + 722 + ], + "score": 1.0, + "content": "pages 8665–8675, 2019.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 46, + "bbox_fs": [ + 105, + 688, + 506, + 722 + ] + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/train/IhiU6AJYpDs/IhiU6AJYpDs_model.json b/parse/train/IhiU6AJYpDs/IhiU6AJYpDs_model.json new file mode 100644 index 0000000000000000000000000000000000000000..0434797ea82f0564d8528382d80459e2575a3ee3 --- /dev/null +++ b/parse/train/IhiU6AJYpDs/IhiU6AJYpDs_model.json @@ -0,0 +1,24274 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 398, + 772, + 1303, + 772, + 1303, + 1288, + 398, + 1288 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 298, + 1418, + 1404, + 1418, + 1404, + 1754, + 298, + 1754 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1768, + 1403, + 1768, + 1403, + 1981, + 298, + 1981 + ], + "score": 0.977 + }, + { + "category_id": 0, + "poly": [ + 320, + 271, + 1382, + 271, + 1382, + 381, + 320, + 381 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 971, + 500, + 1219, + 500, + 1219, + 621, + 971, + 621 + ], + "score": 0.933 + }, + { + "category_id": 0, + "poly": [ + 299, + 1348, + 530, + 1348, + 530, + 1386, + 299, + 1386 + ], + "score": 0.91 + }, + { + "category_id": 0, + "poly": [ + 788, + 701, + 912, + 701, + 912, + 737, + 788, + 737 + ], + "score": 0.894 + }, + { + "category_id": 1, + "poly": [ + 478, + 499, + 775, + 499, + 775, + 622, + 478, + 622 + ], + "score": 0.841 + }, + { + "category_id": 2, + "poly": [ + 298, + 2033, + 1070, + 2033, + 1070, + 2061, + 298, + 2061 + ], + "score": 0.82 + }, + { + "category_id": 0, + "poly": [ + 566, + 500, + 689, + 500, + 689, + 530, + 566, + 530 + ], + "score": 0.192 + }, + { + "category_id": 15, + "poly": [ + 321.0, + 271.0, + 1381.0, + 271.0, + 1381.0, + 329.0, + 321.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 723.0, + 324.0, + 980.0, + 324.0, + 980.0, + 389.0, + 723.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1345.0, + 536.0, + 1345.0, + 536.0, + 1393.0, + 292.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 782.0, + 698.0, + 919.0, + 698.0, + 919.0, + 742.0, + 782.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2028.0, + 1075.0, + 2028.0, + 1075.0, + 2067.0, + 293.0, + 2067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 493.0, + 695.0, + 493.0, + 695.0, + 540.0, + 562.0, + 540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 772.0, + 1309.0, + 772.0, + 1309.0, + 809.0, + 394.0, + 809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 802.0, + 1306.0, + 802.0, + 1306.0, + 839.0, + 394.0, + 839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 832.0, + 1307.0, + 832.0, + 1307.0, + 869.0, + 393.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 863.0, + 1307.0, + 863.0, + 1307.0, + 900.0, + 393.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 894.0, + 1305.0, + 894.0, + 1305.0, + 927.0, + 393.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 924.0, + 1306.0, + 924.0, + 1306.0, + 957.0, + 394.0, + 957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 954.0, + 1305.0, + 954.0, + 1305.0, + 989.0, + 393.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 983.0, + 1305.0, + 983.0, + 1305.0, + 1019.0, + 393.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1014.0, + 1308.0, + 1014.0, + 1308.0, + 1050.0, + 393.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1045.0, + 1306.0, + 1045.0, + 1306.0, + 1081.0, + 392.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1076.0, + 1304.0, + 1076.0, + 1304.0, + 1109.0, + 394.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1106.0, + 1307.0, + 1106.0, + 1307.0, + 1142.0, + 393.0, + 1142.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1135.0, + 1307.0, + 1135.0, + 1307.0, + 1170.0, + 393.0, + 1170.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1167.0, + 1306.0, + 1167.0, + 1306.0, + 1202.0, + 393.0, + 1202.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1197.0, + 1305.0, + 1197.0, + 1305.0, + 1230.0, + 394.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1225.0, + 1305.0, + 1225.0, + 1305.0, + 1263.0, + 392.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1259.0, + 1175.0, + 1259.0, + 1175.0, + 1292.0, + 393.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1418.0, + 1405.0, + 1418.0, + 1405.0, + 1455.0, + 293.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1447.0, + 1405.0, + 1447.0, + 1405.0, + 1488.0, + 292.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1481.0, + 1405.0, + 1481.0, + 1405.0, + 1515.0, + 294.0, + 1515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1509.0, + 1405.0, + 1509.0, + 1405.0, + 1546.0, + 293.0, + 1546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1539.0, + 1404.0, + 1539.0, + 1404.0, + 1577.0, + 292.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1570.0, + 1408.0, + 1570.0, + 1408.0, + 1606.0, + 293.0, + 1606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1600.0, + 1408.0, + 1600.0, + 1408.0, + 1638.0, + 293.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1632.0, + 1406.0, + 1632.0, + 1406.0, + 1665.0, + 293.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1660.0, + 1407.0, + 1660.0, + 1407.0, + 1697.0, + 293.0, + 1697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1694.0, + 1404.0, + 1694.0, + 1404.0, + 1728.0, + 293.0, + 1728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1722.0, + 958.0, + 1722.0, + 958.0, + 1761.0, + 293.0, + 1761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1768.0, + 1405.0, + 1768.0, + 1405.0, + 1802.0, + 293.0, + 1802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1800.0, + 1407.0, + 1800.0, + 1407.0, + 1834.0, + 294.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1830.0, + 1405.0, + 1830.0, + 1405.0, + 1865.0, + 294.0, + 1865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1857.0, + 1407.0, + 1857.0, + 1407.0, + 1895.0, + 293.0, + 1895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1890.0, + 1405.0, + 1890.0, + 1405.0, + 1925.0, + 294.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1920.0, + 1405.0, + 1920.0, + 1405.0, + 1954.0, + 294.0, + 1954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1949.0, + 1403.0, + 1949.0, + 1403.0, + 1986.0, + 293.0, + 1986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1007.0, + 498.0, + 1182.0, + 498.0, + 1182.0, + 533.0, + 1007.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 529.0, + 1216.0, + 529.0, + 1216.0, + 563.0, + 973.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 560.0, + 1206.0, + 560.0, + 1206.0, + 591.0, + 982.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 967.0, + 591.0, + 1221.0, + 591.0, + 1221.0, + 622.0, + 967.0, + 622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 492.0, + 696.0, + 492.0, + 696.0, + 537.0, + 561.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 506.0, + 529.0, + 750.0, + 529.0, + 750.0, + 563.0, + 506.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 515.0, + 560.0, + 738.0, + 560.0, + 738.0, + 591.0, + 515.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 591.0, + 777.0, + 591.0, + 777.0, + 623.0, + 475.0, + 623.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1565, + 1405, + 1565, + 1405, + 1810, + 297, + 1810 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 298, + 203, + 1404, + 203, + 1404, + 386, + 298, + 386 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 299, + 1823, + 1403, + 1823, + 1403, + 2007, + 299, + 2007 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 297, + 400, + 1402, + 400, + 1402, + 461, + 297, + 461 + ], + "score": 0.944 + }, + { + "category_id": 0, + "poly": [ + 299, + 1508, + 528, + 1508, + 528, + 1540, + 299, + 1540 + ], + "score": 0.906 + }, + { + "category_id": 1, + "poly": [ + 299, + 1347, + 1406, + 1347, + 1406, + 1469, + 299, + 1469 + ], + "score": 0.905 + }, + { + "category_id": 1, + "poly": [ + 297, + 1031, + 1405, + 1031, + 1405, + 1334, + 297, + 1334 + ], + "score": 0.791 + }, + { + "category_id": 2, + "poly": [ + 841, + 2062, + 858, + 2062, + 858, + 2084, + 841, + 2084 + ], + "score": 0.721 + }, + { + "category_id": 1, + "poly": [ + 297, + 488, + 1405, + 488, + 1405, + 732, + 297, + 732 + ], + "score": 0.64 + }, + { + "category_id": 1, + "poly": [ + 298, + 743, + 1404, + 743, + 1404, + 1019, + 298, + 1019 + ], + "score": 0.627 + }, + { + "category_id": 2, + "poly": [ + 841, + 2062, + 859, + 2062, + 859, + 2084, + 841, + 2084 + ], + "score": 0.127 + }, + { + "category_id": 13, + "poly": [ + 1139, + 520, + 1163, + 520, + 1163, + 546, + 1139, + 546 + ], + "score": 0.81, + "latex": "R" + }, + { + "category_id": 13, + "poly": [ + 426, + 581, + 451, + 581, + 451, + 607, + 426, + 607 + ], + "score": 0.75, + "latex": "R" + }, + { + "category_id": 13, + "poly": [ + 613, + 554, + 627, + 554, + 627, + 577, + 613, + 577 + ], + "score": 0.65, + "latex": "\\epsilon" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1505.0, + 533.0, + 1505.0, + 533.0, + 1545.0, + 294.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2059.0, + 862.0, + 2059.0, + 862.0, + 2093.0, + 839.0, + 2093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2059.0, + 862.0, + 2059.0, + 862.0, + 2093.0, + 838.0, + 2093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1564.0, + 1407.0, + 1564.0, + 1407.0, + 1600.0, + 294.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1596.0, + 1408.0, + 1596.0, + 1408.0, + 1630.0, + 294.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1625.0, + 1405.0, + 1625.0, + 1405.0, + 1661.0, + 292.0, + 1661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1658.0, + 1405.0, + 1658.0, + 1405.0, + 1691.0, + 294.0, + 1691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1685.0, + 1406.0, + 1685.0, + 1406.0, + 1723.0, + 291.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1716.0, + 1405.0, + 1716.0, + 1405.0, + 1751.0, + 292.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1747.0, + 1405.0, + 1747.0, + 1405.0, + 1783.0, + 291.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1780.0, + 742.0, + 1780.0, + 742.0, + 1813.0, + 296.0, + 1813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 205.0, + 1404.0, + 205.0, + 1404.0, + 236.0, + 294.0, + 236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 233.0, + 1405.0, + 233.0, + 1405.0, + 267.0, + 293.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 263.0, + 1406.0, + 263.0, + 1406.0, + 298.0, + 293.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 295.0, + 1408.0, + 295.0, + 1408.0, + 330.0, + 294.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 321.0, + 1408.0, + 321.0, + 1408.0, + 361.0, + 292.0, + 361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 354.0, + 462.0, + 354.0, + 462.0, + 387.0, + 294.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1822.0, + 1405.0, + 1822.0, + 1405.0, + 1858.0, + 295.0, + 1858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1852.0, + 1407.0, + 1852.0, + 1407.0, + 1892.0, + 290.0, + 1892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1883.0, + 1408.0, + 1883.0, + 1408.0, + 1920.0, + 292.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1912.0, + 1408.0, + 1912.0, + 1408.0, + 1949.0, + 293.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1943.0, + 1407.0, + 1943.0, + 1407.0, + 1980.0, + 293.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1977.0, + 1404.0, + 1977.0, + 1404.0, + 2009.0, + 295.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 395.0, + 1406.0, + 395.0, + 1406.0, + 438.0, + 292.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 431.0, + 903.0, + 431.0, + 903.0, + 464.0, + 293.0, + 464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1344.0, + 1410.0, + 1344.0, + 1410.0, + 1384.0, + 294.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1376.0, + 1409.0, + 1376.0, + 1409.0, + 1416.0, + 317.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1410.0, + 1404.0, + 1410.0, + 1404.0, + 1443.0, + 319.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 1436.0, + 1304.0, + 1436.0, + 1304.0, + 1476.0, + 316.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1029.0, + 1408.0, + 1029.0, + 1408.0, + 1067.0, + 292.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 1062.0, + 1408.0, + 1062.0, + 1408.0, + 1097.0, + 316.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1092.0, + 1406.0, + 1092.0, + 1406.0, + 1127.0, + 317.0, + 1127.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1118.0, + 1405.0, + 1118.0, + 1405.0, + 1159.0, + 319.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1152.0, + 1407.0, + 1152.0, + 1407.0, + 1188.0, + 319.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1183.0, + 1407.0, + 1183.0, + 1407.0, + 1218.0, + 317.0, + 1218.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1214.0, + 1406.0, + 1214.0, + 1406.0, + 1245.0, + 319.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 1242.0, + 1408.0, + 1242.0, + 1408.0, + 1279.0, + 316.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 1272.0, + 1407.0, + 1272.0, + 1407.0, + 1308.0, + 316.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1305.0, + 585.0, + 1305.0, + 585.0, + 1335.0, + 319.0, + 1335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 485.0, + 1408.0, + 485.0, + 1408.0, + 525.0, + 294.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 517.0, + 1138.0, + 517.0, + 1138.0, + 551.0, + 316.0, + 551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1164.0, + 517.0, + 1405.0, + 517.0, + 1405.0, + 551.0, + 1164.0, + 551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 550.0, + 612.0, + 550.0, + 612.0, + 584.0, + 317.0, + 584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 550.0, + 1405.0, + 550.0, + 1405.0, + 584.0, + 628.0, + 584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 580.0, + 425.0, + 580.0, + 425.0, + 614.0, + 317.0, + 614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 580.0, + 1405.0, + 580.0, + 1405.0, + 614.0, + 452.0, + 614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 610.0, + 1405.0, + 610.0, + 1405.0, + 643.0, + 319.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 640.0, + 1405.0, + 640.0, + 1405.0, + 673.0, + 317.0, + 673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 671.0, + 1405.0, + 671.0, + 1405.0, + 704.0, + 319.0, + 704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 702.0, + 1081.0, + 702.0, + 1081.0, + 735.0, + 319.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 741.0, + 1405.0, + 741.0, + 1405.0, + 779.0, + 293.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 775.0, + 1406.0, + 775.0, + 1406.0, + 808.0, + 318.0, + 808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 803.0, + 1405.0, + 803.0, + 1405.0, + 841.0, + 316.0, + 841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 834.0, + 1404.0, + 834.0, + 1404.0, + 871.0, + 317.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 863.0, + 1406.0, + 863.0, + 1406.0, + 902.0, + 316.0, + 902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 894.0, + 1404.0, + 894.0, + 1404.0, + 932.0, + 316.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 928.0, + 1405.0, + 928.0, + 1405.0, + 961.0, + 320.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 955.0, + 1409.0, + 955.0, + 1409.0, + 992.0, + 317.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 985.0, + 773.0, + 985.0, + 773.0, + 1025.0, + 316.0, + 1025.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1464, + 1404, + 1464, + 1404, + 1820, + 298, + 1820 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 297, + 520, + 1405, + 520, + 1405, + 916, + 297, + 916 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 297, + 930, + 1406, + 930, + 1406, + 1355, + 297, + 1355 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 202, + 1405, + 202, + 1405, + 507, + 298, + 507 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 299, + 1833, + 1403, + 1833, + 1403, + 1957, + 299, + 1957 + ], + "score": 0.97 + }, + { + "category_id": 0, + "poly": [ + 299, + 1397, + 542, + 1397, + 542, + 1435, + 299, + 1435 + ], + "score": 0.906 + }, + { + "category_id": 2, + "poly": [ + 320, + 1976, + 1347, + 1976, + 1347, + 2010, + 320, + 2010 + ], + "score": 0.903 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.654 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.435 + }, + { + "category_id": 13, + "poly": [ + 713, + 1562, + 905, + 1562, + 905, + 1600, + 713, + 1600 + ], + "score": 0.95, + "latex": "p _ { s } ^ { a } = \\{ p _ { s , s ^ { \\prime } } ^ { a } \\} _ { s ^ { \\prime } \\in \\mathcal S }" + }, + { + "category_id": 13, + "poly": [ + 1044, + 1691, + 1150, + 1691, + 1150, + 1726, + 1044, + 1726 + ], + "score": 0.93, + "latex": "c ( S _ { t } , A _ { t } )" + }, + { + "category_id": 13, + "poly": [ + 640, + 1496, + 998, + 1496, + 998, + 1535, + 640, + 1535 + ], + "score": 0.93, + "latex": "\\mathsf { P } = \\{ p _ { s } ^ { a } \\in \\varDelta _ { | \\mathcal { S } | } , a \\in \\mathcal { A } , s \\in \\mathcal { S } \\}" + }, + { + "category_id": 13, + "poly": [ + 541, + 1689, + 634, + 1689, + 634, + 1732, + 541, + 1732 + ], + "score": 0.92, + "latex": "p _ { S _ { t } , S _ { t + 1 } } ^ { A _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 1131, + 1864, + 1316, + 1864, + 1316, + 1897, + 1131, + 1897 + ], + "score": 0.92, + "latex": "\\kappa = ( \\mathsf { P } _ { 0 } , \\mathsf { P } _ { 1 } , \\ldots )" + }, + { + "category_id": 13, + "poly": [ + 402, + 1628, + 477, + 1628, + 477, + 1661, + 402, + 1661 + ], + "score": 0.92, + "latex": "c ( s , a )" + }, + { + "category_id": 13, + "poly": [ + 454, + 1531, + 566, + 1531, + 566, + 1565, + 454, + 1565 + ], + "score": 0.91, + "latex": "\\gamma \\in [ 0 , 1 )" + }, + { + "category_id": 13, + "poly": [ + 600, + 1660, + 689, + 1660, + 689, + 1689, + 600, + 1689 + ], + "score": 0.91, + "latex": "A _ { t } \\in \\mathcal A" + }, + { + "category_id": 13, + "poly": [ + 992, + 1565, + 1045, + 1565, + 1045, + 1600, + 992, + 1600 + ], + "score": 0.91, + "latex": "p _ { s , s ^ { \\prime } } ^ { a }" + }, + { + "category_id": 13, + "poly": [ + 1068, + 1465, + 1221, + 1465, + 1221, + 1499, + 1068, + 1499 + ], + "score": 0.91, + "latex": "( \\mathcal { S } , \\mathcal { A } , \\mathsf { P } , c , \\gamma )" + }, + { + "category_id": 13, + "poly": [ + 781, + 1659, + 860, + 1659, + 860, + 1689, + 781, + 1689 + ], + "score": 0.91, + "latex": "S _ { t } \\in \\mathcal S" + }, + { + "category_id": 13, + "poly": [ + 298, + 1692, + 355, + 1692, + 355, + 1726, + 298, + 1726 + ], + "score": 0.91, + "latex": "S _ { t + 1 }" + }, + { + "category_id": 13, + "poly": [ + 446, + 1757, + 836, + 1757, + 836, + 1793, + 446, + 1793 + ], + "score": 0.9, + "latex": "\\pi \\colon \\bar { \\mathbb { E } } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\dot { \\gamma } ^ { t } c ( S _ { t } , A _ { t } ) | S _ { 0 } = s , \\pi \\right]" + }, + { + "category_id": 13, + "poly": [ + 504, + 1978, + 584, + 1978, + 584, + 2008, + 504, + 2008 + ], + "score": 0.88, + "latex": "( n - 1 )" + }, + { + "category_id": 13, + "poly": [ + 734, + 1727, + 798, + 1727, + 798, + 1756, + 734, + 1756 + ], + "score": 0.88, + "latex": "s \\in \\mathcal { S }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1895, + 472, + 1895, + 472, + 1926, + 298, + 1926 + ], + "score": 0.88, + "latex": "\\mathsf { P } _ { t } \\in \\mathbf { P } , \\forall t \\ge 0" + }, + { + "category_id": 13, + "poly": [ + 791, + 1791, + 855, + 1791, + 855, + 1817, + 791, + 1817 + ], + "score": 0.88, + "latex": "s \\in \\mathcal { S }" + }, + { + "category_id": 13, + "poly": [ + 969, + 1536, + 1000, + 1536, + 1000, + 1565, + 969, + 1565 + ], + "score": 0.88, + "latex": "p _ { s } ^ { a }" + }, + { + "category_id": 13, + "poly": [ + 1005, + 1865, + 1036, + 1865, + 1036, + 1895, + 1005, + 1895 + ], + "score": 0.87, + "latex": "\\mathsf { P } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 622, + 1597, + 646, + 1597, + 646, + 1625, + 622, + 1625 + ], + "score": 0.86, + "latex": "s ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 928, + 1975, + 1342, + 1975, + 1342, + 2010, + 928, + 2010 + ], + "score": 0.86, + "latex": "\\begin{array} { r } { \\{ ( p _ { 1 } , . . . , p _ { n } ) | 0 \\leq p _ { i } \\leq 1 , \\sum _ { i = 1 } ^ { n } p _ { i } = 1 \\} . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 714, + 1633, + 735, + 1633, + 735, + 1655, + 714, + 1655 + ], + "score": 0.79, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 1217, + 1902, + 1237, + 1902, + 1237, + 1922, + 1217, + 1922 + ], + "score": 0.78, + "latex": "\\kappa" + }, + { + "category_id": 13, + "poly": [ + 828, + 1605, + 846, + 1605, + 846, + 1625, + 828, + 1625 + ], + "score": 0.78, + "latex": "a" + }, + { + "category_id": 13, + "poly": [ + 1359, + 1603, + 1375, + 1603, + 1375, + 1625, + 1359, + 1625 + ], + "score": 0.76, + "latex": "s" + }, + { + "category_id": 13, + "poly": [ + 343, + 1976, + 380, + 1976, + 380, + 2005, + 343, + 2005 + ], + "score": 0.76, + "latex": "\\varDelta _ { n }" + }, + { + "category_id": 13, + "poly": [ + 1376, + 1467, + 1401, + 1467, + 1401, + 1493, + 1376, + 1493 + ], + "score": 0.76, + "latex": "\\mathcal { A }" + }, + { + "category_id": 13, + "poly": [ + 1209, + 1763, + 1231, + 1763, + 1231, + 1786, + 1209, + 1786 + ], + "score": 0.76, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 935, + 1603, + 951, + 1603, + 951, + 1625, + 935, + 1625 + ], + "score": 0.75, + "latex": "s" + }, + { + "category_id": 13, + "poly": [ + 953, + 1867, + 966, + 1867, + 966, + 1892, + 953, + 1892 + ], + "score": 0.74, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 533, + 1570, + 551, + 1570, + 551, + 1590, + 533, + 1590 + ], + "score": 0.74, + "latex": "a" + }, + { + "category_id": 13, + "poly": [ + 529, + 1896, + 555, + 1896, + 555, + 1922, + 529, + 1922 + ], + "score": 0.74, + "latex": "\\mathbf { P }" + }, + { + "category_id": 13, + "poly": [ + 1251, + 1603, + 1269, + 1603, + 1269, + 1625, + 1251, + 1625 + ], + "score": 0.73, + "latex": "a" + }, + { + "category_id": 13, + "poly": [ + 390, + 1734, + 410, + 1734, + 410, + 1755, + 390, + 1755 + ], + "score": 0.73, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 298, + 1662, + 311, + 1662, + 311, + 1687, + 298, + 1687 + ], + "score": 0.73, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 1269, + 1503, + 1286, + 1503, + 1286, + 1526, + 1269, + 1526 + ], + "score": 0.73, + "latex": "c" + }, + { + "category_id": 13, + "poly": [ + 641, + 1569, + 656, + 1569, + 656, + 1590, + 641, + 1590 + ], + "score": 0.7, + "latex": "s" + }, + { + "category_id": 13, + "poly": [ + 1215, + 1629, + 1240, + 1629, + 1240, + 1655, + 1215, + 1655 + ], + "score": 0.69, + "latex": "\\mathcal A" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1393.0, + 545.0, + 1393.0, + 545.0, + 1441.0, + 291.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 1969.0, + 342.0, + 1969.0, + 342.0, + 2018.0, + 331.0, + 2018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 1969.0, + 503.0, + 1969.0, + 503.0, + 2018.0, + 381.0, + 2018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 1969.0, + 927.0, + 1969.0, + 927.0, + 2018.0, + 585.0, + 2018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1343.0, + 1969.0, + 1354.0, + 1969.0, + 1354.0, + 2018.0, + 1343.0, + 2018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2091.0, + 838.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2091.0, + 838.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1463.0, + 1067.0, + 1463.0, + 1067.0, + 1500.0, + 294.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1222.0, + 1463.0, + 1375.0, + 1463.0, + 1375.0, + 1500.0, + 1222.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1463.0, + 1405.0, + 1463.0, + 1405.0, + 1500.0, + 1402.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1496.0, + 639.0, + 1496.0, + 639.0, + 1534.0, + 292.0, + 1534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 1496.0, + 1268.0, + 1496.0, + 1268.0, + 1534.0, + 999.0, + 1534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1287.0, + 1496.0, + 1407.0, + 1496.0, + 1407.0, + 1534.0, + 1287.0, + 1534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1531.0, + 453.0, + 1531.0, + 453.0, + 1567.0, + 294.0, + 1567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.0, + 1531.0, + 968.0, + 1531.0, + 968.0, + 1567.0, + 567.0, + 1567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1531.0, + 1406.0, + 1531.0, + 1406.0, + 1567.0, + 1001.0, + 1567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1558.0, + 532.0, + 1558.0, + 532.0, + 1607.0, + 291.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 1558.0, + 640.0, + 1558.0, + 640.0, + 1607.0, + 552.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 657.0, + 1558.0, + 712.0, + 1558.0, + 712.0, + 1607.0, + 657.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.0, + 1558.0, + 991.0, + 1558.0, + 991.0, + 1607.0, + 906.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 1558.0, + 1408.0, + 1558.0, + 1408.0, + 1607.0, + 1046.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1598.0, + 621.0, + 1598.0, + 621.0, + 1630.0, + 296.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 647.0, + 1598.0, + 827.0, + 1598.0, + 827.0, + 1630.0, + 647.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1598.0, + 934.0, + 1598.0, + 934.0, + 1630.0, + 847.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 952.0, + 1598.0, + 1250.0, + 1598.0, + 1250.0, + 1630.0, + 952.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1270.0, + 1598.0, + 1358.0, + 1598.0, + 1358.0, + 1630.0, + 1270.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1376.0, + 1598.0, + 1405.0, + 1598.0, + 1405.0, + 1630.0, + 1376.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1626.0, + 401.0, + 1626.0, + 401.0, + 1663.0, + 293.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 1626.0, + 713.0, + 1626.0, + 713.0, + 1663.0, + 478.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 1626.0, + 1214.0, + 1626.0, + 1214.0, + 1663.0, + 736.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1241.0, + 1626.0, + 1406.0, + 1626.0, + 1406.0, + 1663.0, + 1241.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1659.0, + 297.0, + 1659.0, + 297.0, + 1693.0, + 292.0, + 1693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 1659.0, + 599.0, + 1659.0, + 599.0, + 1693.0, + 312.0, + 1693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 1659.0, + 780.0, + 1659.0, + 780.0, + 1693.0, + 690.0, + 1693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 861.0, + 1659.0, + 1406.0, + 1659.0, + 1406.0, + 1693.0, + 861.0, + 1693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 1683.0, + 297.0, + 1683.0, + 297.0, + 1744.0, + 287.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 1683.0, + 540.0, + 1683.0, + 540.0, + 1744.0, + 356.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 635.0, + 1683.0, + 1043.0, + 1683.0, + 1043.0, + 1744.0, + 635.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1151.0, + 1683.0, + 1414.0, + 1683.0, + 1414.0, + 1744.0, + 1151.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 1726.0, + 389.0, + 1726.0, + 389.0, + 1823.0, + 284.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 1726.0, + 445.0, + 1726.0, + 445.0, + 1823.0, + 411.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 856.0, + 1726.0, + 1208.0, + 1726.0, + 1208.0, + 1823.0, + 856.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1232.0, + 1726.0, + 1414.0, + 1726.0, + 1414.0, + 1823.0, + 1232.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 522.0, + 1403.0, + 522.0, + 1403.0, + 554.0, + 296.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 553.0, + 1405.0, + 553.0, + 1405.0, + 585.0, + 295.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 581.0, + 1406.0, + 581.0, + 1406.0, + 619.0, + 294.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 612.0, + 1405.0, + 612.0, + 1405.0, + 648.0, + 295.0, + 648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 643.0, + 1406.0, + 643.0, + 1406.0, + 679.0, + 294.0, + 679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 671.0, + 1406.0, + 671.0, + 1406.0, + 709.0, + 292.0, + 709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 704.0, + 1405.0, + 704.0, + 1405.0, + 736.0, + 295.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 734.0, + 1408.0, + 734.0, + 1408.0, + 768.0, + 292.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 763.0, + 1405.0, + 763.0, + 1405.0, + 799.0, + 294.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 794.0, + 1405.0, + 794.0, + 1405.0, + 830.0, + 294.0, + 830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 823.0, + 1403.0, + 823.0, + 1403.0, + 859.0, + 294.0, + 859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 854.0, + 1403.0, + 854.0, + 1403.0, + 889.0, + 294.0, + 889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 887.0, + 1388.0, + 887.0, + 1388.0, + 920.0, + 297.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 930.0, + 1404.0, + 930.0, + 1404.0, + 962.0, + 296.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 960.0, + 1408.0, + 960.0, + 1408.0, + 995.0, + 294.0, + 995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 991.0, + 1408.0, + 991.0, + 1408.0, + 1026.0, + 295.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1019.0, + 1406.0, + 1019.0, + 1406.0, + 1058.0, + 294.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1050.0, + 1406.0, + 1050.0, + 1406.0, + 1086.0, + 295.0, + 1086.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1082.0, + 1406.0, + 1082.0, + 1406.0, + 1116.0, + 292.0, + 1116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1113.0, + 1406.0, + 1113.0, + 1406.0, + 1145.0, + 296.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1141.0, + 1406.0, + 1141.0, + 1406.0, + 1177.0, + 295.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1173.0, + 1407.0, + 1173.0, + 1407.0, + 1208.0, + 294.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1197.0, + 1409.0, + 1197.0, + 1409.0, + 1242.0, + 291.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1230.0, + 1407.0, + 1230.0, + 1407.0, + 1270.0, + 292.0, + 1270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1262.0, + 1404.0, + 1262.0, + 1404.0, + 1298.0, + 295.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1293.0, + 1407.0, + 1293.0, + 1407.0, + 1332.0, + 294.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1325.0, + 772.0, + 1325.0, + 772.0, + 1357.0, + 296.0, + 1357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 201.0, + 1410.0, + 201.0, + 1410.0, + 241.0, + 293.0, + 241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 232.0, + 1409.0, + 232.0, + 1409.0, + 270.0, + 293.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 262.0, + 1407.0, + 262.0, + 1407.0, + 301.0, + 293.0, + 301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 295.0, + 1405.0, + 295.0, + 1405.0, + 330.0, + 295.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 325.0, + 1402.0, + 325.0, + 1402.0, + 357.0, + 295.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 356.0, + 1403.0, + 356.0, + 1403.0, + 387.0, + 293.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 382.0, + 1405.0, + 382.0, + 1405.0, + 420.0, + 292.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 415.0, + 1406.0, + 415.0, + 1406.0, + 451.0, + 295.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 443.0, + 1408.0, + 443.0, + 1408.0, + 481.0, + 293.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 473.0, + 1367.0, + 473.0, + 1367.0, + 511.0, + 293.0, + 511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1831.0, + 1407.0, + 1831.0, + 1407.0, + 1868.0, + 293.0, + 1868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1863.0, + 952.0, + 1863.0, + 952.0, + 1900.0, + 293.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 967.0, + 1863.0, + 1004.0, + 1863.0, + 1004.0, + 1900.0, + 967.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 1863.0, + 1130.0, + 1863.0, + 1130.0, + 1900.0, + 1037.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1317.0, + 1863.0, + 1407.0, + 1863.0, + 1407.0, + 1900.0, + 1317.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1894.0, + 297.0, + 1894.0, + 297.0, + 1930.0, + 294.0, + 1930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 473.0, + 1894.0, + 528.0, + 1894.0, + 528.0, + 1930.0, + 473.0, + 1930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 1894.0, + 1216.0, + 1894.0, + 1216.0, + 1930.0, + 556.0, + 1930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1894.0, + 1407.0, + 1894.0, + 1407.0, + 1930.0, + 1238.0, + 1930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1924.0, + 1405.0, + 1924.0, + 1405.0, + 1960.0, + 294.0, + 1960.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 418, + 1406, + 418, + 1406, + 540, + 296, + 540 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1117, + 1405, + 1117, + 1405, + 1245, + 298, + 1245 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 670, + 1406, + 670, + 1406, + 862, + 298, + 862 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 1399, + 1406, + 1399, + 1406, + 1556, + 297, + 1556 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 1567, + 1406, + 1567, + 1406, + 1661, + 297, + 1661 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 297, + 1731, + 1405, + 1731, + 1405, + 1830, + 297, + 1830 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 299, + 201, + 1404, + 201, + 1404, + 295, + 299, + 295 + ], + "score": 0.962 + }, + { + "category_id": 8, + "poly": [ + 578, + 310, + 1119, + 310, + 1119, + 404, + 578, + 404 + ], + "score": 0.959 + }, + { + "category_id": 1, + "poly": [ + 296, + 1838, + 1406, + 1838, + 1406, + 2016, + 296, + 2016 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 297, + 1024, + 1403, + 1024, + 1403, + 1095, + 297, + 1095 + ], + "score": 0.954 + }, + { + "category_id": 8, + "poly": [ + 342, + 916, + 1324, + 916, + 1324, + 1012, + 342, + 1012 + ], + "score": 0.948 + }, + { + "category_id": 8, + "poly": [ + 682, + 556, + 1015, + 556, + 1015, + 603, + 682, + 603 + ], + "score": 0.942 + }, + { + "category_id": 1, + "poly": [ + 296, + 1354, + 1206, + 1354, + 1206, + 1388, + 296, + 1388 + ], + "score": 0.937 + }, + { + "category_id": 0, + "poly": [ + 298, + 1284, + 1184, + 1284, + 1184, + 1325, + 298, + 1325 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 297, + 868, + 1404, + 868, + 1404, + 904, + 297, + 904 + ], + "score": 0.921 + }, + { + "category_id": 8, + "poly": [ + 440, + 1674, + 1254, + 1674, + 1254, + 1716, + 440, + 1716 + ], + "score": 0.919 + }, + { + "category_id": 1, + "poly": [ + 296, + 624, + 1303, + 624, + 1303, + 661, + 296, + 661 + ], + "score": 0.918 + }, + { + "category_id": 9, + "poly": [ + 1366, + 342, + 1400, + 342, + 1400, + 373, + 1366, + 373 + ], + "score": 0.889 + }, + { + "category_id": 9, + "poly": [ + 1366, + 561, + 1400, + 561, + 1400, + 591, + 1366, + 591 + ], + "score": 0.887 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1681, + 1400, + 1681, + 1400, + 1710, + 1366, + 1710 + ], + "score": 0.874 + }, + { + "category_id": 9, + "poly": [ + 1366, + 950, + 1400, + 950, + 1400, + 979, + 1366, + 979 + ], + "score": 0.835 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 858, + 2061, + 858, + 2084, + 841, + 2084 + ], + "score": 0.782 + }, + { + "category_id": 14, + "poly": [ + 576, + 309, + 1118, + 309, + 1118, + 405, + 576, + 405 + ], + "score": 0.94, + "latex": "V ^ { \\pi } ( s ) = \\operatorname* { m a x } _ { \\kappa } \\mathbb { E } _ { \\kappa } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } c ( S _ { t } , A _ { t } ) | S _ { 0 } = s , \\pi \\right] ," + }, + { + "category_id": 13, + "poly": [ + 1113, + 1960, + 1397, + 1960, + 1397, + 2017, + 1113, + 2017 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { d _ { T V } ( p , q ) \\leq \\sqrt { \\frac 1 2 d _ { K L } ( p , q ) } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 336, + 916, + 1329, + 916, + 1329, + 1011, + 336, + 1011 + ], + "score": 0.94, + "latex": "\\operatorname* { m i n } _ { \\pi } \\operatorname* { m a x } _ { \\kappa } \\mathbb { E } _ { \\kappa } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } c ( S _ { t } , A _ { t } ) \\big | \\pi , S _ { 0 } = s \\right] = \\operatorname* { m a x } _ { \\kappa } \\operatorname* { m i n } _ { \\pi } \\mathbb { E } _ { \\kappa } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } c ( S _ { t } , A _ { t } ) \\big | \\pi , S _ { 0 } = s \\right] ." + }, + { + "category_id": 13, + "poly": [ + 704, + 1794, + 923, + 1794, + 923, + 1832, + 704, + 1832 + ], + "score": 0.93, + "latex": "\\mathbf { P } = \\bigotimes _ { s \\in \\mathcal { S } , a \\in \\mathcal { A } } \\mathcal { P } _ { s } ^ { a }" + }, + { + "category_id": 13, + "poly": [ + 472, + 623, + 521, + 623, + 521, + 656, + 472, + 656 + ], + "score": 0.92, + "latex": "V ^ { \\pi ^ { * } }" + }, + { + "category_id": 14, + "poly": [ + 681, + 556, + 1018, + 556, + 1018, + 606, + 681, + 606 + ], + "score": 0.92, + "latex": "\\pi ^ { * } = \\arg \\operatorname* { m i n } _ { \\pi } V ^ { \\pi } ( s ) , \\forall s \\in \\mathcal { S } ." + }, + { + "category_id": 13, + "poly": [ + 993, + 626, + 1296, + 626, + 1296, + 662, + 993, + 662 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { V ^ { * } ( s ) = \\operatorname* { m i n } _ { a \\in \\mathcal { A } } Q ^ { * } ( s , a ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 779, + 1120, + 1308, + 1120, + 1308, + 1156, + 779, + 1156 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { \\mathbf { T } Q ( s , a ) = c ( s , a ) + \\gamma \\sigma _ { \\mathcal { P } _ { s } ^ { a } } ( \\operatorname* { m i n } _ { a \\in \\mathcal { A } } Q ( s , a ) ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1132, + 736, + 1404, + 736, + 1404, + 772, + 1132, + 772 + ], + "score": 0.91, + "latex": "\\sigma _ { \\mathcal { P } } ( v ) \\triangleq \\operatorname* { m a x } _ { p \\in \\mathcal { P } } ( p ^ { \\top } v )" + }, + { + "category_id": 13, + "poly": [ + 575, + 623, + 624, + 623, + 624, + 660, + 575, + 660 + ], + "score": 0.91, + "latex": "Q ^ { \\pi ^ { * } }" + }, + { + "category_id": 13, + "poly": [ + 885, + 1733, + 947, + 1733, + 947, + 1767, + 885, + 1767 + ], + "score": 0.9, + "latex": "( s , a )" + }, + { + "category_id": 13, + "poly": [ + 343, + 1401, + 628, + 1401, + 628, + 1435, + 343, + 1435 + ], + "score": 0.9, + "latex": "\\mathsf { P } = \\{ p _ { s } ^ { a } , s \\in \\mathscr { S } , a \\in \\mathscr { A } \\}" + }, + { + "category_id": 13, + "poly": [ + 1065, + 704, + 1099, + 704, + 1099, + 737, + 1065, + 737 + ], + "score": 0.9, + "latex": "\\mathcal { P } _ { s } ^ { a }" + }, + { + "category_id": 13, + "poly": [ + 816, + 1734, + 851, + 1734, + 851, + 1766, + 816, + 1766 + ], + "score": 0.9, + "latex": "\\mathcal { P } _ { s } ^ { a }" + }, + { + "category_id": 13, + "poly": [ + 753, + 628, + 790, + 628, + 790, + 660, + 753, + 660 + ], + "score": 0.9, + "latex": "Q ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 366, + 1735, + 397, + 1735, + 397, + 1766, + 366, + 1766 + ], + "score": 0.89, + "latex": "p _ { s } ^ { a }" + }, + { + "category_id": 13, + "poly": [ + 723, + 740, + 755, + 740, + 755, + 771, + 723, + 771 + ], + "score": 0.89, + "latex": "p _ { s } ^ { a }" + }, + { + "category_id": 13, + "poly": [ + 708, + 1061, + 745, + 1061, + 745, + 1092, + 708, + 1092 + ], + "score": 0.89, + "latex": "Q ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 838, + 1024, + 1298, + 1024, + 1298, + 1061, + 838, + 1061 + ], + "score": 0.89, + "latex": "\\begin{array} { r } { V ^ { * } ( s ) = \\operatorname* { m i n } _ { a \\in \\mathcal { A } } \\bigl ( c ( s , a ) + \\gamma \\sigma _ { \\mathscr { P } _ { s } ^ { a } } ( V ^ { * } ) \\bigr ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 843, + 1062, + 1205, + 1062, + 1205, + 1096, + 843, + 1096 + ], + "score": 0.89, + "latex": "Q ^ { * } ( s , a ) = c ( s , a ) + \\gamma \\sigma _ { \\mathscr P _ { s } ^ { a } } ( V ^ { * } )" + }, + { + "category_id": 13, + "poly": [ + 370, + 420, + 406, + 420, + 406, + 449, + 370, + 449 + ], + "score": 0.89, + "latex": "\\mathbb { E } _ { \\kappa }" + }, + { + "category_id": 13, + "poly": [ + 694, + 1027, + 733, + 1027, + 733, + 1055, + 694, + 1055 + ], + "score": 0.88, + "latex": "V ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1212, + 335, + 1212, + 335, + 1243, + 298, + 1243 + ], + "score": 0.88, + "latex": "Q ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 1332, + 871, + 1396, + 871, + 1396, + 899, + 1332, + 899 + ], + "score": 0.88, + "latex": "s \\in \\mathcal { S }" + }, + { + "category_id": 13, + "poly": [ + 462, + 1182, + 498, + 1182, + 498, + 1213, + 462, + 1213 + ], + "score": 0.88, + "latex": "Q ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 982, + 1185, + 1029, + 1185, + 1029, + 1216, + 982, + 1216 + ], + "score": 0.87, + "latex": "\\sigma _ { \\mathcal P _ { s } ^ { a } }" + }, + { + "category_id": 13, + "poly": [ + 297, + 738, + 332, + 738, + 332, + 771, + 297, + 771 + ], + "score": 0.87, + "latex": "\\mathcal { P } _ { s } ^ { a }" + }, + { + "category_id": 13, + "poly": [ + 1175, + 704, + 1320, + 704, + 1320, + 735, + 1175, + 735 + ], + "score": 0.86, + "latex": "s \\in \\mathcal { S } , a \\in \\mathcal { A }" + }, + { + "category_id": 13, + "poly": [ + 1317, + 1181, + 1354, + 1181, + 1354, + 1208, + 1317, + 1208 + ], + "score": 0.86, + "latex": "V ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 935, + 483, + 966, + 483, + 966, + 507, + 935, + 507 + ], + "score": 0.85, + "latex": "\\pi ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 664, + 627, + 702, + 627, + 702, + 656, + 664, + 656 + ], + "score": 0.85, + "latex": "V ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 479, + 1899, + 1132, + 1899, + 1132, + 1935, + 479, + 1935 + ], + "score": 0.85, + "latex": "\\{ ( 1 - R ) p + R q | q \\in \\varDelta _ { | \\mathcal { S } | } \\} \\subset \\{ q \\in \\varDelta _ { | \\mathcal { S } | } | d _ { T V } ( p , q ) \\leq R \\}" + }, + { + "category_id": 13, + "poly": [ + 684, + 448, + 1394, + 448, + 1394, + 484, + 684, + 484 + ], + "score": 0.84, + "latex": "\\begin{array} { r } { \\pi \\colon Q ^ { \\pi } ( s , a ) = \\operatorname* { m a x } _ { \\kappa } \\mathbb { E } _ { \\kappa } \\left[ \\sum _ { t = 0 } ^ { \\infty } \\gamma ^ { t } c ( S _ { t } , A _ { t } ) | \\dot { S _ { 0 } } = s , A _ { 0 } = a , \\pi \\right] } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1227, + 1735, + 1252, + 1735, + 1252, + 1761, + 1227, + 1761 + ], + "score": 0.83, + "latex": "R" + }, + { + "category_id": 14, + "poly": [ + 442, + 1674, + 1257, + 1674, + 1257, + 1716, + 442, + 1716 + ], + "score": 0.83, + "latex": "\\begin{array} { r } { \\mathfrak { P } _ { s } ^ { a } = \\big \\{ ( 1 - R ) p _ { s } ^ { a } + R q | q \\in \\mathcal { A } _ { | \\mathcal { S } | } \\big \\} , s \\in \\mathcal { S } , a \\in \\mathcal { A } , \\mathrm { f o r ~ s o m e ~ } 0 \\le R \\le 1 . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 776, + 1630, + 800, + 1630, + 800, + 1656, + 776, + 1656 + ], + "score": 0.83, + "latex": "R" + }, + { + "category_id": 13, + "poly": [ + 932, + 1357, + 956, + 1357, + 956, + 1383, + 932, + 1383 + ], + "score": 0.81, + "latex": "R" + }, + { + "category_id": 13, + "poly": [ + 1099, + 1600, + 1123, + 1600, + 1123, + 1626, + 1099, + 1626 + ], + "score": 0.8, + "latex": "R" + }, + { + "category_id": 13, + "poly": [ + 850, + 1935, + 873, + 1935, + 873, + 1957, + 850, + 1957 + ], + "score": 0.78, + "latex": "R" + }, + { + "category_id": 13, + "poly": [ + 1117, + 770, + 1139, + 770, + 1139, + 796, + 1117, + 796 + ], + "score": 0.78, + "latex": "\\mathcal { P }" + }, + { + "category_id": 13, + "poly": [ + 811, + 1872, + 833, + 1872, + 833, + 1897, + 811, + 1897 + ], + "score": 0.78, + "latex": "R" + }, + { + "category_id": 13, + "poly": [ + 936, + 1153, + 960, + 1153, + 960, + 1178, + 936, + 1178 + ], + "score": 0.76, + "latex": "\\mathbf { T }" + }, + { + "category_id": 13, + "poly": [ + 1180, + 210, + 1200, + 210, + 1200, + 231, + 1180, + 231 + ], + "score": 0.75, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 653, + 775, + 671, + 775, + 671, + 796, + 653, + 796 + ], + "score": 0.74, + "latex": "v" + }, + { + "category_id": 13, + "poly": [ + 1015, + 239, + 1035, + 239, + 1035, + 261, + 1015, + 261 + ], + "score": 0.73, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 435, + 1840, + 459, + 1840, + 459, + 1867, + 435, + 1867 + ], + "score": 0.73, + "latex": "R" + }, + { + "category_id": 13, + "poly": [ + 1070, + 425, + 1087, + 425, + 1087, + 446, + 1070, + 446 + ], + "score": 0.72, + "latex": "\\kappa" + }, + { + "category_id": 13, + "poly": [ + 712, + 1121, + 739, + 1121, + 739, + 1149, + 712, + 1149 + ], + "score": 0.64, + "latex": "\\mathbf { T }" + }, + { + "category_id": 13, + "poly": [ + 297, + 704, + 322, + 704, + 322, + 731, + 297, + 731 + ], + "score": 0.64, + "latex": "\\mathbf { P }" + }, + { + "category_id": 13, + "poly": [ + 1174, + 1876, + 1191, + 1876, + 1191, + 1902, + 1174, + 1902 + ], + "score": 0.5, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 738, + 1462, + 761, + 1462, + 761, + 1490, + 738, + 1490 + ], + "score": 0.48, + "latex": "\\mathsf { P }" + }, + { + "category_id": 13, + "poly": [ + 713, + 1433, + 735, + 1433, + 735, + 1459, + 713, + 1459 + ], + "score": 0.43, + "latex": "\\mathsf { P }" + }, + { + "category_id": 13, + "poly": [ + 397, + 1871, + 435, + 1871, + 435, + 1897, + 397, + 1897 + ], + "score": 0.32, + "latex": "K L" + }, + { + "category_id": 13, + "poly": [ + 678, + 1524, + 700, + 1524, + 700, + 1550, + 678, + 1550 + ], + "score": 0.3, + "latex": "\\mathsf { P }" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1281.0, + 1187.0, + 1281.0, + 1187.0, + 1332.0, + 289.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2059.0, + 862.0, + 2059.0, + 862.0, + 2091.0, + 839.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 418.0, + 369.0, + 418.0, + 369.0, + 450.0, + 296.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 418.0, + 1069.0, + 418.0, + 1069.0, + 450.0, + 407.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 418.0, + 1406.0, + 418.0, + 1406.0, + 450.0, + 1088.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 441.0, + 683.0, + 441.0, + 683.0, + 494.0, + 286.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1395.0, + 441.0, + 1412.0, + 441.0, + 1412.0, + 494.0, + 1395.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 477.0, + 934.0, + 477.0, + 934.0, + 515.0, + 293.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 967.0, + 477.0, + 1411.0, + 477.0, + 1411.0, + 515.0, + 967.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 509.0, + 544.0, + 509.0, + 544.0, + 542.0, + 294.0, + 542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1117.0, + 711.0, + 1117.0, + 711.0, + 1157.0, + 292.0, + 1157.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 1117.0, + 778.0, + 1117.0, + 778.0, + 1157.0, + 740.0, + 1157.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1309.0, + 1117.0, + 1406.0, + 1117.0, + 1406.0, + 1157.0, + 1309.0, + 1157.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1150.0, + 935.0, + 1150.0, + 935.0, + 1183.0, + 296.0, + 1183.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 1150.0, + 1403.0, + 1150.0, + 1403.0, + 1183.0, + 961.0, + 1183.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1180.0, + 461.0, + 1180.0, + 461.0, + 1217.0, + 293.0, + 1217.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 1180.0, + 981.0, + 1180.0, + 981.0, + 1217.0, + 499.0, + 1217.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 1180.0, + 1316.0, + 1180.0, + 1316.0, + 1217.0, + 1030.0, + 1217.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1355.0, + 1180.0, + 1406.0, + 1180.0, + 1406.0, + 1217.0, + 1355.0, + 1217.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 1211.0, + 1292.0, + 1211.0, + 1292.0, + 1248.0, + 336.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 668.0, + 1406.0, + 668.0, + 1406.0, + 707.0, + 293.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 700.0, + 296.0, + 700.0, + 296.0, + 739.0, + 292.0, + 739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 700.0, + 1064.0, + 700.0, + 1064.0, + 739.0, + 323.0, + 739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 700.0, + 1174.0, + 700.0, + 1174.0, + 739.0, + 1100.0, + 739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1321.0, + 700.0, + 1404.0, + 700.0, + 1404.0, + 739.0, + 1321.0, + 739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 733.0, + 296.0, + 733.0, + 296.0, + 775.0, + 292.0, + 775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 333.0, + 733.0, + 722.0, + 733.0, + 722.0, + 775.0, + 333.0, + 775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 756.0, + 733.0, + 1131.0, + 733.0, + 1131.0, + 775.0, + 756.0, + 775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 766.0, + 652.0, + 766.0, + 652.0, + 806.0, + 293.0, + 806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 766.0, + 1116.0, + 766.0, + 1116.0, + 806.0, + 672.0, + 806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 766.0, + 1407.0, + 766.0, + 1407.0, + 806.0, + 1140.0, + 806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 796.0, + 1408.0, + 796.0, + 1408.0, + 832.0, + 293.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 829.0, + 474.0, + 829.0, + 474.0, + 865.0, + 293.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1400.0, + 342.0, + 1400.0, + 342.0, + 1437.0, + 294.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 1400.0, + 1407.0, + 1400.0, + 1407.0, + 1437.0, + 629.0, + 1437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1431.0, + 712.0, + 1431.0, + 712.0, + 1465.0, + 292.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 1431.0, + 1408.0, + 1431.0, + 1408.0, + 1465.0, + 736.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1460.0, + 737.0, + 1460.0, + 737.0, + 1495.0, + 292.0, + 1495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 1460.0, + 1406.0, + 1460.0, + 1406.0, + 1495.0, + 762.0, + 1495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1492.0, + 1404.0, + 1492.0, + 1404.0, + 1526.0, + 294.0, + 1526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1522.0, + 677.0, + 1522.0, + 677.0, + 1556.0, + 292.0, + 1556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1522.0, + 1393.0, + 1522.0, + 1393.0, + 1556.0, + 701.0, + 1556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1568.0, + 1408.0, + 1568.0, + 1408.0, + 1602.0, + 295.0, + 1602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1599.0, + 1098.0, + 1599.0, + 1098.0, + 1630.0, + 296.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 1599.0, + 1404.0, + 1599.0, + 1404.0, + 1630.0, + 1124.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1629.0, + 775.0, + 1629.0, + 775.0, + 1663.0, + 295.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 801.0, + 1629.0, + 1368.0, + 1629.0, + 1368.0, + 1663.0, + 801.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1728.0, + 365.0, + 1728.0, + 365.0, + 1771.0, + 292.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 398.0, + 1728.0, + 815.0, + 1728.0, + 815.0, + 1771.0, + 398.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1728.0, + 884.0, + 1728.0, + 884.0, + 1771.0, + 852.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 948.0, + 1728.0, + 1226.0, + 1728.0, + 1226.0, + 1771.0, + 948.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1253.0, + 1728.0, + 1405.0, + 1728.0, + 1405.0, + 1771.0, + 1253.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1763.0, + 1405.0, + 1763.0, + 1405.0, + 1797.0, + 295.0, + 1797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 1779.0, + 703.0, + 1779.0, + 703.0, + 1843.0, + 286.0, + 1843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 924.0, + 1779.0, + 943.0, + 1779.0, + 943.0, + 1843.0, + 924.0, + 1843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 202.0, + 1179.0, + 202.0, + 1179.0, + 238.0, + 294.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1201.0, + 202.0, + 1404.0, + 202.0, + 1404.0, + 238.0, + 1201.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 232.0, + 1014.0, + 232.0, + 1014.0, + 267.0, + 293.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 232.0, + 1405.0, + 232.0, + 1405.0, + 267.0, + 1036.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 265.0, + 475.0, + 265.0, + 475.0, + 300.0, + 295.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1838.0, + 434.0, + 1838.0, + 434.0, + 1874.0, + 294.0, + 1874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 460.0, + 1838.0, + 1408.0, + 1838.0, + 1408.0, + 1874.0, + 460.0, + 1874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1871.0, + 396.0, + 1871.0, + 396.0, + 1902.0, + 296.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 1871.0, + 810.0, + 1871.0, + 810.0, + 1902.0, + 436.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 834.0, + 1871.0, + 1173.0, + 1871.0, + 1173.0, + 1902.0, + 834.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1192.0, + 1871.0, + 1404.0, + 1871.0, + 1404.0, + 1902.0, + 1192.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1895.0, + 478.0, + 1895.0, + 478.0, + 1939.0, + 293.0, + 1939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1133.0, + 1895.0, + 1411.0, + 1895.0, + 1411.0, + 1939.0, + 1133.0, + 1939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1929.0, + 849.0, + 1929.0, + 849.0, + 1965.0, + 294.0, + 1965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 1929.0, + 1404.0, + 1929.0, + 1404.0, + 1965.0, + 874.0, + 1965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1968.0, + 1112.0, + 1968.0, + 1112.0, + 2014.0, + 290.0, + 2014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 1968.0, + 1408.0, + 1968.0, + 1408.0, + 2014.0, + 1398.0, + 2014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1022.0, + 693.0, + 1022.0, + 693.0, + 1064.0, + 293.0, + 1064.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 734.0, + 1022.0, + 837.0, + 1022.0, + 837.0, + 1064.0, + 734.0, + 1064.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1299.0, + 1022.0, + 1405.0, + 1022.0, + 1405.0, + 1064.0, + 1299.0, + 1064.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1054.0, + 707.0, + 1054.0, + 707.0, + 1100.0, + 293.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 1054.0, + 842.0, + 1054.0, + 842.0, + 1100.0, + 746.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 1054.0, + 1218.0, + 1054.0, + 1218.0, + 1100.0, + 1206.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1352.0, + 931.0, + 1352.0, + 931.0, + 1390.0, + 294.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 1352.0, + 1210.0, + 1352.0, + 1210.0, + 1390.0, + 957.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 866.0, + 1331.0, + 866.0, + 1331.0, + 906.0, + 296.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1397.0, + 866.0, + 1407.0, + 866.0, + 1407.0, + 906.0, + 1397.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 618.0, + 471.0, + 618.0, + 471.0, + 668.0, + 291.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 618.0, + 574.0, + 618.0, + 574.0, + 668.0, + 522.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 625.0, + 618.0, + 663.0, + 618.0, + 663.0, + 668.0, + 625.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 618.0, + 752.0, + 618.0, + 752.0, + 668.0, + 703.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 791.0, + 618.0, + 992.0, + 618.0, + 992.0, + 668.0, + 791.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1297.0, + 618.0, + 1308.0, + 618.0, + 1308.0, + 668.0, + 1297.0, + 668.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1224, + 1405, + 1224, + 1405, + 1470, + 296, + 1470 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1806, + 1404, + 1806, + 1404, + 1961, + 298, + 1961 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 296, + 405, + 1404, + 405, + 1404, + 680, + 296, + 680 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 297, + 268, + 1404, + 268, + 1404, + 393, + 297, + 393 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 300, + 1645, + 1398, + 1645, + 1398, + 1709, + 300, + 1709 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 299, + 1482, + 1400, + 1482, + 1400, + 1546, + 299, + 1546 + ], + "score": 0.947 + }, + { + "category_id": 1, + "poly": [ + 295, + 1718, + 1401, + 1718, + 1401, + 1783, + 295, + 1783 + ], + "score": 0.947 + }, + { + "category_id": 1, + "poly": [ + 300, + 1555, + 1397, + 1555, + 1397, + 1620, + 300, + 1620 + ], + "score": 0.936 + }, + { + "category_id": 0, + "poly": [ + 298, + 197, + 840, + 197, + 840, + 238, + 298, + 238 + ], + "score": 0.933 + }, + { + "category_id": 8, + "poly": [ + 377, + 1139, + 1319, + 1139, + 1319, + 1190, + 377, + 1190 + ], + "score": 0.919 + }, + { + "category_id": 1, + "poly": [ + 291, + 1088, + 1366, + 1088, + 1366, + 1124, + 291, + 1124 + ], + "score": 0.917 + }, + { + "category_id": 1, + "poly": [ + 288, + 1974, + 1377, + 1974, + 1377, + 2008, + 288, + 2008 + ], + "score": 0.908 + }, + { + "category_id": 1, + "poly": [ + 299, + 758, + 1245, + 758, + 1245, + 792, + 299, + 792 + ], + "score": 0.891 + }, + { + "category_id": 0, + "poly": [ + 298, + 719, + 666, + 719, + 666, + 751, + 298, + 751 + ], + "score": 0.855 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1144, + 1400, + 1144, + 1400, + 1175, + 1366, + 1175 + ], + "score": 0.848 + }, + { + "category_id": 1, + "poly": [ + 307, + 795, + 1015, + 795, + 1015, + 998, + 307, + 998 + ], + "score": 0.847 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 858, + 2061, + 858, + 2085, + 841, + 2085 + ], + "score": 0.732 + }, + { + "category_id": 1, + "poly": [ + 298, + 1019, + 440, + 1019, + 440, + 1050, + 298, + 1050 + ], + "score": 0.495 + }, + { + "category_id": 1, + "poly": [ + 310, + 985, + 435, + 985, + 435, + 1011, + 310, + 1011 + ], + "score": 0.182 + }, + { + "category_id": 13, + "poly": [ + 397, + 360, + 473, + 360, + 473, + 397, + 397, + 397 + ], + "score": 0.94, + "latex": "\\sigma _ { \\mathcal { P } _ { s } ^ { a } } ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 1279, + 643, + 1405, + 643, + 1405, + 682, + 1279, + 682 + ], + "score": 0.94, + "latex": "\\sigma _ { \\hat { \\Phi } _ { t } } ( V _ { t } ) =" + }, + { + "category_id": 13, + "poly": [ + 958, + 1257, + 1180, + 1257, + 1180, + 1289, + 958, + 1289 + ], + "score": 0.93, + "latex": "p _ { s } ^ { a } , \\forall s \\in \\mathcal { S } , a \\in \\mathcal { A }" + }, + { + "category_id": 13, + "poly": [ + 831, + 1317, + 913, + 1317, + 913, + 1351, + 831, + 1351 + ], + "score": 0.93, + "latex": "| \\dot { \\mathcal { S } } | ^ { 2 } | \\mathcal { A } |" + }, + { + "category_id": 13, + "poly": [ + 352, + 1588, + 463, + 1588, + 463, + 1619, + 352, + 1619 + ], + "score": 0.92, + "latex": "Q _ { t } \\to Q ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 751, + 1286, + 855, + 1286, + 855, + 1319, + 751, + 1319 + ], + "score": 0.92, + "latex": "| \\mathcal { S } | \\times | \\mathcal { A } |" + }, + { + "category_id": 13, + "poly": [ + 297, + 471, + 512, + 471, + 512, + 505, + 297, + 505 + ], + "score": 0.92, + "latex": "O _ { t } = ( s _ { t } , a _ { t } , s _ { t + 1 } )" + }, + { + "category_id": 13, + "poly": [ + 907, + 643, + 1000, + 643, + 1000, + 682, + 907, + 682 + ], + "score": 0.92, + "latex": "\\sigma _ { \\hat { \\mathcal { P } } _ { t } } ( V _ { t } )" + }, + { + "category_id": 13, + "poly": [ + 440, + 571, + 535, + 571, + 535, + 603, + 440, + 603 + ], + "score": 0.92, + "latex": "A _ { t } = a _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1011, + 566, + 1401, + 566, + 1401, + 608, + 1011, + 608 + ], + "score": 0.92, + "latex": "\\hat { \\mathcal { P } } _ { t } \\triangleq \\left\\{ ( 1 - R ) \\hat { p } _ { t } + R q | q \\in \\Delta _ { | \\mathcal { S } | } \\right\\}" + }, + { + "category_id": 13, + "poly": [ + 298, + 571, + 387, + 571, + 387, + 603, + 298, + 603 + ], + "score": 0.92, + "latex": "S _ { t } = s _ { t }" + }, + { + "category_id": 13, + "poly": [ + 698, + 504, + 760, + 504, + 760, + 536, + 698, + 536 + ], + "score": 0.91, + "latex": "\\mathbb { 1 } _ { s _ { t + 1 } }" + }, + { + "category_id": 13, + "poly": [ + 678, + 758, + 843, + 758, + 843, + 792, + 678, + 792 + ], + "score": 0.91, + "latex": "( s , a ) \\in \\mathcal { S } \\times \\mathcal { A }" + }, + { + "category_id": 13, + "poly": [ + 373, + 951, + 631, + 951, + 631, + 985, + 373, + 985 + ], + "score": 0.91, + "latex": "Q _ { t + 1 } ( s , a ) \\gets Q _ { t } ( s , a )" + }, + { + "category_id": 13, + "poly": [ + 500, + 758, + 598, + 758, + 598, + 792, + 500, + 792 + ], + "score": 0.91, + "latex": "Q _ { 0 } ( s , a )" + }, + { + "category_id": 13, + "poly": [ + 1230, + 1555, + 1399, + 1555, + 1399, + 1593, + 1230, + 1593 + ], + "score": 0.9, + "latex": "\\textstyle \\sum _ { t = 0 } ^ { \\infty } \\alpha _ { t } ^ { 2 } < \\infty" + }, + { + "category_id": 13, + "poly": [ + 298, + 1751, + 499, + 1751, + 499, + 1784, + 298, + 1784 + ], + "score": 0.9, + "latex": "p _ { s } ^ { a } , \\forall s \\in \\mathcal { S } , a \\in \\mathcal { A }" + }, + { + "category_id": 13, + "poly": [ + 602, + 437, + 671, + 437, + 671, + 466, + 602, + 466 + ], + "score": 0.9, + "latex": "s \\in \\mathcal { S }" + }, + { + "category_id": 14, + "poly": [ + 377, + 1139, + 1321, + 1139, + 1321, + 1192, + 377, + 1192 + ], + "score": 0.9, + "latex": "Q _ { t + 1 } ( s _ { t } , a _ { t } ) \\gets ( 1 - \\alpha _ { t } ) Q _ { t } ( s _ { t } , a _ { t } ) + \\alpha _ { t } ( c _ { t } + \\gamma R \\operatorname* { m a x } _ { s \\in \\mathcal { S } } V _ { t } ( s ) + \\gamma ( 1 - R ) V _ { t } ( s _ { t + 1 } ) ) ." + }, + { + "category_id": 13, + "poly": [ + 1009, + 1318, + 1213, + 1318, + 1213, + 1350, + 1009, + 1350 + ], + "score": 0.9, + "latex": "p _ { s } ^ { a } , \\forall s \\in \\mathcal { S } , a \\in \\mathcal { A }" + }, + { + "category_id": 13, + "poly": [ + 1132, + 1809, + 1343, + 1809, + 1343, + 1841, + 1132, + 1841 + ], + "score": 0.9, + "latex": "p _ { s } ^ { a } , \\forall s \\in \\mathcal { S } , a \\in \\mathcal { A }" + }, + { + "category_id": 13, + "poly": [ + 544, + 611, + 586, + 611, + 586, + 647, + 544, + 647 + ], + "score": 0.9, + "latex": "\\mathcal { P } _ { s _ { t } } ^ { a _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 725, + 438, + 802, + 438, + 802, + 466, + 725, + 466 + ], + "score": 0.9, + "latex": "a \\in { \\mathcal { A } }" + }, + { + "category_id": 13, + "poly": [ + 638, + 824, + 727, + 824, + 727, + 858, + 638, + 858 + ], + "score": 0.9, + "latex": "\\pi _ { b } ( \\cdot | s _ { t } )" + }, + { + "category_id": 13, + "poly": [ + 566, + 503, + 605, + 503, + 605, + 538, + 566, + 538 + ], + "score": 0.9, + "latex": "p _ { s _ { t } } ^ { a _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 472, + 437, + 507, + 437, + 507, + 470, + 472, + 470 + ], + "score": 0.89, + "latex": "\\mathcal { P } _ { s } ^ { a }" + }, + { + "category_id": 13, + "poly": [ + 834, + 1810, + 909, + 1810, + 909, + 1837, + 834, + 1837 + ], + "score": 0.89, + "latex": "\\mathcal { S } \\times \\mathcal { A }" + }, + { + "category_id": 13, + "poly": [ + 747, + 1515, + 784, + 1515, + 784, + 1547, + 747, + 1547 + ], + "score": 0.89, + "latex": "Q ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 1015, + 1556, + 1181, + 1556, + 1181, + 1593, + 1015, + 1593 + ], + "score": 0.89, + "latex": "\\textstyle \\sum _ { t = 0 } ^ { \\infty } \\alpha _ { t } = \\infty" + }, + { + "category_id": 13, + "poly": [ + 510, + 1840, + 868, + 1840, + 868, + 1875, + 510, + 1875 + ], + "score": 0.89, + "latex": "\\begin{array} { r } { \\mu _ { \\mathrm { m i n } } = \\operatorname* { m i n } _ { ( s , a ) \\in \\mathcal { S } \\times \\mathcal { A } } \\mu _ { \\pi _ { b } } ( s , a ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1122, + 408, + 1152, + 408, + 1152, + 440, + 1122, + 440 + ], + "score": 0.89, + "latex": "p _ { s } ^ { a }" + }, + { + "category_id": 13, + "poly": [ + 499, + 1590, + 584, + 1590, + 584, + 1616, + 499, + 1616 + ], + "score": 0.89, + "latex": "t \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 296, + 1089, + 706, + 1089, + 706, + 1125, + 296, + 1125 + ], + "score": 0.89, + "latex": "\\begin{array} { r } { R \\operatorname* { m a x } _ { s \\in \\mathcal { S } } V _ { t } ( s ) + ( 1 - R ) V _ { t } ( s _ { t + 1 } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1052, + 606, + 1083, + 606, + 1083, + 641, + 1052, + 641 + ], + "score": 0.88, + "latex": "\\hat { \\mathcal { P } } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1349, + 510, + 1402, + 510, + 1402, + 535, + 1349, + 535 + ], + "score": 0.88, + "latex": "s _ { t + 1 }" + }, + { + "category_id": 13, + "poly": [ + 1170, + 535, + 1209, + 535, + 1209, + 567, + 1170, + 567 + ], + "score": 0.87, + "latex": "p _ { s _ { t } } ^ { a _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 343, + 1813, + 386, + 1813, + 386, + 1841, + 343, + 1841 + ], + "score": 0.87, + "latex": "\\mu _ { \\pi _ { b } }" + }, + { + "category_id": 13, + "poly": [ + 667, + 472, + 701, + 472, + 701, + 502, + 667, + 502 + ], + "score": 0.87, + "latex": "O _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1356, + 1900, + 1402, + 1900, + 1402, + 1931, + 1356, + 1931 + ], + "score": 0.87, + "latex": "d _ { \\mathrm { T V } }" + }, + { + "category_id": 13, + "poly": [ + 470, + 861, + 525, + 861, + 525, + 888, + 470, + 888 + ], + "score": 0.87, + "latex": "s _ { t + 1 }" + }, + { + "category_id": 13, + "poly": [ + 400, + 1017, + 442, + 1017, + 442, + 1050, + 400, + 1050 + ], + "score": 0.86, + "latex": "Q _ { T }" + }, + { + "category_id": 13, + "poly": [ + 613, + 1898, + 1265, + 1898, + 1265, + 1937, + 613, + 1937 + ], + "score": 0.85, + "latex": "\\begin{array} { r } { t _ { \\mathrm { m i x } } = \\operatorname* { m i n } \\left. t : \\operatorname* { m a x } _ { s \\in \\mathcal { S } } \\mathsf { \\bar { d } _ { \\mathrm { T V } } } ( \\mu _ { \\pi } , P ( s _ { t } = \\cdot | s _ { 0 } = s ) ) \\breve { \\leq } \\frac { 1 } { 4 } \\right. } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 386, + 795, + 617, + 795, + 617, + 826, + 386, + 826 + ], + "score": 0.85, + "latex": "t = 0 , 1 , 2 , . . . , T - 1" + }, + { + "category_id": 13, + "poly": [ + 1074, + 1725, + 1105, + 1725, + 1105, + 1751, + 1074, + 1751 + ], + "score": 0.85, + "latex": "\\pi _ { b }" + }, + { + "category_id": 13, + "poly": [ + 1048, + 1814, + 1078, + 1814, + 1078, + 1839, + 1048, + 1839 + ], + "score": 0.85, + "latex": "\\pi _ { b }" + }, + { + "category_id": 13, + "poly": [ + 461, + 830, + 489, + 830, + 489, + 855, + 461, + 855 + ], + "score": 0.85, + "latex": "a _ { t }" + }, + { + "category_id": 13, + "poly": [ + 859, + 1562, + 889, + 1562, + 889, + 1588, + 859, + 1588 + ], + "score": 0.84, + "latex": "\\alpha _ { t }" + }, + { + "category_id": 13, + "poly": [ + 677, + 951, + 858, + 951, + 858, + 985, + 677, + 985 + ], + "score": 0.83, + "latex": "( s , a ) \\neq ( s _ { t } , a _ { t } )" + }, + { + "category_id": 13, + "poly": [ + 1278, + 468, + 1401, + 468, + 1401, + 506, + 1278, + 506 + ], + "score": 0.83, + "latex": "\\hat { p } _ { t } \\triangleq \\mathbb { 1 } _ { s _ { t + 1 } }" + }, + { + "category_id": 13, + "poly": [ + 575, + 861, + 600, + 861, + 600, + 886, + 575, + 886 + ], + "score": 0.81, + "latex": "c _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1208, + 440, + 1222, + 440, + 1222, + 464, + 1208, + 464 + ], + "score": 0.76, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 371, + 887, + 750, + 887, + 750, + 917, + 371, + 917 + ], + "score": 0.74, + "latex": "\\begin{array} { r } { V _ { t } ( s ) \\operatorname* { m i n } _ { a \\in \\mathcal { A } } Q _ { t } ( s , a ) , \\forall s \\in \\mathfrak { s } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1216, + 765, + 1245, + 765, + 1245, + 789, + 1216, + 789 + ], + "score": 0.7, + "latex": "\\alpha _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1034, + 766, + 1063, + 766, + 1063, + 790, + 1034, + 790 + ], + "score": 0.69, + "latex": "\\pi _ { b }" + }, + { + "category_id": 13, + "poly": [ + 1075, + 766, + 1102, + 766, + 1102, + 790, + 1075, + 790 + ], + "score": 0.61, + "latex": "s _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 883, + 646, + 897, + 646, + 897, + 671, + 883, + 671 + ], + "score": 0.59, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 466, + 759, + 489, + 759, + 489, + 786, + 466, + 786 + ], + "score": 0.58, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 1099, + 331, + 1124, + 331, + 1124, + 359, + 1099, + 359 + ], + "score": 0.51, + "latex": "\\mathbf { P }" + }, + { + "category_id": 13, + "poly": [ + 771, + 1590, + 787, + 1590, + 787, + 1615, + 771, + 1615 + ], + "score": 0.49, + "latex": "^ { l }" + }, + { + "category_id": 13, + "poly": [ + 450, + 916, + 1008, + 916, + 1008, + 950, + 450, + 950 + ], + "score": 0.28, + "latex": "s _ { t } , a _ { t } ) \\gets ( 1 - \\alpha _ { t } ) Q _ { t } ( s _ { t } , a _ { t } ) + \\alpha _ { t } ( c _ { t } + \\gamma \\sigma _ { \\hat { \\mathcal P } _ { t } } ( V _ { t } ) )" + }, + { + "category_id": 13, + "poly": [ + 678, + 886, + 761, + 886, + 761, + 915, + 678, + 915 + ], + "score": 0.26, + "latex": "\\forall s \\in \\mathcal { S }" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 190.0, + 844.0, + 190.0, + 844.0, + 247.0, + 290.0, + 247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 713.0, + 668.0, + 713.0, + 668.0, + 760.0, + 294.0, + 760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2058.0, + 861.0, + 2058.0, + 861.0, + 2093.0, + 839.0, + 2093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1224.0, + 1406.0, + 1224.0, + 1406.0, + 1261.0, + 296.0, + 1261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1253.0, + 957.0, + 1253.0, + 957.0, + 1294.0, + 293.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 1253.0, + 1407.0, + 1253.0, + 1407.0, + 1294.0, + 1181.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1287.0, + 750.0, + 1287.0, + 750.0, + 1321.0, + 294.0, + 1321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 856.0, + 1287.0, + 1408.0, + 1287.0, + 1408.0, + 1321.0, + 856.0, + 1321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1316.0, + 830.0, + 1316.0, + 830.0, + 1353.0, + 294.0, + 1353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 1316.0, + 1008.0, + 1316.0, + 1008.0, + 1353.0, + 914.0, + 1353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1214.0, + 1316.0, + 1406.0, + 1316.0, + 1406.0, + 1353.0, + 1214.0, + 1353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1349.0, + 1405.0, + 1349.0, + 1405.0, + 1381.0, + 294.0, + 1381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1378.0, + 1405.0, + 1378.0, + 1405.0, + 1412.0, + 294.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1408.0, + 1405.0, + 1408.0, + 1405.0, + 1442.0, + 294.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1440.0, + 892.0, + 1440.0, + 892.0, + 1473.0, + 295.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1805.0, + 342.0, + 1805.0, + 342.0, + 1846.0, + 293.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 387.0, + 1805.0, + 833.0, + 1805.0, + 833.0, + 1846.0, + 387.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 910.0, + 1805.0, + 1047.0, + 1805.0, + 1047.0, + 1846.0, + 910.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1079.0, + 1805.0, + 1131.0, + 1805.0, + 1131.0, + 1846.0, + 1079.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1344.0, + 1805.0, + 1406.0, + 1805.0, + 1406.0, + 1846.0, + 1344.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1836.0, + 509.0, + 1836.0, + 509.0, + 1881.0, + 292.0, + 1881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 1836.0, + 1409.0, + 1836.0, + 1409.0, + 1881.0, + 869.0, + 1881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1870.0, + 1405.0, + 1870.0, + 1405.0, + 1902.0, + 293.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1894.0, + 612.0, + 1894.0, + 612.0, + 1939.0, + 292.0, + 1939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1266.0, + 1894.0, + 1355.0, + 1894.0, + 1355.0, + 1939.0, + 1266.0, + 1939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1403.0, + 1894.0, + 1406.0, + 1894.0, + 1406.0, + 1939.0, + 1403.0, + 1939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1927.0, + 627.0, + 1927.0, + 627.0, + 1963.0, + 292.0, + 1963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 406.0, + 1121.0, + 406.0, + 1121.0, + 438.0, + 296.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1153.0, + 406.0, + 1402.0, + 406.0, + 1402.0, + 438.0, + 1153.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 432.0, + 471.0, + 432.0, + 471.0, + 475.0, + 291.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 432.0, + 601.0, + 432.0, + 601.0, + 475.0, + 508.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 432.0, + 724.0, + 432.0, + 724.0, + 475.0, + 672.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 803.0, + 432.0, + 1207.0, + 432.0, + 1207.0, + 475.0, + 803.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1223.0, + 432.0, + 1405.0, + 432.0, + 1405.0, + 475.0, + 1223.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 462.0, + 296.0, + 462.0, + 296.0, + 517.0, + 291.0, + 517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 462.0, + 666.0, + 462.0, + 666.0, + 517.0, + 513.0, + 517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 462.0, + 1277.0, + 462.0, + 1277.0, + 517.0, + 702.0, + 517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 462.0, + 1406.0, + 462.0, + 1406.0, + 517.0, + 1402.0, + 517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 497.0, + 565.0, + 497.0, + 565.0, + 544.0, + 291.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 497.0, + 697.0, + 497.0, + 697.0, + 544.0, + 606.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 497.0, + 1348.0, + 497.0, + 1348.0, + 544.0, + 761.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1403.0, + 497.0, + 1409.0, + 497.0, + 1409.0, + 544.0, + 1403.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 532.0, + 1169.0, + 532.0, + 1169.0, + 569.0, + 294.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 532.0, + 1406.0, + 532.0, + 1406.0, + 569.0, + 1210.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 563.0, + 297.0, + 563.0, + 297.0, + 613.0, + 293.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 563.0, + 439.0, + 563.0, + 439.0, + 613.0, + 388.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 563.0, + 1010.0, + 563.0, + 1010.0, + 613.0, + 536.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 563.0, + 1405.0, + 563.0, + 1405.0, + 613.0, + 1402.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 605.0, + 543.0, + 605.0, + 543.0, + 649.0, + 293.0, + 649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 605.0, + 1051.0, + 605.0, + 1051.0, + 649.0, + 587.0, + 649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1084.0, + 605.0, + 1405.0, + 605.0, + 1405.0, + 649.0, + 1084.0, + 649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 638.0, + 882.0, + 638.0, + 882.0, + 684.0, + 291.0, + 684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 898.0, + 638.0, + 906.0, + 638.0, + 906.0, + 684.0, + 898.0, + 684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 638.0, + 1278.0, + 638.0, + 1278.0, + 684.0, + 1001.0, + 684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 271.0, + 1401.0, + 271.0, + 1401.0, + 300.0, + 296.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 299.0, + 1404.0, + 299.0, + 1404.0, + 334.0, + 294.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 330.0, + 1098.0, + 330.0, + 1098.0, + 367.0, + 295.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.0, + 330.0, + 1406.0, + 330.0, + 1406.0, + 367.0, + 1125.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 359.0, + 396.0, + 359.0, + 396.0, + 396.0, + 292.0, + 396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 474.0, + 359.0, + 1033.0, + 359.0, + 1033.0, + 396.0, + 474.0, + 396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1645.0, + 1401.0, + 1645.0, + 1401.0, + 1681.0, + 296.0, + 1681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1674.0, + 1357.0, + 1674.0, + 1357.0, + 1715.0, + 292.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1477.0, + 1405.0, + 1477.0, + 1405.0, + 1522.0, + 293.0, + 1522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1514.0, + 746.0, + 1514.0, + 746.0, + 1550.0, + 295.0, + 1550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 1514.0, + 793.0, + 1514.0, + 793.0, + 1550.0, + 785.0, + 1550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1716.0, + 1073.0, + 1716.0, + 1073.0, + 1756.0, + 294.0, + 1756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1106.0, + 1716.0, + 1407.0, + 1716.0, + 1407.0, + 1756.0, + 1106.0, + 1756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1749.0, + 297.0, + 1749.0, + 297.0, + 1789.0, + 293.0, + 1789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 500.0, + 1749.0, + 740.0, + 1749.0, + 740.0, + 1789.0, + 500.0, + 1789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 1542.0, + 858.0, + 1542.0, + 858.0, + 1607.0, + 287.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 1542.0, + 1014.0, + 1542.0, + 1014.0, + 1607.0, + 890.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1182.0, + 1542.0, + 1229.0, + 1542.0, + 1229.0, + 1607.0, + 1182.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1400.0, + 1542.0, + 1410.0, + 1542.0, + 1410.0, + 1607.0, + 1400.0, + 1607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1586.0, + 351.0, + 1586.0, + 351.0, + 1622.0, + 296.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 1586.0, + 498.0, + 1586.0, + 498.0, + 1622.0, + 464.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 1586.0, + 770.0, + 1586.0, + 770.0, + 1622.0, + 585.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 1586.0, + 797.0, + 1586.0, + 797.0, + 1622.0, + 788.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 1088.0, + 1369.0, + 1088.0, + 1369.0, + 1127.0, + 707.0, + 1127.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1973.0, + 1380.0, + 1973.0, + 1380.0, + 2012.0, + 295.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 751.0, + 465.0, + 751.0, + 465.0, + 799.0, + 294.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 751.0, + 499.0, + 751.0, + 499.0, + 799.0, + 490.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 599.0, + 751.0, + 677.0, + 751.0, + 677.0, + 799.0, + 599.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 844.0, + 751.0, + 1033.0, + 751.0, + 1033.0, + 799.0, + 844.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 751.0, + 1074.0, + 751.0, + 1074.0, + 799.0, + 1064.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 751.0, + 1215.0, + 751.0, + 1215.0, + 799.0, + 1103.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 751.0, + 1250.0, + 751.0, + 1250.0, + 799.0, + 1246.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 793.0, + 385.0, + 793.0, + 385.0, + 825.0, + 307.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 618.0, + 793.0, + 658.0, + 793.0, + 658.0, + 825.0, + 618.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 825.0, + 338.0, + 825.0, + 338.0, + 857.0, + 310.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 823.0, + 460.0, + 823.0, + 460.0, + 860.0, + 369.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 823.0, + 637.0, + 823.0, + 637.0, + 860.0, + 490.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 853.0, + 338.0, + 853.0, + 338.0, + 887.0, + 310.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 853.0, + 469.0, + 853.0, + 469.0, + 890.0, + 370.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 853.0, + 574.0, + 853.0, + 574.0, + 890.0, + 526.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 853.0, + 605.0, + 853.0, + 605.0, + 890.0, + 601.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 885.0, + 338.0, + 885.0, + 338.0, + 916.0, + 309.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 367.0, + 881.0, + 370.0, + 881.0, + 370.0, + 922.0, + 367.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 881.0, + 765.0, + 881.0, + 765.0, + 922.0, + 762.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 915.0, + 338.0, + 915.0, + 338.0, + 948.0, + 309.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 914.0, + 449.0, + 914.0, + 449.0, + 951.0, + 370.0, + 951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1009.0, + 914.0, + 1013.0, + 914.0, + 1013.0, + 951.0, + 1009.0, + 951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 951.0, + 338.0, + 951.0, + 338.0, + 983.0, + 309.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 949.0, + 372.0, + 949.0, + 372.0, + 987.0, + 368.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 632.0, + 949.0, + 676.0, + 949.0, + 676.0, + 987.0, + 632.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 981.0, + 341.0, + 981.0, + 341.0, + 1005.0, + 310.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 982.0, + 437.0, + 982.0, + 437.0, + 1006.0, + 342.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1009.0, + 399.0, + 1009.0, + 399.0, + 1060.0, + 294.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 1009.0, + 448.0, + 1009.0, + 448.0, + 1060.0, + 443.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 979.0, + 438.0, + 979.0, + 438.0, + 1015.0, + 307.0, + 1015.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 820, + 1407, + 820, + 1407, + 1007, + 296, + 1007 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 296, + 1017, + 1406, + 1017, + 1406, + 1266, + 296, + 1266 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 296, + 498, + 1407, + 498, + 1407, + 707, + 296, + 707 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1724, + 1403, + 1724, + 1403, + 1892, + 298, + 1892 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 295, + 383, + 1408, + 383, + 1408, + 478, + 295, + 478 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 297, + 1277, + 1405, + 1277, + 1405, + 1390, + 297, + 1390 + ], + "score": 0.972 + }, + { + "category_id": 8, + "poly": [ + 409, + 1452, + 1289, + 1452, + 1289, + 1594, + 409, + 1594 + ], + "score": 0.963 + }, + { + "category_id": 8, + "poly": [ + 412, + 1907, + 1283, + 1907, + 1283, + 2002, + 412, + 2002 + ], + "score": 0.954 + }, + { + "category_id": 0, + "poly": [ + 299, + 750, + 905, + 750, + 905, + 790, + 299, + 790 + ], + "score": 0.945 + }, + { + "category_id": 8, + "poly": [ + 410, + 289, + 1287, + 289, + 1287, + 370, + 410, + 370 + ], + "score": 0.943 + }, + { + "category_id": 1, + "poly": [ + 295, + 1609, + 1179, + 1609, + 1179, + 1643, + 295, + 1643 + ], + "score": 0.934 + }, + { + "category_id": 8, + "poly": [ + 643, + 1660, + 1053, + 1660, + 1053, + 1708, + 643, + 1708 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 297, + 1400, + 1040, + 1400, + 1040, + 1435, + 297, + 1435 + ], + "score": 0.918 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1669, + 1400, + 1669, + 1400, + 1699, + 1366, + 1699 + ], + "score": 0.879 + }, + { + "category_id": 1, + "poly": [ + 289, + 201, + 1404, + 201, + 1404, + 270, + 289, + 270 + ], + "score": 0.87 + }, + { + "category_id": 9, + "poly": [ + 1365, + 314, + 1401, + 314, + 1401, + 344, + 1365, + 344 + ], + "score": 0.86 + }, + { + "category_id": 9, + "poly": [ + 1365, + 1938, + 1401, + 1938, + 1401, + 1969, + 1365, + 1969 + ], + "score": 0.859 + }, + { + "category_id": 9, + "poly": [ + 1365, + 1529, + 1401, + 1529, + 1401, + 1560, + 1365, + 1560 + ], + "score": 0.849 + }, + { + "category_id": 2, + "poly": [ + 839, + 2061, + 859, + 2061, + 859, + 2086, + 839, + 2086 + ], + "score": 0.79 + }, + { + "category_id": 13, + "poly": [ + 442, + 384, + 966, + 384, + 966, + 446, + 442, + 446 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\alpha _ { t } = \\frac { c _ { 1 } } { \\log \\left( \\frac { T | \\mathcal { S } | | \\mathcal { A } | } { \\delta } \\right) } \\operatorname* { m i n } \\left( \\frac { 1 } { t _ { m i x } } , \\frac { \\epsilon ^ { 2 } ( 1 - \\gamma ) ^ { 4 } } { \\gamma ^ { 2 } } \\right) , \\forall t \\geq 0 } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1216, + 501, + 1408, + 501, + 1408, + 548, + 1216, + 548 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\tilde { \\mathcal { O } } ( \\frac { 1 } { \\mu _ { \\mathrm { m i n } } ( 1 - \\gamma ) ^ { 5 } \\epsilon ^ { 2 } } + } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 414, + 1906, + 1283, + 1906, + 1283, + 2003, + 414, + 2003 + ], + "score": 0.93, + "latex": "\\hat { \\mathbf { T } } _ { \\pi } V ( s ) = \\mathbb { E } _ { A \\sim \\pi ( \\cdot | s ) } \\left[ c ( s , A ) + \\gamma ( 1 - R ) \\sum _ { s ^ { \\prime } \\in \\mathcal { S } } p _ { s , s ^ { \\prime } } ^ { A } V ( s ^ { \\prime } ) + \\gamma R \\cdot \\mathrm { L S E } ( V ) \\right] ," + }, + { + "category_id": 13, + "poly": [ + 470, + 232, + 567, + 232, + 567, + 274, + 470, + 274 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\epsilon < \\frac { 1 } { 1 - \\gamma } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 297, + 1348, + 524, + 1348, + 524, + 1392, + 297, + 1392 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\sum _ { i = 1 } ^ { N } \\theta ^ { i } \\phi _ { s } ^ { ( i ) } = \\phi _ { s } ^ { \\top } \\theta } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 406, + 1449, + 1292, + 1449, + 1292, + 1596, + 406, + 1596 + ], + "score": 0.92, + "latex": "\\begin{array} { r l } & { \\mathrm { \\bf T } _ { \\pi } V ( s ) \\triangleq \\mathbb E _ { A \\sim \\pi ( \\cdot | s ) } [ c ( s , A ) + \\gamma \\sigma _ { \\mathcal { P } _ { s } ^ { A } } ( V ) ] } \\\\ & { \\quad \\quad \\quad = \\mathbb E _ { A \\sim \\pi ( \\cdot | s ) } \\left[ c ( s , A ) + \\gamma ( 1 - R ) \\displaystyle \\sum _ { s ^ { \\prime } \\in \\mathcal { S } } p _ { s , s ^ { \\prime } } ^ { A } V ( s ^ { \\prime } ) + \\gamma R \\operatorname* { m a x } _ { s ^ { \\prime } \\in \\mathcal { S } } V ( s ^ { \\prime } ) \\right] . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1255, + 674, + 1341, + 674, + 1341, + 707, + 1255, + 707 + ], + "score": 0.92, + "latex": "( 1 - \\gamma )" + }, + { + "category_id": 13, + "poly": [ + 371, + 1726, + 643, + 1726, + 643, + 1766, + 371, + 1766 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\| v \\| _ { \\mu _ { \\pi } } ^ { 2 } = \\int v ^ { 2 } ( s ) \\mu _ { \\pi } ( d s ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1199, + 1280, + 1350, + 1280, + 1350, + 1314, + 1199, + 1314 + ], + "score": 0.92, + "latex": "N \\ll | \\mathcal S | | \\mathcal A |" + }, + { + "category_id": 14, + "poly": [ + 410, + 290, + 1286, + 290, + 1286, + 367, + 410, + 367 + ], + "score": 0.92, + "latex": "T \\geq c _ { 0 } \\left( \\frac { 1 } { \\mu _ { m i n } ( 1 - \\gamma ) ^ { 5 } \\epsilon ^ { 2 } } + \\frac { t _ { m i x } } { \\mu _ { m i n } ( 1 - \\gamma ) } \\right) \\log { \\left( \\frac { T | \\mathcal { S } | | \\mathcal { A } | } { \\delta } \\right) } \\log { \\left( \\frac { 1 } { \\epsilon ( 1 - \\gamma ) ^ { 2 } } \\right) } ," + }, + { + "category_id": 14, + "poly": [ + 643, + 1659, + 1053, + 1659, + 1053, + 1708, + 643, + 1708 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\mathrm { M S P R B E } ( \\theta ) = \\| \\mathbf { I I T } _ { \\pi } V _ { \\theta } - V _ { \\theta } \\| _ { \\mu _ { \\pi } } ^ { 2 } , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 609, + 1353, + 700, + 1353, + 700, + 1385, + 609, + 1385 + ], + "score": 0.91, + "latex": "\\boldsymbol { \\theta } \\in \\mathbb { R } ^ { N }" + }, + { + "category_id": 13, + "poly": [ + 684, + 1762, + 756, + 1762, + 756, + 1796, + 684, + 1796 + ], + "score": 0.91, + "latex": "\\| \\cdot \\| _ { \\mu _ { \\pi } }" + }, + { + "category_id": 13, + "poly": [ + 688, + 1854, + 727, + 1854, + 727, + 1889, + 688, + 1889 + ], + "score": 0.91, + "latex": "\\hat { \\mathbf { T } } _ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 299, + 543, + 412, + 543, + 412, + 588, + 299, + 588 + ], + "score": 0.91, + "latex": "\\frac { t _ { \\operatorname* { m i x } } } { \\mu _ { \\operatorname* { m i n } } ( 1 - \\gamma ) } )" + }, + { + "category_id": 13, + "poly": [ + 347, + 1277, + 689, + 1277, + 689, + 1317, + 347, + 1317 + ], + "score": 0.9, + "latex": "\\left\\{ \\phi ^ { ( i ) } : \\mathbb { S } \\to \\mathbb { R } , i = 1 , \\ldots , N \\right\\}" + }, + { + "category_id": 13, + "poly": [ + 1228, + 1315, + 1272, + 1315, + 1272, + 1350, + 1228, + 1350 + ], + "score": 0.89, + "latex": "\\phi ^ { ( i ) }" + }, + { + "category_id": 13, + "poly": [ + 1304, + 1316, + 1406, + 1316, + 1406, + 1351, + 1304, + 1351 + ], + "score": 0.89, + "latex": "V _ { \\theta } ( s ) =" + }, + { + "category_id": 13, + "poly": [ + 299, + 442, + 522, + 442, + 522, + 478, + 299, + 478 + ], + "score": 0.89, + "latex": "\\| Q _ { T } - Q ^ { * } \\| _ { \\infty } \\leq 3 \\epsilon" + }, + { + "category_id": 13, + "poly": [ + 345, + 234, + 411, + 234, + 411, + 263, + 345, + 263 + ], + "score": 0.87, + "latex": "\\delta < 1" + }, + { + "category_id": 13, + "poly": [ + 1226, + 207, + 1253, + 207, + 1253, + 234, + 1226, + 234 + ], + "score": 0.86, + "latex": "c _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1146, + 209, + 1173, + 209, + 1173, + 234, + 1146, + 234 + ], + "score": 0.85, + "latex": "c _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 916, + 1792, + 1048, + 1792, + 1048, + 1826, + 916, + 1826 + ], + "score": 0.83, + "latex": "\\operatorname* { m a x } _ { s } V _ { \\theta } ( s )" + }, + { + "category_id": 13, + "poly": [ + 1360, + 1020, + 1400, + 1020, + 1400, + 1048, + 1360, + 1048 + ], + "score": 0.81, + "latex": "V ^ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 833, + 1282, + 861, + 1282, + 861, + 1308, + 833, + 1308 + ], + "score": 0.81, + "latex": "N" + }, + { + "category_id": 13, + "poly": [ + 1304, + 1794, + 1319, + 1794, + 1319, + 1820, + 1304, + 1820 + ], + "score": 0.79, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 626, + 235, + 649, + 235, + 649, + 261, + 626, + 261 + ], + "score": 0.78, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 655, + 1731, + 690, + 1731, + 690, + 1761, + 655, + 1761 + ], + "score": 0.78, + "latex": "\\mu _ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 1134, + 1734, + 1154, + 1734, + 1154, + 1757, + 1134, + 1757 + ], + "score": 0.76, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 1211, + 1730, + 1239, + 1730, + 1239, + 1757, + 1211, + 1757 + ], + "score": 0.76, + "latex": "\\mathbf { I I }" + }, + { + "category_id": 13, + "poly": [ + 653, + 1056, + 674, + 1056, + 674, + 1078, + 653, + 1078 + ], + "score": 0.72, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 1010, + 1409, + 1029, + 1409, + 1029, + 1429, + 1010, + 1429 + ], + "score": 0.72, + "latex": "\\pi" + }, + { + "category_id": 13, + "poly": [ + 845, + 514, + 859, + 514, + 859, + 534, + 845, + 534 + ], + "score": 0.68, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 1326, + 396, + 1398, + 396, + 1398, + 425, + 1326, + 425 + ], + "score": 0.66, + "latex": "1 - 6 \\delta" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 748.0, + 907.0, + 748.0, + 907.0, + 794.0, + 290.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2060.0, + 861.0, + 2060.0, + 861.0, + 2090.0, + 840.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 821.0, + 1404.0, + 821.0, + 1404.0, + 856.0, + 294.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 853.0, + 1407.0, + 853.0, + 1407.0, + 889.0, + 294.0, + 889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 883.0, + 1405.0, + 883.0, + 1405.0, + 919.0, + 293.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 912.0, + 1406.0, + 912.0, + 1406.0, + 949.0, + 291.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 943.0, + 1409.0, + 943.0, + 1409.0, + 978.0, + 293.0, + 978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 974.0, + 959.0, + 974.0, + 959.0, + 1008.0, + 294.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1017.0, + 1359.0, + 1017.0, + 1359.0, + 1053.0, + 291.0, + 1053.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 1017.0, + 1404.0, + 1017.0, + 1404.0, + 1053.0, + 1401.0, + 1053.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1047.0, + 652.0, + 1047.0, + 652.0, + 1087.0, + 291.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 675.0, + 1047.0, + 1409.0, + 1047.0, + 1409.0, + 1087.0, + 675.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1077.0, + 1407.0, + 1077.0, + 1407.0, + 1117.0, + 291.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1109.0, + 1406.0, + 1109.0, + 1406.0, + 1145.0, + 295.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1139.0, + 1406.0, + 1139.0, + 1406.0, + 1175.0, + 293.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1168.0, + 1406.0, + 1168.0, + 1406.0, + 1207.0, + 293.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1202.0, + 1404.0, + 1202.0, + 1404.0, + 1236.0, + 295.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1230.0, + 461.0, + 1230.0, + 461.0, + 1269.0, + 293.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 505.0, + 844.0, + 505.0, + 844.0, + 542.0, + 295.0, + 542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 505.0, + 1215.0, + 505.0, + 1215.0, + 542.0, + 860.0, + 542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 580.0, + 1407.0, + 580.0, + 1407.0, + 618.0, + 291.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 613.0, + 1405.0, + 613.0, + 1405.0, + 647.0, + 294.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 644.0, + 1405.0, + 644.0, + 1405.0, + 677.0, + 294.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 670.0, + 1254.0, + 670.0, + 1254.0, + 711.0, + 293.0, + 711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1342.0, + 670.0, + 1352.0, + 670.0, + 1352.0, + 711.0, + 1342.0, + 711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1240.0, + 513.5, + 1412.0, + 513.5, + 1412.0, + 552.5, + 1240.0, + 552.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 532.0, + 1416.0, + 532.0, + 1416.0, + 597.0, + 285.0, + 597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 542.0, + 369.0, + 542.0, + 369.0, + 572.5, + 331.0, + 572.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1725.0, + 370.0, + 1725.0, + 370.0, + 1766.0, + 293.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 644.0, + 1725.0, + 654.0, + 1725.0, + 654.0, + 1766.0, + 644.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 1725.0, + 1133.0, + 1725.0, + 1133.0, + 1766.0, + 691.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1155.0, + 1725.0, + 1210.0, + 1725.0, + 1210.0, + 1766.0, + 1155.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1240.0, + 1725.0, + 1404.0, + 1725.0, + 1404.0, + 1766.0, + 1240.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1761.0, + 683.0, + 1761.0, + 683.0, + 1798.0, + 293.0, + 1798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 1761.0, + 1407.0, + 1761.0, + 1407.0, + 1798.0, + 757.0, + 1798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1792.0, + 915.0, + 1792.0, + 915.0, + 1826.0, + 294.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1049.0, + 1792.0, + 1303.0, + 1792.0, + 1303.0, + 1826.0, + 1049.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1320.0, + 1792.0, + 1405.0, + 1792.0, + 1405.0, + 1826.0, + 1320.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1821.0, + 1407.0, + 1821.0, + 1407.0, + 1861.0, + 292.0, + 1861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1858.0, + 687.0, + 1858.0, + 687.0, + 1895.0, + 294.0, + 1895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 728.0, + 1858.0, + 1281.0, + 1858.0, + 1281.0, + 1895.0, + 728.0, + 1895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 967.0, + 379.0, + 1325.0, + 379.0, + 1325.0, + 444.0, + 967.0, + 444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 379.0, + 1408.0, + 379.0, + 1408.0, + 444.0, + 1399.0, + 444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 443.0, + 533.0, + 443.0, + 533.0, + 478.0, + 523.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 392.5, + 514.0, + 392.5, + 514.0, + 432.5, + 294.0, + 432.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 389.5, + 733.0, + 389.5, + 733.0, + 411.5, + 718.0, + 411.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1274.0, + 346.0, + 1274.0, + 346.0, + 1318.0, + 292.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 1274.0, + 832.0, + 1274.0, + 832.0, + 1318.0, + 690.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 862.0, + 1274.0, + 1198.0, + 1274.0, + 1198.0, + 1318.0, + 862.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1351.0, + 1274.0, + 1406.0, + 1274.0, + 1406.0, + 1318.0, + 1351.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1311.0, + 1227.0, + 1311.0, + 1227.0, + 1356.0, + 292.0, + 1356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 1311.0, + 1303.0, + 1311.0, + 1303.0, + 1356.0, + 1273.0, + 1356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 282.0, + 1330.0, + 296.0, + 1330.0, + 296.0, + 1409.0, + 282.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 1330.0, + 608.0, + 1330.0, + 608.0, + 1409.0, + 525.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1330.0, + 947.0, + 1330.0, + 947.0, + 1409.0, + 701.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1606.0, + 1181.0, + 1606.0, + 1181.0, + 1648.0, + 294.0, + 1648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1395.0, + 1009.0, + 1395.0, + 1009.0, + 1442.0, + 294.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1030.0, + 1395.0, + 1042.0, + 1395.0, + 1042.0, + 1442.0, + 1030.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 200.0, + 1145.0, + 200.0, + 1145.0, + 239.0, + 295.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1174.0, + 200.0, + 1225.0, + 200.0, + 1225.0, + 239.0, + 1174.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1254.0, + 200.0, + 1403.0, + 200.0, + 1403.0, + 239.0, + 1254.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 230.0, + 344.0, + 230.0, + 344.0, + 274.0, + 293.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 412.0, + 230.0, + 469.0, + 230.0, + 469.0, + 274.0, + 412.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 230.0, + 625.0, + 230.0, + 625.0, + 274.0, + 568.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 230.0, + 766.0, + 230.0, + 766.0, + 274.0, + 650.0, + 274.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1383, + 1408, + 1383, + 1408, + 1637, + 296, + 1637 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 294, + 798, + 1407, + 798, + 1407, + 1085, + 294, + 1085 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 296, + 194, + 1406, + 194, + 1406, + 383, + 296, + 383 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 296, + 1725, + 1406, + 1725, + 1406, + 1849, + 296, + 1849 + ], + "score": 0.971 + }, + { + "category_id": 8, + "poly": [ + 432, + 1227, + 1273, + 1227, + 1273, + 1371, + 432, + 1371 + ], + "score": 0.969 + }, + { + "category_id": 8, + "poly": [ + 583, + 491, + 1115, + 491, + 1115, + 567, + 583, + 567 + ], + "score": 0.949 + }, + { + "category_id": 0, + "poly": [ + 296, + 1669, + 852, + 1669, + 852, + 1704, + 296, + 1704 + ], + "score": 0.936 + }, + { + "category_id": 8, + "poly": [ + 427, + 1098, + 1269, + 1098, + 1269, + 1166, + 427, + 1166 + ], + "score": 0.931 + }, + { + "category_id": 1, + "poly": [ + 295, + 594, + 1270, + 594, + 1270, + 630, + 295, + 630 + ], + "score": 0.93 + }, + { + "category_id": 0, + "poly": [ + 298, + 744, + 649, + 744, + 649, + 778, + 298, + 778 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 298, + 1180, + 696, + 1180, + 696, + 1212, + 298, + 1212 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 294, + 441, + 1092, + 441, + 1092, + 480, + 294, + 480 + ], + "score": 0.907 + }, + { + "category_id": 9, + "poly": [ + 1351, + 662, + 1400, + 662, + 1400, + 694, + 1351, + 694 + ], + "score": 0.902 + }, + { + "category_id": 9, + "poly": [ + 1352, + 513, + 1400, + 513, + 1400, + 545, + 1352, + 545 + ], + "score": 0.89 + }, + { + "category_id": 1, + "poly": [ + 297, + 393, + 849, + 393, + 849, + 430, + 297, + 430 + ], + "score": 0.887 + }, + { + "category_id": 9, + "poly": [ + 1351, + 1116, + 1401, + 1116, + 1401, + 1148, + 1351, + 1148 + ], + "score": 0.876 + }, + { + "category_id": 1, + "poly": [ + 296, + 1943, + 1405, + 1943, + 1405, + 2009, + 296, + 2009 + ], + "score": 0.835 + }, + { + "category_id": 8, + "poly": [ + 633, + 645, + 1062, + 645, + 1062, + 713, + 633, + 713 + ], + "score": 0.785 + }, + { + "category_id": 2, + "poly": [ + 841, + 2060, + 858, + 2060, + 858, + 2084, + 841, + 2084 + ], + "score": 0.739 + }, + { + "category_id": 1, + "poly": [ + 293, + 1899, + 1160, + 1899, + 1160, + 1935, + 293, + 1935 + ], + "score": 0.733 + }, + { + "category_id": 1, + "poly": [ + 293, + 1858, + 913, + 1858, + 913, + 1892, + 293, + 1892 + ], + "score": 0.715 + }, + { + "category_id": 8, + "poly": [ + 638, + 646, + 1057, + 646, + 1057, + 714, + 638, + 714 + ], + "score": 0.235 + }, + { + "category_id": 14, + "poly": [ + 635, + 642, + 1061, + 642, + 1061, + 714, + 635, + 714 + ], + "score": 0.94, + "latex": "\\operatorname* { m i n } _ { \\theta } J ( \\theta ) = \\operatorname* { m i n } _ { \\theta } \\Big \\| \\mathbf { H } \\hat { \\mathbf { T } } _ { \\pi } V _ { \\theta } - V _ { \\theta } \\Big \\| _ { \\mu _ { \\pi } } ^ { 2 } ." + }, + { + "category_id": 14, + "poly": [ + 582, + 491, + 1117, + 491, + 1117, + 567, + 582, + 567 + ], + "score": 0.94, + "latex": "\\| V _ { 1 } - V _ { 2 } \\| _ { \\infty } \\leq \\frac { \\gamma R } { 1 - \\gamma } \\frac { \\log | \\mathcal { S } | } { \\varrho } 0 , a s \\varrho \\infty ." + }, + { + "category_id": 13, + "poly": [ + 1098, + 929, + 1311, + 929, + 1311, + 969, + 1098, + 969 + ], + "score": 0.94, + "latex": "C \\triangleq \\mathbb { E } _ { \\mu _ { \\pi } } \\left[ \\phi _ { S } ^ { \\intercal } \\phi _ { S } \\right]" + }, + { + "category_id": 13, + "poly": [ + 369, + 196, + 657, + 196, + 657, + 252, + 369, + 252 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { \\mathrm { L S E } ( V ) = \\frac { \\log \\left( \\sum _ { s } e ^ { \\varrho V ( s ) } \\right) } { \\varrho } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 843, + 1943, + 1034, + 1943, + 1034, + 1980, + 843, + 1980 + ], + "score": 0.93, + "latex": "C = \\mathbb { E } _ { \\mu _ { \\pi } } [ \\phi _ { S } \\phi _ { S } ^ { \\top } ]" + }, + { + "category_id": 13, + "poly": [ + 869, + 1757, + 1042, + 1757, + 1042, + 1791, + 869, + 1791 + ], + "score": 0.93, + "latex": "\\| \\nabla J ( \\theta ) \\| 0" + }, + { + "category_id": 13, + "poly": [ + 880, + 1017, + 1262, + 1017, + 1262, + 1055, + 880, + 1055 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\pmb { \\Pi } ^ { \\top } D \\pmb { \\Pi } = D ^ { \\top } \\pmb { \\varPhi } ( \\pmb { \\varPhi } ^ { \\top } D \\pmb { \\varPhi } ) ^ { - 1 } \\pmb { \\varPhi } ^ { \\top } D } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 371, + 1385, + 725, + 1385, + 725, + 1421, + 371, + 1421 + ], + "score": 0.92, + "latex": "\\omega ( \\theta ) = C ^ { - 1 } \\mathbb { E } _ { \\mu _ { \\pi } } [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } ]" + }, + { + "category_id": 14, + "poly": [ + 428, + 1096, + 1271, + 1096, + 1271, + 1167, + 428, + 1167 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\boldsymbol J ( \\theta ) = \\Big \\| \\mathbf { \\hat { I } } \\mathbf { \\hat { T } } _ { \\boldsymbol \\pi } V _ { \\theta } - V _ { \\theta } \\Big \\| _ { \\mu _ { \\boldsymbol \\pi } } ^ { 2 } = \\mathbb { E } _ { \\mu _ { \\boldsymbol \\pi } } [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } ] ^ { \\top } \\boldsymbol C ^ { - 1 } \\mathbb { E } _ { \\mu _ { \\boldsymbol \\pi } } [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } ] . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 832, + 249, + 951, + 249, + 951, + 286, + 832, + 286 + ], + "score": 0.92, + "latex": "\\hat { \\mathbf { T } } _ { \\pi } \\to \\mathbf { T } _ { \\pi }" + }, + { + "category_id": 14, + "poly": [ + 427, + 1224, + 1270, + 1224, + 1270, + 1372, + 427, + 1372 + ], + "score": 0.92, + "latex": "\\begin{array} { l } { \\displaystyle - \\frac { 1 } { 2 } \\nabla J ( \\theta ) = - \\mathbb { E } _ { \\mu _ { \\pi } } \\big [ ( \\nabla \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) ) \\phi _ { S } \\big ] ^ { \\top } C ^ { - 1 } \\mathbb { E } _ { \\mu _ { \\pi } } \\big [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } \\big ] } \\\\ { = \\mathbb { E } _ { \\mu _ { \\pi } } \\big [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } \\big ] - \\gamma \\mathbb { E } _ { \\mu _ { \\pi } } \\bigg [ \\bigg ( ( 1 - R ) \\phi _ { S ^ { \\prime } } + R \\cdot \\nabla \\mathrm { L S E } ( V _ { \\theta } ) \\bigg ) \\phi _ { S } ^ { \\top } \\bigg ] \\omega ( \\theta ) , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 510, + 597, + 565, + 597, + 565, + 630, + 510, + 630 + ], + "score": 0.92, + "latex": "J ( \\theta )" + }, + { + "category_id": 13, + "poly": [ + 813, + 930, + 943, + 930, + 943, + 964, + 813, + 964 + ], + "score": 0.92, + "latex": "V _ { \\theta } = \\theta ^ { \\top } \\phi" + }, + { + "category_id": 13, + "poly": [ + 1319, + 1386, + 1396, + 1386, + 1396, + 1420, + 1319, + 1420 + ], + "score": 0.91, + "latex": "\\nabla J ( \\theta )" + }, + { + "category_id": 13, + "poly": [ + 1037, + 1508, + 1115, + 1508, + 1115, + 1541, + 1037, + 1541 + ], + "score": 0.91, + "latex": "\\nabla J ( \\theta )" + }, + { + "category_id": 13, + "poly": [ + 297, + 1540, + 680, + 1540, + 680, + 1579, + 297, + 1579 + ], + "score": 0.91, + "latex": "\\Pi _ { K } ( x ) \\triangleq \\arg \\operatorname* { m i n } _ { \\| y \\| \\leq K } \\| y - x \\|" + }, + { + "category_id": 13, + "poly": [ + 606, + 1508, + 660, + 1508, + 660, + 1541, + 606, + 1541 + ], + "score": 0.91, + "latex": "\\omega ( \\theta )" + }, + { + "category_id": 13, + "poly": [ + 702, + 1727, + 758, + 1727, + 758, + 1761, + 702, + 1761 + ], + "score": 0.91, + "latex": "J ( \\theta )" + }, + { + "category_id": 13, + "poly": [ + 926, + 831, + 981, + 831, + 981, + 865, + 926, + 865 + ], + "score": 0.91, + "latex": "J ( \\theta )" + }, + { + "category_id": 13, + "poly": [ + 741, + 440, + 781, + 440, + 781, + 478, + 741, + 478 + ], + "score": 0.9, + "latex": "\\hat { \\mathbf { T } } _ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 689, + 1858, + 901, + 1858, + 901, + 1894, + 689, + 1894 + ], + "score": 0.9, + "latex": "\\| \\phi _ { s } \\| _ { 2 } \\leq 1 , \\forall s \\in \\mathcal { S }" + }, + { + "category_id": 13, + "poly": [ + 1128, + 316, + 1167, + 316, + 1167, + 350, + 1128, + 350 + ], + "score": 0.9, + "latex": "\\hat { \\mathbf { T } } _ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 1152, + 213, + 1219, + 213, + 1219, + 243, + 1152, + 243 + ], + "score": 0.9, + "latex": "\\varrho > 0" + }, + { + "category_id": 13, + "poly": [ + 727, + 351, + 766, + 351, + 766, + 380, + 727, + 380 + ], + "score": 0.9, + "latex": "\\mathbf { T } _ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 772, + 1540, + 864, + 1540, + 864, + 1571, + 772, + 1571 + ], + "score": 0.89, + "latex": "\\boldsymbol { x } \\in \\mathbb { R } ^ { N }" + }, + { + "category_id": 13, + "poly": [ + 883, + 975, + 1347, + 975, + 1347, + 1012, + 883, + 1012 + ], + "score": 0.89, + "latex": "\\textit { D } = \\ \\mathrm { d i a g } ( \\mu _ { \\pi } ( s _ { 1 } ) , \\mu _ { \\pi } ( s _ { 2 } ) , . . . , \\mu _ { \\pi } ( s _ { | \\mathcal { S } | } ) )" + }, + { + "category_id": 13, + "poly": [ + 833, + 445, + 872, + 445, + 872, + 478, + 833, + 478 + ], + "score": 0.89, + "latex": "\\mathbf { T } _ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 298, + 259, + 385, + 259, + 385, + 286, + 298, + 286 + ], + "score": 0.89, + "latex": "\\varrho \\to \\infty" + }, + { + "category_id": 13, + "poly": [ + 576, + 894, + 1312, + 894, + 1312, + 929, + 576, + 929 + ], + "score": 0.88, + "latex": "\\delta _ { s , a , s ^ { \\prime } } ( \\theta ) \\ \\triangleq \\ c ( s , a ) + \\gamma ( 1 - R ) V _ { \\theta } ( s ^ { \\prime } ) + \\gamma R \\mathrm { L S E } ( V _ { \\theta } ) - V _ { \\theta } ( s )" + }, + { + "category_id": 13, + "poly": [ + 1263, + 1946, + 1334, + 1946, + 1334, + 1975, + 1263, + 1975 + ], + "score": 0.87, + "latex": "\\lambda > 0" + }, + { + "category_id": 13, + "poly": [ + 298, + 1605, + 372, + 1605, + 372, + 1631, + 298, + 1631 + ], + "score": 0.87, + "latex": "R = 0" + }, + { + "category_id": 13, + "poly": [ + 1080, + 1902, + 1152, + 1902, + 1152, + 1929, + 1080, + 1929 + ], + "score": 0.87, + "latex": "a \\in { \\mathcal { A } }" + }, + { + "category_id": 13, + "poly": [ + 470, + 446, + 501, + 446, + 501, + 477, + 470, + 477 + ], + "score": 0.87, + "latex": "V _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 385, + 446, + 417, + 446, + 417, + 477, + 385, + 477 + ], + "score": 0.87, + "latex": "V _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1017, + 707, + 1017, + 707, + 1059, + 297, + 1059 + ], + "score": 0.86, + "latex": "\\boldsymbol { \\varPhi } = ( \\phi _ { s _ { 1 } } , \\phi _ { s _ { 2 } } , . . . , \\phi _ { s _ { | \\mathcal { S } | } } ) ^ { \\top } \\in \\dot { \\mathbb { R } } ^ { | \\mathcal { S } | \\times N }" + }, + { + "category_id": 13, + "poly": [ + 296, + 965, + 785, + 965, + 785, + 1019, + 296, + 1019 + ], + "score": 0.85, + "latex": "\\mathbb { E } _ { \\mu } [ \\delta _ { S , A , S ^ { \\prime } } ( \\theta ) \\phi _ { S } ] \\ = \\ \\varPhi ^ { \\top } D \\left( \\hat { \\mathbf { T } } _ { \\pi } V _ { \\theta } - V _ { \\theta } \\right)" + }, + { + "category_id": 13, + "poly": [ + 755, + 1900, + 1027, + 1900, + 1027, + 1934, + 755, + 1934 + ], + "score": 0.83, + "latex": "| c ( s , a ) | \\leq c _ { \\mathrm { m a x } } , \\forall s \\in \\mathcal { S }" + }, + { + "category_id": 13, + "poly": [ + 612, + 390, + 653, + 390, + 653, + 428, + 612, + 428 + ], + "score": 0.83, + "latex": "\\hat { \\mathbf { T } } _ { \\pi }" + }, + { + "category_id": 13, + "poly": [ + 296, + 930, + 400, + 930, + 400, + 965, + 296, + 965 + ], + "score": 0.82, + "latex": "\\operatorname { L S E } ( V _ { \\theta } )" + }, + { + "category_id": 13, + "poly": [ + 932, + 212, + 959, + 212, + 959, + 239, + 932, + 239 + ], + "score": 0.77, + "latex": "V" + }, + { + "category_id": 13, + "poly": [ + 873, + 356, + 890, + 356, + 890, + 381, + 873, + 381 + ], + "score": 0.76, + "latex": "\\varrho" + }, + { + "category_id": 13, + "poly": [ + 1330, + 326, + 1346, + 326, + 1346, + 351, + 1330, + 351 + ], + "score": 0.74, + "latex": "\\varrho" + }, + { + "category_id": 13, + "poly": [ + 586, + 404, + 603, + 404, + 603, + 428, + 586, + 428 + ], + "score": 0.56, + "latex": "\\varrho" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1670.0, + 853.0, + 1670.0, + 853.0, + 1707.0, + 295.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 740.0, + 652.0, + 740.0, + 652.0, + 785.0, + 292.0, + 785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2059.0, + 860.0, + 2059.0, + 860.0, + 2090.0, + 839.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1382.0, + 370.0, + 1382.0, + 370.0, + 1424.0, + 294.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 726.0, + 1382.0, + 1318.0, + 1382.0, + 1318.0, + 1424.0, + 726.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1397.0, + 1382.0, + 1409.0, + 1382.0, + 1409.0, + 1424.0, + 1397.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1417.0, + 1407.0, + 1417.0, + 1407.0, + 1451.0, + 294.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1447.0, + 1405.0, + 1447.0, + 1405.0, + 1480.0, + 293.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1478.0, + 1406.0, + 1478.0, + 1406.0, + 1512.0, + 295.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1504.0, + 605.0, + 1504.0, + 605.0, + 1546.0, + 292.0, + 1546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 1504.0, + 1036.0, + 1504.0, + 1036.0, + 1546.0, + 661.0, + 1546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1116.0, + 1504.0, + 1406.0, + 1504.0, + 1406.0, + 1546.0, + 1116.0, + 1546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1537.0, + 296.0, + 1537.0, + 296.0, + 1581.0, + 289.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 1537.0, + 771.0, + 1537.0, + 771.0, + 1581.0, + 681.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 865.0, + 1537.0, + 1409.0, + 1537.0, + 1409.0, + 1581.0, + 865.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1571.0, + 1408.0, + 1571.0, + 1408.0, + 1610.0, + 293.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1603.0, + 1006.0, + 1603.0, + 1006.0, + 1638.0, + 373.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 799.0, + 1405.0, + 799.0, + 1405.0, + 836.0, + 293.0, + 836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 833.0, + 925.0, + 833.0, + 925.0, + 863.0, + 297.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 833.0, + 1404.0, + 833.0, + 1404.0, + 863.0, + 982.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 862.0, + 1409.0, + 862.0, + 1409.0, + 896.0, + 296.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 894.0, + 575.0, + 894.0, + 575.0, + 932.0, + 295.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1313.0, + 894.0, + 1406.0, + 894.0, + 1406.0, + 932.0, + 1313.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 926.0, + 295.0, + 926.0, + 295.0, + 969.0, + 292.0, + 969.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 401.0, + 926.0, + 812.0, + 926.0, + 812.0, + 969.0, + 401.0, + 969.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 926.0, + 1097.0, + 926.0, + 1097.0, + 969.0, + 944.0, + 969.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1312.0, + 926.0, + 1409.0, + 926.0, + 1409.0, + 969.0, + 1312.0, + 969.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 968.0, + 295.0, + 968.0, + 295.0, + 1016.0, + 289.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 968.0, + 882.0, + 968.0, + 882.0, + 1016.0, + 786.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1348.0, + 968.0, + 1409.0, + 968.0, + 1409.0, + 1016.0, + 1348.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1012.0, + 296.0, + 1012.0, + 296.0, + 1060.0, + 288.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 708.0, + 1012.0, + 879.0, + 1012.0, + 879.0, + 1060.0, + 708.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1263.0, + 1012.0, + 1411.0, + 1012.0, + 1411.0, + 1060.0, + 1263.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1048.0, + 556.0, + 1048.0, + 556.0, + 1086.0, + 294.0, + 1086.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 196.0, + 368.0, + 196.0, + 368.0, + 251.0, + 293.0, + 251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 196.0, + 931.0, + 196.0, + 931.0, + 251.0, + 658.0, + 251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 960.0, + 196.0, + 1151.0, + 196.0, + 1151.0, + 251.0, + 960.0, + 251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1220.0, + 196.0, + 1407.0, + 196.0, + 1407.0, + 251.0, + 1220.0, + 251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 252.0, + 297.0, + 252.0, + 297.0, + 290.0, + 291.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 386.0, + 252.0, + 831.0, + 252.0, + 831.0, + 290.0, + 386.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 952.0, + 252.0, + 1405.0, + 252.0, + 1405.0, + 290.0, + 952.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 285.0, + 1404.0, + 285.0, + 1404.0, + 318.0, + 294.0, + 318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 318.0, + 1127.0, + 318.0, + 1127.0, + 354.0, + 294.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1168.0, + 318.0, + 1329.0, + 318.0, + 1329.0, + 354.0, + 1168.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 318.0, + 1404.0, + 318.0, + 1404.0, + 354.0, + 1347.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 346.0, + 726.0, + 346.0, + 726.0, + 388.0, + 294.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 346.0, + 872.0, + 346.0, + 872.0, + 388.0, + 767.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 346.0, + 900.0, + 346.0, + 900.0, + 388.0, + 891.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1726.0, + 701.0, + 1726.0, + 701.0, + 1763.0, + 295.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 1726.0, + 1406.0, + 1726.0, + 1406.0, + 1763.0, + 759.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1757.0, + 868.0, + 1757.0, + 868.0, + 1793.0, + 294.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 1757.0, + 1406.0, + 1757.0, + 1406.0, + 1793.0, + 1043.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1787.0, + 1406.0, + 1787.0, + 1406.0, + 1822.0, + 291.0, + 1822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1817.0, + 1138.0, + 1817.0, + 1138.0, + 1853.0, + 294.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 592.0, + 509.0, + 592.0, + 509.0, + 634.0, + 293.0, + 634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 566.0, + 592.0, + 1274.0, + 592.0, + 1274.0, + 634.0, + 566.0, + 634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1176.0, + 699.0, + 1176.0, + 699.0, + 1216.0, + 295.0, + 1216.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 440.0, + 384.0, + 440.0, + 384.0, + 484.0, + 293.0, + 484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 418.0, + 440.0, + 469.0, + 440.0, + 469.0, + 484.0, + 418.0, + 484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 502.0, + 440.0, + 740.0, + 440.0, + 740.0, + 484.0, + 502.0, + 484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 782.0, + 440.0, + 832.0, + 440.0, + 832.0, + 484.0, + 782.0, + 484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 440.0, + 1094.0, + 440.0, + 1094.0, + 484.0, + 873.0, + 484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 389.0, + 585.0, + 389.0, + 585.0, + 435.0, + 293.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 604.0, + 389.0, + 611.0, + 389.0, + 611.0, + 435.0, + 604.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 654.0, + 389.0, + 852.0, + 389.0, + 852.0, + 435.0, + 654.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1938.0, + 842.0, + 1938.0, + 842.0, + 1984.0, + 294.0, + 1984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 1938.0, + 1262.0, + 1938.0, + 1262.0, + 1984.0, + 1035.0, + 1984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1335.0, + 1938.0, + 1406.0, + 1938.0, + 1406.0, + 1984.0, + 1335.0, + 1984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1975.0, + 561.0, + 1975.0, + 561.0, + 2011.0, + 295.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1896.0, + 754.0, + 1896.0, + 754.0, + 1938.0, + 294.0, + 1938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 1896.0, + 1079.0, + 1896.0, + 1079.0, + 1938.0, + 1028.0, + 1938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1153.0, + 1896.0, + 1163.0, + 1896.0, + 1163.0, + 1938.0, + 1153.0, + 1938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1855.0, + 688.0, + 1855.0, + 688.0, + 1897.0, + 293.0, + 1897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 1855.0, + 911.0, + 1855.0, + 911.0, + 1897.0, + 902.0, + 1897.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1437, + 1406, + 1437, + 1406, + 1841, + 297, + 1841 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 298, + 1854, + 1403, + 1854, + 1403, + 2008, + 298, + 2008 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 297, + 1148, + 1406, + 1148, + 1406, + 1272, + 297, + 1272 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 298, + 803, + 1403, + 803, + 1403, + 897, + 298, + 897 + ], + "score": 0.958 + }, + { + "category_id": 8, + "poly": [ + 479, + 983, + 1216, + 983, + 1216, + 1056, + 479, + 1056 + ], + "score": 0.942 + }, + { + "category_id": 1, + "poly": [ + 296, + 717, + 1399, + 717, + 1399, + 784, + 296, + 784 + ], + "score": 0.941 + }, + { + "category_id": 1, + "poly": [ + 295, + 903, + 1405, + 903, + 1405, + 969, + 295, + 969 + ], + "score": 0.938 + }, + { + "category_id": 0, + "poly": [ + 299, + 1382, + 595, + 1382, + 595, + 1416, + 299, + 1416 + ], + "score": 0.93 + }, + { + "category_id": 0, + "poly": [ + 298, + 1314, + 533, + 1314, + 533, + 1352, + 298, + 1352 + ], + "score": 0.916 + }, + { + "category_id": 1, + "poly": [ + 296, + 1072, + 1020, + 1072, + 1020, + 1128, + 296, + 1128 + ], + "score": 0.914 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1004, + 1400, + 1004, + 1400, + 1035, + 1352, + 1035 + ], + "score": 0.872 + }, + { + "category_id": 2, + "poly": [ + 840, + 2062, + 858, + 2062, + 858, + 2085, + 840, + 2085 + ], + "score": 0.802 + }, + { + "category_id": 0, + "poly": [ + 297, + 201, + 1010, + 201, + 1010, + 235, + 297, + 235 + ], + "score": 0.665 + }, + { + "category_id": 1, + "poly": [ + 298, + 623, + 440, + 623, + 440, + 655, + 298, + 655 + ], + "score": 0.504 + }, + { + "category_id": 1, + "poly": [ + 296, + 240, + 1362, + 240, + 1362, + 593, + 296, + 593 + ], + "score": 0.432 + }, + { + "category_id": 8, + "poly": [ + 306, + 281, + 1360, + 281, + 1360, + 593, + 306, + 593 + ], + "score": 0.398 + }, + { + "category_id": 2, + "poly": [ + 297, + 201, + 1010, + 201, + 1010, + 235, + 297, + 235 + ], + "score": 0.222 + }, + { + "category_id": 1, + "poly": [ + 309, + 589, + 435, + 589, + 435, + 617, + 309, + 617 + ], + "score": 0.203 + }, + { + "category_id": 13, + "poly": [ + 1201, + 1649, + 1362, + 1649, + 1362, + 1691, + 1201, + 1691 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { \\pi _ { b } ( a | s ) ^ { \\bullet } = \\frac { 1 } { | \\mathcal { A } | } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 841, + 903, + 981, + 903, + 981, + 944, + 841, + 944 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { \\beta = \\mathcal { O } \\left( \\frac { 1 } { T ^ { b } } \\right) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1045, + 903, + 1187, + 903, + 1187, + 944, + 1045, + 944 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\alpha = \\mathcal { O } \\left( \\frac { 1 } { T ^ { a } } \\right) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 519, + 1748, + 824, + 1748, + 824, + 1782, + 519, + 1782 + ], + "score": 0.92, + "latex": "\\pi _ { t } ( s ) = \\arg \\operatorname* { m a x } _ { a } Q _ { t } ( s , a )" + }, + { + "category_id": 14, + "poly": [ + 480, + 981, + 1216, + 981, + 1216, + 1058, + 480, + 1058 + ], + "score": 0.92, + "latex": "\\mathbb { E } [ \\| \\nabla J ( \\theta _ { W } ) \\| ^ { 2 } ] = \\odot \\left( \\frac { 1 } { T \\alpha } + \\alpha \\log ( 1 / \\alpha ) + \\frac { 1 } { T \\beta } + \\beta \\log ( 1 / \\beta ) \\right) ," + }, + { + "category_id": 13, + "poly": [ + 651, + 366, + 730, + 366, + 730, + 400, + 651, + 400 + ], + "score": 0.92, + "latex": "\\pi ( \\cdot | s _ { t } )" + }, + { + "category_id": 13, + "poly": [ + 1290, + 718, + 1403, + 718, + 1403, + 752, + 1290, + 752 + ], + "score": 0.92, + "latex": "\\rho \\in ( 0 , 1 )" + }, + { + "category_id": 13, + "poly": [ + 1278, + 904, + 1404, + 904, + 1404, + 943, + 1278, + 943 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\frac { 1 } { 2 } < a \\leq 1 } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 367, + 423, + 1358, + 423, + 1358, + 596, + 367, + 596 + ], + "score": 0.91, + "latex": "\\begin{array} { r l } & { \\ddot { \\delta _ { t } } ( \\theta _ { t } ) \\gets c _ { t } + \\gamma ( 1 - R ) V _ { \\theta _ { t } } ( s _ { t + 1 } ) + \\gamma R \\frac { \\log ( \\sum _ { s } e ^ { \\varrho \\theta ^ { \\top } \\phi _ { s } } ) } { \\varrho } - V _ { \\theta _ { t } } ( s _ { t } ) } \\\\ & { \\theta _ { t + 1 } \\gets \\Pi _ { K } \\left( \\theta _ { t } + \\alpha \\left( \\delta _ { t } ( \\theta _ { t } ) \\phi _ { t } - \\gamma \\bigg ( ( 1 - R ) \\phi _ { t + 1 } + R \\sum _ { s \\in \\mathcal { S } } \\left( \\frac { e ^ { \\sigma V _ { \\theta } ( s ) } \\phi _ { s } } { \\sum _ { j \\in \\mathcal { S } } e ^ { \\sigma V _ { \\theta } ( j ) } } \\right) \\bigg ) \\phi _ { t } ^ { \\top } \\omega _ { t } \\right) \\right) } \\\\ & { \\omega _ { t + 1 } \\gets \\Pi _ { K } ( \\omega _ { t } + \\beta ( \\delta _ { t } ( \\theta _ { t } ) - \\phi _ { t } ^ { \\top } \\omega _ { t } ) \\phi _ { t } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 346, + 938, + 463, + 938, + 463, + 968, + 346, + 968 + ], + "score": 0.91, + "latex": "0 < b \\leq a" + }, + { + "category_id": 13, + "poly": [ + 874, + 372, + 927, + 372, + 927, + 399, + 874, + 399 + ], + "score": 0.9, + "latex": "s _ { t + 1 }" + }, + { + "category_id": 13, + "poly": [ + 345, + 1687, + 410, + 1687, + 410, + 1717, + 345, + 1717 + ], + "score": 0.89, + "latex": "s \\in \\mathcal { S }" + }, + { + "category_id": 13, + "poly": [ + 461, + 1687, + 534, + 1687, + 534, + 1716, + 461, + 1716 + ], + "score": 0.89, + "latex": "a \\in { \\mathcal { A } }" + }, + { + "category_id": 13, + "poly": [ + 570, + 240, + 705, + 240, + 705, + 271, + 570, + 271 + ], + "score": 0.89, + "latex": "i = 1 , . . . , N" + }, + { + "category_id": 13, + "poly": [ + 1125, + 1592, + 1191, + 1592, + 1191, + 1622, + 1125, + 1622 + ], + "score": 0.89, + "latex": "1 - p" + }, + { + "category_id": 13, + "poly": [ + 1156, + 720, + 1236, + 720, + 1236, + 748, + 1156, + 748 + ], + "score": 0.89, + "latex": "m > 0" + }, + { + "category_id": 13, + "poly": [ + 386, + 337, + 627, + 337, + 627, + 368, + 386, + 368 + ], + "score": 0.88, + "latex": "t = 0 , 1 , 2 , . . . , W - 1" + }, + { + "category_id": 13, + "poly": [ + 490, + 750, + 554, + 750, + 554, + 778, + 490, + 778 + ], + "score": 0.83, + "latex": "t > 0" + }, + { + "category_id": 13, + "poly": [ + 562, + 748, + 999, + 748, + 999, + 783, + 562, + 783 + ], + "score": 0.83, + "latex": "\\begin{array} { r } { \\operatorname* { m a x } _ { s \\in \\mathcal { S } } d _ { T V } ( \\bar { \\mathbb { P } } ( s _ { t } | s _ { 0 } = s ) , \\mu _ { \\pi } ) \\leq m \\rho ^ { t } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 478, + 1083, + 615, + 1083, + 615, + 1114, + 478, + 1114 + ], + "score": 0.82, + "latex": "a = b = 0 . 5" + }, + { + "category_id": 13, + "poly": [ + 434, + 305, + 778, + 305, + 778, + 338, + 434, + 338 + ], + "score": 0.82, + "latex": "W \\sim \\mathrm { U n i f o r m } ( 0 , 1 , . . . , T - 1 )" + }, + { + "category_id": 13, + "poly": [ + 401, + 1889, + 420, + 1889, + 420, + 1916, + 401, + 1916 + ], + "score": 0.81, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 472, + 1886, + 497, + 1886, + 497, + 1912, + 472, + 1912 + ], + "score": 0.81, + "latex": "R" + }, + { + "category_id": 13, + "poly": [ + 1224, + 837, + 1247, + 837, + 1247, + 863, + 1224, + 863 + ], + "score": 0.81, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 682, + 1071, + 1015, + 1071, + 1015, + 1129, + 682, + 1129 + ], + "score": 0.8, + "latex": "\\begin{array} { r } { \\mathbb { E } [ \\| \\nabla J ( \\theta _ { W } ) \\| ^ { 2 } ] = \\odot \\left( \\frac { \\log T } { \\sqrt { T } } \\right) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 881, + 1565, + 899, + 1565, + 899, + 1592, + 881, + 1592 + ], + "score": 0.8, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 1227, + 1981, + 1245, + 1981, + 1245, + 2008, + 1227, + 2008 + ], + "score": 0.79, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 1294, + 1977, + 1319, + 1977, + 1319, + 2004, + 1294, + 2004 + ], + "score": 0.78, + "latex": "R" + }, + { + "category_id": 13, + "poly": [ + 978, + 372, + 1003, + 372, + 1003, + 397, + 978, + 397 + ], + "score": 0.77, + "latex": "c _ { t }" + }, + { + "category_id": 13, + "poly": [ + 465, + 272, + 533, + 272, + 533, + 302, + 465, + 302 + ], + "score": 0.77, + "latex": "\\theta _ { 0 } , w _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 426, + 1596, + 444, + 1596, + 444, + 1622, + 426, + 1622 + ], + "score": 0.74, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 372, + 397, + 481, + 397, + 481, + 431, + 372, + 431 + ], + "score": 0.74, + "latex": "\\phi _ { t } \\phi _ { s _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 399, + 622, + 440, + 622, + 440, + 654, + 399, + 654 + ], + "score": 0.73, + "latex": "\\theta _ { W }" + }, + { + "category_id": 13, + "poly": [ + 910, + 241, + 939, + 241, + 939, + 267, + 910, + 267 + ], + "score": 0.7, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 498, + 241, + 526, + 241, + 526, + 271, + 498, + 271 + ], + "score": 0.66, + "latex": "\\phi _ { i }" + }, + { + "category_id": 13, + "poly": [ + 544, + 278, + 572, + 278, + 572, + 302, + 544, + 302 + ], + "score": 0.5, + "latex": "s _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 371, + 556, + 826, + 556, + 826, + 591, + 371, + 591 + ], + "score": 0.45, + "latex": "\\omega _ { t + 1 } \\gets \\Pi _ { K } ( \\omega _ { t } + \\beta ( \\delta _ { t } ( \\theta _ { t } ) - \\phi _ { t } ^ { \\top } \\omega _ { t } ) \\phi _ { t } )" + }, + { + "category_id": 13, + "poly": [ + 382, + 241, + 428, + 241, + 428, + 269, + 382, + 269 + ], + "score": 0.28, + "latex": "T , \\alpha" + }, + { + "category_id": 13, + "poly": [ + 382, + 240, + 527, + 240, + 527, + 271, + 382, + 271 + ], + "score": 0.27, + "latex": "T , \\alpha , \\beta , \\varrho , \\phi _ { i }" + }, + { + "category_id": 13, + "poly": [ + 437, + 241, + 458, + 241, + 458, + 271, + 437, + 271 + ], + "score": 0.27, + "latex": "\\beta" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1376.0, + 599.0, + 1376.0, + 599.0, + 1423.0, + 293.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1311.0, + 537.0, + 1311.0, + 537.0, + 1360.0, + 291.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2090.0, + 838.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 196.0, + 1013.0, + 196.0, + 1013.0, + 242.0, + 293.0, + 242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 196.0, + 1013.0, + 196.0, + 1013.0, + 242.0, + 293.0, + 242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1436.0, + 1406.0, + 1436.0, + 1406.0, + 1475.0, + 292.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1470.0, + 1404.0, + 1470.0, + 1404.0, + 1502.0, + 295.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1499.0, + 1406.0, + 1499.0, + 1406.0, + 1535.0, + 294.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1530.0, + 1406.0, + 1530.0, + 1406.0, + 1563.0, + 295.0, + 1563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1559.0, + 880.0, + 1559.0, + 880.0, + 1596.0, + 295.0, + 1596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 900.0, + 1559.0, + 1404.0, + 1559.0, + 1404.0, + 1596.0, + 900.0, + 1596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1590.0, + 425.0, + 1590.0, + 425.0, + 1626.0, + 294.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 445.0, + 1590.0, + 1124.0, + 1590.0, + 1124.0, + 1626.0, + 445.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1192.0, + 1590.0, + 1406.0, + 1590.0, + 1406.0, + 1626.0, + 1192.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1622.0, + 1404.0, + 1622.0, + 1404.0, + 1655.0, + 295.0, + 1655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1647.0, + 1200.0, + 1647.0, + 1200.0, + 1690.0, + 291.0, + 1690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1363.0, + 1652.0, + 1406.0, + 1652.0, + 1406.0, + 1685.0, + 1363.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1686.0, + 344.0, + 1686.0, + 344.0, + 1722.0, + 292.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 1686.0, + 460.0, + 1686.0, + 460.0, + 1722.0, + 411.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 1686.0, + 1406.0, + 1686.0, + 1406.0, + 1722.0, + 535.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1716.0, + 1407.0, + 1716.0, + 1407.0, + 1752.0, + 292.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1748.0, + 518.0, + 1748.0, + 518.0, + 1785.0, + 295.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 825.0, + 1748.0, + 1406.0, + 1748.0, + 1406.0, + 1785.0, + 825.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1777.0, + 1406.0, + 1777.0, + 1406.0, + 1814.0, + 295.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1808.0, + 1134.0, + 1808.0, + 1134.0, + 1845.0, + 295.0, + 1845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1853.0, + 1405.0, + 1853.0, + 1405.0, + 1890.0, + 294.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1882.0, + 400.0, + 1882.0, + 400.0, + 1920.0, + 293.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 1882.0, + 471.0, + 1882.0, + 471.0, + 1920.0, + 421.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 1882.0, + 1405.0, + 1882.0, + 1405.0, + 1920.0, + 498.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1911.0, + 1407.0, + 1911.0, + 1407.0, + 1950.0, + 293.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1946.0, + 1404.0, + 1946.0, + 1404.0, + 1979.0, + 297.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1973.0, + 1226.0, + 1973.0, + 1226.0, + 2013.0, + 293.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 1973.0, + 1293.0, + 1973.0, + 1293.0, + 2013.0, + 1246.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1320.0, + 1973.0, + 1405.0, + 1973.0, + 1405.0, + 2013.0, + 1320.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1146.0, + 1409.0, + 1146.0, + 1409.0, + 1184.0, + 292.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1178.0, + 1404.0, + 1178.0, + 1404.0, + 1211.0, + 295.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1211.0, + 1406.0, + 1211.0, + 1406.0, + 1244.0, + 295.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1240.0, + 661.0, + 1240.0, + 661.0, + 1273.0, + 295.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 804.0, + 1405.0, + 804.0, + 1405.0, + 838.0, + 294.0, + 838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 835.0, + 1223.0, + 835.0, + 1223.0, + 869.0, + 294.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1248.0, + 835.0, + 1405.0, + 835.0, + 1405.0, + 869.0, + 1248.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 863.0, + 829.0, + 863.0, + 829.0, + 900.0, + 293.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 713.0, + 1155.0, + 713.0, + 1155.0, + 756.0, + 294.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 713.0, + 1289.0, + 713.0, + 1289.0, + 756.0, + 1237.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 745.0, + 489.0, + 745.0, + 489.0, + 784.0, + 292.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 745.0, + 561.0, + 745.0, + 561.0, + 784.0, + 555.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1000.0, + 745.0, + 1012.0, + 745.0, + 1012.0, + 784.0, + 1000.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 900.0, + 840.0, + 900.0, + 840.0, + 948.0, + 292.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 982.0, + 900.0, + 1044.0, + 900.0, + 1044.0, + 948.0, + 982.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 900.0, + 1277.0, + 900.0, + 1277.0, + 948.0, + 1188.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1405.0, + 900.0, + 1408.0, + 900.0, + 1408.0, + 948.0, + 1405.0, + 948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 936.0, + 345.0, + 936.0, + 345.0, + 972.0, + 295.0, + 972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 936.0, + 686.0, + 936.0, + 686.0, + 972.0, + 464.0, + 972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 283.0, + 1061.0, + 477.0, + 1061.0, + 477.0, + 1138.0, + 283.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 1061.0, + 681.0, + 1061.0, + 681.0, + 1138.0, + 616.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1016.0, + 1061.0, + 1025.0, + 1061.0, + 1025.0, + 1138.0, + 1016.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 614.0, + 398.0, + 614.0, + 398.0, + 664.0, + 294.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 441.0, + 614.0, + 445.0, + 614.0, + 445.0, + 664.0, + 441.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 234.0, + 381.0, + 234.0, + 381.0, + 274.0, + 296.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 528.0, + 234.0, + 569.0, + 234.0, + 569.0, + 274.0, + 528.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.0, + 234.0, + 909.0, + 234.0, + 909.0, + 274.0, + 706.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 298.0, + 433.0, + 298.0, + 433.0, + 340.0, + 305.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 333.0, + 385.0, + 333.0, + 385.0, + 368.0, + 306.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 333.0, + 670.0, + 333.0, + 670.0, + 368.0, + 628.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 360.0, + 650.0, + 360.0, + 650.0, + 404.0, + 301.0, + 404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 731.0, + 360.0, + 873.0, + 360.0, + 873.0, + 404.0, + 731.0, + 404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 360.0, + 977.0, + 360.0, + 977.0, + 404.0, + 928.0, + 404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1004.0, + 360.0, + 1009.0, + 360.0, + 1009.0, + 404.0, + 1004.0, + 404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 392.0, + 371.0, + 392.0, + 371.0, + 437.0, + 306.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 392.0, + 487.0, + 392.0, + 487.0, + 437.0, + 482.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 412.0, + 366.0, + 412.0, + 366.0, + 563.0, + 306.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1359.0, + 412.0, + 1366.0, + 412.0, + 1366.0, + 563.0, + 1359.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 551.0, + 366.0, + 551.0, + 366.0, + 595.0, + 302.0, + 595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.25, + 268.5, + 577.25, + 268.5, + 577.25, + 302.5, + 296.25, + 302.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 584.0, + 438.0, + 584.0, + 438.0, + 620.0, + 307.0, + 620.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1320, + 1405, + 1320, + 1405, + 1597, + 296, + 1597 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 1030, + 1404, + 1030, + 1404, + 1124, + 298, + 1124 + ], + "score": 0.972 + }, + { + "category_id": 3, + "poly": [ + 325, + 1664, + 1365, + 1664, + 1365, + 1940, + 325, + 1940 + ], + "score": 0.971 + }, + { + "category_id": 3, + "poly": [ + 324, + 638, + 1365, + 638, + 1365, + 913, + 324, + 913 + ], + "score": 0.969 + }, + { + "category_id": 3, + "poly": [ + 330, + 237, + 1369, + 237, + 1369, + 510, + 330, + 510 + ], + "score": 0.968 + }, + { + "category_id": 0, + "poly": [ + 297, + 1159, + 960, + 1159, + 960, + 1193, + 297, + 1193 + ], + "score": 0.93 + }, + { + "category_id": 4, + "poly": [ + 432, + 1965, + 1268, + 1965, + 1268, + 2001, + 432, + 2001 + ], + "score": 0.929 + }, + { + "category_id": 4, + "poly": [ + 441, + 536, + 1254, + 536, + 1254, + 569, + 441, + 569 + ], + "score": 0.92 + }, + { + "category_id": 4, + "poly": [ + 460, + 939, + 1236, + 939, + 1236, + 972, + 460, + 972 + ], + "score": 0.912 + }, + { + "category_id": 1, + "poly": [ + 302, + 1214, + 1402, + 1214, + 1402, + 1308, + 302, + 1308 + ], + "score": 0.891 + }, + { + "category_id": 2, + "poly": [ + 840, + 2061, + 859, + 2061, + 859, + 2084, + 840, + 2084 + ], + "score": 0.803 + }, + { + "category_id": 13, + "poly": [ + 970, + 1381, + 1060, + 1381, + 1060, + 1416, + 970, + 1416 + ], + "score": 0.93, + "latex": "\\{ \\theta _ { t } ^ { i } \\} _ { t = 1 } ^ { \\infty }" + }, + { + "category_id": 13, + "poly": [ + 299, + 1381, + 417, + 1381, + 417, + 1416, + 299, + 1416 + ], + "score": 0.92, + "latex": "\\| \\nabla J ( \\theta ) \\| ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 738, + 1964, + 856, + 1964, + 856, + 2000, + 738, + 2000 + ], + "score": 0.92, + "latex": "\\| \\nabla J ( \\theta ) \\| ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1282, + 1410, + 1401, + 1410, + 1401, + 1446, + 1282, + 1446 + ], + "score": 0.92, + "latex": "\\| \\nabla J ( \\theta ) \\| ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 548, + 1247, + 615, + 1247, + 615, + 1275, + 548, + 1275 + ], + "score": 0.89, + "latex": "4 \\times 4" + }, + { + "category_id": 13, + "poly": [ + 575, + 1351, + 665, + 1351, + 665, + 1386, + 575, + 1386 + ], + "score": 0.88, + "latex": "\\{ \\theta _ { t } ^ { i } \\} _ { t = 1 } ^ { \\infty }" + }, + { + "category_id": 13, + "poly": [ + 681, + 1353, + 857, + 1353, + 857, + 1384, + 681, + 1384 + ], + "score": 0.83, + "latex": "i = 1 , 2 , . . . , 3 0" + }, + { + "category_id": 13, + "poly": [ + 437, + 1448, + 456, + 1448, + 456, + 1475, + 437, + 1475 + ], + "score": 0.8, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 505, + 1445, + 529, + 1445, + 529, + 1471, + 505, + 1471 + ], + "score": 0.72, + "latex": "R" + }, + { + "category_id": 13, + "poly": [ + 1230, + 477, + 1298, + 477, + 1298, + 506, + 1230, + 506 + ], + "score": 0.49, + "latex": "\\mathrm { R } { = } 0 . 2" + }, + { + "category_id": 13, + "poly": [ + 505, + 1907, + 572, + 1907, + 572, + 1936, + 505, + 1936 + ], + "score": 0.39, + "latex": "\\mathrm { R } { = } 0 . 1" + }, + { + "category_id": 13, + "poly": [ + 434, + 1908, + 496, + 1908, + 496, + 1939, + 434, + 1939 + ], + "score": 0.31, + "latex": "\\mathrm { p { = } 0 . 1 }" + }, + { + "category_id": 13, + "poly": [ + 1233, + 881, + 1300, + 881, + 1300, + 910, + 1233, + 910 + ], + "score": 0.29, + "latex": "\\mathrm { R } { = } 0 . 2" + }, + { + "category_id": 13, + "poly": [ + 1160, + 478, + 1222, + 478, + 1222, + 508, + 1160, + 508 + ], + "score": 0.26, + "latex": "\\mathrm { p { = } } 0 . 1" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1668.0, + 364.0, + 1668.0, + 364.0, + 1682.0, + 347.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 1667.0, + 1091.0, + 1667.0, + 1091.0, + 1681.0, + 1075.0, + 1681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 1677.0, + 602.0, + 1677.0, + 602.0, + 1696.0, + 527.0, + 1696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 1676.0, + 727.0, + 1676.0, + 727.0, + 1689.0, + 711.0, + 1689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 1677.0, + 966.0, + 1677.0, + 966.0, + 1696.0, + 891.0, + 1696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 1677.0, + 1331.0, + 1677.0, + 1331.0, + 1696.0, + 1255.0, + 1696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1693.0, + 365.0, + 1693.0, + 365.0, + 1710.0, + 345.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 1694.0, + 624.0, + 1694.0, + 624.0, + 1713.0, + 527.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 1694.0, + 989.0, + 1694.0, + 989.0, + 1713.0, + 891.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 1688.0, + 1091.0, + 1688.0, + 1091.0, + 1704.0, + 1075.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 1694.0, + 1353.0, + 1694.0, + 1353.0, + 1713.0, + 1255.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 1706.0, + 727.0, + 1706.0, + 727.0, + 1720.0, + 710.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 1714.0, + 1092.0, + 1714.0, + 1092.0, + 1729.0, + 1075.0, + 1729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1720.0, + 365.0, + 1720.0, + 365.0, + 1737.0, + 345.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1742.0, + 364.0, + 1742.0, + 364.0, + 1801.0, + 322.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 1736.0, + 727.0, + 1736.0, + 727.0, + 1800.0, + 685.0, + 1800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 1733.0, + 1090.0, + 1733.0, + 1090.0, + 1798.0, + 1051.0, + 1798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1074.0, + 1755.0, + 1092.0, + 1755.0, + 1092.0, + 1778.0, + 1074.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1774.0, + 365.0, + 1774.0, + 365.0, + 1791.0, + 345.0, + 1791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1074.0, + 1784.0, + 1090.0, + 1784.0, + 1090.0, + 1800.0, + 1074.0, + 1800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 1800.0, + 365.0, + 1800.0, + 365.0, + 1818.0, + 345.0, + 1818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 1795.0, + 727.0, + 1795.0, + 727.0, + 1810.0, + 711.0, + 1810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 1809.0, + 1092.0, + 1809.0, + 1092.0, + 1824.0, + 1075.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 1829.0, + 364.0, + 1829.0, + 364.0, + 1843.0, + 346.0, + 1843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 1826.0, + 727.0, + 1826.0, + 727.0, + 1840.0, + 711.0, + 1840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 1833.0, + 1092.0, + 1833.0, + 1092.0, + 1847.0, + 1075.0, + 1847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1857.0, + 363.0, + 1857.0, + 363.0, + 1868.0, + 352.0, + 1868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 1856.0, + 730.0, + 1856.0, + 730.0, + 1870.0, + 716.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1080.0, + 1858.0, + 1091.0, + 1858.0, + 1091.0, + 1870.0, + 1080.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 1874.0, + 382.0, + 1874.0, + 382.0, + 1884.0, + 372.0, + 1884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 1870.0, + 429.0, + 1870.0, + 429.0, + 1888.0, + 403.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 444.0, + 1871.0, + 471.0, + 1871.0, + 471.0, + 1888.0, + 444.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 485.0, + 1870.0, + 511.0, + 1870.0, + 511.0, + 1888.0, + 485.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 1871.0, + 554.0, + 1871.0, + 554.0, + 1888.0, + 524.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 565.0, + 1871.0, + 594.0, + 1871.0, + 594.0, + 1888.0, + 565.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 605.0, + 1871.0, + 634.0, + 1871.0, + 634.0, + 1888.0, + 605.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 1874.0, + 746.0, + 1874.0, + 746.0, + 1885.0, + 736.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 1871.0, + 795.0, + 1871.0, + 795.0, + 1888.0, + 769.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 811.0, + 1871.0, + 838.0, + 1871.0, + 838.0, + 1888.0, + 811.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 1870.0, + 879.0, + 1870.0, + 879.0, + 1888.0, + 852.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 1870.0, + 921.0, + 1870.0, + 921.0, + 1888.0, + 893.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 1871.0, + 964.0, + 1871.0, + 964.0, + 1888.0, + 934.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 976.0, + 1871.0, + 1004.0, + 1871.0, + 1004.0, + 1888.0, + 976.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1874.0, + 1111.0, + 1874.0, + 1111.0, + 1884.0, + 1099.0, + 1884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1134.0, + 1868.0, + 1159.0, + 1868.0, + 1159.0, + 1890.0, + 1134.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1175.0, + 1870.0, + 1202.0, + 1870.0, + 1202.0, + 1888.0, + 1175.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 1870.0, + 1245.0, + 1870.0, + 1245.0, + 1888.0, + 1218.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1259.0, + 1871.0, + 1288.0, + 1871.0, + 1288.0, + 1888.0, + 1259.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 1871.0, + 1331.0, + 1871.0, + 1331.0, + 1888.0, + 1302.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 1878.0, + 558.0, + 1878.0, + 558.0, + 1903.0, + 436.0, + 1903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 800.0, + 1878.0, + 922.0, + 1878.0, + 922.0, + 1903.0, + 800.0, + 1903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1164.0, + 1881.0, + 1287.0, + 1881.0, + 1287.0, + 1903.0, + 1164.0, + 1903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 1907.0, + 433.0, + 1907.0, + 433.0, + 1937.0, + 396.0, + 1937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 497.0, + 1907.0, + 504.0, + 1907.0, + 504.0, + 1937.0, + 497.0, + 1937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 752.0, + 1906.0, + 947.0, + 1906.0, + 947.0, + 1937.0, + 752.0, + 1937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 1907.0, + 1304.0, + 1907.0, + 1304.0, + 1937.0, + 1124.0, + 1937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 713.0, + 1764.0, + 723.0, + 1764.0, + 723.0, + 1780.0, + 713.0, + 1780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 647.0, + 364.0, + 647.0, + 364.0, + 664.0, + 341.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 648.0, + 728.0, + 648.0, + 728.0, + 664.0, + 704.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 648.0, + 1092.0, + 648.0, + 1092.0, + 664.0, + 1068.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 649.0, + 347.0, + 649.0, + 347.0, + 842.0, + 324.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 651.0, + 709.0, + 651.0, + 709.0, + 841.0, + 690.0, + 841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 651.0, + 1072.0, + 651.0, + 1072.0, + 840.0, + 1052.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 670.0, + 363.0, + 670.0, + 363.0, + 689.0, + 342.0, + 689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 670.0, + 727.0, + 670.0, + 727.0, + 689.0, + 705.0, + 689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1070.0, + 670.0, + 1091.0, + 670.0, + 1091.0, + 689.0, + 1070.0, + 689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 696.0, + 363.0, + 696.0, + 363.0, + 713.0, + 342.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 696.0, + 727.0, + 696.0, + 727.0, + 713.0, + 705.0, + 713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1070.0, + 695.0, + 1092.0, + 695.0, + 1092.0, + 712.0, + 1070.0, + 712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 720.0, + 362.0, + 720.0, + 362.0, + 738.0, + 342.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 720.0, + 727.0, + 720.0, + 727.0, + 738.0, + 707.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1072.0, + 720.0, + 1091.0, + 720.0, + 1091.0, + 734.0, + 1072.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 744.0, + 363.0, + 744.0, + 363.0, + 761.0, + 342.0, + 761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 746.0, + 726.0, + 746.0, + 726.0, + 760.0, + 707.0, + 760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1070.0, + 743.0, + 1091.0, + 743.0, + 1091.0, + 760.0, + 1070.0, + 760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 770.0, + 362.0, + 770.0, + 362.0, + 784.0, + 348.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 771.0, + 726.0, + 771.0, + 726.0, + 785.0, + 711.0, + 785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1076.0, + 767.0, + 1091.0, + 767.0, + 1091.0, + 784.0, + 1076.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 789.0, + 777.0, + 804.0, + 777.0, + 804.0, + 793.0, + 789.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 793.0, + 362.0, + 793.0, + 362.0, + 808.0, + 347.0, + 808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 800.0, + 602.0, + 800.0, + 602.0, + 825.0, + 491.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 795.0, + 726.0, + 795.0, + 726.0, + 809.0, + 711.0, + 809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 854.0, + 800.0, + 967.0, + 800.0, + 967.0, + 825.0, + 854.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 792.0, + 1089.0, + 792.0, + 1089.0, + 808.0, + 1075.0, + 808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 800.0, + 1331.0, + 800.0, + 1331.0, + 825.0, + 1218.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 820.0, + 360.0, + 820.0, + 360.0, + 831.0, + 348.0, + 831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 815.0, + 626.0, + 815.0, + 626.0, + 841.0, + 490.0, + 841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 820.0, + 726.0, + 820.0, + 726.0, + 834.0, + 711.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 854.0, + 816.0, + 990.0, + 816.0, + 990.0, + 841.0, + 854.0, + 841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 815.0, + 1089.0, + 815.0, + 1089.0, + 831.0, + 1075.0, + 831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1218.0, + 815.0, + 1355.0, + 815.0, + 1355.0, + 841.0, + 1218.0, + 841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 848.0, + 382.0, + 848.0, + 382.0, + 859.0, + 373.0, + 859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 845.0, + 417.0, + 845.0, + 417.0, + 862.0, + 395.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 429.0, + 846.0, + 444.0, + 846.0, + 444.0, + 861.0, + 429.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 461.0, + 846.0, + 476.0, + 846.0, + 476.0, + 860.0, + 461.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 487.0, + 845.0, + 541.0, + 845.0, + 541.0, + 861.0, + 487.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 845.0, + 571.0, + 845.0, + 571.0, + 861.0, + 547.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 845.0, + 603.0, + 845.0, + 603.0, + 862.0, + 577.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 845.0, + 632.0, + 845.0, + 632.0, + 862.0, + 606.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 848.0, + 746.0, + 848.0, + 746.0, + 859.0, + 736.0, + 859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 846.0, + 779.0, + 846.0, + 779.0, + 861.0, + 762.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 793.0, + 846.0, + 809.0, + 846.0, + 809.0, + 860.0, + 793.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 824.0, + 846.0, + 840.0, + 846.0, + 840.0, + 860.0, + 824.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 845.0, + 875.0, + 845.0, + 875.0, + 861.0, + 852.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 880.0, + 845.0, + 904.0, + 845.0, + 904.0, + 862.0, + 880.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 911.0, + 845.0, + 936.0, + 845.0, + 936.0, + 862.0, + 911.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 940.0, + 843.0, + 997.0, + 843.0, + 997.0, + 863.0, + 940.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 848.0, + 1111.0, + 848.0, + 1111.0, + 859.0, + 1100.0, + 859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.0, + 845.0, + 1145.0, + 845.0, + 1145.0, + 862.0, + 1125.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1156.0, + 846.0, + 1173.0, + 846.0, + 1173.0, + 861.0, + 1156.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 845.0, + 1205.0, + 845.0, + 1205.0, + 860.0, + 1188.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 845.0, + 1240.0, + 845.0, + 1240.0, + 861.0, + 1215.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 845.0, + 1269.0, + 845.0, + 1269.0, + 861.0, + 1245.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1275.0, + 845.0, + 1302.0, + 845.0, + 1302.0, + 862.0, + 1275.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 843.0, + 1362.0, + 843.0, + 1362.0, + 863.0, + 1303.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 856.0, + 556.0, + 856.0, + 556.0, + 875.0, + 439.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 803.0, + 856.0, + 919.0, + 856.0, + 919.0, + 875.0, + 803.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1167.0, + 856.0, + 1284.0, + 856.0, + 1284.0, + 875.0, + 1167.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 881.0, + 574.0, + 881.0, + 574.0, + 911.0, + 395.0, + 911.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 753.0, + 880.0, + 946.0, + 880.0, + 946.0, + 910.0, + 753.0, + 910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 881.0, + 1232.0, + 881.0, + 1232.0, + 911.0, + 1124.0, + 911.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 881.0, + 1304.0, + 881.0, + 1304.0, + 911.0, + 1301.0, + 911.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 754.75, + 794.5, + 800.75, + 794.5, + 800.75, + 811.5, + 754.75, + 811.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 243.0, + 359.0, + 243.0, + 359.0, + 441.0, + 328.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 246.0, + 1077.0, + 246.0, + 1077.0, + 438.0, + 1054.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 248.0, + 714.0, + 248.0, + 714.0, + 439.0, + 692.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 250.0, + 367.0, + 250.0, + 367.0, + 267.0, + 346.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 251.0, + 728.0, + 251.0, + 728.0, + 269.0, + 707.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1071.0, + 252.0, + 1088.0, + 252.0, + 1088.0, + 267.0, + 1071.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 281.0, + 364.0, + 281.0, + 364.0, + 297.0, + 349.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 281.0, + 726.0, + 281.0, + 726.0, + 297.0, + 711.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1070.0, + 276.0, + 1090.0, + 276.0, + 1090.0, + 294.0, + 1070.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 304.0, + 1089.0, + 304.0, + 1089.0, + 318.0, + 1073.0, + 318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 311.0, + 367.0, + 311.0, + 367.0, + 328.0, + 346.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 313.0, + 726.0, + 313.0, + 726.0, + 327.0, + 711.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1070.0, + 328.0, + 1092.0, + 328.0, + 1092.0, + 346.0, + 1070.0, + 346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 342.0, + 365.0, + 342.0, + 365.0, + 357.0, + 349.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 342.0, + 727.0, + 342.0, + 727.0, + 356.0, + 711.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 356.0, + 1089.0, + 356.0, + 1089.0, + 370.0, + 1073.0, + 370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 372.0, + 364.0, + 372.0, + 364.0, + 387.0, + 349.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 371.0, + 727.0, + 371.0, + 727.0, + 386.0, + 711.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 380.0, + 1089.0, + 380.0, + 1089.0, + 396.0, + 1073.0, + 396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 401.0, + 365.0, + 401.0, + 365.0, + 420.0, + 346.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 492.0, + 400.0, + 628.0, + 400.0, + 628.0, + 438.0, + 492.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 709.0, + 401.0, + 726.0, + 401.0, + 726.0, + 417.0, + 709.0, + 417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 855.0, + 400.0, + 990.0, + 400.0, + 990.0, + 438.0, + 855.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1070.0, + 405.0, + 1089.0, + 405.0, + 1089.0, + 424.0, + 1070.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 397.0, + 1352.0, + 397.0, + 1352.0, + 438.0, + 1217.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 433.0, + 365.0, + 433.0, + 365.0, + 447.0, + 348.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 432.0, + 728.0, + 432.0, + 728.0, + 446.0, + 711.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1071.0, + 433.0, + 1090.0, + 433.0, + 1090.0, + 447.0, + 1071.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 443.0, + 384.0, + 443.0, + 384.0, + 455.0, + 375.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 441.0, + 443.0, + 441.0, + 443.0, + 459.0, + 415.0, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 465.0, + 441.0, + 495.0, + 441.0, + 495.0, + 457.0, + 465.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 441.0, + 546.0, + 441.0, + 546.0, + 457.0, + 514.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 564.0, + 441.0, + 596.0, + 441.0, + 596.0, + 457.0, + 564.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 612.0, + 441.0, + 645.0, + 441.0, + 645.0, + 457.0, + 612.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 737.0, + 445.0, + 747.0, + 445.0, + 747.0, + 455.0, + 737.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 441.0, + 804.0, + 441.0, + 804.0, + 459.0, + 777.0, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 826.0, + 441.0, + 857.0, + 441.0, + 857.0, + 457.0, + 826.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 441.0, + 908.0, + 441.0, + 908.0, + 457.0, + 877.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 441.0, + 957.0, + 441.0, + 957.0, + 457.0, + 925.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 975.0, + 441.0, + 1007.0, + 441.0, + 1007.0, + 457.0, + 975.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 445.0, + 1108.0, + 445.0, + 1108.0, + 455.0, + 1099.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1139.0, + 441.0, + 1166.0, + 441.0, + 1166.0, + 459.0, + 1139.0, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 441.0, + 1218.0, + 441.0, + 1218.0, + 457.0, + 1188.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 441.0, + 1271.0, + 441.0, + 1271.0, + 457.0, + 1237.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1287.0, + 441.0, + 1321.0, + 441.0, + 1321.0, + 457.0, + 1287.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1337.0, + 442.0, + 1370.0, + 442.0, + 1370.0, + 457.0, + 1337.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 450.0, + 559.0, + 450.0, + 559.0, + 473.0, + 439.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 801.0, + 450.0, + 921.0, + 450.0, + 921.0, + 473.0, + 801.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1164.0, + 450.0, + 1283.0, + 450.0, + 1283.0, + 473.0, + 1164.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 476.0, + 576.0, + 476.0, + 576.0, + 506.0, + 397.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 753.0, + 476.0, + 946.0, + 476.0, + 946.0, + 507.0, + 753.0, + 507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 476.0, + 1159.0, + 476.0, + 1159.0, + 507.0, + 1121.0, + 507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1223.0, + 476.0, + 1229.0, + 476.0, + 1229.0, + 507.0, + 1223.0, + 507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1154.0, + 962.0, + 1154.0, + 962.0, + 1200.0, + 293.0, + 1200.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 429.0, + 1963.0, + 737.0, + 1963.0, + 737.0, + 2003.0, + 429.0, + 2003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 857.0, + 1963.0, + 1269.0, + 1963.0, + 1269.0, + 2003.0, + 857.0, + 2003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 529.0, + 1258.0, + 529.0, + 1258.0, + 576.0, + 439.0, + 576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 458.0, + 930.0, + 1239.0, + 930.0, + 1239.0, + 981.0, + 458.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2060.0, + 861.0, + 2060.0, + 861.0, + 2090.0, + 839.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1320.0, + 1406.0, + 1320.0, + 1406.0, + 1356.0, + 294.0, + 1356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1346.0, + 574.0, + 1346.0, + 574.0, + 1394.0, + 290.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 666.0, + 1346.0, + 680.0, + 1346.0, + 680.0, + 1394.0, + 666.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 1346.0, + 1411.0, + 1346.0, + 1411.0, + 1394.0, + 858.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1376.0, + 298.0, + 1376.0, + 298.0, + 1426.0, + 293.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 418.0, + 1376.0, + 969.0, + 1376.0, + 969.0, + 1426.0, + 418.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1061.0, + 1376.0, + 1410.0, + 1376.0, + 1410.0, + 1426.0, + 1061.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1409.0, + 1281.0, + 1409.0, + 1281.0, + 1449.0, + 290.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1409.0, + 1406.0, + 1409.0, + 1406.0, + 1449.0, + 1402.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1443.0, + 436.0, + 1443.0, + 436.0, + 1479.0, + 294.0, + 1479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 457.0, + 1443.0, + 504.0, + 1443.0, + 504.0, + 1479.0, + 457.0, + 1479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 530.0, + 1443.0, + 1407.0, + 1443.0, + 1407.0, + 1479.0, + 530.0, + 1479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1472.0, + 1406.0, + 1472.0, + 1406.0, + 1507.0, + 293.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1504.0, + 1405.0, + 1504.0, + 1405.0, + 1537.0, + 296.0, + 1537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1533.0, + 1407.0, + 1533.0, + 1407.0, + 1570.0, + 294.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1561.0, + 1129.0, + 1561.0, + 1129.0, + 1601.0, + 293.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1028.0, + 1405.0, + 1028.0, + 1405.0, + 1067.0, + 292.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1059.0, + 1405.0, + 1059.0, + 1405.0, + 1098.0, + 292.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1093.0, + 546.0, + 1093.0, + 546.0, + 1128.0, + 296.0, + 1128.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1215.0, + 1405.0, + 1215.0, + 1405.0, + 1249.0, + 294.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1246.0, + 547.0, + 1246.0, + 547.0, + 1280.0, + 294.0, + 1280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 1246.0, + 1406.0, + 1246.0, + 1406.0, + 1280.0, + 616.0, + 1280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1277.0, + 1217.0, + 1277.0, + 1217.0, + 1311.0, + 296.0, + 1311.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 257, + 1405, + 257, + 1405, + 506, + 296, + 506 + ], + "score": 0.978 + }, + { + "category_id": 3, + "poly": [ + 338, + 937, + 1347, + 937, + 1347, + 1203, + 338, + 1203 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 299, + 520, + 1404, + 520, + 1404, + 612, + 299, + 612 + ], + "score": 0.969 + }, + { + "category_id": 1, + "poly": [ + 298, + 1800, + 1401, + 1800, + 1401, + 1861, + 298, + 1861 + ], + "score": 0.938 + }, + { + "category_id": 1, + "poly": [ + 297, + 626, + 1406, + 626, + 1406, + 871, + 297, + 871 + ], + "score": 0.937 + }, + { + "category_id": 0, + "poly": [ + 296, + 201, + 1002, + 201, + 1002, + 236, + 296, + 236 + ], + "score": 0.92 + }, + { + "category_id": 0, + "poly": [ + 298, + 1340, + 510, + 1340, + 510, + 1376, + 298, + 1376 + ], + "score": 0.916 + }, + { + "category_id": 0, + "poly": [ + 299, + 1731, + 597, + 1731, + 597, + 1769, + 299, + 1769 + ], + "score": 0.911 + }, + { + "category_id": 2, + "poly": [ + 837, + 2062, + 865, + 2062, + 865, + 2085, + 837, + 2085 + ], + "score": 0.851 + }, + { + "category_id": 4, + "poly": [ + 573, + 1230, + 1121, + 1230, + 1121, + 1262, + 573, + 1262 + ], + "score": 0.821 + }, + { + "category_id": 1, + "poly": [ + 298, + 1384, + 1404, + 1384, + 1404, + 1568, + 298, + 1568 + ], + "score": 0.625 + }, + { + "category_id": 1, + "poly": [ + 299, + 1568, + 1406, + 1568, + 1406, + 1689, + 299, + 1689 + ], + "score": 0.448 + }, + { + "category_id": 4, + "poly": [ + 574, + 1229, + 1121, + 1229, + 1121, + 1262, + 574, + 1262 + ], + "score": 0.143 + }, + { + "category_id": 13, + "poly": [ + 788, + 351, + 890, + 351, + 890, + 384, + 788, + 384 + ], + "score": 0.94, + "latex": "\\mathcal { A } _ { a d } \\triangleq \\mathcal { S }" + }, + { + "category_id": 13, + "poly": [ + 1307, + 749, + 1398, + 749, + 1398, + 780, + 1307, + 780 + ], + "score": 0.9, + "latex": "\\gamma = 0 . 9" + }, + { + "category_id": 13, + "poly": [ + 297, + 354, + 347, + 354, + 347, + 384, + 297, + 384 + ], + "score": 0.9, + "latex": "\\mathcal { A } _ { a d }" + }, + { + "category_id": 13, + "poly": [ + 1164, + 748, + 1258, + 748, + 1258, + 777, + 1164, + 777 + ], + "score": 0.89, + "latex": "\\alpha = 0 . 2" + }, + { + "category_id": 13, + "poly": [ + 527, + 445, + 600, + 445, + 600, + 474, + 527, + 474 + ], + "score": 0.88, + "latex": "1 - R" + }, + { + "category_id": 13, + "poly": [ + 297, + 446, + 338, + 446, + 338, + 478, + 297, + 478 + ], + "score": 0.88, + "latex": "p _ { s } ^ { a _ { a } }" + }, + { + "category_id": 13, + "poly": [ + 1214, + 388, + 1296, + 388, + 1296, + 415, + 1214, + 415 + ], + "score": 0.88, + "latex": "a _ { a } , a _ { a d }" + }, + { + "category_id": 13, + "poly": [ + 1294, + 551, + 1328, + 551, + 1328, + 583, + 1294, + 583 + ], + "score": 0.88, + "latex": "Q _ { t }" + }, + { + "category_id": 13, + "poly": [ + 656, + 418, + 700, + 418, + 700, + 445, + 656, + 445 + ], + "score": 0.87, + "latex": "a _ { a d }" + }, + { + "category_id": 13, + "poly": [ + 889, + 416, + 914, + 416, + 914, + 443, + 889, + 443 + ], + "score": 0.79, + "latex": "R" + }, + { + "category_id": 13, + "poly": [ + 707, + 723, + 724, + 723, + 724, + 750, + 707, + 750 + ], + "score": 0.74, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 299, + 717, + 510, + 717, + 510, + 751, + 299, + 751 + ], + "score": 0.62, + "latex": "( \\mathrm { a r g m i n } _ { s \\in \\mathcal { S } } V _ { t } ( s ) )" + }, + { + "category_id": 13, + "poly": [ + 518, + 1171, + 585, + 1171, + 585, + 1200, + 518, + 1200 + ], + "score": 0.33, + "latex": "\\mathrm { R } { = } 0 . 1" + }, + { + "category_id": 13, + "poly": [ + 1220, + 1171, + 1287, + 1171, + 1287, + 1200, + 1220, + 1200 + ], + "score": 0.27, + "latex": "\\mathrm { R } { = } 0 . 2" + }, + { + "category_id": 13, + "poly": [ + 440, + 717, + 505, + 717, + 505, + 751, + 440, + 751 + ], + "score": 0.26, + "latex": "V _ { t } ( s ) \\big )" + }, + { + "category_id": 15, + "poly": [ + 340.0, + 949.0, + 358.0, + 949.0, + 358.0, + 1132.0, + 340.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 961.0, + 379.0, + 961.0, + 379.0, + 972.0, + 371.0, + 972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 949.0, + 708.0, + 949.0, + 708.0, + 1132.0, + 690.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 961.0, + 729.0, + 961.0, + 729.0, + 970.0, + 721.0, + 970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1039.0, + 948.0, + 1061.0, + 948.0, + 1061.0, + 1133.0, + 1039.0, + 1133.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1072.0, + 961.0, + 1082.0, + 961.0, + 1082.0, + 972.0, + 1072.0, + 972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 991.0, + 380.0, + 991.0, + 380.0, + 1005.0, + 363.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 713.0, + 991.0, + 730.0, + 991.0, + 730.0, + 1005.0, + 713.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 993.0, + 1082.0, + 993.0, + 1082.0, + 1006.0, + 1064.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 1023.0, + 380.0, + 1023.0, + 380.0, + 1037.0, + 364.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 1024.0, + 730.0, + 1024.0, + 730.0, + 1037.0, + 714.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 1027.0, + 1082.0, + 1027.0, + 1082.0, + 1040.0, + 1064.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1056.0, + 380.0, + 1056.0, + 380.0, + 1069.0, + 362.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 712.0, + 1056.0, + 730.0, + 1056.0, + 730.0, + 1069.0, + 712.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1063.0, + 1060.0, + 1082.0, + 1060.0, + 1082.0, + 1073.0, + 1063.0, + 1073.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1088.0, + 379.0, + 1088.0, + 379.0, + 1102.0, + 363.0, + 1102.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 1090.0, + 526.0, + 1090.0, + 526.0, + 1117.0, + 416.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 713.0, + 1088.0, + 729.0, + 1088.0, + 729.0, + 1102.0, + 713.0, + 1102.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 1090.0, + 875.0, + 1090.0, + 875.0, + 1117.0, + 766.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1063.0, + 1094.0, + 1080.0, + 1094.0, + 1080.0, + 1107.0, + 1063.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1117.0, + 1090.0, + 1226.0, + 1090.0, + 1226.0, + 1117.0, + 1117.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1119.0, + 382.0, + 1119.0, + 382.0, + 1135.0, + 354.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 1110.0, + 456.0, + 1110.0, + 456.0, + 1131.0, + 417.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 704.0, + 1119.0, + 733.0, + 1119.0, + 733.0, + 1136.0, + 704.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 1110.0, + 806.0, + 1110.0, + 806.0, + 1131.0, + 767.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 1126.0, + 1084.0, + 1126.0, + 1084.0, + 1142.0, + 1059.0, + 1142.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1117.0, + 1110.0, + 1157.0, + 1110.0, + 1157.0, + 1131.0, + 1117.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 1137.0, + 402.0, + 1137.0, + 402.0, + 1151.0, + 389.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 1135.0, + 641.0, + 1135.0, + 641.0, + 1153.0, + 410.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 1137.0, + 752.0, + 1137.0, + 752.0, + 1151.0, + 739.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 760.0, + 1135.0, + 786.0, + 1135.0, + 786.0, + 1152.0, + 760.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 1135.0, + 992.0, + 1135.0, + 992.0, + 1153.0, + 788.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 1137.0, + 1104.0, + 1137.0, + 1104.0, + 1151.0, + 1090.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1112.0, + 1135.0, + 1343.0, + 1135.0, + 1343.0, + 1153.0, + 1112.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 457.0, + 1147.0, + 564.0, + 1147.0, + 564.0, + 1165.0, + 457.0, + 1165.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 1147.0, + 915.0, + 1147.0, + 915.0, + 1165.0, + 808.0, + 1165.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1158.0, + 1147.0, + 1265.0, + 1147.0, + 1265.0, + 1165.0, + 1158.0, + 1165.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 1169.0, + 517.0, + 1169.0, + 517.0, + 1201.0, + 408.0, + 1201.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 1169.0, + 947.0, + 1169.0, + 947.0, + 1201.0, + 751.0, + 1201.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 1169.0, + 1219.0, + 1169.0, + 1219.0, + 1203.0, + 1109.0, + 1203.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 200.0, + 1006.0, + 200.0, + 1006.0, + 243.0, + 293.0, + 243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1331.0, + 514.0, + 1331.0, + 514.0, + 1386.0, + 291.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1727.0, + 603.0, + 1727.0, + 603.0, + 1775.0, + 292.0, + 1775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 1227.0, + 1123.0, + 1227.0, + 1123.0, + 1267.0, + 574.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 1227.0, + 1123.0, + 1227.0, + 1123.0, + 1266.0, + 574.0, + 1266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 255.0, + 1406.0, + 255.0, + 1406.0, + 294.0, + 293.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 287.0, + 1406.0, + 287.0, + 1406.0, + 323.0, + 293.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 317.0, + 1403.0, + 317.0, + 1403.0, + 356.0, + 294.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 351.0, + 296.0, + 351.0, + 296.0, + 389.0, + 293.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 351.0, + 787.0, + 351.0, + 787.0, + 389.0, + 348.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 351.0, + 1405.0, + 351.0, + 1405.0, + 389.0, + 891.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 380.0, + 1213.0, + 380.0, + 1213.0, + 420.0, + 293.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1297.0, + 380.0, + 1405.0, + 380.0, + 1405.0, + 420.0, + 1297.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 415.0, + 655.0, + 415.0, + 655.0, + 447.0, + 293.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 415.0, + 888.0, + 415.0, + 888.0, + 447.0, + 701.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 415.0, + 1405.0, + 415.0, + 1405.0, + 447.0, + 915.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 442.0, + 296.0, + 442.0, + 296.0, + 481.0, + 291.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 442.0, + 526.0, + 442.0, + 526.0, + 481.0, + 339.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 442.0, + 1407.0, + 442.0, + 1407.0, + 481.0, + 601.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 474.0, + 577.0, + 474.0, + 577.0, + 508.0, + 294.0, + 508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 518.0, + 1405.0, + 518.0, + 1405.0, + 555.0, + 294.0, + 555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 549.0, + 1293.0, + 549.0, + 1293.0, + 585.0, + 293.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1329.0, + 549.0, + 1405.0, + 549.0, + 1405.0, + 585.0, + 1329.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 579.0, + 975.0, + 579.0, + 975.0, + 618.0, + 293.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1798.0, + 1405.0, + 1798.0, + 1405.0, + 1837.0, + 294.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1830.0, + 688.0, + 1830.0, + 688.0, + 1862.0, + 297.0, + 1862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 625.0, + 1407.0, + 625.0, + 1407.0, + 662.0, + 294.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 657.0, + 1406.0, + 657.0, + 1406.0, + 691.0, + 295.0, + 691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 686.0, + 1406.0, + 686.0, + 1406.0, + 720.0, + 294.0, + 720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 717.0, + 298.0, + 717.0, + 298.0, + 752.0, + 292.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 511.0, + 717.0, + 706.0, + 717.0, + 706.0, + 752.0, + 511.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 725.0, + 717.0, + 1409.0, + 717.0, + 1409.0, + 752.0, + 725.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 747.0, + 1163.0, + 747.0, + 1163.0, + 781.0, + 292.0, + 781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1259.0, + 747.0, + 1306.0, + 747.0, + 1306.0, + 781.0, + 1259.0, + 781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 747.0, + 1409.0, + 747.0, + 1409.0, + 781.0, + 1399.0, + 781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 776.0, + 1406.0, + 776.0, + 1406.0, + 812.0, + 292.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 810.0, + 1406.0, + 810.0, + 1406.0, + 843.0, + 294.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 840.0, + 855.0, + 840.0, + 855.0, + 873.0, + 295.0, + 873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1383.0, + 1404.0, + 1383.0, + 1404.0, + 1418.0, + 293.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1415.0, + 1404.0, + 1415.0, + 1404.0, + 1451.0, + 294.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1443.0, + 1406.0, + 1443.0, + 1406.0, + 1481.0, + 293.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1475.0, + 1405.0, + 1475.0, + 1405.0, + 1509.0, + 294.0, + 1509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1505.0, + 1405.0, + 1505.0, + 1405.0, + 1539.0, + 295.0, + 1539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1538.0, + 1408.0, + 1538.0, + 1408.0, + 1570.0, + 294.0, + 1570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1566.0, + 1404.0, + 1566.0, + 1404.0, + 1601.0, + 294.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1595.0, + 1406.0, + 1595.0, + 1406.0, + 1631.0, + 293.0, + 1631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1624.0, + 1404.0, + 1624.0, + 1404.0, + 1664.0, + 293.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1658.0, + 793.0, + 1658.0, + 793.0, + 1696.0, + 294.0, + 1696.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 0, + "poly": [ + 298, + 203, + 431, + 203, + 431, + 234, + 298, + 234 + ], + "score": 0.877 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 863, + 2061, + 863, + 2085, + 836, + 2085 + ], + "score": 0.786 + }, + { + "category_id": 1, + "poly": [ + 297, + 1473, + 1401, + 1473, + 1401, + 1538, + 297, + 1538 + ], + "score": 0.651 + }, + { + "category_id": 1, + "poly": [ + 295, + 1064, + 1400, + 1064, + 1400, + 1129, + 295, + 1129 + ], + "score": 0.628 + }, + { + "category_id": 1, + "poly": [ + 301, + 1269, + 1403, + 1269, + 1403, + 1366, + 301, + 1366 + ], + "score": 0.61 + }, + { + "category_id": 1, + "poly": [ + 294, + 364, + 1404, + 364, + 1404, + 459, + 294, + 459 + ], + "score": 0.609 + }, + { + "category_id": 1, + "poly": [ + 296, + 482, + 1403, + 482, + 1403, + 575, + 296, + 575 + ], + "score": 0.602 + }, + { + "category_id": 1, + "poly": [ + 290, + 977, + 1403, + 977, + 1403, + 1042, + 290, + 1042 + ], + "score": 0.597 + }, + { + "category_id": 1, + "poly": [ + 296, + 1560, + 1406, + 1560, + 1406, + 1655, + 296, + 1655 + ], + "score": 0.56 + }, + { + "category_id": 1, + "poly": [ + 293, + 1386, + 1401, + 1386, + 1401, + 1452, + 293, + 1452 + ], + "score": 0.557 + }, + { + "category_id": 1, + "poly": [ + 297, + 1678, + 1405, + 1678, + 1405, + 1772, + 297, + 1772 + ], + "score": 0.55 + }, + { + "category_id": 1, + "poly": [ + 295, + 1151, + 1402, + 1151, + 1402, + 1247, + 295, + 1247 + ], + "score": 0.544 + }, + { + "category_id": 1, + "poly": [ + 290, + 891, + 1402, + 891, + 1402, + 955, + 290, + 955 + ], + "score": 0.544 + }, + { + "category_id": 1, + "poly": [ + 300, + 599, + 1401, + 599, + 1401, + 693, + 300, + 693 + ], + "score": 0.538 + }, + { + "category_id": 1, + "poly": [ + 298, + 717, + 1399, + 717, + 1399, + 811, + 298, + 811 + ], + "score": 0.517 + }, + { + "category_id": 1, + "poly": [ + 293, + 1795, + 1403, + 1795, + 1403, + 1890, + 293, + 1890 + ], + "score": 0.502 + }, + { + "category_id": 1, + "poly": [ + 289, + 832, + 1405, + 832, + 1405, + 870, + 289, + 870 + ], + "score": 0.485 + }, + { + "category_id": 1, + "poly": [ + 298, + 248, + 1407, + 248, + 1407, + 341, + 298, + 341 + ], + "score": 0.38 + }, + { + "category_id": 1, + "poly": [ + 298, + 1914, + 1406, + 1914, + 1406, + 2008, + 298, + 2008 + ], + "score": 0.336 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 200.0, + 433.0, + 200.0, + 433.0, + 237.0, + 295.0, + 237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 868.0, + 2058.0, + 868.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1475.0, + 1405.0, + 1475.0, + 1405.0, + 1511.0, + 296.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1504.0, + 1168.0, + 1504.0, + 1168.0, + 1540.0, + 322.0, + 1540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1066.0, + 1404.0, + 1066.0, + 1404.0, + 1102.0, + 295.0, + 1102.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1097.0, + 707.0, + 1097.0, + 707.0, + 1130.0, + 320.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1271.0, + 1407.0, + 1271.0, + 1407.0, + 1305.0, + 296.0, + 1305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1302.0, + 1407.0, + 1302.0, + 1407.0, + 1336.0, + 322.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1330.0, + 1408.0, + 1330.0, + 1408.0, + 1368.0, + 319.0, + 1368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 362.0, + 1405.0, + 362.0, + 1405.0, + 403.0, + 293.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 394.0, + 1404.0, + 394.0, + 1404.0, + 432.0, + 323.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 427.0, + 659.0, + 427.0, + 659.0, + 458.0, + 323.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 483.0, + 1405.0, + 483.0, + 1405.0, + 516.0, + 296.0, + 516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 510.0, + 1408.0, + 510.0, + 1408.0, + 548.0, + 320.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 539.0, + 399.0, + 539.0, + 399.0, + 580.0, + 320.0, + 580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 979.0, + 1405.0, + 979.0, + 1405.0, + 1015.0, + 297.0, + 1015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1008.0, + 782.0, + 1008.0, + 782.0, + 1042.0, + 322.0, + 1042.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1559.0, + 1411.0, + 1559.0, + 1411.0, + 1600.0, + 295.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1593.0, + 1408.0, + 1593.0, + 1408.0, + 1629.0, + 320.0, + 1629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1625.0, + 631.0, + 1625.0, + 631.0, + 1656.0, + 321.0, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1387.0, + 1404.0, + 1387.0, + 1404.0, + 1424.0, + 297.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1416.0, + 1382.0, + 1416.0, + 1382.0, + 1453.0, + 324.0, + 1453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1678.0, + 1407.0, + 1678.0, + 1407.0, + 1715.0, + 296.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1708.0, + 1407.0, + 1708.0, + 1407.0, + 1746.0, + 320.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1734.0, + 397.0, + 1734.0, + 397.0, + 1775.0, + 320.0, + 1775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1150.0, + 1406.0, + 1150.0, + 1406.0, + 1190.0, + 293.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1182.0, + 1406.0, + 1182.0, + 1406.0, + 1222.0, + 320.0, + 1222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1215.0, + 786.0, + 1215.0, + 786.0, + 1249.0, + 322.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 891.0, + 1405.0, + 891.0, + 1405.0, + 927.0, + 294.0, + 927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 922.0, + 1013.0, + 922.0, + 1013.0, + 956.0, + 321.0, + 956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 599.0, + 1403.0, + 599.0, + 1403.0, + 635.0, + 295.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 628.0, + 1406.0, + 628.0, + 1406.0, + 669.0, + 320.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 661.0, + 787.0, + 661.0, + 787.0, + 694.0, + 321.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 714.0, + 1405.0, + 714.0, + 1405.0, + 756.0, + 293.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 748.0, + 1405.0, + 748.0, + 1405.0, + 785.0, + 322.0, + 785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 777.0, + 975.0, + 777.0, + 975.0, + 815.0, + 321.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1794.0, + 1405.0, + 1794.0, + 1405.0, + 1833.0, + 295.0, + 1833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1827.0, + 1403.0, + 1827.0, + 1403.0, + 1861.0, + 324.0, + 1861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1857.0, + 611.0, + 1857.0, + 611.0, + 1890.0, + 322.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 831.0, + 1409.0, + 831.0, + 1409.0, + 870.0, + 292.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 244.0, + 1407.0, + 244.0, + 1407.0, + 285.0, + 293.0, + 285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 278.0, + 1404.0, + 278.0, + 1404.0, + 315.0, + 321.0, + 315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 310.0, + 610.0, + 310.0, + 610.0, + 339.0, + 322.0, + 339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1911.0, + 1406.0, + 1911.0, + 1406.0, + 1950.0, + 293.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1945.0, + 1402.0, + 1945.0, + 1402.0, + 1979.0, + 324.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1976.0, + 720.0, + 1976.0, + 720.0, + 2010.0, + 324.0, + 2010.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 985, + 1402, + 985, + 1402, + 1050, + 296, + 1050 + ], + "score": 0.8 + }, + { + "category_id": 1, + "poly": [ + 295, + 202, + 1401, + 202, + 1401, + 295, + 295, + 295 + ], + "score": 0.794 + }, + { + "category_id": 1, + "poly": [ + 297, + 1182, + 1401, + 1182, + 1401, + 1247, + 297, + 1247 + ], + "score": 0.781 + }, + { + "category_id": 1, + "poly": [ + 299, + 1377, + 1401, + 1377, + 1401, + 1472, + 299, + 1472 + ], + "score": 0.764 + }, + { + "category_id": 1, + "poly": [ + 295, + 902, + 1404, + 902, + 1404, + 967, + 295, + 967 + ], + "score": 0.75 + }, + { + "category_id": 1, + "poly": [ + 290, + 1489, + 1402, + 1489, + 1402, + 1557, + 290, + 1557 + ], + "score": 0.731 + }, + { + "category_id": 1, + "poly": [ + 297, + 1068, + 1406, + 1068, + 1406, + 1164, + 297, + 1164 + ], + "score": 0.725 + }, + { + "category_id": 1, + "poly": [ + 298, + 314, + 1401, + 314, + 1401, + 409, + 298, + 409 + ], + "score": 0.721 + }, + { + "category_id": 1, + "poly": [ + 296, + 1264, + 1408, + 1264, + 1408, + 1357, + 296, + 1357 + ], + "score": 0.714 + }, + { + "category_id": 1, + "poly": [ + 299, + 1573, + 1404, + 1573, + 1404, + 1668, + 299, + 1668 + ], + "score": 0.703 + }, + { + "category_id": 1, + "poly": [ + 296, + 706, + 1399, + 706, + 1399, + 772, + 296, + 772 + ], + "score": 0.696 + }, + { + "category_id": 1, + "poly": [ + 297, + 789, + 1407, + 789, + 1407, + 883, + 297, + 883 + ], + "score": 0.692 + }, + { + "category_id": 1, + "poly": [ + 297, + 510, + 1400, + 510, + 1400, + 576, + 297, + 576 + ], + "score": 0.663 + }, + { + "category_id": 1, + "poly": [ + 300, + 1686, + 1404, + 1686, + 1404, + 1783, + 300, + 1783 + ], + "score": 0.63 + }, + { + "category_id": 1, + "poly": [ + 294, + 428, + 1402, + 428, + 1402, + 492, + 294, + 492 + ], + "score": 0.629 + }, + { + "category_id": 1, + "poly": [ + 298, + 1799, + 1401, + 1799, + 1401, + 1896, + 298, + 1896 + ], + "score": 0.599 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2086, + 836, + 2086 + ], + "score": 0.568 + }, + { + "category_id": 1, + "poly": [ + 295, + 594, + 1403, + 594, + 1403, + 689, + 295, + 689 + ], + "score": 0.556 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2086, + 836, + 2086 + ], + "score": 0.503 + }, + { + "category_id": 1, + "poly": [ + 298, + 1913, + 1406, + 1913, + 1406, + 2008, + 298, + 2008 + ], + "score": 0.455 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 984.0, + 1408.0, + 984.0, + 1408.0, + 1024.0, + 295.0, + 1024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1016.0, + 506.0, + 1016.0, + 506.0, + 1050.0, + 323.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 199.0, + 1405.0, + 199.0, + 1405.0, + 240.0, + 294.0, + 240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 232.0, + 1405.0, + 232.0, + 1405.0, + 269.0, + 322.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 264.0, + 531.0, + 264.0, + 531.0, + 294.0, + 324.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1182.0, + 1406.0, + 1182.0, + 1406.0, + 1218.0, + 295.0, + 1218.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1212.0, + 948.0, + 1212.0, + 948.0, + 1248.0, + 322.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1376.0, + 1405.0, + 1376.0, + 1405.0, + 1416.0, + 294.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1410.0, + 1405.0, + 1410.0, + 1405.0, + 1444.0, + 324.0, + 1444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1438.0, + 742.0, + 1438.0, + 742.0, + 1476.0, + 323.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 898.0, + 1409.0, + 898.0, + 1409.0, + 942.0, + 292.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 935.0, + 602.0, + 935.0, + 602.0, + 967.0, + 320.0, + 967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1491.0, + 1405.0, + 1491.0, + 1405.0, + 1528.0, + 296.0, + 1528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1523.0, + 1290.0, + 1523.0, + 1290.0, + 1557.0, + 322.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1069.0, + 1404.0, + 1069.0, + 1404.0, + 1104.0, + 295.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1099.0, + 1404.0, + 1099.0, + 1404.0, + 1133.0, + 324.0, + 1133.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1128.0, + 1146.0, + 1128.0, + 1146.0, + 1165.0, + 324.0, + 1165.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 313.0, + 1406.0, + 313.0, + 1406.0, + 353.0, + 293.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 343.0, + 1408.0, + 343.0, + 1408.0, + 383.0, + 321.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 379.0, + 602.0, + 379.0, + 602.0, + 408.0, + 320.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1262.0, + 1405.0, + 1262.0, + 1405.0, + 1302.0, + 293.0, + 1302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1295.0, + 1408.0, + 1295.0, + 1408.0, + 1331.0, + 322.0, + 1331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1321.0, + 399.0, + 1321.0, + 399.0, + 1362.0, + 320.0, + 1362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1573.0, + 1404.0, + 1573.0, + 1404.0, + 1610.0, + 294.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1606.0, + 1408.0, + 1606.0, + 1408.0, + 1640.0, + 322.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1633.0, + 397.0, + 1633.0, + 397.0, + 1669.0, + 318.0, + 1669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 707.0, + 1404.0, + 707.0, + 1404.0, + 743.0, + 294.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 736.0, + 1195.0, + 736.0, + 1195.0, + 771.0, + 321.0, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 791.0, + 1404.0, + 791.0, + 1404.0, + 825.0, + 296.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 823.0, + 1408.0, + 823.0, + 1408.0, + 857.0, + 321.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 849.0, + 397.0, + 849.0, + 397.0, + 883.0, + 321.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 507.0, + 1406.0, + 507.0, + 1406.0, + 551.0, + 293.0, + 551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 542.0, + 769.0, + 542.0, + 769.0, + 575.0, + 323.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1684.0, + 1408.0, + 1684.0, + 1408.0, + 1725.0, + 295.0, + 1725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1717.0, + 1405.0, + 1717.0, + 1405.0, + 1754.0, + 322.0, + 1754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1749.0, + 1120.0, + 1749.0, + 1120.0, + 1784.0, + 325.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 428.0, + 1404.0, + 428.0, + 1404.0, + 464.0, + 296.0, + 464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 460.0, + 1276.0, + 460.0, + 1276.0, + 493.0, + 322.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1801.0, + 1403.0, + 1801.0, + 1403.0, + 1835.0, + 296.0, + 1835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1833.0, + 1405.0, + 1833.0, + 1405.0, + 1867.0, + 323.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1861.0, + 997.0, + 1861.0, + 997.0, + 1899.0, + 319.0, + 1899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 592.0, + 1404.0, + 592.0, + 1404.0, + 632.0, + 294.0, + 632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 626.0, + 1404.0, + 626.0, + 1404.0, + 660.0, + 323.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 656.0, + 953.0, + 656.0, + 953.0, + 690.0, + 323.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1912.0, + 1407.0, + 1912.0, + 1407.0, + 1951.0, + 295.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1945.0, + 1404.0, + 1945.0, + 1404.0, + 1982.0, + 322.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1975.0, + 885.0, + 1975.0, + 885.0, + 2009.0, + 322.0, + 2009.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 299, + 202, + 1404, + 202, + 1404, + 295, + 299, + 295 + ], + "score": 0.833 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2086, + 836, + 2086 + ], + "score": 0.802 + }, + { + "category_id": 1, + "poly": [ + 292, + 783, + 1404, + 783, + 1404, + 848, + 292, + 848 + ], + "score": 0.794 + }, + { + "category_id": 1, + "poly": [ + 289, + 1655, + 1405, + 1655, + 1405, + 1719, + 289, + 1719 + ], + "score": 0.789 + }, + { + "category_id": 1, + "poly": [ + 297, + 1102, + 1406, + 1102, + 1406, + 1197, + 297, + 1197 + ], + "score": 0.786 + }, + { + "category_id": 1, + "poly": [ + 298, + 1335, + 1406, + 1335, + 1406, + 1430, + 298, + 1430 + ], + "score": 0.776 + }, + { + "category_id": 1, + "poly": [ + 296, + 318, + 1402, + 318, + 1402, + 382, + 296, + 382 + ], + "score": 0.773 + }, + { + "category_id": 1, + "poly": [ + 298, + 520, + 1407, + 520, + 1407, + 644, + 298, + 644 + ], + "score": 0.769 + }, + { + "category_id": 1, + "poly": [ + 297, + 1452, + 1406, + 1452, + 1406, + 1547, + 297, + 1547 + ], + "score": 0.765 + }, + { + "category_id": 1, + "poly": [ + 296, + 1568, + 1403, + 1568, + 1403, + 1634, + 296, + 1634 + ], + "score": 0.764 + }, + { + "category_id": 1, + "poly": [ + 299, + 868, + 1406, + 868, + 1406, + 964, + 299, + 964 + ], + "score": 0.761 + }, + { + "category_id": 1, + "poly": [ + 298, + 666, + 1401, + 666, + 1401, + 761, + 298, + 761 + ], + "score": 0.742 + }, + { + "category_id": 1, + "poly": [ + 299, + 1218, + 1404, + 1218, + 1404, + 1315, + 299, + 1315 + ], + "score": 0.737 + }, + { + "category_id": 1, + "poly": [ + 299, + 986, + 1405, + 986, + 1405, + 1081, + 299, + 1081 + ], + "score": 0.725 + }, + { + "category_id": 1, + "poly": [ + 291, + 1740, + 1406, + 1740, + 1406, + 1807, + 291, + 1807 + ], + "score": 0.721 + }, + { + "category_id": 1, + "poly": [ + 297, + 403, + 1406, + 403, + 1406, + 499, + 297, + 499 + ], + "score": 0.72 + }, + { + "category_id": 1, + "poly": [ + 299, + 1826, + 1402, + 1826, + 1402, + 1922, + 299, + 1922 + ], + "score": 0.624 + }, + { + "category_id": 1, + "poly": [ + 299, + 1944, + 1400, + 1944, + 1400, + 2008, + 299, + 2008 + ], + "score": 0.615 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2096.0, + 832.0, + 2096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 203.0, + 1403.0, + 203.0, + 1403.0, + 236.0, + 297.0, + 236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 234.0, + 1408.0, + 234.0, + 1408.0, + 268.0, + 322.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 262.0, + 397.0, + 262.0, + 397.0, + 296.0, + 322.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 782.0, + 1407.0, + 782.0, + 1407.0, + 822.0, + 293.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 812.0, + 399.0, + 812.0, + 399.0, + 850.0, + 319.0, + 850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1655.0, + 1408.0, + 1655.0, + 1408.0, + 1691.0, + 295.0, + 1691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1688.0, + 818.0, + 1688.0, + 818.0, + 1718.0, + 324.0, + 1718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1102.0, + 1408.0, + 1102.0, + 1408.0, + 1139.0, + 294.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1135.0, + 1408.0, + 1135.0, + 1408.0, + 1169.0, + 323.0, + 1169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1165.0, + 485.0, + 1165.0, + 485.0, + 1197.0, + 323.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1335.0, + 1404.0, + 1335.0, + 1404.0, + 1372.0, + 295.0, + 1372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1367.0, + 1407.0, + 1367.0, + 1407.0, + 1401.0, + 324.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1394.0, + 396.0, + 1394.0, + 396.0, + 1432.0, + 321.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 313.0, + 1408.0, + 313.0, + 1408.0, + 358.0, + 293.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 346.0, + 399.0, + 346.0, + 399.0, + 385.0, + 319.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 521.0, + 1407.0, + 521.0, + 1407.0, + 556.0, + 293.0, + 556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 553.0, + 1407.0, + 553.0, + 1407.0, + 587.0, + 321.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 582.0, + 1408.0, + 582.0, + 1408.0, + 618.0, + 320.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 610.0, + 398.0, + 610.0, + 398.0, + 645.0, + 319.0, + 645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1447.0, + 1406.0, + 1447.0, + 1406.0, + 1493.0, + 292.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1485.0, + 1404.0, + 1485.0, + 1404.0, + 1519.0, + 323.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1514.0, + 780.0, + 1514.0, + 780.0, + 1548.0, + 323.0, + 1548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1571.0, + 1401.0, + 1571.0, + 1401.0, + 1604.0, + 298.0, + 1604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1600.0, + 1346.0, + 1600.0, + 1346.0, + 1636.0, + 323.0, + 1636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 871.0, + 1403.0, + 871.0, + 1403.0, + 905.0, + 296.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 898.0, + 1406.0, + 898.0, + 1406.0, + 939.0, + 321.0, + 939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 932.0, + 504.0, + 932.0, + 504.0, + 964.0, + 323.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 666.0, + 1406.0, + 666.0, + 1406.0, + 705.0, + 293.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 696.0, + 1406.0, + 696.0, + 1406.0, + 738.0, + 321.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 731.0, + 787.0, + 731.0, + 787.0, + 762.0, + 321.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1218.0, + 1406.0, + 1218.0, + 1406.0, + 1257.0, + 294.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1248.0, + 1405.0, + 1248.0, + 1405.0, + 1290.0, + 321.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1280.0, + 1139.0, + 1280.0, + 1139.0, + 1318.0, + 322.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 988.0, + 1403.0, + 988.0, + 1403.0, + 1023.0, + 297.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1018.0, + 1404.0, + 1018.0, + 1404.0, + 1054.0, + 322.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1047.0, + 1108.0, + 1047.0, + 1108.0, + 1083.0, + 319.0, + 1083.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1737.0, + 1409.0, + 1737.0, + 1409.0, + 1781.0, + 292.0, + 1781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1771.0, + 1242.0, + 1771.0, + 1242.0, + 1807.0, + 322.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 398.0, + 1406.0, + 398.0, + 1406.0, + 445.0, + 291.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 436.0, + 1404.0, + 436.0, + 1404.0, + 471.0, + 324.0, + 471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 467.0, + 780.0, + 467.0, + 780.0, + 501.0, + 324.0, + 501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1825.0, + 1407.0, + 1825.0, + 1407.0, + 1867.0, + 293.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1861.0, + 1407.0, + 1861.0, + 1407.0, + 1895.0, + 324.0, + 1895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1890.0, + 1234.0, + 1890.0, + 1234.0, + 1924.0, + 324.0, + 1924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1943.0, + 1402.0, + 1943.0, + 1402.0, + 1979.0, + 295.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1976.0, + 882.0, + 1976.0, + 882.0, + 2008.0, + 325.0, + 2008.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2085, + 836, + 2085 + ], + "score": 0.831 + }, + { + "category_id": 1, + "poly": [ + 295, + 764, + 1403, + 764, + 1403, + 829, + 295, + 829 + ], + "score": 0.725 + }, + { + "category_id": 1, + "poly": [ + 298, + 202, + 1401, + 202, + 1401, + 296, + 298, + 296 + ], + "score": 0.719 + }, + { + "category_id": 1, + "poly": [ + 298, + 1185, + 1401, + 1185, + 1401, + 1251, + 298, + 1251 + ], + "score": 0.674 + }, + { + "category_id": 1, + "poly": [ + 296, + 1267, + 1402, + 1267, + 1402, + 1332, + 296, + 1332 + ], + "score": 0.669 + }, + { + "category_id": 1, + "poly": [ + 296, + 1072, + 1403, + 1072, + 1403, + 1166, + 296, + 1166 + ], + "score": 0.66 + }, + { + "category_id": 1, + "poly": [ + 299, + 1462, + 1402, + 1462, + 1402, + 1557, + 299, + 1557 + ], + "score": 0.644 + }, + { + "category_id": 1, + "poly": [ + 298, + 1350, + 1402, + 1350, + 1402, + 1445, + 298, + 1445 + ], + "score": 0.639 + }, + { + "category_id": 1, + "poly": [ + 297, + 539, + 1405, + 539, + 1405, + 632, + 297, + 632 + ], + "score": 0.636 + }, + { + "category_id": 1, + "poly": [ + 298, + 1688, + 1402, + 1688, + 1402, + 1783, + 298, + 1783 + ], + "score": 0.632 + }, + { + "category_id": 1, + "poly": [ + 298, + 456, + 1398, + 456, + 1398, + 523, + 298, + 523 + ], + "score": 0.632 + }, + { + "category_id": 1, + "poly": [ + 298, + 1800, + 1401, + 1800, + 1401, + 1895, + 298, + 1895 + ], + "score": 0.63 + }, + { + "category_id": 1, + "poly": [ + 301, + 1574, + 1404, + 1574, + 1404, + 1671, + 301, + 1671 + ], + "score": 0.626 + }, + { + "category_id": 1, + "poly": [ + 297, + 313, + 1405, + 313, + 1405, + 438, + 297, + 438 + ], + "score": 0.624 + }, + { + "category_id": 1, + "poly": [ + 297, + 846, + 1405, + 846, + 1405, + 941, + 297, + 941 + ], + "score": 0.605 + }, + { + "category_id": 1, + "poly": [ + 296, + 651, + 1407, + 651, + 1407, + 747, + 296, + 747 + ], + "score": 0.597 + }, + { + "category_id": 1, + "poly": [ + 298, + 960, + 1406, + 960, + 1406, + 1055, + 298, + 1055 + ], + "score": 0.534 + }, + { + "category_id": 1, + "poly": [ + 299, + 1913, + 1406, + 1913, + 1406, + 2007, + 299, + 2007 + ], + "score": 0.46 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2057.0, + 871.0, + 2057.0, + 871.0, + 2095.0, + 832.0, + 2095.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 762.0, + 1404.0, + 762.0, + 1404.0, + 802.0, + 293.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 795.0, + 610.0, + 795.0, + 610.0, + 826.0, + 322.0, + 826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 202.0, + 1406.0, + 202.0, + 1406.0, + 239.0, + 294.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 234.0, + 1405.0, + 234.0, + 1405.0, + 268.0, + 324.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 262.0, + 1401.0, + 262.0, + 1401.0, + 300.0, + 321.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1183.0, + 1405.0, + 1183.0, + 1405.0, + 1222.0, + 296.0, + 1222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1218.0, + 926.0, + 1218.0, + 926.0, + 1249.0, + 323.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1269.0, + 1404.0, + 1269.0, + 1404.0, + 1305.0, + 298.0, + 1305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1298.0, + 949.0, + 1298.0, + 949.0, + 1333.0, + 323.0, + 1333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1074.0, + 1404.0, + 1074.0, + 1404.0, + 1108.0, + 298.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1102.0, + 1405.0, + 1102.0, + 1405.0, + 1143.0, + 321.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1135.0, + 748.0, + 1135.0, + 748.0, + 1169.0, + 323.0, + 1169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1462.0, + 1406.0, + 1462.0, + 1406.0, + 1500.0, + 295.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1491.0, + 1406.0, + 1491.0, + 1406.0, + 1533.0, + 319.0, + 1533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1524.0, + 853.0, + 1524.0, + 853.0, + 1558.0, + 322.0, + 1558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1350.0, + 1407.0, + 1350.0, + 1407.0, + 1388.0, + 294.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1382.0, + 1404.0, + 1382.0, + 1404.0, + 1417.0, + 321.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1413.0, + 795.0, + 1413.0, + 795.0, + 1445.0, + 321.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 537.0, + 1408.0, + 537.0, + 1408.0, + 578.0, + 294.0, + 578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 570.0, + 1410.0, + 570.0, + 1410.0, + 607.0, + 320.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 596.0, + 397.0, + 596.0, + 397.0, + 637.0, + 320.0, + 637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1688.0, + 1406.0, + 1688.0, + 1406.0, + 1726.0, + 296.0, + 1726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1720.0, + 1404.0, + 1720.0, + 1404.0, + 1754.0, + 321.0, + 1754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1749.0, + 1028.0, + 1749.0, + 1028.0, + 1787.0, + 322.0, + 1787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 457.0, + 1404.0, + 457.0, + 1404.0, + 493.0, + 295.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 488.0, + 1362.0, + 488.0, + 1362.0, + 523.0, + 322.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1800.0, + 1406.0, + 1800.0, + 1406.0, + 1835.0, + 294.0, + 1835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1832.0, + 1406.0, + 1832.0, + 1406.0, + 1867.0, + 322.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1861.0, + 888.0, + 1861.0, + 888.0, + 1899.0, + 322.0, + 1899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1574.0, + 1406.0, + 1574.0, + 1406.0, + 1613.0, + 295.0, + 1613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1608.0, + 1404.0, + 1608.0, + 1404.0, + 1642.0, + 322.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1636.0, + 1093.0, + 1636.0, + 1093.0, + 1674.0, + 322.0, + 1674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 316.0, + 1405.0, + 316.0, + 1405.0, + 349.0, + 295.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 347.0, + 1407.0, + 347.0, + 1407.0, + 380.0, + 324.0, + 380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 377.0, + 1403.0, + 377.0, + 1403.0, + 413.0, + 323.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 406.0, + 686.0, + 406.0, + 686.0, + 442.0, + 321.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 845.0, + 1408.0, + 845.0, + 1408.0, + 885.0, + 294.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 875.0, + 1405.0, + 875.0, + 1405.0, + 918.0, + 321.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 909.0, + 670.0, + 909.0, + 670.0, + 943.0, + 323.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 651.0, + 1405.0, + 651.0, + 1405.0, + 689.0, + 294.0, + 689.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 684.0, + 1405.0, + 684.0, + 1405.0, + 720.0, + 322.0, + 720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 715.0, + 1081.0, + 715.0, + 1081.0, + 749.0, + 323.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 960.0, + 1407.0, + 960.0, + 1407.0, + 998.0, + 297.0, + 998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 990.0, + 1407.0, + 990.0, + 1407.0, + 1028.0, + 321.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1021.0, + 851.0, + 1021.0, + 851.0, + 1059.0, + 322.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1913.0, + 1407.0, + 1913.0, + 1407.0, + 1950.0, + 294.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1947.0, + 1407.0, + 1947.0, + 1407.0, + 1981.0, + 323.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1977.0, + 603.0, + 1977.0, + 603.0, + 2006.0, + 320.0, + 2006.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 13, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/train/NiM9Q7Z95z/NiM9Q7Z95z.md b/parse/train/NiM9Q7Z95z/NiM9Q7Z95z.md new file mode 100644 index 0000000000000000000000000000000000000000..a6f66705f415381c4cea370cb568dc5d85b05fb0 --- /dev/null +++ b/parse/train/NiM9Q7Z95z/NiM9Q7Z95z.md @@ -0,0 +1,479 @@ +# Secure Quantized Training for Deep Learning + +Anonymous Author(s) +Affiliation +Address +email + +# Abstract + +1 We have implemented training of neural networks in secure multi-party computa +2 tion (MPC) using quantization commonly used in said setting. To the best of our +3 knowledge, we are the first to present training of MNIST purely implemented in +4 MPC that comes within one percent of accuracy of training using plaintext compu +5 tation. We found that training with MPC is possible, but it takes more epochs and +6 achieves a lower accuracy than the usual CPU/GPU computation. More concretely, +7 we have trained a network with two convolution and two dense layers to $9 8 . 5 \%$ +8 accuracy in 150 epochs. This took a day in our MPC implementation. + +# 9 1 Introduction + +10 Secure multi-party computation (MPC) is a cryptographic technique that allows a set of parties to +11 compute a public output on private inputs without revealing the inputs or any intermediate results. +12 This makes it a potential solution to federated learning where the sample data stays private and only +13 the model or even only inference results are revealed. +14 Imagine a set of healthcare providers holding sensitive patient data. MPC allows them to collabo +15 ratively train a model. This model could then either be released or even kept private for inference +16 using MPC again. See Figure for an illustration. A more conceptual example is the well-known +17 millionaires’ problem where two people want to find out who is richer without revealing their wealth. +18 There is clearly a difference between the one bit of information desired and the full figures. +19 There has been a sustained interest in applying secure computation to machine learning and neural +20 networks going back to at least Barni et al. [2006]. More recent advantages in practical MPC have +21 led to an increased effort in implementing both inference and training. +22 A number of works such as Mohassel and Zhang [2017], Mohassel and Rindal [2018], Wagh et al. +23 [2019], Wagh et al. [2021] implement neural network training with MPC at least in parts. However, +24 they either give accuracy figures below $9 5 \%$ or figures that have been obtained using plaintext training. +25 For the latter case, the works do not clarify how close the computation for plaintext training matches +26 the lower precision and other differences in the MPC setting. Agrawal et al. [2019] claim a higher +27 accuracy in a comparable setting for a convolutional neural network with more channels than we +28 use. However, they have only implemented dense layers, and we achieve comparable accuracy to +29 them with only dense layers. All works use quantization in the sense that a fractional number $x$ is +30 represented as $\lfloor x \cdot 2 ^ { - f } \rceil$ . This makes addition considerably faster in the secure computation setting +31 because it reduces to integer addition. Furthermore, some of the works suggest to replace the softmax +32 function that uses exponentiation with a ReLU-based replacement. Keller and Sun [2020] have found +33 that this softmax replacement deteriorates accuracy in dense neural networks to the extent that it does +34 not justify the performance gains. +35 The concurrent work of Tan et al. [2021] gives some figures on the learning curve when run using +36 secure computation. However, they stop at five epochs for MNIST training where they achieve $94 \%$ +37 accuracy whereas we present the figures up to 150 epochs and $9 8 . 5 \%$ accuracy. Furthermore, their +38 choice of fixed-point precision 20 is considerably below 32, which we found to be optimal. We also +39 found that our implementation is $40 \%$ faster than theirs. Note that we use the CPU of one AWS +40 c5.9xlarge instance per party whereas Tan et al. use one NVIDIA Tesla V100 GPU per party. We +41 believe this somewhat counter-intuitive result comes from MPC heavily relying on communication, +42 which is an aspect where GPUs do not have an advantage over CPUs. +43 In this paper, we present an extensible framework for implementing deep learning training based on +44 MP-SPDZ by Keller [2020], a framework for multi-party computation.1 Similar to TensorFlow and +45 PyTorch, our approach allows representing deep learning models as succession of layers. We then +46 use this implementation to obtain accuracy figures for MNIST training by utilizing the MP-SPDZ +47 emulator, which allows to run the plaintext equivalent of secure computation, that is, the same +48 algorithms with the same precision. Finally, we run one of the most promising instantiation in real +49 secure computation in order to benchmark it confirm the result from the plaintext emulator. +50 There are a number of projects that integrate secure computation directly into popular machine +51 learning frameworks such as CrypTen by Gunning et al. [2019], PySyft by Ryffel et al. [2018], and +52 TF Encrypted by Dahl et al. [2018]. Our approach differs from all of them by running the protocol as +53 native CPU code (implemented using $\mathrm { C } { + + }$ ). This allows for much faster execution. For example, +54 CrypTen provides an MNIST training example (mpc_autograd_cnn) that takes over one minute +55 to run one epoch with 100 samples on one machine. In comparison, our implementation takes 11 +56 minutes to run one epoch with the full dataset of 60,000 samples. +57 Another line of work (e.g., Quoc et al. [2021]) uses trusted execution environments that provide +58 computation outside the reach of the operating system. This is a different security model to multi-party +59 computation that works with distributing the information among several entities. +60 The paper is structured as follows: After introducing the basics of the protocol we use in Section 2, +61 we will explain the mathematical building blocks in Section 3 and their use in the context of deep +62 learning in Section 4. Finally, we will present our implementation in Section 5 and our experimental +63 results for MNIST in Section 6. + +![](images/bb0f136189595323f73d65ddc0b25170a7b7ffb1f9107b9e37395cb5ee7bab1c.jpg) +Figure 1: Outsourced computation: Data holders (on the left) secret-share their data to a number of computing parties (on the right), who then return the desired the result (e.g., a model or inference results on further queries). All communication except outputs are secret-shared and thus secure if no two computing parties collude. + +# 64 2 An Efficient Secure Multi-Party Computation Protocol + +There is a wide range of MPC protocols with a variety of security properties (see Keller [2020] for an overview). In this paper we focus on the setting of three-party computation with one semi-honest corruption. This means that out of the three parties two are expected to behave honestly, i.e., they follow the protocol and keep their view of the protocol secret, and one party is expected to follow the protocol but might try extract information from their view. The reason for choosing this setting is that it allows an efficient MPC protocol while still allowing secure outsourced computation. The concrete protocol we use goes back to Benaloh and Leichter [1990] with further aspects by Araki et al. [2016], Mohassel and Rindal [2018], and Eerikson et al. [2020]. We summarize the core protocol below. The mathematical building blocks in the next section mostly use the aspects below. + +74 Secret sharing All intermediate values in our protocol are stored using replicated secret sharing. A +75 secret value $x$ is a represented as a random sum $x = x _ { 0 } + x _ { 1 } + x _ { 2 }$ , and party $P _ { i }$ holds $( x _ { i - 1 } , x _ { i + 1 } )$ +76 where the indices are computed modulo three. Clearly, each party is missing one value to compute +77 the sum. On the other hand, each pair of parties hold all necessary to reconstruct the secret. For a +78 uniformly random generation of shares, the computation domain has to be finite. Most commonly, +79 this domain is defined by integer computation modulo a number. We use $2 ^ { k }$ for $k$ being a multiple 64 +80 and 2 as the moduli. The first case corresponds to an extension of 64-bit arithmetic found on most +81 processors. We will refer to the two settings as arithmetic and binary secret sharing throughout the +82 paper. +83 Input sharing The secret sharing scheme implies a protocol to share inputs where the inputting +84 party samples the shares and distributes them accordingly. Eerikson et al. [2020] have proposed a +85 more efficient protocol where the inputting party only needs to send one value instead of two pairs of +86 values. If $P _ { i }$ would like to input $x$ , $x _ { i }$ is set to zero, and $x _ { i - 1 }$ is generated with a pseudo-random +87 generator using a key previously shared between $P _ { i }$ and $P _ { i + 1 }$ . $P _ { i }$ can compute $x _ { i + 1 } = x - x _ { i - 1 }$ and +88 send it to $P _ { i - 1 }$ . While the resulting secret sharing is not entirely random, the fact that $P _ { i }$ already +89 knows $x$ makes randomizing $x _ { i }$ obsolete. +90 Addition The commutative nature of addition allows to add secret sharings without communication. +91 More concretely, secret sharings $x = x _ { 0 } + x _ { 1 } + x _ { 2 }$ and $y = y _ { 0 } + y _ { 1 } + y _ { 2 }$ imply the secret sharing +92 $x + y = ( x _ { 0 } + y _ { 0 } ) + ( x _ { 1 } + y _ { 1 } ) + ( x _ { 2 } + y _ { 2 } )$ . + +93 Multiplication The product of $x = x _ { 0 } + x _ { 1 } + x _ { 2 }$ and $y = y _ { 0 } + y _ { 1 } + y _ { 2 }$ is + +$$ +{ \begin{array} { r l } & { x \cdot y = ( x _ { 0 } + x _ { 1 } + x _ { 2 } ) \cdot ( y _ { 0 } + y _ { 1 } + y _ { 2 } ) } \\ & { \qquad = ( x _ { 0 } y _ { 0 } + x _ { 0 } y _ { 1 } + x _ { 1 } y _ { 0 } ) + ( x _ { 1 } y _ { 1 } + x _ { 1 } y _ { 2 } + x _ { 1 } y _ { 1 } ) + ( x _ { 2 } y _ { 2 } + x _ { 2 } y _ { 0 } + x _ { 0 } y _ { 2 } ) . } \end{array} } +$$ + +94 Each of the brackets only contains shares known by one of the parties. They can thus compute an +95 additive secret sharing (one summand per party) of the product. However, every party only holding +96 one share does not satisfy the replication requirement for further multiplications. It is not secure +97 for every party to pass their value on to another party because the summands are not distributed +98 randomly. This can be fixed by rerandomization: Let $x y = z _ { 0 } + z _ { 1 } + z _ { 2 }$ where $z _ { i }$ is know to $P _ { i }$ . +99 Every party $P _ { i }$ computes $z _ { i } ^ { \prime } = z _ { i } + r _ { i , i + 1 } - r _ { i - 1 , i }$ where $r _ { i , i + 1 }$ is generated with a pseudo-random +100 generator using a key pre-shared between $P _ { i }$ and $P _ { i + 1 }$ . The resulting sum $x y = \hat { z } _ { 0 } ^ { \prime } + z _ { 1 } ^ { \prime } + z _ { 2 } ^ { \prime }$ is +101 pseudo-random, and it is thus secure for $P _ { i }$ to send $z _ { i } ^ { \prime }$ to $P _ { i + 1 }$ in order to create a replicated secret +102 sharing $\left( ( x y ) _ { i - 1 } , ( x y ) _ { i + 1 } \right) = \left( z _ { i } ^ { \prime } , z _ { i - 1 } ^ { \prime } \right)$ . + +# 103 3 Secure Computation Building Blocks + +104 In this section, we will discuss how to implement computation with MPC with a focus on how +105 it differs from computation on CPUs or GPUs. Most of the techniques below are already known +106 individually. To the best of our knowledge however, we are the first to put them together in an efficient +107 and extensible framework for secure computation of deep learning training. +108 Domain conversion Recall we that we use computation modulo $2 ^ { k }$ for $k$ being a multiple of 64 +109 as well as 1. Given that the main operations are just addition and multiplication in the respective +110 domain, it is desirable to compute integer arithmetic in the large domain but operations with a +111 straight-forward binary circuit modulo two. There has been a long-running interest in this going +112 back to least Kolesnikov et al. [2013]. We mainly rely on the approach proposed by Mohassel and +113 114 $\{ x _ { 0 } ^ { ( i ) } \} _ { j = 0 } ^ { k - 1 }$ 018] and Araki et al. [20the bit decomposition of $x _ { 0 }$ . Recal, that is, $x _ { 0 } ^ { ( i ) } \in \{ 0 , 1 \}$ $x \in 2 ^ { k }$ is shand $\begin{array} { r } { x _ { 0 } = \sum _ { i = 0 } ^ { k - 1 } x _ { 0 } ^ { ( i ) } 2 ^ { i } } \end{array}$ $x = x _ { 0 } + x _ { 1 } + x _ { 2 }$ . Now letlf-evident +115 that $x _ { 0 } ^ { ( i ) } = x _ { 0 } ^ { ( i ) } + 0 + 0$ is a valid secret sharing modulo two (albeit not a random one). Furthermore, +116 every party holding $x _ { 0 }$ can generate $x _ { 0 } ^ { ( i ) }$ . It is therefore possible for the parties to generate a secret +117 sharing modulo two of a single share modulo $2 ^ { k }$ . Repeating this for all shares and the computing +118 the addition as a binary circuit allows the parties to generate a secret sharing modulo two from a +119 secret sharing modulo $\bar { 2 } ^ { k }$ . Conversion in the other direction can be achieved using a similar technique +120 or using daBits as described by Rotaru and Wood [2019]. In the following we will use the term +121 mixed-circuit computation for any technique that works over both computation domains. +122 Quantization While Aliasgari et al. [2013] showed that it is possible to implement floating-point +123 computation, the cost is far higher than integer computation. It is therefore common to represent +124 fractional numbers using quantization (also called fixed-point representation) as suggested by Catrina +125 and Saxena [2010]. A real number $x$ is represented as $\bar { \boldsymbol { x } } = \lfloor \boldsymbol { x } \cdot \mathbf { \bar { 2 } } ^ { f } \rceil$ where $f$ is an integer specifying +126 the precision. The linearity of the representation allows to compute addition by simply adding the +127 representing integers. Multiplication however requires to adjust the result because it will have twice +128 the precision: $( x \cdot 2 ^ { f } ) \cdot ( y \cdot \dot { 2 } ^ { f } ) = x y \cdot 2 ^ { 2 f }$ . There are two ways to rectify this: + +• An obvious correction would be to shift the result by $f$ bits after adding $2 ^ { f - 1 }$ to the integer representation. This ensures rounding to the nearest number possible in the representation, with the tie being broken by rounding up. Dalskov et al. [2021] presented an efficient implementation of the truncation using mixed-circuit computation. + +• However, Catrina and Saxena have found that in the context of secure computation it is more efficient to use probabilistic truncation. This method rounds up or down probabilistically depending on the input. For example, probabilistically rounding 0.75 to an integer would see it rounded of up with probability 0.75 and down with probability 0.25. The probabilistic truncation is an effect of the fact that the operation involves the truncation of a randomized value, that is the computation of $\left\lfloor ( x + r ) / 2 ^ { m } \right\rfloor$ for a random $m$ -bit value $r$ . It is easy to see that + +$$ +\lfloor ( x + r ) / 2 ^ { m } \rfloor = \left\{ \begin{array} { l l } { \lfloor x / 2 ^ { m } \rfloor } & { ( x \bmod 2 ^ { m } + r ) < 2 ^ { m } } \\ { \lfloor x / 2 ^ { m } \rfloor + 1 } & { ( x \bmod 2 ^ { m } + r ) \ge 2 ^ { m } . } \end{array} \right. +$$ + +Therefore, the larger $( x \bmod 2 ^ { m } )$ ) is, the more likely the latter condition is true. Dalskov et al. [2020] present an efficient protocol in our security model. + +142 Our quantization scheme is related to quantized neural networks (see e.g. Hubara et al. [2016]). +143 However, our consideration is not to compress the model, but to improve the computational speed +144 and save communication cost. +145 Dot products Dot products are an essential building block of linear computation such as matrix +146 multiplication. In the light of quantization, it is possible to reduce the usage of truncation by deferring +147 after the summation. In other words, the dot product in the integer representations is computed before +148 truncating. This not only reduces the truncation error, it is also more efficient because the truncation +149 is the most expensive part in quantized secure multiplication. Similarly, our protocol allows to defer +150 the communication needed for multiplication. Let $\vec { x }$ and $\vec { y }$ be two vectors where the elements are +151 secret shared, that is, $\{ x ^ { ( i ) } \} = x _ { 0 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } + x _ { 2 } ^ { ( i ) }$ and similarly for $\boldsymbol y ^ { ( i ) }$ . The inner product then is + +$$ +\begin{array} { r l } & { \displaystyle \sum _ { i } x ^ { ( i ) } \cdot y ^ { ( i ) } = \sum _ { i } ( x _ { 0 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } + x _ { 2 } ^ { ( i ) } ) \cdot ( y _ { 0 } ^ { ( i ) } + y _ { 1 } ^ { ( i ) } + y _ { 2 } ^ { ( i ) } ) } \\ & { \quad \quad \quad \quad = \displaystyle \sum _ { i } ( x _ { 0 } ^ { ( i ) } y _ { 0 } ^ { ( i ) } + x _ { 0 } ^ { ( i ) } y _ { 1 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } y _ { 0 } ^ { ( i ) } ) + \sum _ { i } ( x _ { 1 } ^ { ( i ) } y _ { 1 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } y _ { 2 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } y _ { 1 } ^ { ( i ) } ) } \\ & { \quad \quad \quad \quad + \displaystyle \sum _ { i } ( x _ { 2 } ^ { ( i ) } y _ { 2 } ^ { ( i ) } + x _ { 2 } ^ { ( i ) } y _ { 0 } ^ { ( i ) } + x _ { 0 } ^ { ( i ) } y _ { 2 } ^ { ( i ) } ) . } \end{array} +$$ + +152 The three sums in the last term can be compute locally by one party each before applying the same +153 protocol as for a single multiplication. +154 Comparisons Arithmetic secret sharing does not allow to access the individual bits directly. It +155 is therefore not straightforward to compute comparisons such as “less than”. There is a long line +156 of literature on how to achieve this going back to at least Damgård et al. [2006]. More recently, +157 most attention has been given to combine the power of arithmetic and binary secret sharing in +158 order to combine the best of worlds. One possibility to do so is to plainly convert to the binary +159 domain and compute the comparison circuit there. In our concrete implementation we use the more +160 efficient approach by Mohassel and Rindal [2018]. It starts by taking the difference of the two +161 inputs. Computing the comparison then reduces to comparing to zero, which in turn is equivalent +162 to extracting the most significant bit as it indicates the sign. The latter is achieved by converting +163 the shares locally to bit-wise sharing of the arithmetic shares, which sum up to the secret value. It +164 remains to compute the sum of the binary shares in order to come up with the most significant bit. +165 Oblivious Selection Plain secure computation does not allow branching because the parties would +166 need to be aware which branch is followed. Conditional assignment can be implemented as follows +167 however. If $b \in \{ 0 , 1 \}$ denotes the condition, $x + b \cdot ( y - x )$ is either $x$ or $y$ depending on $b$ . If the +168 condition is available in binary secret sharing but $x$ and $y$ in arithmetic secret sharing, $b$ has to be +169 converted to the latter. This can be done using a daBit as introduced by Rotaru and Wood [2019], +170 which is a secret random bit shared both in arithmetic and binary. It allows to mask a bit in one world +171 by XORing it. The result is then revealed and the masking is undone in the other world. +172 Division Catrina and Saxena [2010] have shown how to implement quantized division using the +173 algorithm by Goldschmidt [1964]. It mainly uses arithmetic and the probabilistic truncation already +174 explained. In addition, the initial approximation requires a full bit decomposition as described above. +175 The error of the output depends on the error in the multiplications used for Goldschmidt’s iteration, +176 which compounds in particular when using probabilistic truncation. Due to the nature of secure +177 computation, the result of division by zero is undefined. One could obtain a secret failure bit by +178 testing the divisor to zero. However, we found that not to be necessary in our algorithm. This is +179 because we only use division by secret value only for the softmax function where the it is guaranteed +180 to strictly positive. +181 Logarithm Computation logarithm with any public base can be reduced to logarithm to base two +182 using $\log _ { x } y = \log _ { 2 } y \cdot \log _ { x } 2$ . Aly and Smart [2019] have proposed computing $y = a \cdot 2 ^ { b }$ where +183 $a \in [ 0 . 5 , 1 )$ and $b \in \mathbb { Z }$ . This then allows to compute $\log _ { 2 } y = \log _ { 2 } a + b$ . Given the restricted range +184 of $a$ , $\log _ { 2 } a$ can be approximated using a division of polynomials. Numerical stability and input range +185 control are less of an issue here because we only use logarithm for the loss computation, which does +186 not influence the training. +187 Exponentiation By using $x ^ { y } = 2 ^ { y \log _ { 2 } x }$ , any exponentiation can be reduced to exponentiation +188 with base two. Aly and Smart [2019] have shown how to compute $2 ^ { a } = 2 ^ { \lfloor a \rfloor } \cdot 2 ^ { a - \lfloor a \rfloor }$ by computing +189 the two exponents using bit decomposition and the second factor using a polynomial approximation. +190 Regarding the first factor, if $\begin{array} { r } { b = \sum _ { i \geq 0 } b _ { i } 2 ^ { i } } \end{array}$ is an integer with $b _ { i } \in \{ 0 , 1 \}$ , + +$$ +2 ^ { b } = 2 ^ { \sum _ { i \geq 0 } b _ { i } 2 ^ { i } } = \prod _ { i \geq 0 } 2 ^ { b _ { i } 2 ^ { i } } = \prod _ { i \geq 0 } ( 1 + b _ { i } \cdot ( 2 ^ { 2 ^ { i } } - 1 ) ) . +$$ + +191 As with division the numerical stability depends on the truncation used for multiplication. + +Inverse square root Aly and Smart [2019] have proposed to compute square root using Goldschmidt and Raphson-Newton iterations. We could combine this with the division above. However, Lu et al. [2020] have proposed a more direct computation that avoids running two successive iterations. + +Uniformly random fractional number Limiting ourselves to intervals of the form $[ x , x + 2 ^ { e } ]$ for a potentially negative integer $e$ , we can reduce the problem to generate a random $( f + { \dot { e } } )$ -bit number where $f$ is the fixed-point precision. Recall that we represent a fractional number $x$ as $\lfloor x \cdot 2 ^ { - f } \rceil$ Generating a random $n$ -bit number is straight-forward using random bits, which in our protocol can be generated as presented by Damgård et al. [2019]. In the context of our protocol however, Dalskov et al. [2021] have a presented a more efficient approach that involves mixed-circuit computation. + +Communication cost Table 1 show the total communication cost of some of the building blocks in our protocol for $f = 3 2$ . This setting mandates the modulus $2 ^ { 1 2 8 }$ because the division protocol requires bit length $4 f$ . + +# 4 Machine Learning Building Blocks + +In this section, we will use the building blocks in Section 3 to construct high-level computational modules for deep learning. + +Fully connected layers Both forward and back-propagation of fully connected layers can be seen as matrix multiplications and thus can be implemented using dot products. A particular challenge in secure computation is to compute a number of outputs in parallel in order to save communication + +Table 1: Communication cost of select computation for $f = 3 2$ and integer modulus $2 ^ { 1 2 8 }$ + +
Bits
Integer multiplication384
Probabilistic truncation Nearest truncation1,536 4,462
Comparison1,369
Division (prob. truncation)29,866
Division (nearest truncation)57,798
Exponentiation (prob. truncation)77,684
Exponentiation (nearest truncation)
171,638
Invert square root (prob. truncation) Invert square root (nearest truncation)20,073 27,699
+ +10 rounds. We solve this by having a dedicated infrastructure in our implementation that computes all dot +11 products for a matrix multiplication in a single batch, thus combining all necessary communication. + +2D convolution layers Similar to fully connected layers, 2D convolution and its corresponding gradient can be implemented using only dot products, and we again compute several output values in parallel. + +215 Rectified Linear Unit (ReLU) ReLU Nair and Hinton [2010] is defined as follows: + +$$ +{ \mathsf { R e L U } } ( x ) : = { \left\{ \begin{array} { l l } { x , } & { { \mathrm { i f ~ } } x > 0 } \\ { 0 . } & { { \mathrm { o t h e r w i s e } } } \end{array} \right. } +$$ + +216 It can thus be implemented as a comparison followed by an oblivious selection. For back-propagation, +217 it is advantageous to reuse the comparison results from forward propagation due to the relatively high +218 cost in secure computation. Note that the comparison results are stored in secret-shared form and +219 thus there is no reduction in security. +220 Max pooling Similar to ReLU, max pooling can be reduced to comparison and oblivious selection. +221 In secure computation, it saves communication rounds if the process uses a balanced tree rather than +222 iterating over all input values of one maximum computation. For back-propagation it again pays off +223 to the store intermediate results from forward propagation, again in secret-shared form. +224 Softmax and cross entropy loss This combination requires computing the following gradient for +225 back-propagation: + +$$ +\nabla i : = \frac { \partial \ell } { \partial x _ { i } } = \frac { \partial } { \partial x _ { i } } \Big ( - \sum _ { k } y _ { k } \cdot x _ { k } + \log \sum _ { j } e ^ { x _ { j } } \Big ) = - y _ { i } + \frac { e ^ { x _ { i } } } { \sum _ { j } e ^ { x _ { j } } } , +$$ + +226 where $y _ { i }$ denotes an element of the ground truth as a one-hot vector, and $x _ { i }$ denotes the output of the +227 last layer. +228 On the right hand side of eq. (1), the values in the denominator are potentially large due to the use +229 of the exponential. This is prone to numerical overflow in our quantized representation because the +230 latter puts relatively strict limits on the values. We therefore optimize the computation by computing +231 the maximum of the input values: + +$$ +m = \operatorname* { m a x } _ { j } ( \{ x _ { j } \} ) . +$$ + +232 Then we compute + +$$ +{ \frac { e ^ { x _ { i } - m } } { \sum _ { j } e ^ { x _ { j } - m } } } = { \frac { e ^ { x _ { i } } e ^ { - m } } { ( \sum _ { j } e ^ { x _ { j } } ) e ^ { - m } } } = { \frac { e ^ { x _ { i } } } { \sum _ { j } e ^ { x _ { j } } } } . +$$ + +233 All the exponents on the left-most term are at most zero, and thus the dividend is at most one and the +234 divisor is at most the number of labels (which is 10 in MNIST). The same technique can be used to +235 compute the sigmoid activation function, as sigmoid(x) = 11+exp(−x) $\begin{array} { r } { \mathrm { \begin{array} { r } { \mathrm { i d } ( x ) = \frac { 1 } { 1 + \exp ( - x ) } = \frac { \exp ( 0 ) } { \exp ( 0 ) + \exp ( x ) } } } \end{array} } \end{array}$ is a special +236 case of softmax. +237 Stochastic gradient descent The model parameter update in SGD only involves basic arithmetic: +238 $\begin{array} { r } { \theta _ { j } \theta _ { j } - \frac { \gamma } { B } \sum _ { i = 1 } ^ { B } \bigtriangledown _ { i j } } \end{array}$ where $\theta _ { j }$ is the parameter indexed by $j , B$ is the mini-batch size, $\gamma > 0$ is +239 the learning rate, and $\nabla i j$ is the gradient of the loss with respect to the ’th sample in the mini-batch +240 and the parameter $\theta _ { j }$ . In order to tackle the limited precision with quantization, we defer dividing by +241 the batch size to the model update. This means that we do not divide the gradient value by the batch +242 size when computing them as described in the previous paragraph. Instead, we divide the model +243 update by the batch size. Since we use a batch size that is a power of two (128), it is sufficient to use +244 probabilistic truncation instead of full-blown division. This saves both time and decreases the error. + +Adam The main difference to SGD in terms of basic computational operations is the additional use of an inverse square root. We again defer the division by the batch size to just before the model update. + +Parameter initialization We use the Glorot initialization by Glorot and Bengio [2010]. Besides basic operations, it mainly involves generating a uniformly random fractional value in a given interval. + +# 250 5 Implementation + +51 We built our implementation on MP-SPDZ by Keller [2020]. MP-SPDZ not only implements a +52 range of MPC protocols, it also comes with a high-level library containing the building blocks in +253 Section 3. MP-SPDZ already featured capabilities to train dense neural networks as well as inference +254 for convolutional neural networks. We have added backward propagation for a number of layer types, +255 including 2D convolution. Furthermore, we have corrected a bug in the backward propagation for +256 dense layers. + +MP-SPDZ allows implementing the computation in Python code, which is then compiled a specific bytecode. This code can be execute by a virtual machine executing the actual secure computation. The process allows to optimize the computation in the context of MPC. + +The framework also features an emulator that executes the exact computation that could be done securely in the clear. This allowed us to collect the accuracy figures in the next section at a lower cost. + +262 It is licensed under a BSD-style license, which allows to extend the code. + +# 63 6 MNIST + +For a concrete measurement of accuracy and running times, we have implemented training for the well-known MNIST dataset by LeCun et al. [2010]. We work mainly with the models that have been used by Wagh et al. [2019] with secure computation, and we will reuse their numbering (A–D). The models contain up to four linear layers. Network C is a convolutional neural network going back to the seminal work by LeCun et al. [1998] whereas the others are simpler networks that have been proposed by works on secure computation such as Mohassel and Zhang [2017], Liu et al. [2017], and Riazi et al. [2018]. We present the networks as Keras code in the supplemental material. + +Figure 2 shows the results for various quantization precisions and and the two rounding options. We have used SGD with learning rate 0.01, batch size 128, and the usual MNIST training/test split. $f = 6 4$ is the best option with probabilistic rounding, improving on both $f = 1 6$ and $f = 3 2$ Furthermore, nearest rounding performs worse that probabilistic for $f = 1 6$ and $f = 3 2$ . Due to the high cost, we only ran $f = 3 2$ with probabilistic rounding several times. The range is indicate by the shaded area. We focus on $f = 3 2$ because it offers the faster convergence. + +Figure 3 then shows the result with a variety of optimizers. While increasing the learning rate for SGD leads to a lower stability, Adam exposes a smoother learning learning curve albeit not a faster process. + +280 Finally, Figures 4 shows our results for all networks used by Wagh et al. [2019]. As one would expect, +281 the most sophisticated network performs best. Somewhat surprisingly, however, Network A (without +282 convolutional layers) performs better than the simpler networks containing convolutional layers. + +![](images/3a38cf90fbe641b15f4921c674ee686f614e2a42472767f16d74facf9f248ace.jpg) +Figure 2: Loss and accuracy for network C and precision options when running SGD with rate 0.01. + +![](images/ca865aaa95abbc18690e028b3ca52b226046ffa5e278339ad08eb557c5491c91.jpg) +Figure 3: Loss and accuracy for network C with various optimizer options, $f = 3 2$ , and probabilistic truncation. + +Resources We ran the emulator on AWS c5.9xlarge instances. One epoch takes a few second to several minutes depending on the model. Overall, we estimate that we have used a few weeks worth of computing time including experiments not included here because of bugs in the code. + +# 6.1 Secure computation + +In order to verify our emulation results, we have run Network C with precision $f \ = \ 3 2$ and probabilistic rounding in our actual multi-party computation protocol. We could verify that it converges on $9 8 . 5 \%$ accuracy at 150 epochs, taking 20 hours. Table 2 compares our result to previous works in a LAN setting. Note that Wagh et al. [2019] and Wagh et al. [2021] give accuracy figures. From personal communication with the authors and the fact that the source repository for the latter work2 says that their “code has not run end-to-end training”, we derive our assessment that their figures do not reflect the secure computation. + +![](images/700a37281e56f21023ee263453c7785f933f7176bea7c10046675acc845cf598.jpg) +Figure 4: Loss and accuracy for various networks, $f = 3 2$ , and probabilistic truncation. + +Table 2: Comparison to previous work in the LAN setting. $( ^ { * } )$ Mohassel and Zhang [2017] and Agrawal et al. [2019] use a different security model and are thus incomparable. We include them for completeness. Two numbers refer to online and offline time. Accuracy N/A means that the accuracy figures were not given or computed in a way that does not reflect the secure computation. + +
NetworkEpoch time (s)Acc. (# epochs)Precision(f)
AMohassel and Zhang [2017]283/19333*93.4% (15)13
Mohassel and Rindal [2018]18094.0% (15)N/A
Agrawal et al. [2019]31392*95.0% (10)N/A
Wagh et al. [2019]247N/A13
Wagh et al. [2021]41N/A13
Ours3197.9% (15)16
Ours5097.7% (15)32
BWagh et al. [2019]4176N/A13
Wagh et al. [2021]101N/A13
Ours14493.6% (15)16
Ours24994.7% (15)32
CWagh et al. [2019]7188N/A13
Wagh et al. [2021]891N/A13
Tan et al. [2021]103694.0% (5)20
Ours34494.9% (5)16
Ours64393.8% (5)32
DMohassel and Rindal [2018]234N/AN/A
Ours4196.8% (15)16
Ours6896.8% (15)32
+ +# 294 7 Conclusions + +295 We have presented an implementation of deep learning training purely in multi-party computation +296 with extensive results on the accuracy. We have found that the lower precision of MPC computation +297 increases the error considerably. We only have considered one particular implementation of more +298 complex computation such as division and exponentiation, which are crucial to the learning process +299 as part of softmax. Future work might consider different approximations of these building blocks. + +# + +00 References +301 N. Agrawal, A. S. Shamsabadi, M. J. Kusner, and A. Gascón. QUOTIENT: Two-party secure neural +302 network training and prediction. In L. Cavallaro, J. Kinder, X. Wang, and J. Katz, editors, ACM +303 CCS 2019, pages 1231–1247. ACM Press, Nov. 2019. doi: 10.1145/3319535.3339819. +304 M. Aliasgari, M. Blanton, Y. Zhang, and A. Steele. Secure computation on floating point numbers. +305 In NDSS 2013. The Internet Society, Feb. 2013. +306 A. Aly and N. P. Smart. Benchmarking privacy preserving scientific operations. In R. H. Deng, +307 V. Gauthier-Umaña, M. Ochoa, and M. Yung, editors, ACNS 19, volume 11464 of LNCS, pages +308 509–529. Springer, Heidelberg, June 2019. doi: 10.1007/978-3-030-21568-2_25. +309 T. Araki, J. Furukawa, Y. Lindell, A. Nof, and K. Ohara. High-throughput semi-honest secure +310 three-party computation with an honest majority. In E. R. Weippl, S. Katzenbeisser, C. Kruegel, +311 A. C. Myers, and S. Halevi, editors, ACM CCS 2016, pages 805–817. ACM Press, Oct. 2016. doi: +312 10.1145/2976749.2978331. +313 T. Araki, A. Barak, J. Furukawa, M. Keller, Y. Lindell, K. Ohara, and H. Tsuchida. Generalizing +314 the SPDZ compiler for other protocols. In D. Lie, M. Mannan, M. Backes, and X. Wang, editors, +315 ACM CCS 2018, pages 880–895. ACM Press, Oct. 2018. doi: 10.1145/3243734.3243854. +316 M. Barni, C. Orlandi, and A. Piva. A privacy-preserving protocol for neural-network-based computa +317 tion. In Proceedings of the 8th workshop on Multimedia and security, pages 146–151, 2006. +318 J. C. Benaloh and J. Leichter. Generalized secret sharing and monotone functions. In S. Goldwasser, +319 editor, CRYPTO’88, volume 403 of LNCS, pages 27–35. Springer, Heidelberg, Aug. 1990. doi: +320 10.1007/0-387-34799-2_3. +321 O. Catrina and A. Saxena. Secure computation with fixed-point numbers. In R. Sion, editor, FC +322 2010, volume 6052 of LNCS, pages 35–50. Springer, Heidelberg, Jan. 2010. +323 M. Dahl, J. Mancuso, Y. Dupis, B. Decoste, M. Giraud, I. Livingstone, J. Patriquin, and G. Uhma. +324 Private machine learning in tensorflow using secure computation. CoRR, abs/1810.08130, 2018. +325 URL http://arxiv.org/abs/1810.08130. +326 A. Dalskov, D. Escudero, and M. Keller. Fantastic four: Honest-majority four-party secure com +327 putation with malicious security. In 30th USENIX Security Symposium (USENIX Security 21), +328 2021. +329 A. P. K. Dalskov, D. Escudero, and M. Keller. Secure evaluation of quantized neural networks. +330 PoPETs, 2020(4):355–375, Oct. 2020. doi: 10.2478/popets-2020-0077. +331 I. Damgård, M. Fitzi, E. Kiltz, J. B. Nielsen, and T. Toft. Unconditionally secure constant-rounds +332 multi-party computation for equality, comparison, bits and exponentiation. In S. Halevi and +333 T. Rabin, editors, TCC 2006, volume 3876 of LNCS, pages 285–304. Springer, Heidelberg, Mar. +334 2006. doi: 10.1007/11681878_15. +335 I. Damgård, D. Escudero, T. K. Frederiksen, M. Keller, P. Scholl, and N. Volgushev. New primitives +336 for actively-secure MPC over rings with applications to private machine learning. In 2019 IEEE +337 Symposium on Security and Privacy, pages 1102–1120. IEEE Computer Society Press, May 2019. +338 doi: 10.1109/SP.2019.00078. +339 H. Eerikson, M. Keller, C. Orlandi, P. Pullonen, J. Puura, and M. Simkin. Use your brain! Arithmetic +340 3PC for any modulus with active security. In Y. T. Kalai, A. D. Smith, and D. Wichs, editors, ITC +341 2020, pages 5:1–5:24. Schloss Dagstuhl, June 2020. doi: 10.4230/LIPIcs.ITC.2020.5. +342 X. Glorot and Y. Bengio. Understanding the difficulty of training deep feedforward neural networks. +343 In Proceedings of the thirteenth international conference on artificial intelligence and statistics, +344 pages 249–256. JMLR Workshop and Conference Proceedings, 2010. +345 R. E. Goldschmidt. Applications of division by convergence. Master’s thesis, MIT, 1964. +346 D. Gunning, A. Hannun, M. Ibrahim, B. Knott, L. van der Maaten, V. Reis, S. Sengupta, S. Venkatara +347 man, and X. Zhou. Crypten: A new research tool for secure machine learning with pytorch, +348 2019. +349 I. Hubara, M. Courbariaux, D. Soudry, R. El-Yaniv, and Y. Bengio. Binarized neural networks. In +350 Advances in Neural Information Processing Systems, volume 29. Curran Associates, Inc., 2016. +351 M. Keller. MP-SPDZ: A versatile framework for multi-party computation. In J. Ligatti, X. Ou, +352 J. Katz, and G. Vigna, editors, ACM CCS 20, pages 1575–1590. ACM Press, Nov. 2020. doi: +353 10.1145/3372297.3417872. +354 M. Keller and K. Sun. Effectiveness of MPC-friendly softmax replacement, 2020. +355 V. Kolesnikov, A.-R. Sadeghi, and T. Schneider. A systematic approach to practically efficient general +356 two-party secure function evaluation protocols and their modular design. Journal of Computer +357 Security, 21(2):283–315, 2013. +358 Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner. Gradient-based learning applied to document +359 recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998. +360 Y. LeCun, C. Cortes, and C. Burges. Mnist handwritten digit database. ATT Labs [Online]. Avail +361 able: http: // yann. lecun. com/ exdb/ mnist , 2, 2010. Creative Commons Attribution-Share +362 Alike 3.0 license, https://creativecommons.org/licenses/by-sa/3.0/. +363 J. Liu, M. Juuti, Y. Lu, and N. Asokan. Oblivious neural network predictions via MiniONN +364 transformations. In B. M. Thuraisingham, D. Evans, T. Malkin, and D. Xu, editors, ACM CCS +365 2017, pages 619–631. ACM Press, Oct. / Nov. 2017. doi: 10.1145/3133956.3134056. +366 W.-j. Lu, Y. Fang, Z. Huang, C. Hong, C. Chen, H. Qu, Y. Zhou, and K. Ren. Faster secure multiparty +367 computation of adaptive gradient descent. In Proceedings of the 2020 Workshop on Privacy +368 Preserving Machine Learning in Practice, PPMLP’20, page 47–49, New York, NY, USA, 2020. +369 Association for Computing Machinery. ISBN 9781450380881. doi: 10.1145/3411501.3419427. +370 URL https://doi.org/10.1145/3411501.3419427. +371 P. Mohassel and P. Rindal. ABY3: A mixed protocol framework for machine learning. In D. Lie, +372 M. Mannan, M. Backes, and X. Wang, editors, ACM CCS 2018, pages 35–52. ACM Press, Oct. +373 2018. doi: 10.1145/3243734.3243760. +374 P. Mohassel and Y. Zhang. SecureML: A system for scalable privacy-preserving machine learning. +375 In 2017 IEEE Symposium on Security and Privacy, pages 19–38. IEEE Computer Society Press, +376 May 2017. doi: 10.1109/SP.2017.12. +377 V. Nair and G. E. Hinton. Rectified linear units improve Restricted Boltzmann machines. In Pro +378 ceedings of the 27th International Conference on International Conference on Machine Learning, +379 ICML’10, pages 807–814, 2010. +380 D. L. Quoc, F. Gregor, S. Arnautov, R. Kunkel, P. Bhatotia, and C. Fetzer. securetf: A secure +381 tensorflow framework. CoRR, abs/2101.08204, 2021. URL https://arxiv.org/abs/2101. +382 08204. +383 M. S. Riazi, C. Weinert, O. Tkachenko, E. M. Songhori, T. Schneider, and F. Koushanfar. Chameleon: +384 A hybrid secure computation framework for machine learning applications. In J. Kim, G.-J. Ahn, +385 S. Kim, Y. Kim, J. López, and T. Kim, editors, ASIACCS 18, pages 707–721. ACM Press, Apr. +386 2018. +387 D. Rotaru and T. Wood. MArBled circuits: Mixing arithmetic and Boolean circuits with active +388 security. In F. Hao, S. Ruj, and S. Sen Gupta, editors, INDOCRYPT 2019, volume 11898 of LNCS, +389 pages 227–249. Springer, Heidelberg, Dec. 2019. doi: 10.1007/978-3-030-35423-7_12. +390 T. Ryffel, A. Trask, M. Dahl, B. Wagner, J. Mancuso, D. Rueckert, and J. Passerat-Palmbach. A +391 generic framework for privacy preserving deep learning. CoRR, abs/1811.04017, 2018. URL +392 http://arxiv.org/abs/1811.04017. + +3 S. Tan, B. Knott, Y. Tian, and D. J. Wu. CryptGPU: Fast privacy-preserving machine learning on the GPU, 2021. +S. Wagh, D. Gupta, and N. Chandran. SecureNN: 3-party secure computation for neural network training. PoPETs, 2019(3):26–49, July 2019. doi: 10.2478/popets-2019-0035. +S. Wagh, S. Tople, F. Benhamouda, E. Kushilevitz, P. Mittal, and T. Rabin. Falcon: Honest-majority maliciously secure framework for private deep learning. PoPETs, 2021(1):188–208, Jan. 2021. doi: 10.2478/popets-2021-0011. + +# Checklist + +1. For all authors... + +(a) Do the main claims made in the abstract and introduction accurately reflect the paper’s contributions and scope? [Yes] +(b) Did you describe the limitations of your work? [Yes] See Section 7. +(c) Did you discuss any potential negative societal impacts of your work? [No] We only work with MNIST, and we believe that the potential negative impacts associated to it are very low. +(d) Have you read the ethics review guidelines and ensured that your paper conforms to them? [Yes] + +2. If you are including theoretical results... + +(a) Did you state the full set of assumptions of all theoretical results? [N/A] (b) Did you include complete proofs of all theoretical results? [N/A] + +3. If you ran experiments... + +(a) Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes] We include all code in the supplemental material, and the references include the URL for MNIST. +(b) Did you specify all the training details (e.g., data splits, hyperparameters, how they were chosen)? [Yes] See Section 6. +(c) Did you report error bars (e.g., with respect to the random seed after running experiments multiple times)? [Yes] Only in Figure 2. +(d) Did you include the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [Yes] See Section 6. + +4. If you are using existing assets (e.g., code, data, models) or curating/releasing new assets... + +(a) If your work uses existing assets, did you cite the creators? [Yes] See Section 6. +(b) Did you mention the license of the assets? [Yes] See the references and and Section 5. +(c) Did you include any new assets either in the supplemental material or as a URL? [Yes] Our code is included the supplemental material. +(d) Did you discuss whether and how consent was obtained from people whose data you’re using/curating? [N/A] +(e) Did you discuss whether the data you are using/curating contains personally identifiable information or offensive content? [N/A] + +5. If you used crowdsourcing or conducted research with human subjects... + +(a) Did you include the full text of instructions given to participants and screenshots, if applicable? [N/A] +(b) Did you describe any potential participant risks, with links to Institutional Review Board (IRB) approvals, if applicable? [N/A] +(c) Did you include the estimated hourly wage paid to participants and the total amount spent on participant compensation? [N/A] \ No newline at end of file diff --git a/parse/train/NiM9Q7Z95z/NiM9Q7Z95z_content_list.json b/parse/train/NiM9Q7Z95z/NiM9Q7Z95z_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..33a53e2a45cd5b52c06c7f34622b3d791da5eed2 --- /dev/null +++ b/parse/train/NiM9Q7Z95z/NiM9Q7Z95z_content_list.json @@ -0,0 +1,1167 @@ +[ + { + "type": "text", + "text": "Secure Quantized Training for Deep Learning ", + "text_level": 1, + "bbox": [ + 218, + 122, + 779, + 148 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Anonymous Author(s) \nAffiliation \nAddress \nemail ", + "bbox": [ + 423, + 200, + 580, + 256 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Abstract ", + "text_level": 1, + "bbox": [ + 462, + 292, + 535, + 309 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 We have implemented training of neural networks in secure multi-party computa \n2 tion (MPC) using quantization commonly used in said setting. To the best of our \n3 knowledge, we are the first to present training of MNIST purely implemented in \n4 MPC that comes within one percent of accuracy of training using plaintext compu \n5 tation. We found that training with MPC is possible, but it takes more epochs and \n6 achieves a lower accuracy than the usual CPU/GPU computation. More concretely, \n7 we have trained a network with two convolution and two dense layers to $9 8 . 5 \\%$ \n8 accuracy in 150 epochs. This took a day in our MPC implementation. ", + "bbox": [ + 150, + 321, + 767, + 434 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "9 1 Introduction ", + "text_level": 1, + "bbox": [ + 151, + 458, + 312, + 474 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "10 Secure multi-party computation (MPC) is a cryptographic technique that allows a set of parties to \n11 compute a public output on private inputs without revealing the inputs or any intermediate results. \n12 This makes it a potential solution to federated learning where the sample data stays private and only \n13 the model or even only inference results are revealed. \n14 Imagine a set of healthcare providers holding sensitive patient data. MPC allows them to collabo \n15 ratively train a model. This model could then either be released or even kept private for inference \n16 using MPC again. See Figure for an illustration. A more conceptual example is the well-known \n17 millionaires’ problem where two people want to find out who is richer without revealing their wealth. \n18 There is clearly a difference between the one bit of information desired and the full figures. \n19 There has been a sustained interest in applying secure computation to machine learning and neural \n20 networks going back to at least Barni et al. [2006]. More recent advantages in practical MPC have \n21 led to an increased effort in implementing both inference and training. \n22 A number of works such as Mohassel and Zhang [2017], Mohassel and Rindal [2018], Wagh et al. \n23 [2019], Wagh et al. [2021] implement neural network training with MPC at least in parts. However, \n24 they either give accuracy figures below $9 5 \\%$ or figures that have been obtained using plaintext training. \n25 For the latter case, the works do not clarify how close the computation for plaintext training matches \n26 the lower precision and other differences in the MPC setting. Agrawal et al. [2019] claim a higher \n27 accuracy in a comparable setting for a convolutional neural network with more channels than we \n28 use. However, they have only implemented dense layers, and we achieve comparable accuracy to \n29 them with only dense layers. All works use quantization in the sense that a fractional number $x$ is \n30 represented as $\\lfloor x \\cdot 2 ^ { - f } \\rceil$ . This makes addition considerably faster in the secure computation setting \n31 because it reduces to integer addition. Furthermore, some of the works suggest to replace the softmax \n32 function that uses exponentiation with a ReLU-based replacement. Keller and Sun [2020] have found \n33 that this softmax replacement deteriorates accuracy in dense neural networks to the extent that it does \n34 not justify the performance gains. \n35 The concurrent work of Tan et al. [2021] gives some figures on the learning curve when run using \n36 secure computation. However, they stop at five epochs for MNIST training where they achieve $94 \\%$ \n37 accuracy whereas we present the figures up to 150 epochs and $9 8 . 5 \\%$ accuracy. Furthermore, their \n38 choice of fixed-point precision 20 is considerably below 32, which we found to be optimal. We also \n39 found that our implementation is $40 \\%$ faster than theirs. Note that we use the CPU of one AWS \n40 c5.9xlarge instance per party whereas Tan et al. use one NVIDIA Tesla V100 GPU per party. We \n41 believe this somewhat counter-intuitive result comes from MPC heavily relying on communication, \n42 which is an aspect where GPUs do not have an advantage over CPUs. \n43 In this paper, we present an extensible framework for implementing deep learning training based on \n44 MP-SPDZ by Keller [2020], a framework for multi-party computation.1 Similar to TensorFlow and \n45 PyTorch, our approach allows representing deep learning models as succession of layers. We then \n46 use this implementation to obtain accuracy figures for MNIST training by utilizing the MP-SPDZ \n47 emulator, which allows to run the plaintext equivalent of secure computation, that is, the same \n48 algorithms with the same precision. Finally, we run one of the most promising instantiation in real \n49 secure computation in order to benchmark it confirm the result from the plaintext emulator. \n50 There are a number of projects that integrate secure computation directly into popular machine \n51 learning frameworks such as CrypTen by Gunning et al. [2019], PySyft by Ryffel et al. [2018], and \n52 TF Encrypted by Dahl et al. [2018]. Our approach differs from all of them by running the protocol as \n53 native CPU code (implemented using $\\mathrm { C } { + + }$ ). This allows for much faster execution. For example, \n54 CrypTen provides an MNIST training example (mpc_autograd_cnn) that takes over one minute \n55 to run one epoch with 100 samples on one machine. In comparison, our implementation takes 11 \n56 minutes to run one epoch with the full dataset of 60,000 samples. \n57 Another line of work (e.g., Quoc et al. [2021]) uses trusted execution environments that provide \n58 computation outside the reach of the operating system. This is a different security model to multi-party \n59 computation that works with distributing the information among several entities. \n60 The paper is structured as follows: After introducing the basics of the protocol we use in Section 2, \n61 we will explain the mathematical building blocks in Section 3 and their use in the context of deep \n62 learning in Section 4. Finally, we will present our implementation in Section 5 and our experimental \n63 results for MNIST in Section 6. ", + "bbox": [ + 148, + 488, + 825, + 544 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "", + "bbox": [ + 148, + 550, + 825, + 619 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "", + "bbox": [ + 148, + 626, + 823, + 669 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "", + "bbox": [ + 147, + 674, + 826, + 854 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "", + "bbox": [ + 148, + 859, + 823, + 902 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/bb0f136189595323f73d65ddc0b25170a7b7ffb1f9107b9e37395cb5ee7bab1c.jpg", + "image_caption": [ + "Figure 1: Outsourced computation: Data holders (on the left) secret-share their data to a number of computing parties (on the right), who then return the desired the result (e.g., a model or inference results on further queries). All communication except outputs are secret-shared and thus secure if no two computing parties collude. " + ], + "image_footnote": [], + "bbox": [ + 254, + 93, + 741, + 226 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 147, + 323, + 825, + 392 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 147, + 400, + 825, + 497 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 147, + 502, + 825, + 599 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 147, + 606, + 825, + 648 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 147, + 655, + 825, + 710 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "64 2 An Efficient Secure Multi-Party Computation Protocol ", + "text_level": 1, + "bbox": [ + 148, + 729, + 661, + 747 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "There is a wide range of MPC protocols with a variety of security properties (see Keller [2020] for an overview). In this paper we focus on the setting of three-party computation with one semi-honest corruption. This means that out of the three parties two are expected to behave honestly, i.e., they follow the protocol and keep their view of the protocol secret, and one party is expected to follow the protocol but might try extract information from their view. The reason for choosing this setting is that it allows an efficient MPC protocol while still allowing secure outsourced computation. The concrete protocol we use goes back to Benaloh and Leichter [1990] with further aspects by Araki et al. [2016], Mohassel and Rindal [2018], and Eerikson et al. [2020]. We summarize the core protocol below. The mathematical building blocks in the next section mostly use the aspects below. ", + "bbox": [ + 173, + 761, + 825, + 887 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "74 Secret sharing All intermediate values in our protocol are stored using replicated secret sharing. A \n75 secret value $x$ is a represented as a random sum $x = x _ { 0 } + x _ { 1 } + x _ { 2 }$ , and party $P _ { i }$ holds $( x _ { i - 1 } , x _ { i + 1 } )$ \n76 where the indices are computed modulo three. Clearly, each party is missing one value to compute \n77 the sum. On the other hand, each pair of parties hold all necessary to reconstruct the secret. For a \n78 uniformly random generation of shares, the computation domain has to be finite. Most commonly, \n79 this domain is defined by integer computation modulo a number. We use $2 ^ { k }$ for $k$ being a multiple 64 \n80 and 2 as the moduli. The first case corresponds to an extension of 64-bit arithmetic found on most \n81 processors. We will refer to the two settings as arithmetic and binary secret sharing throughout the \n82 paper. \n83 Input sharing The secret sharing scheme implies a protocol to share inputs where the inputting \n84 party samples the shares and distributes them accordingly. Eerikson et al. [2020] have proposed a \n85 more efficient protocol where the inputting party only needs to send one value instead of two pairs of \n86 values. If $P _ { i }$ would like to input $x$ , $x _ { i }$ is set to zero, and $x _ { i - 1 }$ is generated with a pseudo-random \n87 generator using a key previously shared between $P _ { i }$ and $P _ { i + 1 }$ . $P _ { i }$ can compute $x _ { i + 1 } = x - x _ { i - 1 }$ and \n88 send it to $P _ { i - 1 }$ . While the resulting secret sharing is not entirely random, the fact that $P _ { i }$ already \n89 knows $x$ makes randomizing $x _ { i }$ obsolete. \n90 Addition The commutative nature of addition allows to add secret sharings without communication. \n91 More concretely, secret sharings $x = x _ { 0 } + x _ { 1 } + x _ { 2 }$ and $y = y _ { 0 } + y _ { 1 } + y _ { 2 }$ imply the secret sharing \n92 $x + y = ( x _ { 0 } + y _ { 0 } ) + ( x _ { 1 } + y _ { 1 } ) + ( x _ { 2 } + y _ { 2 } )$ . ", + "bbox": [ + 147, + 90, + 825, + 217 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "", + "bbox": [ + 145, + 229, + 825, + 328 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "", + "bbox": [ + 147, + 342, + 825, + 386 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "93 Multiplication The product of $x = x _ { 0 } + x _ { 1 } + x _ { 2 }$ and $y = y _ { 0 } + y _ { 1 } + y _ { 2 }$ is ", + "bbox": [ + 147, + 398, + 678, + 414 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/0e562fbeff8dbe326579bda5ca198bdb7384ec8bfd3048b083e4bf0b03a38815.jpg", + "text": "$$\n{ \\begin{array} { r l } & { x \\cdot y = ( x _ { 0 } + x _ { 1 } + x _ { 2 } ) \\cdot ( y _ { 0 } + y _ { 1 } + y _ { 2 } ) } \\\\ & { \\qquad = ( x _ { 0 } y _ { 0 } + x _ { 0 } y _ { 1 } + x _ { 1 } y _ { 0 } ) + ( x _ { 1 } y _ { 1 } + x _ { 1 } y _ { 2 } + x _ { 1 } y _ { 1 } ) + ( x _ { 2 } y _ { 2 } + x _ { 2 } y _ { 0 } + x _ { 0 } y _ { 2 } ) . } \\end{array} }\n$$", + "text_format": "latex", + "bbox": [ + 225, + 417, + 771, + 455 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "94 Each of the brackets only contains shares known by one of the parties. They can thus compute an \n95 additive secret sharing (one summand per party) of the product. However, every party only holding \n96 one share does not satisfy the replication requirement for further multiplications. It is not secure \n97 for every party to pass their value on to another party because the summands are not distributed \n98 randomly. This can be fixed by rerandomization: Let $x y = z _ { 0 } + z _ { 1 } + z _ { 2 }$ where $z _ { i }$ is know to $P _ { i }$ . \n99 Every party $P _ { i }$ computes $z _ { i } ^ { \\prime } = z _ { i } + r _ { i , i + 1 } - r _ { i - 1 , i }$ where $r _ { i , i + 1 }$ is generated with a pseudo-random \n100 generator using a key pre-shared between $P _ { i }$ and $P _ { i + 1 }$ . The resulting sum $x y = \\hat { z } _ { 0 } ^ { \\prime } + z _ { 1 } ^ { \\prime } + z _ { 2 } ^ { \\prime }$ is \n101 pseudo-random, and it is thus secure for $P _ { i }$ to send $z _ { i } ^ { \\prime }$ to $P _ { i + 1 }$ in order to create a replicated secret \n102 sharing $\\left( ( x y ) _ { i - 1 } , ( x y ) _ { i + 1 } \\right) = \\left( z _ { i } ^ { \\prime } , z _ { i - 1 } ^ { \\prime } \\right)$ . ", + "bbox": [ + 143, + 458, + 825, + 585 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "103 3 Secure Computation Building Blocks ", + "text_level": 1, + "bbox": [ + 142, + 602, + 514, + 619 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "104 In this section, we will discuss how to implement computation with MPC with a focus on how \n105 it differs from computation on CPUs or GPUs. Most of the techniques below are already known \n106 individually. To the best of our knowledge however, we are the first to put them together in an efficient \n107 and extensible framework for secure computation of deep learning training. \n108 Domain conversion Recall we that we use computation modulo $2 ^ { k }$ for $k$ being a multiple of 64 \n109 as well as 1. Given that the main operations are just addition and multiplication in the respective \n110 domain, it is desirable to compute integer arithmetic in the large domain but operations with a \n111 straight-forward binary circuit modulo two. There has been a long-running interest in this going \n112 back to least Kolesnikov et al. [2013]. We mainly rely on the approach proposed by Mohassel and \n113 114 $\\{ x _ { 0 } ^ { ( i ) } \\} _ { j = 0 } ^ { k - 1 }$ 018] and Araki et al. [20the bit decomposition of $x _ { 0 }$ . Recal, that is, $x _ { 0 } ^ { ( i ) } \\in \\{ 0 , 1 \\}$ $x \\in 2 ^ { k }$ is shand $\\begin{array} { r } { x _ { 0 } = \\sum _ { i = 0 } ^ { k - 1 } x _ { 0 } ^ { ( i ) } 2 ^ { i } } \\end{array}$ $x = x _ { 0 } + x _ { 1 } + x _ { 2 }$ . Now letlf-evident \n115 that $x _ { 0 } ^ { ( i ) } = x _ { 0 } ^ { ( i ) } + 0 + 0$ is a valid secret sharing modulo two (albeit not a random one). Furthermore, \n116 every party holding $x _ { 0 }$ can generate $x _ { 0 } ^ { ( i ) }$ . It is therefore possible for the parties to generate a secret \n117 sharing modulo two of a single share modulo $2 ^ { k }$ . Repeating this for all shares and the computing \n118 the addition as a binary circuit allows the parties to generate a secret sharing modulo two from a \n119 secret sharing modulo $\\bar { 2 } ^ { k }$ . Conversion in the other direction can be achieved using a similar technique \n120 or using daBits as described by Rotaru and Wood [2019]. In the following we will use the term \n121 mixed-circuit computation for any technique that works over both computation domains. \n122 Quantization While Aliasgari et al. [2013] showed that it is possible to implement floating-point \n123 computation, the cost is far higher than integer computation. It is therefore common to represent \n124 fractional numbers using quantization (also called fixed-point representation) as suggested by Catrina \n125 and Saxena [2010]. A real number $x$ is represented as $\\bar { \\boldsymbol { x } } = \\lfloor \\boldsymbol { x } \\cdot \\mathbf { \\bar { 2 } } ^ { f } \\rceil$ where $f$ is an integer specifying \n126 the precision. The linearity of the representation allows to compute addition by simply adding the \n127 representing integers. Multiplication however requires to adjust the result because it will have twice \n128 the precision: $( x \\cdot 2 ^ { f } ) \\cdot ( y \\cdot \\dot { 2 } ^ { f } ) = x y \\cdot 2 ^ { 2 f }$ . There are two ways to rectify this: ", + "bbox": [ + 142, + 633, + 825, + 689 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "", + "bbox": [ + 140, + 703, + 825, + 911 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "", + "bbox": [ + 142, + 90, + 825, + 189 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "• An obvious correction would be to shift the result by $f$ bits after adding $2 ^ { f - 1 }$ to the integer representation. This ensures rounding to the nearest number possible in the representation, with the tie being broken by rounding up. Dalskov et al. [2021] presented an efficient implementation of the truncation using mixed-circuit computation. ", + "bbox": [ + 218, + 200, + 825, + 257 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "• However, Catrina and Saxena have found that in the context of secure computation it is more efficient to use probabilistic truncation. This method rounds up or down probabilistically depending on the input. For example, probabilistically rounding 0.75 to an integer would see it rounded of up with probability 0.75 and down with probability 0.25. The probabilistic truncation is an effect of the fact that the operation involves the truncation of a randomized value, that is the computation of $\\left\\lfloor ( x + r ) / 2 ^ { m } \\right\\rfloor$ for a random $m$ -bit value $r$ . It is easy to see that ", + "bbox": [ + 217, + 262, + 825, + 359 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/8c5c9712adea6b5b528c5e11a746ad687533c04d65eda8941ed882721c5e70cc.jpg", + "text": "$$\n\\lfloor ( x + r ) / 2 ^ { m } \\rfloor = \\left\\{ \\begin{array} { l l } { \\lfloor x / 2 ^ { m } \\rfloor } & { ( x \\bmod 2 ^ { m } + r ) < 2 ^ { m } } \\\\ { \\lfloor x / 2 ^ { m } \\rfloor + 1 } & { ( x \\bmod 2 ^ { m } + r ) \\ge 2 ^ { m } . } \\end{array} \\right.\n$$", + "text_format": "latex", + "bbox": [ + 334, + 358, + 720, + 393 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Therefore, the larger $( x \\bmod 2 ^ { m } )$ ) is, the more likely the latter condition is true. Dalskov et al. [2020] present an efficient protocol in our security model. ", + "bbox": [ + 230, + 398, + 823, + 428 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "142 Our quantization scheme is related to quantized neural networks (see e.g. Hubara et al. [2016]). \n143 However, our consideration is not to compress the model, but to improve the computational speed \n144 and save communication cost. \n145 Dot products Dot products are an essential building block of linear computation such as matrix \n146 multiplication. In the light of quantization, it is possible to reduce the usage of truncation by deferring \n147 after the summation. In other words, the dot product in the integer representations is computed before \n148 truncating. This not only reduces the truncation error, it is also more efficient because the truncation \n149 is the most expensive part in quantized secure multiplication. Similarly, our protocol allows to defer \n150 the communication needed for multiplication. Let $\\vec { x }$ and $\\vec { y }$ be two vectors where the elements are \n151 secret shared, that is, $\\{ x ^ { ( i ) } \\} = x _ { 0 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } + x _ { 2 } ^ { ( i ) }$ and similarly for $\\boldsymbol y ^ { ( i ) }$ . The inner product then is ", + "bbox": [ + 142, + 438, + 826, + 481 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "", + "bbox": [ + 140, + 496, + 825, + 599 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/36540aeb1a5d2f89955c3e16fde0f520a4fd1af5a546d08dd4d4fe06ee6c50ec.jpg", + "text": "$$\n\\begin{array} { r l } & { \\displaystyle \\sum _ { i } x ^ { ( i ) } \\cdot y ^ { ( i ) } = \\sum _ { i } ( x _ { 0 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } + x _ { 2 } ^ { ( i ) } ) \\cdot ( y _ { 0 } ^ { ( i ) } + y _ { 1 } ^ { ( i ) } + y _ { 2 } ^ { ( i ) } ) } \\\\ & { \\quad \\quad \\quad \\quad = \\displaystyle \\sum _ { i } ( x _ { 0 } ^ { ( i ) } y _ { 0 } ^ { ( i ) } + x _ { 0 } ^ { ( i ) } y _ { 1 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } y _ { 0 } ^ { ( i ) } ) + \\sum _ { i } ( x _ { 1 } ^ { ( i ) } y _ { 1 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } y _ { 2 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } y _ { 1 } ^ { ( i ) } ) } \\\\ & { \\quad \\quad \\quad \\quad + \\displaystyle \\sum _ { i } ( x _ { 2 } ^ { ( i ) } y _ { 2 } ^ { ( i ) } + x _ { 2 } ^ { ( i ) } y _ { 0 } ^ { ( i ) } + x _ { 0 } ^ { ( i ) } y _ { 2 } ^ { ( i ) } ) . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 207, + 604, + 789, + 708 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "152 The three sums in the last term can be compute locally by one party each before applying the same \n153 protocol as for a single multiplication. \n154 Comparisons Arithmetic secret sharing does not allow to access the individual bits directly. It \n155 is therefore not straightforward to compute comparisons such as “less than”. There is a long line \n156 of literature on how to achieve this going back to at least Damgård et al. [2006]. More recently, \n157 most attention has been given to combine the power of arithmetic and binary secret sharing in \n158 order to combine the best of worlds. One possibility to do so is to plainly convert to the binary \n159 domain and compute the comparison circuit there. In our concrete implementation we use the more \n160 efficient approach by Mohassel and Rindal [2018]. It starts by taking the difference of the two \n161 inputs. Computing the comparison then reduces to comparing to zero, which in turn is equivalent \n162 to extracting the most significant bit as it indicates the sign. The latter is achieved by converting \n163 the shares locally to bit-wise sharing of the arithmetic shares, which sum up to the secret value. It \n164 remains to compute the sum of the binary shares in order to come up with the most significant bit. \n165 Oblivious Selection Plain secure computation does not allow branching because the parties would \n166 need to be aware which branch is followed. Conditional assignment can be implemented as follows \n167 however. If $b \\in \\{ 0 , 1 \\}$ denotes the condition, $x + b \\cdot ( y - x )$ is either $x$ or $y$ depending on $b$ . If the \n168 condition is available in binary secret sharing but $x$ and $y$ in arithmetic secret sharing, $b$ has to be \n169 converted to the latter. This can be done using a daBit as introduced by Rotaru and Wood [2019], \n170 which is a secret random bit shared both in arithmetic and binary. It allows to mask a bit in one world \n171 by XORing it. The result is then revealed and the masking is undone in the other world. \n172 Division Catrina and Saxena [2010] have shown how to implement quantized division using the \n173 algorithm by Goldschmidt [1964]. It mainly uses arithmetic and the probabilistic truncation already \n174 explained. In addition, the initial approximation requires a full bit decomposition as described above. \n175 The error of the output depends on the error in the multiplications used for Goldschmidt’s iteration, \n176 which compounds in particular when using probabilistic truncation. Due to the nature of secure \n177 computation, the result of division by zero is undefined. One could obtain a secret failure bit by \n178 testing the divisor to zero. However, we found that not to be necessary in our algorithm. This is \n179 because we only use division by secret value only for the softmax function where the it is guaranteed \n180 to strictly positive. \n181 Logarithm Computation logarithm with any public base can be reduced to logarithm to base two \n182 using $\\log _ { x } y = \\log _ { 2 } y \\cdot \\log _ { x } 2$ . Aly and Smart [2019] have proposed computing $y = a \\cdot 2 ^ { b }$ where \n183 $a \\in [ 0 . 5 , 1 )$ and $b \\in \\mathbb { Z }$ . This then allows to compute $\\log _ { 2 } y = \\log _ { 2 } a + b$ . Given the restricted range \n184 of $a$ , $\\log _ { 2 } a$ can be approximated using a division of polynomials. Numerical stability and input range \n185 control are less of an issue here because we only use logarithm for the loss computation, which does \n186 not influence the training. \n187 Exponentiation By using $x ^ { y } = 2 ^ { y \\log _ { 2 } x }$ , any exponentiation can be reduced to exponentiation \n188 with base two. Aly and Smart [2019] have shown how to compute $2 ^ { a } = 2 ^ { \\lfloor a \\rfloor } \\cdot 2 ^ { a - \\lfloor a \\rfloor }$ by computing \n189 the two exponents using bit decomposition and the second factor using a polynomial approximation. \n190 Regarding the first factor, if $\\begin{array} { r } { b = \\sum _ { i \\geq 0 } b _ { i } 2 ^ { i } } \\end{array}$ is an integer with $b _ { i } \\in \\{ 0 , 1 \\}$ , ", + "bbox": [ + 148, + 714, + 823, + 743 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "", + "bbox": [ + 140, + 758, + 825, + 911 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "", + "bbox": [ + 142, + 90, + 825, + 189 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "", + "bbox": [ + 140, + 203, + 825, + 328 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "", + "bbox": [ + 140, + 342, + 825, + 426 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "", + "bbox": [ + 142, + 439, + 826, + 500 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/8c23276368b1b8c7d81c896fefea584c2c7d3d49e6ccd58490eca1b2a9f47135.jpg", + "text": "$$\n2 ^ { b } = 2 ^ { \\sum _ { i \\geq 0 } b _ { i } 2 ^ { i } } = \\prod _ { i \\geq 0 } 2 ^ { b _ { i } 2 ^ { i } } = \\prod _ { i \\geq 0 } ( 1 + b _ { i } \\cdot ( 2 ^ { 2 ^ { i } } - 1 ) ) .\n$$", + "text_format": "latex", + "bbox": [ + 320, + 506, + 678, + 542 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "191 As with division the numerical stability depends on the truncation used for multiplication. ", + "bbox": [ + 150, + 547, + 759, + 564 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Inverse square root Aly and Smart [2019] have proposed to compute square root using Goldschmidt and Raphson-Newton iterations. We could combine this with the division above. However, Lu et al. [2020] have proposed a more direct computation that avoids running two successive iterations. ", + "bbox": [ + 163, + 577, + 825, + 621 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Uniformly random fractional number Limiting ourselves to intervals of the form $[ x , x + 2 ^ { e } ]$ for a potentially negative integer $e$ , we can reduce the problem to generate a random $( f + { \\dot { e } } )$ -bit number where $f$ is the fixed-point precision. Recall that we represent a fractional number $x$ as $\\lfloor x \\cdot 2 ^ { - f } \\rceil$ Generating a random $n$ -bit number is straight-forward using random bits, which in our protocol can be generated as presented by Damgård et al. [2019]. In the context of our protocol however, Dalskov et al. [2021] have a presented a more efficient approach that involves mixed-circuit computation. ", + "bbox": [ + 171, + 633, + 825, + 719 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Communication cost Table 1 show the total communication cost of some of the building blocks in our protocol for $f = 3 2$ . This setting mandates the modulus $2 ^ { 1 2 8 }$ because the division protocol requires bit length $4 f$ . ", + "bbox": [ + 171, + 733, + 825, + 776 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "4 Machine Learning Building Blocks ", + "text_level": 1, + "bbox": [ + 165, + 795, + 496, + 813 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "In this section, we will use the building blocks in Section 3 to construct high-level computational modules for deep learning. ", + "bbox": [ + 174, + 825, + 823, + 854 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Fully connected layers Both forward and back-propagation of fully connected layers can be seen as matrix multiplications and thus can be implemented using dot products. A particular challenge in secure computation is to compute a number of outputs in parallel in order to save communication ", + "bbox": [ + 173, + 869, + 823, + 911 + ], + "page_idx": 4 + }, + { + "type": "table", + "img_path": "images/9fd53d77e94058a5729b0569126beb6fc6cee5af420d0afe5b03e2771b4647cb.jpg", + "table_caption": [ + "Table 1: Communication cost of select computation for $f = 3 2$ and integer modulus $2 ^ { 1 2 8 }$ " + ], + "table_footnote": [], + "table_body": "
Bits
Integer multiplication384
Probabilistic truncation Nearest truncation1,536 4,462
Comparison1,369
Division (prob. truncation)29,866
Division (nearest truncation)57,798
Exponentiation (prob. truncation)77,684
Exponentiation (nearest truncation)
171,638
Invert square root (prob. truncation) Invert square root (nearest truncation)20,073 27,699
", + "bbox": [ + 328, + 111, + 671, + 280 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "10 rounds. We solve this by having a dedicated infrastructure in our implementation that computes all dot \n11 products for a matrix multiplication in a single batch, thus combining all necessary communication. ", + "bbox": [ + 148, + 304, + 825, + 333 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "2D convolution layers Similar to fully connected layers, 2D convolution and its corresponding gradient can be implemented using only dot products, and we again compute several output values in parallel. ", + "bbox": [ + 169, + 347, + 826, + 390 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "215 Rectified Linear Unit (ReLU) ReLU Nair and Hinton [2010] is defined as follows: ", + "bbox": [ + 140, + 402, + 736, + 420 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/6552a81b7732498e72666ed8446d188466cdfbd1913c006a8ad902a1062afcf8.jpg", + "text": "$$\n{ \\mathsf { R e L U } } ( x ) : = { \\left\\{ \\begin{array} { l l } { x , } & { { \\mathrm { i f ~ } } x > 0 } \\\\ { 0 . } & { { \\mathrm { o t h e r w i s e } } } \\end{array} \\right. }\n$$", + "text_format": "latex", + "bbox": [ + 398, + 424, + 594, + 460 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "216 It can thus be implemented as a comparison followed by an oblivious selection. For back-propagation, \n217 it is advantageous to reuse the comparison results from forward propagation due to the relatively high \n218 cost in secure computation. Note that the comparison results are stored in secret-shared form and \n219 thus there is no reduction in security. \n220 Max pooling Similar to ReLU, max pooling can be reduced to comparison and oblivious selection. \n221 In secure computation, it saves communication rounds if the process uses a balanced tree rather than \n222 iterating over all input values of one maximum computation. For back-propagation it again pays off \n223 to the store intermediate results from forward propagation, again in secret-shared form. \n224 Softmax and cross entropy loss This combination requires computing the following gradient for \n225 back-propagation: ", + "bbox": [ + 140, + 465, + 825, + 522 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "", + "bbox": [ + 140, + 535, + 826, + 593 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "", + "bbox": [ + 148, + 606, + 825, + 635 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/eae777e4bc355b0a5c52d65a7001eadc3f0665c189d84dfb531be701ff493e40.jpg", + "text": "$$\n\\nabla i : = \\frac { \\partial \\ell } { \\partial x _ { i } } = \\frac { \\partial } { \\partial x _ { i } } \\Big ( - \\sum _ { k } y _ { k } \\cdot x _ { k } + \\log \\sum _ { j } e ^ { x _ { j } } \\Big ) = - y _ { i } + \\frac { e ^ { x _ { i } } } { \\sum _ { j } e ^ { x _ { j } } } ,\n$$", + "text_format": "latex", + "bbox": [ + 271, + 638, + 725, + 679 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "226 where $y _ { i }$ denotes an element of the ground truth as a one-hot vector, and $x _ { i }$ denotes the output of the \n227 last layer. \n228 On the right hand side of eq. (1), the values in the denominator are potentially large due to the use \n229 of the exponential. This is prone to numerical overflow in our quantized representation because the \n230 latter puts relatively strict limits on the values. We therefore optimize the computation by computing \n231 the maximum of the input values: ", + "bbox": [ + 143, + 684, + 828, + 713 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "", + "bbox": [ + 140, + 718, + 825, + 775 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/6c2446d51ceae645baff7cf2c22ccb1d25b3d2e64c166c57a212c0dae2a3f332.jpg", + "text": "$$\nm = \\operatorname* { m a x } _ { j } ( \\{ x _ { j } \\} ) .\n$$", + "text_format": "latex", + "bbox": [ + 437, + 772, + 558, + 796 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "232 Then we compute ", + "bbox": [ + 142, + 799, + 294, + 814 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/e97d4b3c6d76abf1abd9d00e5c9538cb1f5ba3ae1df09857309d21f1c004e343.jpg", + "text": "$$\n{ \\frac { e ^ { x _ { i } - m } } { \\sum _ { j } e ^ { x _ { j } - m } } } = { \\frac { e ^ { x _ { i } } e ^ { - m } } { ( \\sum _ { j } e ^ { x _ { j } } ) e ^ { - m } } } = { \\frac { e ^ { x _ { i } } } { \\sum _ { j } e ^ { x _ { j } } } } .\n$$", + "text_format": "latex", + "bbox": [ + 362, + 811, + 635, + 849 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "233 All the exponents on the left-most term are at most zero, and thus the dividend is at most one and the \n234 divisor is at most the number of labels (which is 10 in MNIST). The same technique can be used to \n235 compute the sigmoid activation function, as sigmoid(x) = 11+exp(−x) $\\begin{array} { r } { \\mathrm { \\begin{array} { r } { \\mathrm { i d } ( x ) = \\frac { 1 } { 1 + \\exp ( - x ) } = \\frac { \\exp ( 0 ) } { \\exp ( 0 ) + \\exp ( x ) } } } \\end{array} } \\end{array}$ is a special \n236 case of softmax. \n237 Stochastic gradient descent The model parameter update in SGD only involves basic arithmetic: \n238 $\\begin{array} { r } { \\theta _ { j } \\theta _ { j } - \\frac { \\gamma } { B } \\sum _ { i = 1 } ^ { B } \\bigtriangledown _ { i j } } \\end{array}$ where $\\theta _ { j }$ is the parameter indexed by $j , B$ is the mini-batch size, $\\gamma > 0$ is \n239 the learning rate, and $\\nabla i j$ is the gradient of the loss with respect to the ’th sample in the mini-batch \n240 and the parameter $\\theta _ { j }$ . In order to tackle the limited precision with quantization, we defer dividing by \n241 the batch size to the model update. This means that we do not divide the gradient value by the batch \n242 size when computing them as described in the previous paragraph. Instead, we divide the model \n243 update by the batch size. Since we use a batch size that is a power of two (128), it is sufficient to use \n244 probabilistic truncation instead of full-blown division. This saves both time and decreases the error. ", + "bbox": [ + 140, + 851, + 826, + 912 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "", + "bbox": [ + 140, + 90, + 825, + 205 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Adam The main difference to SGD in terms of basic computational operations is the additional use of an inverse square root. We again defer the division by the batch size to just before the model update. ", + "bbox": [ + 161, + 226, + 825, + 268 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Parameter initialization We use the Glorot initialization by Glorot and Bengio [2010]. Besides basic operations, it mainly involves generating a uniformly random fractional value in a given interval. ", + "bbox": [ + 161, + 290, + 825, + 319 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "250 5 Implementation ", + "text_level": 1, + "bbox": [ + 151, + 344, + 338, + 361 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "51 We built our implementation on MP-SPDZ by Keller [2020]. MP-SPDZ not only implements a \n52 range of MPC protocols, it also comes with a high-level library containing the building blocks in \n253 Section 3. MP-SPDZ already featured capabilities to train dense neural networks as well as inference \n254 for convolutional neural networks. We have added backward propagation for a number of layer types, \n255 including 2D convolution. Furthermore, we have corrected a bug in the backward propagation for \n256 dense layers. ", + "bbox": [ + 150, + 378, + 825, + 463 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "MP-SPDZ allows implementing the computation in Python code, which is then compiled a specific bytecode. This code can be execute by a virtual machine executing the actual secure computation. The process allows to optimize the computation in the context of MPC. ", + "bbox": [ + 173, + 469, + 826, + 511 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "The framework also features an emulator that executes the exact computation that could be done securely in the clear. This allowed us to collect the accuracy figures in the next section at a lower cost. ", + "bbox": [ + 161, + 517, + 825, + 546 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "262 It is licensed under a BSD-style license, which allows to extend the code. ", + "bbox": [ + 147, + 551, + 655, + 566 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "63 6 MNIST ", + "text_level": 1, + "bbox": [ + 155, + 593, + 267, + 609 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "For a concrete measurement of accuracy and running times, we have implemented training for the well-known MNIST dataset by LeCun et al. [2010]. We work mainly with the models that have been used by Wagh et al. [2019] with secure computation, and we will reuse their numbering (A–D). The models contain up to four linear layers. Network C is a convolutional neural network going back to the seminal work by LeCun et al. [1998] whereas the others are simpler networks that have been proposed by works on secure computation such as Mohassel and Zhang [2017], Liu et al. [2017], and Riazi et al. [2018]. We present the networks as Keras code in the supplemental material. ", + "bbox": [ + 166, + 628, + 825, + 726 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Figure 2 shows the results for various quantization precisions and and the two rounding options. We have used SGD with learning rate 0.01, batch size 128, and the usual MNIST training/test split. $f = 6 4$ is the best option with probabilistic rounding, improving on both $f = 1 6$ and $f = 3 2$ Furthermore, nearest rounding performs worse that probabilistic for $f = 1 6$ and $f = 3 2$ . Due to the high cost, we only ran $f = 3 2$ with probabilistic rounding several times. The range is indicate by the shaded area. We focus on $f = 3 2$ because it offers the faster convergence. ", + "bbox": [ + 171, + 731, + 825, + 815 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Figure 3 then shows the result with a variety of optimizers. While increasing the learning rate for SGD leads to a lower stability, Adam exposes a smoother learning learning curve albeit not a faster process. ", + "bbox": [ + 168, + 820, + 825, + 863 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "280 Finally, Figures 4 shows our results for all networks used by Wagh et al. [2019]. As one would expect, \n281 the most sophisticated network performs best. Somewhat surprisingly, however, Network A (without \n282 convolutional layers) performs better than the simpler networks containing convolutional layers. ", + "bbox": [ + 145, + 869, + 825, + 911 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/3a38cf90fbe641b15f4921c674ee686f614e2a42472767f16d74facf9f248ace.jpg", + "image_caption": [ + "Figure 2: Loss and accuracy for network C and precision options when running SGD with rate 0.01. " + ], + "image_footnote": [], + "bbox": [ + 202, + 87, + 797, + 329 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/ca865aaa95abbc18690e028b3ca52b226046ffa5e278339ad08eb557c5491c91.jpg", + "image_caption": [ + "Figure 3: Loss and accuracy for network C with various optimizer options, $f = 3 2$ , and probabilistic truncation. " + ], + "image_footnote": [], + "bbox": [ + 197, + 375, + 772, + 617 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Resources We ran the emulator on AWS c5.9xlarge instances. One epoch takes a few second to several minutes depending on the model. Overall, we estimate that we have used a few weeks worth of computing time including experiments not included here because of bugs in the code. ", + "bbox": [ + 166, + 689, + 825, + 731 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "6.1 Secure computation ", + "text_level": 1, + "bbox": [ + 174, + 752, + 351, + 767 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "In order to verify our emulation results, we have run Network C with precision $f \\ = \\ 3 2$ and probabilistic rounding in our actual multi-party computation protocol. We could verify that it converges on $9 8 . 5 \\%$ accuracy at 150 epochs, taking 20 hours. Table 2 compares our result to previous works in a LAN setting. Note that Wagh et al. [2019] and Wagh et al. [2021] give accuracy figures. From personal communication with the authors and the fact that the source repository for the latter work2 says that their “code has not run end-to-end training”, we derive our assessment that their figures do not reflect the secure computation. ", + "bbox": [ + 173, + 780, + 825, + 877 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/700a37281e56f21023ee263453c7785f933f7176bea7c10046675acc845cf598.jpg", + "image_caption": [ + "Figure 4: Loss and accuracy for various networks, $f = 3 2$ , and probabilistic truncation. " + ], + "image_footnote": [], + "bbox": [ + 217, + 87, + 784, + 329 + ], + "page_idx": 8 + }, + { + "type": "table", + "img_path": "images/97abf13640224b2e7ab8627c268f2fab07076453ba8f2fc10f4cc1bc2f920aae.jpg", + "table_caption": [ + "Table 2: Comparison to previous work in the LAN setting. $( ^ { * } )$ Mohassel and Zhang [2017] and Agrawal et al. [2019] use a different security model and are thus incomparable. We include them for completeness. Two numbers refer to online and offline time. Accuracy N/A means that the accuracy figures were not given or computed in a way that does not reflect the secure computation. " + ], + "table_footnote": [], + "table_body": "
NetworkEpoch time (s)Acc. (# epochs)Precision(f)
AMohassel and Zhang [2017]283/19333*93.4% (15)13
Mohassel and Rindal [2018]18094.0% (15)N/A
Agrawal et al. [2019]31392*95.0% (10)N/A
Wagh et al. [2019]247N/A13
Wagh et al. [2021]41N/A13
Ours3197.9% (15)16
Ours5097.7% (15)32
BWagh et al. [2019]4176N/A13
Wagh et al. [2021]101N/A13
Ours14493.6% (15)16
Ours24994.7% (15)32
CWagh et al. [2019]7188N/A13
Wagh et al. [2021]891N/A13
Tan et al. [2021]103694.0% (5)20
Ours34494.9% (5)16
Ours64393.8% (5)32
DMohassel and Rindal [2018]234N/AN/A
Ours4196.8% (15)16
Ours6896.8% (15)32
", + "bbox": [ + 183, + 449, + 815, + 762 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "294 7 Conclusions ", + "text_level": 1, + "bbox": [ + 140, + 800, + 307, + 818 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "295 We have presented an implementation of deep learning training purely in multi-party computation \n296 with extensive results on the accuracy. We have found that the lower precision of MPC computation \n297 increases the error considerably. We only have considered one particular implementation of more \n298 complex computation such as division and exponentiation, which are crucial to the learning process \n299 as part of softmax. Future work might consider different approximations of these building blocks. ", + "bbox": [ + 140, + 842, + 825, + 911 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "", + "text_level": 1, + "bbox": [ + 150, + 90, + 267, + 106 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "00 References \n301 N. Agrawal, A. S. Shamsabadi, M. J. Kusner, and A. Gascón. QUOTIENT: Two-party secure neural \n302 network training and prediction. In L. Cavallaro, J. Kinder, X. Wang, and J. Katz, editors, ACM \n303 CCS 2019, pages 1231–1247. ACM Press, Nov. 2019. doi: 10.1145/3319535.3339819. \n304 M. Aliasgari, M. Blanton, Y. Zhang, and A. Steele. Secure computation on floating point numbers. \n305 In NDSS 2013. The Internet Society, Feb. 2013. \n306 A. Aly and N. P. Smart. Benchmarking privacy preserving scientific operations. In R. H. Deng, \n307 V. Gauthier-Umaña, M. Ochoa, and M. Yung, editors, ACNS 19, volume 11464 of LNCS, pages \n308 509–529. Springer, Heidelberg, June 2019. doi: 10.1007/978-3-030-21568-2_25. \n309 T. Araki, J. Furukawa, Y. Lindell, A. Nof, and K. Ohara. High-throughput semi-honest secure \n310 three-party computation with an honest majority. In E. R. Weippl, S. Katzenbeisser, C. Kruegel, \n311 A. C. Myers, and S. Halevi, editors, ACM CCS 2016, pages 805–817. ACM Press, Oct. 2016. doi: \n312 10.1145/2976749.2978331. \n313 T. Araki, A. Barak, J. Furukawa, M. Keller, Y. Lindell, K. Ohara, and H. Tsuchida. Generalizing \n314 the SPDZ compiler for other protocols. In D. Lie, M. Mannan, M. Backes, and X. Wang, editors, \n315 ACM CCS 2018, pages 880–895. ACM Press, Oct. 2018. doi: 10.1145/3243734.3243854. \n316 M. Barni, C. Orlandi, and A. Piva. A privacy-preserving protocol for neural-network-based computa \n317 tion. In Proceedings of the 8th workshop on Multimedia and security, pages 146–151, 2006. \n318 J. C. Benaloh and J. Leichter. Generalized secret sharing and monotone functions. In S. Goldwasser, \n319 editor, CRYPTO’88, volume 403 of LNCS, pages 27–35. Springer, Heidelberg, Aug. 1990. doi: \n320 10.1007/0-387-34799-2_3. \n321 O. Catrina and A. Saxena. Secure computation with fixed-point numbers. In R. Sion, editor, FC \n322 2010, volume 6052 of LNCS, pages 35–50. Springer, Heidelberg, Jan. 2010. \n323 M. Dahl, J. Mancuso, Y. Dupis, B. Decoste, M. Giraud, I. Livingstone, J. Patriquin, and G. Uhma. \n324 Private machine learning in tensorflow using secure computation. CoRR, abs/1810.08130, 2018. \n325 URL http://arxiv.org/abs/1810.08130. \n326 A. Dalskov, D. Escudero, and M. Keller. Fantastic four: Honest-majority four-party secure com \n327 putation with malicious security. In 30th USENIX Security Symposium (USENIX Security 21), \n328 2021. \n329 A. P. K. Dalskov, D. Escudero, and M. Keller. Secure evaluation of quantized neural networks. \n330 PoPETs, 2020(4):355–375, Oct. 2020. doi: 10.2478/popets-2020-0077. \n331 I. Damgård, M. Fitzi, E. Kiltz, J. B. Nielsen, and T. Toft. Unconditionally secure constant-rounds \n332 multi-party computation for equality, comparison, bits and exponentiation. In S. Halevi and \n333 T. Rabin, editors, TCC 2006, volume 3876 of LNCS, pages 285–304. Springer, Heidelberg, Mar. \n334 2006. doi: 10.1007/11681878_15. \n335 I. Damgård, D. Escudero, T. K. Frederiksen, M. Keller, P. Scholl, and N. Volgushev. New primitives \n336 for actively-secure MPC over rings with applications to private machine learning. In 2019 IEEE \n337 Symposium on Security and Privacy, pages 1102–1120. IEEE Computer Society Press, May 2019. \n338 doi: 10.1109/SP.2019.00078. \n339 H. Eerikson, M. Keller, C. Orlandi, P. Pullonen, J. Puura, and M. Simkin. Use your brain! Arithmetic \n340 3PC for any modulus with active security. In Y. T. Kalai, A. D. Smith, and D. Wichs, editors, ITC \n341 2020, pages 5:1–5:24. Schloss Dagstuhl, June 2020. doi: 10.4230/LIPIcs.ITC.2020.5. \n342 X. Glorot and Y. Bengio. Understanding the difficulty of training deep feedforward neural networks. \n343 In Proceedings of the thirteenth international conference on artificial intelligence and statistics, \n344 pages 249–256. JMLR Workshop and Conference Proceedings, 2010. \n345 R. E. Goldschmidt. Applications of division by convergence. Master’s thesis, MIT, 1964. \n346 D. Gunning, A. Hannun, M. Ibrahim, B. Knott, L. van der Maaten, V. Reis, S. Sengupta, S. Venkatara \n347 man, and X. Zhou. Crypten: A new research tool for secure machine learning with pytorch, \n348 2019. \n349 I. Hubara, M. Courbariaux, D. Soudry, R. El-Yaniv, and Y. Bengio. Binarized neural networks. In \n350 Advances in Neural Information Processing Systems, volume 29. Curran Associates, Inc., 2016. \n351 M. Keller. MP-SPDZ: A versatile framework for multi-party computation. In J. Ligatti, X. Ou, \n352 J. Katz, and G. Vigna, editors, ACM CCS 20, pages 1575–1590. ACM Press, Nov. 2020. doi: \n353 10.1145/3372297.3417872. \n354 M. Keller and K. Sun. Effectiveness of MPC-friendly softmax replacement, 2020. \n355 V. Kolesnikov, A.-R. Sadeghi, and T. Schneider. A systematic approach to practically efficient general \n356 two-party secure function evaluation protocols and their modular design. Journal of Computer \n357 Security, 21(2):283–315, 2013. \n358 Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner. Gradient-based learning applied to document \n359 recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998. \n360 Y. LeCun, C. Cortes, and C. Burges. Mnist handwritten digit database. ATT Labs [Online]. Avail \n361 able: http: // yann. lecun. com/ exdb/ mnist , 2, 2010. Creative Commons Attribution-Share \n362 Alike 3.0 license, https://creativecommons.org/licenses/by-sa/3.0/. \n363 J. Liu, M. Juuti, Y. Lu, and N. Asokan. Oblivious neural network predictions via MiniONN \n364 transformations. In B. M. Thuraisingham, D. Evans, T. Malkin, and D. Xu, editors, ACM CCS \n365 2017, pages 619–631. ACM Press, Oct. / Nov. 2017. doi: 10.1145/3133956.3134056. \n366 W.-j. Lu, Y. Fang, Z. Huang, C. Hong, C. Chen, H. Qu, Y. Zhou, and K. Ren. Faster secure multiparty \n367 computation of adaptive gradient descent. In Proceedings of the 2020 Workshop on Privacy \n368 Preserving Machine Learning in Practice, PPMLP’20, page 47–49, New York, NY, USA, 2020. \n369 Association for Computing Machinery. ISBN 9781450380881. doi: 10.1145/3411501.3419427. \n370 URL https://doi.org/10.1145/3411501.3419427. \n371 P. Mohassel and P. Rindal. ABY3: A mixed protocol framework for machine learning. In D. Lie, \n372 M. Mannan, M. Backes, and X. Wang, editors, ACM CCS 2018, pages 35–52. ACM Press, Oct. \n373 2018. doi: 10.1145/3243734.3243760. \n374 P. Mohassel and Y. Zhang. SecureML: A system for scalable privacy-preserving machine learning. \n375 In 2017 IEEE Symposium on Security and Privacy, pages 19–38. IEEE Computer Society Press, \n376 May 2017. doi: 10.1109/SP.2017.12. \n377 V. Nair and G. E. Hinton. Rectified linear units improve Restricted Boltzmann machines. In Pro \n378 ceedings of the 27th International Conference on International Conference on Machine Learning, \n379 ICML’10, pages 807–814, 2010. \n380 D. L. Quoc, F. Gregor, S. Arnautov, R. Kunkel, P. Bhatotia, and C. Fetzer. securetf: A secure \n381 tensorflow framework. CoRR, abs/2101.08204, 2021. URL https://arxiv.org/abs/2101. \n382 08204. \n383 M. S. Riazi, C. Weinert, O. Tkachenko, E. M. Songhori, T. Schneider, and F. Koushanfar. Chameleon: \n384 A hybrid secure computation framework for machine learning applications. In J. Kim, G.-J. Ahn, \n385 S. Kim, Y. Kim, J. López, and T. Kim, editors, ASIACCS 18, pages 707–721. ACM Press, Apr. \n386 2018. \n387 D. Rotaru and T. Wood. MArBled circuits: Mixing arithmetic and Boolean circuits with active \n388 security. In F. Hao, S. Ruj, and S. Sen Gupta, editors, INDOCRYPT 2019, volume 11898 of LNCS, \n389 pages 227–249. Springer, Heidelberg, Dec. 2019. doi: 10.1007/978-3-030-35423-7_12. \n390 T. Ryffel, A. Trask, M. Dahl, B. Wagner, J. Mancuso, D. Rueckert, and J. Passerat-Palmbach. A \n391 generic framework for privacy preserving deep learning. CoRR, abs/1811.04017, 2018. URL \n392 http://arxiv.org/abs/1811.04017. ", + "bbox": [ + 140, + 94, + 830, + 919 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "", + "bbox": [ + 137, + 59, + 830, + 917 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "3 S. Tan, B. Knott, Y. Tian, and D. J. Wu. CryptGPU: Fast privacy-preserving machine learning on the GPU, 2021. \nS. Wagh, D. Gupta, and N. Chandran. SecureNN: 3-party secure computation for neural network training. PoPETs, 2019(3):26–49, July 2019. doi: 10.2478/popets-2019-0035. \nS. Wagh, S. Tople, F. Benhamouda, E. Kushilevitz, P. Mittal, and T. Rabin. Falcon: Honest-majority maliciously secure framework for private deep learning. PoPETs, 2021(1):188–208, Jan. 2021. doi: 10.2478/popets-2021-0011. ", + "bbox": [ + 155, + 90, + 826, + 209 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Checklist ", + "text_level": 1, + "bbox": [ + 143, + 234, + 254, + 251 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "1. For all authors... ", + "bbox": [ + 214, + 261, + 339, + 275 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "(a) Do the main claims made in the abstract and introduction accurately reflect the paper’s contributions and scope? [Yes] \n(b) Did you describe the limitations of your work? [Yes] See Section 7. \n(c) Did you discuss any potential negative societal impacts of your work? [No] We only work with MNIST, and we believe that the potential negative impacts associated to it are very low. \n(d) Have you read the ethics review guidelines and ensured that your paper conforms to them? [Yes] ", + "bbox": [ + 238, + 280, + 825, + 398 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "2. If you are including theoretical results... ", + "bbox": [ + 214, + 402, + 493, + 417 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "(a) Did you state the full set of assumptions of all theoretical results? [N/A] (b) Did you include complete proofs of all theoretical results? [N/A] ", + "bbox": [ + 238, + 421, + 738, + 453 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "3. If you ran experiments... ", + "bbox": [ + 212, + 457, + 393, + 472 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "(a) Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes] We include all code in the supplemental material, and the references include the URL for MNIST. \n(b) Did you specify all the training details (e.g., data splits, hyperparameters, how they were chosen)? [Yes] See Section 6. \n(c) Did you report error bars (e.g., with respect to the random seed after running experiments multiple times)? [Yes] Only in Figure 2. \n(d) Did you include the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [Yes] See Section 6. ", + "bbox": [ + 238, + 476, + 825, + 608 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "4. If you are using existing assets (e.g., code, data, models) or curating/releasing new assets... ", + "bbox": [ + 215, + 612, + 823, + 627 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "(a) If your work uses existing assets, did you cite the creators? [Yes] See Section 6. \n(b) Did you mention the license of the assets? [Yes] See the references and and Section 5. \n(c) Did you include any new assets either in the supplemental material or as a URL? [Yes] Our code is included the supplemental material. \n(d) Did you discuss whether and how consent was obtained from people whose data you’re using/curating? [N/A] \n(e) Did you discuss whether the data you are using/curating contains personally identifiable information or offensive content? [N/A] ", + "bbox": [ + 238, + 631, + 825, + 752 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "5. If you used crowdsourcing or conducted research with human subjects... ", + "bbox": [ + 214, + 756, + 705, + 771 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "(a) Did you include the full text of instructions given to participants and screenshots, if applicable? [N/A] \n(b) Did you describe any potential participant risks, with links to Institutional Review Board (IRB) approvals, if applicable? [N/A] \n(c) Did you include the estimated hourly wage paid to participants and the total amount spent on participant compensation? [N/A] ", + "bbox": [ + 238, + 775, + 825, + 864 + ], + "page_idx": 11 + } +] \ No newline at end of file diff --git a/parse/train/NiM9Q7Z95z/NiM9Q7Z95z_middle.json b/parse/train/NiM9Q7Z95z/NiM9Q7Z95z_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..77adfed8b91b393f3ba9d2ad3cec8f5967e2dbba --- /dev/null +++ b/parse/train/NiM9Q7Z95z/NiM9Q7Z95z_middle.json @@ -0,0 +1,38691 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 134, + 97, + 477, + 118 + ], + "lines": [ + { + "bbox": [ + 133, + 94, + 478, + 122 + ], + "spans": [ + { + "bbox": [ + 133, + 94, + 478, + 122 + ], + "score": 1.0, + "content": "Secure Quantized Training for Deep Learning", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 259, + 159, + 355, + 203 + ], + "lines": [ + { + "bbox": [ + 258, + 159, + 356, + 171 + ], + "spans": [ + { + "bbox": [ + 258, + 159, + 356, + 171 + ], + "score": 1.0, + "content": "Anonymous Author(s)", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 283, + 169, + 328, + 182 + ], + "spans": [ + { + "bbox": [ + 283, + 169, + 328, + 182 + ], + "score": 1.0, + "content": "Affiliation", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 287, + 180, + 324, + 193 + ], + "spans": [ + { + "bbox": [ + 287, + 180, + 324, + 193 + ], + "score": 1.0, + "content": "Address", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 290, + 193, + 320, + 203 + ], + "spans": [ + { + "bbox": [ + 290, + 193, + 320, + 203 + ], + "score": 1.0, + "content": "email", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 283, + 232, + 328, + 245 + ], + "lines": [ + { + "bbox": [ + 281, + 231, + 331, + 246 + ], + "spans": [ + { + "bbox": [ + 281, + 231, + 331, + 246 + ], + "score": 1.0, + "content": "Abstract", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 92, + 255, + 470, + 344 + ], + "lines": [ + { + "bbox": [ + 93, + 255, + 470, + 269 + ], + "spans": [ + { + "bbox": [ + 93, + 259, + 99, + 267 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 142, + 255, + 470, + 269 + ], + "score": 1.0, + "content": "We have implemented training of neural networks in secure multi-party computa-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 93, + 267, + 470, + 280 + ], + "spans": [ + { + "bbox": [ + 93, + 270, + 99, + 277 + ], + "score": 1.0, + "content": "2", + "type": "text" + }, + { + "bbox": [ + 142, + 267, + 470, + 280 + ], + "score": 1.0, + "content": "tion (MPC) using quantization commonly used in said setting. To the best of our", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 93, + 278, + 469, + 290 + ], + "spans": [ + { + "bbox": [ + 93, + 281, + 99, + 289 + ], + "score": 1.0, + "content": "3", + "type": "text" + }, + { + "bbox": [ + 142, + 278, + 469, + 290 + ], + "score": 1.0, + "content": "knowledge, we are the first to present training of MNIST purely implemented in", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 93, + 289, + 470, + 302 + ], + "spans": [ + { + "bbox": [ + 93, + 291, + 100, + 300 + ], + "score": 1.0, + "content": "4", + "type": "text" + }, + { + "bbox": [ + 141, + 289, + 470, + 302 + ], + "score": 1.0, + "content": "MPC that comes within one percent of accuracy of training using plaintext compu-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 93, + 300, + 470, + 312 + ], + "spans": [ + { + "bbox": [ + 93, + 302, + 100, + 311 + ], + "score": 1.0, + "content": "5", + "type": "text" + }, + { + "bbox": [ + 142, + 300, + 470, + 312 + ], + "score": 1.0, + "content": "tation. We found that training with MPC is possible, but it takes more epochs and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 93, + 309, + 470, + 324 + ], + "spans": [ + { + "bbox": [ + 93, + 313, + 100, + 322 + ], + "score": 1.0, + "content": "6", + "type": "text" + }, + { + "bbox": [ + 141, + 309, + 470, + 324 + ], + "score": 1.0, + "content": "achieves a lower accuracy than the usual CPU/GPU computation. More concretely,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 93, + 322, + 469, + 334 + ], + "spans": [ + { + "bbox": [ + 93, + 324, + 99, + 333 + ], + "score": 1.0, + "content": "7", + "type": "text" + }, + { + "bbox": [ + 142, + 322, + 441, + 334 + ], + "score": 1.0, + "content": "we have trained a network with two convolution and two dense layers to", + "type": "text" + }, + { + "bbox": [ + 441, + 322, + 469, + 332 + ], + "score": 0.84, + "content": "9 8 . 5 \\%", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 93, + 332, + 423, + 345 + ], + "spans": [ + { + "bbox": [ + 93, + 335, + 100, + 343 + ], + "score": 1.0, + "content": "8", + "type": "text" + }, + { + "bbox": [ + 141, + 332, + 423, + 345 + ], + "score": 1.0, + "content": "accuracy in 150 epochs. This took a day in our MPC implementation.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 9.5 + }, + { + "type": "title", + "bbox": [ + 93, + 363, + 191, + 376 + ], + "lines": [ + { + "bbox": [ + 90, + 362, + 192, + 379 + ], + "spans": [ + { + "bbox": [ + 90, + 362, + 192, + 379 + ], + "score": 1.0, + "content": "9 1 Introduction", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 91, + 387, + 505, + 431 + ], + "lines": [ + { + "bbox": [ + 90, + 388, + 505, + 400 + ], + "spans": [ + { + "bbox": [ + 90, + 389, + 100, + 399 + ], + "score": 1.0, + "content": "10", + "type": "text" + }, + { + "bbox": [ + 106, + 388, + 505, + 400 + ], + "score": 1.0, + "content": "Secure multi-party computation (MPC) is a cryptographic technique that allows a set of parties to", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 90, + 398, + 507, + 411 + ], + "spans": [ + { + "bbox": [ + 90, + 401, + 99, + 409 + ], + "score": 1.0, + "content": "11", + "type": "text" + }, + { + "bbox": [ + 105, + 398, + 507, + 411 + ], + "score": 1.0, + "content": "compute a public output on private inputs without revealing the inputs or any intermediate results.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 90, + 409, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 90, + 411, + 100, + 421 + ], + "score": 1.0, + "content": "12", + "type": "text" + }, + { + "bbox": [ + 105, + 409, + 505, + 422 + ], + "score": 1.0, + "content": "This makes it a potential solution to federated learning where the sample data stays private and only", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 90, + 420, + 321, + 433 + ], + "spans": [ + { + "bbox": [ + 90, + 422, + 100, + 432 + ], + "score": 1.0, + "content": "13", + "type": "text" + }, + { + "bbox": [ + 105, + 420, + 321, + 433 + ], + "score": 1.0, + "content": "the model or even only inference results are revealed.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16.5 + }, + { + "type": "text", + "bbox": [ + 91, + 436, + 505, + 491 + ], + "lines": [ + { + "bbox": [ + 90, + 437, + 506, + 448 + ], + "spans": [ + { + "bbox": [ + 90, + 439, + 99, + 447 + ], + "score": 1.0, + "content": "14", + "type": "text" + }, + { + "bbox": [ + 106, + 437, + 506, + 448 + ], + "score": 1.0, + "content": "Imagine a set of healthcare providers holding sensitive patient data. MPC allows them to collabo-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 90, + 447, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 90, + 450, + 99, + 460 + ], + "score": 1.0, + "content": "15", + "type": "text" + }, + { + "bbox": [ + 105, + 447, + 505, + 460 + ], + "score": 1.0, + "content": "ratively train a model. This model could then either be released or even kept private for inference", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 90, + 458, + 506, + 471 + ], + "spans": [ + { + "bbox": [ + 90, + 461, + 100, + 470 + ], + "score": 1.0, + "content": "16", + "type": "text" + }, + { + "bbox": [ + 105, + 458, + 506, + 471 + ], + "score": 1.0, + "content": "using MPC again. See Figure for an illustration. A more conceptual example is the well-known", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 90, + 469, + 506, + 482 + ], + "spans": [ + { + "bbox": [ + 90, + 471, + 100, + 481 + ], + "score": 1.0, + "content": "17", + "type": "text" + }, + { + "bbox": [ + 105, + 469, + 506, + 482 + ], + "score": 1.0, + "content": "millionaires’ problem where two people want to find out who is richer without revealing their wealth.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 90, + 480, + 471, + 493 + ], + "spans": [ + { + "bbox": [ + 90, + 483, + 100, + 492 + ], + "score": 1.0, + "content": "18", + "type": "text" + }, + { + "bbox": [ + 105, + 480, + 471, + 493 + ], + "score": 1.0, + "content": "There is clearly a difference between the one bit of information desired and the full figures.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 91, + 496, + 504, + 530 + ], + "lines": [ + { + "bbox": [ + 91, + 496, + 506, + 510 + ], + "spans": [ + { + "bbox": [ + 91, + 499, + 100, + 507 + ], + "score": 1.0, + "content": "19", + "type": "text" + }, + { + "bbox": [ + 103, + 496, + 506, + 510 + ], + "score": 1.0, + "content": "There has been a sustained interest in applying secure computation to machine learning and neural", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 90, + 506, + 506, + 520 + ], + "spans": [ + { + "bbox": [ + 90, + 510, + 100, + 518 + ], + "score": 1.0, + "content": "20", + "type": "text" + }, + { + "bbox": [ + 103, + 506, + 506, + 520 + ], + "score": 1.0, + "content": "networks going back to at least Barni et al. [2006]. More recent advantages in practical MPC have", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 90, + 517, + 389, + 532 + ], + "spans": [ + { + "bbox": [ + 90, + 521, + 100, + 529 + ], + "score": 1.0, + "content": "21", + "type": "text" + }, + { + "bbox": [ + 103, + 517, + 389, + 532 + ], + "score": 1.0, + "content": "led to an increased effort in implementing both inference and training.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 90, + 534, + 506, + 677 + ], + "lines": [ + { + "bbox": [ + 89, + 534, + 507, + 547 + ], + "spans": [ + { + "bbox": [ + 89, + 536, + 99, + 546 + ], + "score": 1.0, + "content": "22", + "type": "text" + }, + { + "bbox": [ + 105, + 534, + 507, + 547 + ], + "score": 1.0, + "content": "A number of works such as Mohassel and Zhang [2017], Mohassel and Rindal [2018], Wagh et al.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 89, + 545, + 507, + 560 + ], + "spans": [ + { + "bbox": [ + 89, + 547, + 99, + 558 + ], + "score": 1.0, + "content": "23", + "type": "text" + }, + { + "bbox": [ + 105, + 545, + 507, + 560 + ], + "score": 1.0, + "content": "[2019], Wagh et al. [2021] implement neural network training with MPC at least in parts. However,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 89, + 555, + 507, + 571 + ], + "spans": [ + { + "bbox": [ + 89, + 559, + 99, + 568 + ], + "score": 1.0, + "content": "24", + "type": "text" + }, + { + "bbox": [ + 104, + 555, + 259, + 571 + ], + "score": 1.0, + "content": "they either give accuracy figures below", + "type": "text" + }, + { + "bbox": [ + 259, + 557, + 279, + 567 + ], + "score": 0.88, + "content": "9 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 555, + 507, + 571 + ], + "score": 1.0, + "content": "or figures that have been obtained using plaintext training.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 89, + 567, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 89, + 569, + 100, + 579 + ], + "score": 1.0, + "content": "25", + "type": "text" + }, + { + "bbox": [ + 105, + 567, + 505, + 579 + ], + "score": 1.0, + "content": "For the latter case, the works do not clarify how close the computation for plaintext training matches", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 88, + 578, + 506, + 591 + ], + "spans": [ + { + "bbox": [ + 88, + 581, + 99, + 590 + ], + "score": 1.0, + "content": "26", + "type": "text" + }, + { + "bbox": [ + 105, + 578, + 506, + 591 + ], + "score": 1.0, + "content": "the lower precision and other differences in the MPC setting. Agrawal et al. [2019] claim a higher", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 89, + 589, + 506, + 602 + ], + "spans": [ + { + "bbox": [ + 89, + 592, + 100, + 601 + ], + "score": 1.0, + "content": "27", + "type": "text" + }, + { + "bbox": [ + 105, + 589, + 506, + 602 + ], + "score": 1.0, + "content": "accuracy in a comparable setting for a convolutional neural network with more channels than we", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 88, + 601, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 88, + 603, + 99, + 612 + ], + "score": 1.0, + "content": "28", + "type": "text" + }, + { + "bbox": [ + 105, + 601, + 505, + 613 + ], + "score": 1.0, + "content": "use. However, they have only implemented dense layers, and we achieve comparable accuracy to", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 89, + 612, + 506, + 623 + ], + "spans": [ + { + "bbox": [ + 89, + 613, + 100, + 623 + ], + "score": 1.0, + "content": "29", + "type": "text" + }, + { + "bbox": [ + 106, + 612, + 487, + 623 + ], + "score": 1.0, + "content": "them with only dense layers. All works use quantization in the sense that a fractional number", + "type": "text" + }, + { + "bbox": [ + 487, + 613, + 495, + 621 + ], + "score": 0.77, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 612, + 506, + 623 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 89, + 621, + 506, + 636 + ], + "spans": [ + { + "bbox": [ + 89, + 624, + 100, + 634 + ], + "score": 1.0, + "content": "30", + "type": "text" + }, + { + "bbox": [ + 104, + 621, + 166, + 636 + ], + "score": 1.0, + "content": "represented as", + "type": "text" + }, + { + "bbox": [ + 167, + 622, + 205, + 634 + ], + "score": 0.93, + "content": "\\lfloor x \\cdot 2 ^ { - f } \\rceil", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 621, + 506, + 636 + ], + "score": 1.0, + "content": ". This makes addition considerably faster in the secure computation setting", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 88, + 633, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 88, + 635, + 99, + 645 + ], + "score": 1.0, + "content": "31", + "type": "text" + }, + { + "bbox": [ + 105, + 633, + 505, + 645 + ], + "score": 1.0, + "content": "because it reduces to integer addition. Furthermore, some of the works suggest to replace the softmax", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 89, + 644, + 506, + 656 + ], + "spans": [ + { + "bbox": [ + 89, + 646, + 100, + 656 + ], + "score": 1.0, + "content": "32", + "type": "text" + }, + { + "bbox": [ + 105, + 644, + 506, + 656 + ], + "score": 1.0, + "content": "function that uses exponentiation with a ReLU-based replacement. Keller and Sun [2020] have found", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 89, + 655, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 89, + 657, + 100, + 666 + ], + "score": 1.0, + "content": "33", + "type": "text" + }, + { + "bbox": [ + 105, + 655, + 506, + 668 + ], + "score": 1.0, + "content": "that this softmax replacement deteriorates accuracy in dense neural networks to the extent that it does", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 89, + 666, + 243, + 678 + ], + "spans": [ + { + "bbox": [ + 89, + 668, + 100, + 677 + ], + "score": 1.0, + "content": "34", + "type": "text" + }, + { + "bbox": [ + 105, + 666, + 243, + 678 + ], + "score": 1.0, + "content": "not justify the performance gains.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 91, + 681, + 504, + 715 + ], + "lines": [ + { + "bbox": [ + 89, + 681, + 505, + 695 + ], + "spans": [ + { + "bbox": [ + 89, + 684, + 99, + 693 + ], + "score": 1.0, + "content": "35", + "type": "text" + }, + { + "bbox": [ + 105, + 681, + 505, + 695 + ], + "score": 1.0, + "content": "The concurrent work of Tan et al. [2021] gives some figures on the learning curve when run using", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 90, + 693, + 505, + 705 + ], + "spans": [ + { + "bbox": [ + 90, + 696, + 99, + 704 + ], + "score": 1.0, + "content": "36", + "type": "text" + }, + { + "bbox": [ + 104, + 693, + 484, + 705 + ], + "score": 1.0, + "content": "secure computation. However, they stop at five epochs for MNIST training where they achieve", + "type": "text" + }, + { + "bbox": [ + 485, + 693, + 505, + 704 + ], + "score": 0.85, + "content": "94 \\%", + "type": "inline_equation" + } + ], + "index": 41 + }, + { + "bbox": [ + 90, + 704, + 505, + 716 + ], + "spans": [ + { + "bbox": [ + 90, + 707, + 100, + 716 + ], + "score": 1.0, + "content": "37", + "type": "text" + }, + { + "bbox": [ + 105, + 704, + 359, + 716 + ], + "score": 1.0, + "content": "accuracy whereas we present the figures up to 150 epochs and", + "type": "text" + }, + { + "bbox": [ + 359, + 704, + 387, + 714 + ], + "score": 0.84, + "content": "9 8 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 704, + 505, + 716 + ], + "score": 1.0, + "content": "accuracy. Furthermore, their", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 102, + 731, + 499, + 742 + ], + "lines": [ + { + "bbox": [ + 106, + 730, + 500, + 743 + ], + "spans": [ + { + "bbox": [ + 106, + 730, + 500, + 743 + ], + "score": 1.0, + "content": "Submitted to 35th Conference on Neural Information Processing Systems (NeurIPS 2021). Do not distribute.", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 134, + 97, + 477, + 118 + ], + "lines": [ + { + "bbox": [ + 133, + 94, + 478, + 122 + ], + "spans": [ + { + "bbox": [ + 133, + 94, + 478, + 122 + ], + "score": 1.0, + "content": "Secure Quantized Training for Deep Learning", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "list", + "bbox": [ + 259, + 159, + 355, + 203 + ], + "lines": [ + { + "bbox": [ + 258, + 159, + 356, + 171 + ], + "spans": [ + { + "bbox": [ + 258, + 159, + 356, + 171 + ], + "score": 1.0, + "content": "Anonymous Author(s)", + "type": "text" + } + ], + "index": 1, + "is_list_start_line": true + }, + { + "bbox": [ + 283, + 169, + 328, + 182 + ], + "spans": [ + { + "bbox": [ + 283, + 169, + 328, + 182 + ], + "score": 1.0, + "content": "Affiliation", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 287, + 180, + 324, + 193 + ], + "spans": [ + { + "bbox": [ + 287, + 180, + 324, + 193 + ], + "score": 1.0, + "content": "Address", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 290, + 193, + 320, + 203 + ], + "spans": [ + { + "bbox": [ + 290, + 193, + 320, + 203 + ], + "score": 1.0, + "content": "email", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true + } + ], + "index": 2.5, + "bbox_fs": [ + 258, + 159, + 356, + 203 + ] + }, + { + "type": "title", + "bbox": [ + 283, + 232, + 328, + 245 + ], + "lines": [ + { + "bbox": [ + 281, + 231, + 331, + 246 + ], + "spans": [ + { + "bbox": [ + 281, + 231, + 331, + 246 + ], + "score": 1.0, + "content": "Abstract", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "index", + "bbox": [ + 92, + 255, + 470, + 344 + ], + "lines": [ + { + "bbox": [ + 93, + 255, + 470, + 269 + ], + "spans": [ + { + "bbox": [ + 93, + 259, + 99, + 267 + ], + "score": 1.0, + "content": "1", + "type": "text" + }, + { + "bbox": [ + 142, + 255, + 470, + 269 + ], + "score": 1.0, + "content": "We have implemented training of neural networks in secure multi-party computa-", + "type": "text" + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 93, + 267, + 470, + 280 + ], + "spans": [ + { + "bbox": [ + 93, + 270, + 99, + 277 + ], + "score": 1.0, + "content": "2", + "type": "text" + }, + { + "bbox": [ + 142, + 267, + 470, + 280 + ], + "score": 1.0, + "content": "tion (MPC) using quantization commonly used in said setting. To the best of our", + "type": "text" + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 93, + 278, + 469, + 290 + ], + "spans": [ + { + "bbox": [ + 93, + 281, + 99, + 289 + ], + "score": 1.0, + "content": "3", + "type": "text" + }, + { + "bbox": [ + 142, + 278, + 469, + 290 + ], + "score": 1.0, + "content": "knowledge, we are the first to present training of MNIST purely implemented in", + "type": "text" + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 93, + 289, + 470, + 302 + ], + "spans": [ + { + "bbox": [ + 93, + 291, + 100, + 300 + ], + "score": 1.0, + "content": "4", + "type": "text" + }, + { + "bbox": [ + 141, + 289, + 470, + 302 + ], + "score": 1.0, + "content": "MPC that comes within one percent of accuracy of training using plaintext compu-", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 93, + 300, + 470, + 312 + ], + "spans": [ + { + "bbox": [ + 93, + 302, + 100, + 311 + ], + "score": 1.0, + "content": "5", + "type": "text" + }, + { + "bbox": [ + 142, + 300, + 470, + 312 + ], + "score": 1.0, + "content": "tation. We found that training with MPC is possible, but it takes more epochs and", + "type": "text" + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 93, + 309, + 470, + 324 + ], + "spans": [ + { + "bbox": [ + 93, + 313, + 100, + 322 + ], + "score": 1.0, + "content": "6", + "type": "text" + }, + { + "bbox": [ + 141, + 309, + 470, + 324 + ], + "score": 1.0, + "content": "achieves a lower accuracy than the usual CPU/GPU computation. More concretely,", + "type": "text" + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 93, + 322, + 469, + 334 + ], + "spans": [ + { + "bbox": [ + 93, + 324, + 99, + 333 + ], + "score": 1.0, + "content": "7", + "type": "text" + }, + { + "bbox": [ + 142, + 322, + 441, + 334 + ], + "score": 1.0, + "content": "we have trained a network with two convolution and two dense layers to", + "type": "text" + }, + { + "bbox": [ + 441, + 322, + 469, + 332 + ], + "score": 0.84, + "content": "9 8 . 5 \\%", + "type": "inline_equation" + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 93, + 332, + 423, + 345 + ], + "spans": [ + { + "bbox": [ + 93, + 335, + 100, + 343 + ], + "score": 1.0, + "content": "8", + "type": "text" + }, + { + "bbox": [ + 141, + 332, + 423, + 345 + ], + "score": 1.0, + "content": "accuracy in 150 epochs. This took a day in our MPC implementation.", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true + } + ], + "index": 9.5, + "bbox_fs": [ + 93, + 255, + 470, + 345 + ] + }, + { + "type": "title", + "bbox": [ + 93, + 363, + 191, + 376 + ], + "lines": [ + { + "bbox": [ + 90, + 362, + 192, + 379 + ], + "spans": [ + { + "bbox": [ + 90, + 362, + 192, + 379 + ], + "score": 1.0, + "content": "9 1 Introduction", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "index", + "bbox": [ + 91, + 387, + 505, + 431 + ], + "lines": [ + { + "bbox": [ + 90, + 388, + 505, + 400 + ], + "spans": [ + { + "bbox": [ + 90, + 389, + 100, + 399 + ], + "score": 1.0, + "content": "10", + "type": "text" + }, + { + "bbox": [ + 106, + 388, + 505, + 400 + ], + "score": 1.0, + "content": "Secure multi-party computation (MPC) is a cryptographic technique that allows a set of parties to", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 398, + 507, + 411 + ], + "spans": [ + { + "bbox": [ + 90, + 401, + 99, + 409 + ], + "score": 1.0, + "content": "11", + "type": "text" + }, + { + "bbox": [ + 105, + 398, + 507, + 411 + ], + "score": 1.0, + "content": "compute a public output on private inputs without revealing the inputs or any intermediate results.", + "type": "text" + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 409, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 90, + 411, + 100, + 421 + ], + "score": 1.0, + "content": "12", + "type": "text" + }, + { + "bbox": [ + 105, + 409, + 505, + 422 + ], + "score": 1.0, + "content": "This makes it a potential solution to federated learning where the sample data stays private and only", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 420, + 321, + 433 + ], + "spans": [ + { + "bbox": [ + 90, + 422, + 100, + 432 + ], + "score": 1.0, + "content": "13", + "type": "text" + }, + { + "bbox": [ + 105, + 420, + 321, + 433 + ], + "score": 1.0, + "content": "the model or even only inference results are revealed.", + "type": "text" + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 437, + 506, + 448 + ], + "spans": [ + { + "bbox": [ + 90, + 439, + 99, + 447 + ], + "score": 1.0, + "content": "14", + "type": "text" + }, + { + "bbox": [ + 106, + 437, + 506, + 448 + ], + "score": 1.0, + "content": "Imagine a set of healthcare providers holding sensitive patient data. MPC allows them to collabo-", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 447, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 90, + 450, + 99, + 460 + ], + "score": 1.0, + "content": "15", + "type": "text" + }, + { + "bbox": [ + 105, + 447, + 505, + 460 + ], + "score": 1.0, + "content": "ratively train a model. This model could then either be released or even kept private for inference", + "type": "text" + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 458, + 506, + 471 + ], + "spans": [ + { + "bbox": [ + 90, + 461, + 100, + 470 + ], + "score": 1.0, + "content": "16", + "type": "text" + }, + { + "bbox": [ + 105, + 458, + 506, + 471 + ], + "score": 1.0, + "content": "using MPC again. See Figure for an illustration. A more conceptual example is the well-known", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 469, + 506, + 482 + ], + "spans": [ + { + "bbox": [ + 90, + 471, + 100, + 481 + ], + "score": 1.0, + "content": "17", + "type": "text" + }, + { + "bbox": [ + 105, + 469, + 506, + 482 + ], + "score": 1.0, + "content": "millionaires’ problem where two people want to find out who is richer without revealing their wealth.", + "type": "text" + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 480, + 471, + 493 + ], + "spans": [ + { + "bbox": [ + 90, + 483, + 100, + 492 + ], + "score": 1.0, + "content": "18", + "type": "text" + }, + { + "bbox": [ + 105, + 480, + 471, + 493 + ], + "score": 1.0, + "content": "There is clearly a difference between the one bit of information desired and the full figures.", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 91, + 496, + 506, + 510 + ], + "spans": [ + { + "bbox": [ + 91, + 499, + 100, + 507 + ], + "score": 1.0, + "content": "19", + "type": "text" + }, + { + "bbox": [ + 103, + 496, + 506, + 510 + ], + "score": 1.0, + "content": "There has been a sustained interest in applying secure computation to machine learning and neural", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 506, + 506, + 520 + ], + "spans": [ + { + "bbox": [ + 90, + 510, + 100, + 518 + ], + "score": 1.0, + "content": "20", + "type": "text" + }, + { + "bbox": [ + 103, + 506, + 506, + 520 + ], + "score": 1.0, + "content": "networks going back to at least Barni et al. [2006]. More recent advantages in practical MPC have", + "type": "text" + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 517, + 389, + 532 + ], + "spans": [ + { + "bbox": [ + 90, + 521, + 100, + 529 + ], + "score": 1.0, + "content": "21", + "type": "text" + }, + { + "bbox": [ + 103, + 517, + 389, + 532 + ], + "score": 1.0, + "content": "led to an increased effort in implementing both inference and training.", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 534, + 507, + 547 + ], + "spans": [ + { + "bbox": [ + 89, + 536, + 99, + 546 + ], + "score": 1.0, + "content": "22", + "type": "text" + }, + { + "bbox": [ + 105, + 534, + 507, + 547 + ], + "score": 1.0, + "content": "A number of works such as Mohassel and Zhang [2017], Mohassel and Rindal [2018], Wagh et al.", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 545, + 507, + 560 + ], + "spans": [ + { + "bbox": [ + 89, + 547, + 99, + 558 + ], + "score": 1.0, + "content": "23", + "type": "text" + }, + { + "bbox": [ + 105, + 545, + 507, + 560 + ], + "score": 1.0, + "content": "[2019], Wagh et al. [2021] implement neural network training with MPC at least in parts. However,", + "type": "text" + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 555, + 507, + 571 + ], + "spans": [ + { + "bbox": [ + 89, + 559, + 99, + 568 + ], + "score": 1.0, + "content": "24", + "type": "text" + }, + { + "bbox": [ + 104, + 555, + 259, + 571 + ], + "score": 1.0, + "content": "they either give accuracy figures below", + "type": "text" + }, + { + "bbox": [ + 259, + 557, + 279, + 567 + ], + "score": 0.88, + "content": "9 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 555, + 507, + 571 + ], + "score": 1.0, + "content": "or figures that have been obtained using plaintext training.", + "type": "text" + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 567, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 89, + 569, + 100, + 579 + ], + "score": 1.0, + "content": "25", + "type": "text" + }, + { + "bbox": [ + 105, + 567, + 505, + 579 + ], + "score": 1.0, + "content": "For the latter case, the works do not clarify how close the computation for plaintext training matches", + "type": "text" + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 578, + 506, + 591 + ], + "spans": [ + { + "bbox": [ + 88, + 581, + 99, + 590 + ], + "score": 1.0, + "content": "26", + "type": "text" + }, + { + "bbox": [ + 105, + 578, + 506, + 591 + ], + "score": 1.0, + "content": "the lower precision and other differences in the MPC setting. Agrawal et al. [2019] claim a higher", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 589, + 506, + 602 + ], + "spans": [ + { + "bbox": [ + 89, + 592, + 100, + 601 + ], + "score": 1.0, + "content": "27", + "type": "text" + }, + { + "bbox": [ + 105, + 589, + 506, + 602 + ], + "score": 1.0, + "content": "accuracy in a comparable setting for a convolutional neural network with more channels than we", + "type": "text" + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 601, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 88, + 603, + 99, + 612 + ], + "score": 1.0, + "content": "28", + "type": "text" + }, + { + "bbox": [ + 105, + 601, + 505, + 613 + ], + "score": 1.0, + "content": "use. However, they have only implemented dense layers, and we achieve comparable accuracy to", + "type": "text" + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 612, + 506, + 623 + ], + "spans": [ + { + "bbox": [ + 89, + 613, + 100, + 623 + ], + "score": 1.0, + "content": "29", + "type": "text" + }, + { + "bbox": [ + 106, + 612, + 487, + 623 + ], + "score": 1.0, + "content": "them with only dense layers. All works use quantization in the sense that a fractional number", + "type": "text" + }, + { + "bbox": [ + 487, + 613, + 495, + 621 + ], + "score": 0.77, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 612, + 506, + 623 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 621, + 506, + 636 + ], + "spans": [ + { + "bbox": [ + 89, + 624, + 100, + 634 + ], + "score": 1.0, + "content": "30", + "type": "text" + }, + { + "bbox": [ + 104, + 621, + 166, + 636 + ], + "score": 1.0, + "content": "represented as", + "type": "text" + }, + { + "bbox": [ + 167, + 622, + 205, + 634 + ], + "score": 0.93, + "content": "\\lfloor x \\cdot 2 ^ { - f } \\rceil", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 621, + 506, + 636 + ], + "score": 1.0, + "content": ". This makes addition considerably faster in the secure computation setting", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 633, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 88, + 635, + 99, + 645 + ], + "score": 1.0, + "content": "31", + "type": "text" + }, + { + "bbox": [ + 105, + 633, + 505, + 645 + ], + "score": 1.0, + "content": "because it reduces to integer addition. Furthermore, some of the works suggest to replace the softmax", + "type": "text" + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 644, + 506, + 656 + ], + "spans": [ + { + "bbox": [ + 89, + 646, + 100, + 656 + ], + "score": 1.0, + "content": "32", + "type": "text" + }, + { + "bbox": [ + 105, + 644, + 506, + 656 + ], + "score": 1.0, + "content": "function that uses exponentiation with a ReLU-based replacement. Keller and Sun [2020] have found", + "type": "text" + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 655, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 89, + 657, + 100, + 666 + ], + "score": 1.0, + "content": "33", + "type": "text" + }, + { + "bbox": [ + 105, + 655, + 506, + 668 + ], + "score": 1.0, + "content": "that this softmax replacement deteriorates accuracy in dense neural networks to the extent that it does", + "type": "text" + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 666, + 243, + 678 + ], + "spans": [ + { + "bbox": [ + 89, + 668, + 100, + 677 + ], + "score": 1.0, + "content": "34", + "type": "text" + }, + { + "bbox": [ + 105, + 666, + 243, + 678 + ], + "score": 1.0, + "content": "not justify the performance gains.", + "type": "text" + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 681, + 505, + 695 + ], + "spans": [ + { + "bbox": [ + 89, + 684, + 99, + 693 + ], + "score": 1.0, + "content": "35", + "type": "text" + }, + { + "bbox": [ + 105, + 681, + 505, + 695 + ], + "score": 1.0, + "content": "The concurrent work of Tan et al. [2021] gives some figures on the learning curve when run using", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 693, + 505, + 705 + ], + "spans": [ + { + "bbox": [ + 90, + 696, + 99, + 704 + ], + "score": 1.0, + "content": "36", + "type": "text" + }, + { + "bbox": [ + 104, + 693, + 484, + 705 + ], + "score": 1.0, + "content": "secure computation. However, they stop at five epochs for MNIST training where they achieve", + "type": "text" + }, + { + "bbox": [ + 485, + 693, + 505, + 704 + ], + "score": 0.85, + "content": "94 \\%", + "type": "inline_equation" + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 704, + 505, + 716 + ], + "spans": [ + { + "bbox": [ + 90, + 707, + 100, + 716 + ], + "score": 1.0, + "content": "37", + "type": "text" + }, + { + "bbox": [ + 105, + 704, + 359, + 716 + ], + "score": 1.0, + "content": "accuracy whereas we present the figures up to 150 epochs and", + "type": "text" + }, + { + "bbox": [ + 359, + 704, + 387, + 714 + ], + "score": 0.84, + "content": "9 8 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 704, + 505, + 716 + ], + "score": 1.0, + "content": "accuracy. Furthermore, their", + "type": "text" + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 257, + 505, + 268 + ], + "spans": [ + { + "bbox": [ + 89, + 259, + 100, + 268 + ], + "score": 1.0, + "content": "38", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 257, + 505, + 268 + ], + "score": 1.0, + "content": "choice of fixed-point precision 20 is considerably below 32, which we found to be optimal. We also", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 268, + 505, + 280 + ], + "spans": [ + { + "bbox": [ + 89, + 270, + 100, + 279 + ], + "score": 1.0, + "content": "39", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 268, + 244, + 280 + ], + "score": 1.0, + "content": "found that our implementation is", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 244, + 268, + 264, + 278 + ], + "score": 0.87, + "content": "40 \\%", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 265, + 268, + 505, + 280 + ], + "score": 1.0, + "content": "faster than theirs. Note that we use the CPU of one AWS", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 277, + 506, + 292 + ], + "spans": [ + { + "bbox": [ + 89, + 281, + 100, + 290 + ], + "score": 1.0, + "content": "40", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 277, + 506, + 292 + ], + "score": 1.0, + "content": "c5.9xlarge instance per party whereas Tan et al. use one NVIDIA Tesla V100 GPU per party. We", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 288, + 507, + 303 + ], + "spans": [ + { + "bbox": [ + 89, + 292, + 99, + 301 + ], + "score": 1.0, + "content": "41", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 288, + 507, + 303 + ], + "score": 1.0, + "content": "believe this somewhat counter-intuitive result comes from MPC heavily relying on communication,", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 300, + 386, + 313 + ], + "spans": [ + { + "bbox": [ + 89, + 303, + 100, + 312 + ], + "score": 1.0, + "content": "42", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 300, + 386, + 313 + ], + "score": 1.0, + "content": "which is an aspect where GPUs do not have an advantage over CPUs.", + "type": "text", + "cross_page": true + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 316, + 506, + 330 + ], + "spans": [ + { + "bbox": [ + 90, + 319, + 99, + 328 + ], + "score": 1.0, + "content": "43", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 316, + 506, + 330 + ], + "score": 1.0, + "content": "In this paper, we present an extensible framework for implementing deep learning training based on", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 327, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 89, + 330, + 99, + 339 + ], + "score": 1.0, + "content": "44", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 327, + 505, + 340 + ], + "score": 1.0, + "content": "MP-SPDZ by Keller [2020], a framework for multi-party computation.1 Similar to TensorFlow and", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 339, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 89, + 341, + 100, + 351 + ], + "score": 1.0, + "content": "45", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 339, + 505, + 351 + ], + "score": 1.0, + "content": "PyTorch, our approach allows representing deep learning models as succession of layers. We then", + "type": "text", + "cross_page": true + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 350, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 89, + 352, + 99, + 361 + ], + "score": 1.0, + "content": "46", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 350, + 505, + 362 + ], + "score": 1.0, + "content": "use this implementation to obtain accuracy figures for MNIST training by utilizing the MP-SPDZ", + "type": "text", + "cross_page": true + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 359, + 506, + 374 + ], + "spans": [ + { + "bbox": [ + 89, + 363, + 100, + 372 + ], + "score": 1.0, + "content": "47", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 359, + 506, + 374 + ], + "score": 1.0, + "content": "emulator, which allows to run the plaintext equivalent of secure computation, that is, the same", + "type": "text", + "cross_page": true + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 371, + 505, + 384 + ], + "spans": [ + { + "bbox": [ + 89, + 374, + 99, + 383 + ], + "score": 1.0, + "content": "48", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 371, + 505, + 384 + ], + "score": 1.0, + "content": "algorithms with the same precision. Finally, we run one of the most promising instantiation in real", + "type": "text", + "cross_page": true + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 383, + 472, + 395 + ], + "spans": [ + { + "bbox": [ + 89, + 385, + 99, + 394 + ], + "score": 1.0, + "content": "49", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 383, + 472, + 395 + ], + "score": 1.0, + "content": "secure computation in order to benchmark it confirm the result from the plaintext emulator.", + "type": "text", + "cross_page": true + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 398, + 506, + 412 + ], + "spans": [ + { + "bbox": [ + 89, + 401, + 99, + 410 + ], + "score": 1.0, + "content": "50", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 398, + 506, + 412 + ], + "score": 1.0, + "content": "There are a number of projects that integrate secure computation directly into popular machine", + "type": "text", + "cross_page": true + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 410, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 89, + 412, + 99, + 421 + ], + "score": 1.0, + "content": "51", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 410, + 505, + 423 + ], + "score": 1.0, + "content": "learning frameworks such as CrypTen by Gunning et al. [2019], PySyft by Ryffel et al. [2018], and", + "type": "text", + "cross_page": true + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 421, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 89, + 423, + 100, + 432 + ], + "score": 1.0, + "content": "52", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 421, + 506, + 433 + ], + "score": 1.0, + "content": "TF Encrypted by Dahl et al. [2018]. Our approach differs from all of them by running the protocol as", + "type": "text", + "cross_page": true + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 430, + 506, + 445 + ], + "spans": [ + { + "bbox": [ + 89, + 434, + 99, + 443 + ], + "score": 1.0, + "content": "53", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 430, + 261, + 445 + ], + "score": 1.0, + "content": "native CPU code (implemented using", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 262, + 432, + 282, + 442 + ], + "score": 0.84, + "content": "\\mathrm { C } { + + }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 282, + 430, + 506, + 445 + ], + "score": 1.0, + "content": "). This allows for much faster execution. For example,", + "type": "text", + "cross_page": true + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 442, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 89, + 444, + 100, + 454 + ], + "score": 1.0, + "content": "54", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 442, + 506, + 455 + ], + "score": 1.0, + "content": "CrypTen provides an MNIST training example (mpc_autograd_cnn) that takes over one minute", + "type": "text", + "cross_page": true + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 454, + 506, + 466 + ], + "spans": [ + { + "bbox": [ + 89, + 456, + 100, + 465 + ], + "score": 1.0, + "content": "55", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 454, + 506, + 466 + ], + "score": 1.0, + "content": "to run one epoch with 100 samples on one machine. In comparison, our implementation takes 11", + "type": "text", + "cross_page": true + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 464, + 369, + 477 + ], + "spans": [ + { + "bbox": [ + 89, + 467, + 100, + 476 + ], + "score": 1.0, + "content": "56", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 464, + 369, + 477 + ], + "score": 1.0, + "content": "minutes to run one epoch with the full dataset of 60,000 samples.", + "type": "text", + "cross_page": true + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 480, + 506, + 494 + ], + "spans": [ + { + "bbox": [ + 90, + 483, + 100, + 492 + ], + "score": 1.0, + "content": "57", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 480, + 506, + 494 + ], + "score": 1.0, + "content": "Another line of work (e.g., Quoc et al. [2021]) uses trusted execution environments that provide", + "type": "text", + "cross_page": true + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 491, + 506, + 505 + ], + "spans": [ + { + "bbox": [ + 90, + 494, + 100, + 502 + ], + "score": 1.0, + "content": "58", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 491, + 506, + 505 + ], + "score": 1.0, + "content": "computation outside the reach of the operating system. This is a different security model to multi-party", + "type": "text", + "cross_page": true + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 502, + 429, + 515 + ], + "spans": [ + { + "bbox": [ + 89, + 505, + 100, + 514 + ], + "score": 1.0, + "content": "59", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 502, + 429, + 515 + ], + "score": 1.0, + "content": "computation that works with distributing the information among several entities.", + "type": "text", + "cross_page": true + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 519, + 506, + 531 + ], + "spans": [ + { + "bbox": [ + 89, + 521, + 100, + 531 + ], + "score": 1.0, + "content": "60", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 519, + 506, + 531 + ], + "score": 1.0, + "content": "The paper is structured as follows: After introducing the basics of the protocol we use in Section 2,", + "type": "text", + "cross_page": true + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 528, + 506, + 543 + ], + "spans": [ + { + "bbox": [ + 89, + 532, + 100, + 541 + ], + "score": 1.0, + "content": "61", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 528, + 506, + 543 + ], + "score": 1.0, + "content": "we will explain the mathematical building blocks in Section 3 and their use in the context of deep", + "type": "text", + "cross_page": true + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 540, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 89, + 542, + 100, + 552 + ], + "score": 1.0, + "content": "62", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 540, + 505, + 552 + ], + "score": 1.0, + "content": "learning in Section 4. Finally, we will present our implementation in Section 5 and our experimental", + "type": "text", + "cross_page": true + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 552, + 236, + 564 + ], + "spans": [ + { + "bbox": [ + 89, + 554, + 100, + 563 + ], + "score": 1.0, + "content": "63", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 552, + 236, + 564 + ], + "score": 1.0, + "content": "results for MNIST in Section 6.", + "type": "text", + "cross_page": true + } + ], + "index": 32, + "is_list_start_line": true + } + ], + "index": 16.5, + "bbox_fs": [ + 90, + 388, + 507, + 433 + ] + }, + { + "type": "index", + "bbox": [ + 91, + 436, + 505, + 491 + ], + "lines": [], + "index": 21, + "bbox_fs": [ + 90, + 437, + 506, + 493 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 91, + 496, + 504, + 530 + ], + "lines": [], + "index": 25, + "bbox_fs": [ + 90, + 496, + 506, + 532 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 90, + 534, + 506, + 677 + ], + "lines": [], + "index": 33, + "bbox_fs": [ + 88, + 534, + 507, + 678 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 91, + 681, + 504, + 715 + ], + "lines": [], + "index": 41, + "bbox_fs": [ + 89, + 681, + 505, + 716 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 156, + 74, + 454, + 179 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 156, + 74, + 454, + 179 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 156, + 74, + 454, + 179 + ], + "spans": [ + { + "bbox": [ + 156, + 74, + 454, + 179 + ], + "score": 0.927, + "type": "image", + "image_path": "bb0f136189595323f73d65ddc0b25170a7b7ffb1f9107b9e37395cb5ee7bab1c.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 156, + 74, + 454, + 109.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 156, + 109.0, + 454, + 144.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 156, + 144.0, + 454, + 179.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 191, + 505, + 236 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 191, + 506, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 506, + 204 + ], + "score": 1.0, + "content": "Figure 1: Outsourced computation: Data holders (on the left) secret-share their data to a number of", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 202, + 506, + 215 + ], + "spans": [ + { + "bbox": [ + 105, + 202, + 506, + 215 + ], + "score": 1.0, + "content": "computing parties (on the right), who then return the desired the result (e.g., a model or inference", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 214, + 505, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 505, + 227 + ], + "score": 1.0, + "content": "results on further queries). All communication except outputs are secret-shared and thus secure if no", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 225, + 232, + 236 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 232, + 236 + ], + "score": 1.0, + "content": "two computing parties collude.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 90, + 256, + 505, + 311 + ], + "lines": [ + { + "bbox": [ + 89, + 257, + 505, + 268 + ], + "spans": [ + { + "bbox": [ + 89, + 259, + 100, + 268 + ], + "score": 1.0, + "content": "38", + "type": "text" + }, + { + "bbox": [ + 106, + 257, + 505, + 268 + ], + "score": 1.0, + "content": "choice of fixed-point precision 20 is considerably below 32, which we found to be optimal. We also", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 89, + 268, + 505, + 280 + ], + "spans": [ + { + "bbox": [ + 89, + 270, + 100, + 279 + ], + "score": 1.0, + "content": "39", + "type": "text" + }, + { + "bbox": [ + 106, + 268, + 244, + 280 + ], + "score": 1.0, + "content": "found that our implementation is", + "type": "text" + }, + { + "bbox": [ + 244, + 268, + 264, + 278 + ], + "score": 0.87, + "content": "40 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 265, + 268, + 505, + 280 + ], + "score": 1.0, + "content": "faster than theirs. Note that we use the CPU of one AWS", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 89, + 277, + 506, + 292 + ], + "spans": [ + { + "bbox": [ + 89, + 281, + 100, + 290 + ], + "score": 1.0, + "content": "40", + "type": "text" + }, + { + "bbox": [ + 105, + 277, + 506, + 292 + ], + "score": 1.0, + "content": "c5.9xlarge instance per party whereas Tan et al. use one NVIDIA Tesla V100 GPU per party. We", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 89, + 288, + 507, + 303 + ], + "spans": [ + { + "bbox": [ + 89, + 292, + 99, + 301 + ], + "score": 1.0, + "content": "41", + "type": "text" + }, + { + "bbox": [ + 105, + 288, + 507, + 303 + ], + "score": 1.0, + "content": "believe this somewhat counter-intuitive result comes from MPC heavily relying on communication,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 89, + 300, + 386, + 313 + ], + "spans": [ + { + "bbox": [ + 89, + 303, + 100, + 312 + ], + "score": 1.0, + "content": "42", + "type": "text" + }, + { + "bbox": [ + 104, + 300, + 386, + 313 + ], + "score": 1.0, + "content": "which is an aspect where GPUs do not have an advantage over CPUs.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 90, + 317, + 505, + 394 + ], + "lines": [ + { + "bbox": [ + 90, + 316, + 506, + 330 + ], + "spans": [ + { + "bbox": [ + 90, + 319, + 99, + 328 + ], + "score": 1.0, + "content": "43", + "type": "text" + }, + { + "bbox": [ + 104, + 316, + 506, + 330 + ], + "score": 1.0, + "content": "In this paper, we present an extensible framework for implementing deep learning training based on", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 89, + 327, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 89, + 330, + 99, + 339 + ], + "score": 1.0, + "content": "44", + "type": "text" + }, + { + "bbox": [ + 105, + 327, + 505, + 340 + ], + "score": 1.0, + "content": "MP-SPDZ by Keller [2020], a framework for multi-party computation.1 Similar to TensorFlow and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 89, + 339, + 505, + 351 + ], + "spans": [ + { + "bbox": [ + 89, + 341, + 100, + 351 + ], + "score": 1.0, + "content": "45", + "type": "text" + }, + { + "bbox": [ + 105, + 339, + 505, + 351 + ], + "score": 1.0, + "content": "PyTorch, our approach allows representing deep learning models as succession of layers. We then", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 89, + 350, + 505, + 362 + ], + "spans": [ + { + "bbox": [ + 89, + 352, + 99, + 361 + ], + "score": 1.0, + "content": "46", + "type": "text" + }, + { + "bbox": [ + 105, + 350, + 505, + 362 + ], + "score": 1.0, + "content": "use this implementation to obtain accuracy figures for MNIST training by utilizing the MP-SPDZ", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 89, + 359, + 506, + 374 + ], + "spans": [ + { + "bbox": [ + 89, + 363, + 100, + 372 + ], + "score": 1.0, + "content": "47", + "type": "text" + }, + { + "bbox": [ + 105, + 359, + 506, + 374 + ], + "score": 1.0, + "content": "emulator, which allows to run the plaintext equivalent of secure computation, that is, the same", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 89, + 371, + 505, + 384 + ], + "spans": [ + { + "bbox": [ + 89, + 374, + 99, + 383 + ], + "score": 1.0, + "content": "48", + "type": "text" + }, + { + "bbox": [ + 105, + 371, + 505, + 384 + ], + "score": 1.0, + "content": "algorithms with the same precision. Finally, we run one of the most promising instantiation in real", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 89, + 383, + 472, + 395 + ], + "spans": [ + { + "bbox": [ + 89, + 385, + 99, + 394 + ], + "score": 1.0, + "content": "49", + "type": "text" + }, + { + "bbox": [ + 105, + 383, + 472, + 395 + ], + "score": 1.0, + "content": "secure computation in order to benchmark it confirm the result from the plaintext emulator.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 90, + 398, + 505, + 475 + ], + "lines": [ + { + "bbox": [ + 89, + 398, + 506, + 412 + ], + "spans": [ + { + "bbox": [ + 89, + 401, + 99, + 410 + ], + "score": 1.0, + "content": "50", + "type": "text" + }, + { + "bbox": [ + 105, + 398, + 506, + 412 + ], + "score": 1.0, + "content": "There are a number of projects that integrate secure computation directly into popular machine", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 89, + 410, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 89, + 412, + 99, + 421 + ], + "score": 1.0, + "content": "51", + "type": "text" + }, + { + "bbox": [ + 106, + 410, + 505, + 423 + ], + "score": 1.0, + "content": "learning frameworks such as CrypTen by Gunning et al. [2019], PySyft by Ryffel et al. [2018], and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 89, + 421, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 89, + 423, + 100, + 432 + ], + "score": 1.0, + "content": "52", + "type": "text" + }, + { + "bbox": [ + 105, + 421, + 506, + 433 + ], + "score": 1.0, + "content": "TF Encrypted by Dahl et al. [2018]. Our approach differs from all of them by running the protocol as", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 89, + 430, + 506, + 445 + ], + "spans": [ + { + "bbox": [ + 89, + 434, + 99, + 443 + ], + "score": 1.0, + "content": "53", + "type": "text" + }, + { + "bbox": [ + 105, + 430, + 261, + 445 + ], + "score": 1.0, + "content": "native CPU code (implemented using", + "type": "text" + }, + { + "bbox": [ + 262, + 432, + 282, + 442 + ], + "score": 0.84, + "content": "\\mathrm { C } { + + }", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 430, + 506, + 445 + ], + "score": 1.0, + "content": "). This allows for much faster execution. For example,", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 89, + 442, + 506, + 455 + ], + "spans": [ + { + "bbox": [ + 89, + 444, + 100, + 454 + ], + "score": 1.0, + "content": "54", + "type": "text" + }, + { + "bbox": [ + 105, + 442, + 506, + 455 + ], + "score": 1.0, + "content": "CrypTen provides an MNIST training example (mpc_autograd_cnn) that takes over one minute", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 89, + 454, + 506, + 466 + ], + "spans": [ + { + "bbox": [ + 89, + 456, + 100, + 465 + ], + "score": 1.0, + "content": "55", + "type": "text" + }, + { + "bbox": [ + 105, + 454, + 506, + 466 + ], + "score": 1.0, + "content": "to run one epoch with 100 samples on one machine. In comparison, our implementation takes 11", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 89, + 464, + 369, + 477 + ], + "spans": [ + { + "bbox": [ + 89, + 467, + 100, + 476 + ], + "score": 1.0, + "content": "56", + "type": "text" + }, + { + "bbox": [ + 105, + 464, + 369, + 477 + ], + "score": 1.0, + "content": "minutes to run one epoch with the full dataset of 60,000 samples.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 90, + 480, + 505, + 514 + ], + "lines": [ + { + "bbox": [ + 90, + 480, + 506, + 494 + ], + "spans": [ + { + "bbox": [ + 90, + 483, + 100, + 492 + ], + "score": 1.0, + "content": "57", + "type": "text" + }, + { + "bbox": [ + 104, + 480, + 506, + 494 + ], + "score": 1.0, + "content": "Another line of work (e.g., Quoc et al. [2021]) uses trusted execution environments that provide", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 90, + 491, + 506, + 505 + ], + "spans": [ + { + "bbox": [ + 90, + 494, + 100, + 502 + ], + "score": 1.0, + "content": "58", + "type": "text" + }, + { + "bbox": [ + 104, + 491, + 506, + 505 + ], + "score": 1.0, + "content": "computation outside the reach of the operating system. This is a different security model to multi-party", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 89, + 502, + 429, + 515 + ], + "spans": [ + { + "bbox": [ + 89, + 505, + 100, + 514 + ], + "score": 1.0, + "content": "59", + "type": "text" + }, + { + "bbox": [ + 105, + 502, + 429, + 515 + ], + "score": 1.0, + "content": "computation that works with distributing the information among several entities.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 90, + 519, + 505, + 563 + ], + "lines": [ + { + "bbox": [ + 89, + 519, + 506, + 531 + ], + "spans": [ + { + "bbox": [ + 89, + 521, + 100, + 531 + ], + "score": 1.0, + "content": "60", + "type": "text" + }, + { + "bbox": [ + 106, + 519, + 506, + 531 + ], + "score": 1.0, + "content": "The paper is structured as follows: After introducing the basics of the protocol we use in Section 2,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 89, + 528, + 506, + 543 + ], + "spans": [ + { + "bbox": [ + 89, + 532, + 100, + 541 + ], + "score": 1.0, + "content": "61", + "type": "text" + }, + { + "bbox": [ + 105, + 528, + 506, + 543 + ], + "score": 1.0, + "content": "we will explain the mathematical building blocks in Section 3 and their use in the context of deep", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 89, + 540, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 89, + 542, + 100, + 552 + ], + "score": 1.0, + "content": "62", + "type": "text" + }, + { + "bbox": [ + 106, + 540, + 505, + 552 + ], + "score": 1.0, + "content": "learning in Section 4. Finally, we will present our implementation in Section 5 and our experimental", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 89, + 552, + 236, + 564 + ], + "spans": [ + { + "bbox": [ + 89, + 554, + 100, + 563 + ], + "score": 1.0, + "content": "63", + "type": "text" + }, + { + "bbox": [ + 105, + 552, + 236, + 564 + ], + "score": 1.0, + "content": "results for MNIST in Section 6.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30.5 + }, + { + "type": "title", + "bbox": [ + 91, + 578, + 405, + 592 + ], + "lines": [ + { + "bbox": [ + 87, + 578, + 407, + 595 + ], + "spans": [ + { + "bbox": [ + 87, + 578, + 407, + 595 + ], + "score": 1.0, + "content": "64 2 An Efficient Secure Multi-Party Computation Protocol", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 106, + 603, + 505, + 703 + ], + "lines": [ + { + "bbox": [ + 106, + 604, + 506, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 604, + 506, + 617 + ], + "score": 1.0, + "content": "There is a wide range of MPC protocols with a variety of security properties (see Keller [2020] for an", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 614, + 506, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 506, + 628 + ], + "score": 1.0, + "content": "overview). In this paper we focus on the setting of three-party computation with one semi-honest", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 624, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 506, + 639 + ], + "score": 1.0, + "content": "corruption. This means that out of the three parties two are expected to behave honestly, i.e., they", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 636, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 505, + 650 + ], + "score": 1.0, + "content": "follow the protocol and keep their view of the protocol secret, and one party is expected to follow the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 647, + 506, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 647, + 506, + 660 + ], + "score": 1.0, + "content": "protocol but might try extract information from their view. The reason for choosing this setting is that", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 658, + 505, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 505, + 671 + ], + "score": 1.0, + "content": "it allows an efficient MPC protocol while still allowing secure outsourced computation. The concrete", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 669, + 506, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 669, + 506, + 682 + ], + "score": 1.0, + "content": "protocol we use goes back to Benaloh and Leichter [1990] with further aspects by Araki et al. [2016],", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 679, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 105, + 679, + 505, + 693 + ], + "score": 1.0, + "content": "Mohassel and Rindal [2018], and Eerikson et al. [2020]. We summarize the core protocol below. The", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 691, + 421, + 703 + ], + "spans": [ + { + "bbox": [ + 106, + 691, + 421, + 703 + ], + "score": 1.0, + "content": "mathematical building blocks in the next section mostly use the aspects below.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 38 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 742, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 741, + 310, + 753 + ], + "spans": [ + { + "bbox": [ + 301, + 741, + 310, + 753 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 120, + 712, + 330, + 722 + ], + "lines": [ + { + "bbox": [ + 120, + 709, + 331, + 724 + ], + "spans": [ + { + "bbox": [ + 120, + 709, + 331, + 724 + ], + "score": 1.0, + "content": "1We are committed to publishing our code as open source.", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 156, + 74, + 454, + 179 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 156, + 74, + 454, + 179 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 156, + 74, + 454, + 179 + ], + "spans": [ + { + "bbox": [ + 156, + 74, + 454, + 179 + ], + "score": 0.927, + "type": "image", + "image_path": "bb0f136189595323f73d65ddc0b25170a7b7ffb1f9107b9e37395cb5ee7bab1c.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 156, + 74, + 454, + 109.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 156, + 109.0, + 454, + 144.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 156, + 144.0, + 454, + 179.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 191, + 505, + 236 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 191, + 506, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 506, + 204 + ], + "score": 1.0, + "content": "Figure 1: Outsourced computation: Data holders (on the left) secret-share their data to a number of", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 202, + 506, + 215 + ], + "spans": [ + { + "bbox": [ + 105, + 202, + 506, + 215 + ], + "score": 1.0, + "content": "computing parties (on the right), who then return the desired the result (e.g., a model or inference", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 214, + 505, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 505, + 227 + ], + "score": 1.0, + "content": "results on further queries). All communication except outputs are secret-shared and thus secure if no", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 225, + 232, + 236 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 232, + 236 + ], + "score": 1.0, + "content": "two computing parties collude.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "index", + "bbox": [ + 90, + 256, + 505, + 311 + ], + "lines": [], + "index": 9, + "bbox_fs": [ + 89, + 257, + 507, + 313 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 90, + 317, + 505, + 394 + ], + "lines": [], + "index": 15, + "bbox_fs": [ + 89, + 316, + 506, + 395 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 90, + 398, + 505, + 475 + ], + "lines": [], + "index": 22, + "bbox_fs": [ + 89, + 398, + 506, + 477 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 90, + 480, + 505, + 514 + ], + "lines": [], + "index": 27, + "bbox_fs": [ + 89, + 480, + 506, + 515 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 90, + 519, + 505, + 563 + ], + "lines": [], + "index": 30.5, + "bbox_fs": [ + 89, + 519, + 506, + 564 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 91, + 578, + 405, + 592 + ], + "lines": [ + { + "bbox": [ + 87, + 578, + 407, + 595 + ], + "spans": [ + { + "bbox": [ + 87, + 578, + 407, + 595 + ], + "score": 1.0, + "content": "64 2 An Efficient Secure Multi-Party Computation Protocol", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 106, + 603, + 505, + 703 + ], + "lines": [ + { + "bbox": [ + 106, + 604, + 506, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 604, + 506, + 617 + ], + "score": 1.0, + "content": "There is a wide range of MPC protocols with a variety of security properties (see Keller [2020] for an", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 614, + 506, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 506, + 628 + ], + "score": 1.0, + "content": "overview). In this paper we focus on the setting of three-party computation with one semi-honest", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 624, + 506, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 506, + 639 + ], + "score": 1.0, + "content": "corruption. This means that out of the three parties two are expected to behave honestly, i.e., they", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 636, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 505, + 650 + ], + "score": 1.0, + "content": "follow the protocol and keep their view of the protocol secret, and one party is expected to follow the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 647, + 506, + 660 + ], + "spans": [ + { + "bbox": [ + 105, + 647, + 506, + 660 + ], + "score": 1.0, + "content": "protocol but might try extract information from their view. The reason for choosing this setting is that", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 658, + 505, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 505, + 671 + ], + "score": 1.0, + "content": "it allows an efficient MPC protocol while still allowing secure outsourced computation. The concrete", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 669, + 506, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 669, + 506, + 682 + ], + "score": 1.0, + "content": "protocol we use goes back to Benaloh and Leichter [1990] with further aspects by Araki et al. [2016],", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 679, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 105, + 679, + 505, + 693 + ], + "score": 1.0, + "content": "Mohassel and Rindal [2018], and Eerikson et al. [2020]. We summarize the core protocol below. The", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 691, + 421, + 703 + ], + "spans": [ + { + "bbox": [ + 106, + 691, + 421, + 703 + ], + "score": 1.0, + "content": "mathematical building blocks in the next section mostly use the aspects below.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 38, + "bbox_fs": [ + 105, + 604, + 506, + 703 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 90, + 72, + 505, + 172 + ], + "lines": [ + { + "bbox": [ + 89, + 72, + 506, + 85 + ], + "spans": [ + { + "bbox": [ + 89, + 74, + 99, + 84 + ], + "score": 1.0, + "content": "74", + "type": "text" + }, + { + "bbox": [ + 105, + 72, + 506, + 85 + ], + "score": 1.0, + "content": "Secret sharing All intermediate values in our protocol are stored using replicated secret sharing. A", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 89, + 83, + 505, + 97 + ], + "spans": [ + { + "bbox": [ + 89, + 86, + 100, + 95 + ], + "score": 1.0, + "content": "75", + "type": "text" + }, + { + "bbox": [ + 105, + 83, + 156, + 97 + ], + "score": 1.0, + "content": "secret value", + "type": "text" + }, + { + "bbox": [ + 156, + 86, + 163, + 93 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 83, + 298, + 97 + ], + "score": 1.0, + "content": "is a represented as a random sum", + "type": "text" + }, + { + "bbox": [ + 299, + 84, + 373, + 95 + ], + "score": 0.92, + "content": "x = x _ { 0 } + x _ { 1 } + x _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 83, + 417, + 97 + ], + "score": 1.0, + "content": ", and party", + "type": "text" + }, + { + "bbox": [ + 417, + 84, + 428, + 95 + ], + "score": 0.88, + "content": "P _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 83, + 453, + 97 + ], + "score": 1.0, + "content": "holds", + "type": "text" + }, + { + "bbox": [ + 454, + 84, + 505, + 96 + ], + "score": 0.92, + "content": "( x _ { i - 1 } , x _ { i + 1 } )", + "type": "inline_equation" + } + ], + "index": 1 + }, + { + "bbox": [ + 89, + 93, + 505, + 108 + ], + "spans": [ + { + "bbox": [ + 89, + 97, + 99, + 106 + ], + "score": 1.0, + "content": "76", + "type": "text" + }, + { + "bbox": [ + 105, + 93, + 505, + 108 + ], + "score": 1.0, + "content": "where the indices are computed modulo three. Clearly, each party is missing one value to compute", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 89, + 105, + 506, + 118 + ], + "spans": [ + { + "bbox": [ + 89, + 107, + 100, + 117 + ], + "score": 1.0, + "content": "77", + "type": "text" + }, + { + "bbox": [ + 105, + 105, + 506, + 118 + ], + "score": 1.0, + "content": "the sum. On the other hand, each pair of parties hold all necessary to reconstruct the secret. For a", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 89, + 116, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 89, + 118, + 100, + 128 + ], + "score": 1.0, + "content": "78", + "type": "text" + }, + { + "bbox": [ + 105, + 116, + 506, + 129 + ], + "score": 1.0, + "content": "uniformly random generation of shares, the computation domain has to be finite. Most commonly,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 88, + 126, + 505, + 140 + ], + "spans": [ + { + "bbox": [ + 88, + 128, + 100, + 139 + ], + "score": 1.0, + "content": "79", + "type": "text" + }, + { + "bbox": [ + 105, + 126, + 392, + 140 + ], + "score": 1.0, + "content": "this domain is defined by integer computation modulo a number. We use", + "type": "text" + }, + { + "bbox": [ + 393, + 127, + 404, + 137 + ], + "score": 0.86, + "content": "2 ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 126, + 418, + 140 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 419, + 127, + 426, + 137 + ], + "score": 0.79, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 126, + 505, + 140 + ], + "score": 1.0, + "content": "being a multiple 64", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 88, + 138, + 505, + 150 + ], + "spans": [ + { + "bbox": [ + 88, + 140, + 100, + 149 + ], + "score": 1.0, + "content": "80", + "type": "text" + }, + { + "bbox": [ + 106, + 138, + 505, + 150 + ], + "score": 1.0, + "content": "and 2 as the moduli. The first case corresponds to an extension of 64-bit arithmetic found on most", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 88, + 149, + 505, + 162 + ], + "spans": [ + { + "bbox": [ + 88, + 151, + 99, + 161 + ], + "score": 1.0, + "content": "81", + "type": "text" + }, + { + "bbox": [ + 104, + 149, + 505, + 162 + ], + "score": 1.0, + "content": "processors. We will refer to the two settings as arithmetic and binary secret sharing throughout the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 88, + 162, + 133, + 173 + ], + "spans": [ + { + "bbox": [ + 88, + 162, + 100, + 173 + ], + "score": 1.0, + "content": "82", + "type": "text" + }, + { + "bbox": [ + 104, + 162, + 133, + 173 + ], + "score": 1.0, + "content": "paper.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 89, + 182, + 505, + 260 + ], + "lines": [ + { + "bbox": [ + 89, + 183, + 506, + 196 + ], + "spans": [ + { + "bbox": [ + 89, + 185, + 99, + 195 + ], + "score": 1.0, + "content": "83", + "type": "text" + }, + { + "bbox": [ + 105, + 183, + 506, + 196 + ], + "score": 1.0, + "content": "Input sharing The secret sharing scheme implies a protocol to share inputs where the inputting", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 89, + 194, + 506, + 207 + ], + "spans": [ + { + "bbox": [ + 89, + 196, + 100, + 205 + ], + "score": 1.0, + "content": "84", + "type": "text" + }, + { + "bbox": [ + 105, + 194, + 506, + 207 + ], + "score": 1.0, + "content": "party samples the shares and distributes them accordingly. Eerikson et al. [2020] have proposed a", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 88, + 204, + 506, + 217 + ], + "spans": [ + { + "bbox": [ + 88, + 206, + 100, + 216 + ], + "score": 1.0, + "content": "85", + "type": "text" + }, + { + "bbox": [ + 105, + 204, + 506, + 217 + ], + "score": 1.0, + "content": "more efficient protocol where the inputting party only needs to send one value instead of two pairs of", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 89, + 216, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 89, + 218, + 99, + 227 + ], + "score": 1.0, + "content": "86", + "type": "text" + }, + { + "bbox": [ + 105, + 216, + 147, + 228 + ], + "score": 1.0, + "content": "values. If", + "type": "text" + }, + { + "bbox": [ + 148, + 216, + 158, + 227 + ], + "score": 0.87, + "content": "P _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 216, + 240, + 228 + ], + "score": 1.0, + "content": "would like to input", + "type": "text" + }, + { + "bbox": [ + 241, + 218, + 247, + 226 + ], + "score": 0.57, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 216, + 251, + 228 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 251, + 217, + 262, + 227 + ], + "score": 0.79, + "content": "x _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 216, + 337, + 228 + ], + "score": 1.0, + "content": "is set to zero, and", + "type": "text" + }, + { + "bbox": [ + 338, + 217, + 358, + 227 + ], + "score": 0.88, + "content": "x _ { i - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 216, + 506, + 228 + ], + "score": 1.0, + "content": "is generated with a pseudo-random", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 89, + 226, + 506, + 240 + ], + "spans": [ + { + "bbox": [ + 89, + 229, + 99, + 238 + ], + "score": 1.0, + "content": "87", + "type": "text" + }, + { + "bbox": [ + 105, + 226, + 298, + 240 + ], + "score": 1.0, + "content": "generator using a key previously shared between", + "type": "text" + }, + { + "bbox": [ + 299, + 227, + 309, + 237 + ], + "score": 0.87, + "content": "P _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 310, + 226, + 327, + 240 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 327, + 227, + 348, + 238 + ], + "score": 0.87, + "content": "P _ { i + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 226, + 352, + 240 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 353, + 227, + 363, + 237 + ], + "score": 0.82, + "content": "P _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 364, + 226, + 417, + 240 + ], + "score": 1.0, + "content": "can compute", + "type": "text" + }, + { + "bbox": [ + 417, + 228, + 487, + 238 + ], + "score": 0.9, + "content": "x _ { i + 1 } = x - x _ { i - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 226, + 506, + 240 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 89, + 237, + 505, + 250 + ], + "spans": [ + { + "bbox": [ + 89, + 240, + 99, + 249 + ], + "score": 1.0, + "content": "88", + "type": "text" + }, + { + "bbox": [ + 105, + 237, + 147, + 250 + ], + "score": 1.0, + "content": "send it to", + "type": "text" + }, + { + "bbox": [ + 147, + 238, + 168, + 249 + ], + "score": 0.91, + "content": "P _ { i - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 237, + 460, + 250 + ], + "score": 1.0, + "content": ". While the resulting secret sharing is not entirely random, the fact that", + "type": "text" + }, + { + "bbox": [ + 461, + 238, + 471, + 248 + ], + "score": 0.87, + "content": "P _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 237, + 505, + 250 + ], + "score": 1.0, + "content": "already", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 89, + 248, + 273, + 261 + ], + "spans": [ + { + "bbox": [ + 89, + 250, + 99, + 260 + ], + "score": 1.0, + "content": "89", + "type": "text" + }, + { + "bbox": [ + 105, + 248, + 135, + 261 + ], + "score": 1.0, + "content": "knows", + "type": "text" + }, + { + "bbox": [ + 135, + 251, + 142, + 258 + ], + "score": 0.73, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 248, + 223, + 261 + ], + "score": 1.0, + "content": "makes randomizing", + "type": "text" + }, + { + "bbox": [ + 224, + 250, + 234, + 259 + ], + "score": 0.85, + "content": "x _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 248, + 273, + 261 + ], + "score": 1.0, + "content": "obsolete.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 90, + 271, + 505, + 306 + ], + "lines": [ + { + "bbox": [ + 89, + 271, + 506, + 283 + ], + "spans": [ + { + "bbox": [ + 89, + 273, + 100, + 282 + ], + "score": 1.0, + "content": "90", + "type": "text" + }, + { + "bbox": [ + 105, + 271, + 506, + 283 + ], + "score": 1.0, + "content": "Addition The commutative nature of addition allows to add secret sharings without communication.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 90, + 282, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 90, + 284, + 100, + 293 + ], + "score": 1.0, + "content": "91", + "type": "text" + }, + { + "bbox": [ + 104, + 282, + 239, + 295 + ], + "score": 1.0, + "content": "More concretely, secret sharings", + "type": "text" + }, + { + "bbox": [ + 239, + 283, + 314, + 293 + ], + "score": 0.91, + "content": "x = x _ { 0 } + x _ { 1 } + x _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 282, + 333, + 295 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 333, + 283, + 404, + 294 + ], + "score": 0.92, + "content": "y = y _ { 0 } + y _ { 1 } + y _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 282, + 505, + 295 + ], + "score": 1.0, + "content": "imply the secret sharing", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 88, + 293, + 291, + 306 + ], + "spans": [ + { + "bbox": [ + 88, + 293, + 106, + 306 + ], + "score": 1.0, + "content": "92", + "type": "text" + }, + { + "bbox": [ + 106, + 293, + 286, + 306 + ], + "score": 0.92, + "content": "x + y = ( x _ { 0 } + y _ { 0 } ) + ( x _ { 1 } + y _ { 1 } ) + ( x _ { 2 } + y _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 293, + 291, + 306 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 90, + 316, + 415, + 328 + ], + "lines": [ + { + "bbox": [ + 86, + 314, + 416, + 330 + ], + "spans": [ + { + "bbox": [ + 86, + 314, + 239, + 330 + ], + "score": 1.0, + "content": "93 Multiplication The product of", + "type": "text" + }, + { + "bbox": [ + 239, + 317, + 314, + 327 + ], + "score": 0.91, + "content": "x = x _ { 0 } + x _ { 1 } + x _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 314, + 332, + 330 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 333, + 317, + 404, + 328 + ], + "score": 0.91, + "content": "y = y _ { 0 } + y _ { 1 } + y _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 314, + 416, + 330 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "interline_equation", + "bbox": [ + 138, + 331, + 472, + 361 + ], + "lines": [ + { + "bbox": [ + 138, + 331, + 472, + 361 + ], + "spans": [ + { + "bbox": [ + 138, + 331, + 472, + 361 + ], + "score": 0.89, + "content": "{ \\begin{array} { r l } & { x \\cdot y = ( x _ { 0 } + x _ { 1 } + x _ { 2 } ) \\cdot ( y _ { 0 } + y _ { 1 } + y _ { 2 } ) } \\\\ & { \\qquad = ( x _ { 0 } y _ { 0 } + x _ { 0 } y _ { 1 } + x _ { 1 } y _ { 0 } ) + ( x _ { 1 } y _ { 1 } + x _ { 1 } y _ { 2 } + x _ { 1 } y _ { 1 } ) + ( x _ { 2 } y _ { 2 } + x _ { 2 } y _ { 0 } + x _ { 0 } y _ { 2 } ) . } \\end{array} }", + "type": "interline_equation", + "image_path": "0e562fbeff8dbe326579bda5ca198bdb7384ec8bfd3048b083e4bf0b03a38815.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 138, + 331, + 472, + 341.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 138, + 341.0, + 472, + 351.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 138, + 351.0, + 472, + 361.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 88, + 363, + 505, + 464 + ], + "lines": [ + { + "bbox": [ + 89, + 363, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 89, + 366, + 100, + 375 + ], + "score": 1.0, + "content": "94", + "type": "text" + }, + { + "bbox": [ + 105, + 363, + 505, + 376 + ], + "score": 1.0, + "content": "Each of the brackets only contains shares known by one of the parties. They can thus compute an", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 89, + 374, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 89, + 376, + 100, + 386 + ], + "score": 1.0, + "content": "95", + "type": "text" + }, + { + "bbox": [ + 105, + 374, + 505, + 388 + ], + "score": 1.0, + "content": "additive secret sharing (one summand per party) of the product. However, every party only holding", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 89, + 386, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 89, + 388, + 100, + 397 + ], + "score": 1.0, + "content": "96", + "type": "text" + }, + { + "bbox": [ + 105, + 386, + 505, + 398 + ], + "score": 1.0, + "content": "one share does not satisfy the replication requirement for further multiplications. It is not secure", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 89, + 396, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 89, + 398, + 100, + 408 + ], + "score": 1.0, + "content": "97", + "type": "text" + }, + { + "bbox": [ + 105, + 396, + 505, + 408 + ], + "score": 1.0, + "content": "for every party to pass their value on to another party because the summands are not distributed", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 89, + 407, + 506, + 420 + ], + "spans": [ + { + "bbox": [ + 89, + 409, + 100, + 419 + ], + "score": 1.0, + "content": "98", + "type": "text" + }, + { + "bbox": [ + 105, + 407, + 326, + 420 + ], + "score": 1.0, + "content": "randomly. This can be fixed by rerandomization: Let", + "type": "text" + }, + { + "bbox": [ + 327, + 408, + 406, + 419 + ], + "score": 0.92, + "content": "x y = z _ { 0 } + z _ { 1 } + z _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 407, + 435, + 420 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 435, + 409, + 445, + 419 + ], + "score": 0.86, + "content": "z _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 407, + 492, + 420 + ], + "score": 1.0, + "content": "is know to", + "type": "text" + }, + { + "bbox": [ + 492, + 408, + 503, + 419 + ], + "score": 0.88, + "content": "P _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 407, + 506, + 420 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 88, + 418, + 506, + 432 + ], + "spans": [ + { + "bbox": [ + 88, + 420, + 100, + 431 + ], + "score": 1.0, + "content": "99", + "type": "text" + }, + { + "bbox": [ + 105, + 418, + 155, + 432 + ], + "score": 1.0, + "content": "Every party", + "type": "text" + }, + { + "bbox": [ + 155, + 419, + 167, + 429 + ], + "score": 0.87, + "content": "P _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 418, + 208, + 432 + ], + "score": 1.0, + "content": "computes", + "type": "text" + }, + { + "bbox": [ + 208, + 419, + 309, + 430 + ], + "score": 0.92, + "content": "z _ { i } ^ { \\prime } = z _ { i } + r _ { i , i + 1 } - r _ { i - 1 , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 310, + 418, + 337, + 432 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 338, + 421, + 362, + 430 + ], + "score": 0.91, + "content": "r _ { i , i + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 418, + 506, + 432 + ], + "score": 1.0, + "content": "is generated with a pseudo-random", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 86, + 428, + 506, + 443 + ], + "spans": [ + { + "bbox": [ + 86, + 431, + 100, + 442 + ], + "score": 1.0, + "content": "100", + "type": "text" + }, + { + "bbox": [ + 105, + 428, + 280, + 443 + ], + "score": 1.0, + "content": "generator using a key pre-shared between", + "type": "text" + }, + { + "bbox": [ + 280, + 430, + 291, + 440 + ], + "score": 0.86, + "content": "P _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 428, + 310, + 443 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 311, + 430, + 331, + 441 + ], + "score": 0.9, + "content": "P _ { i + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 428, + 414, + 443 + ], + "score": 1.0, + "content": ". The resulting sum", + "type": "text" + }, + { + "bbox": [ + 414, + 429, + 494, + 441 + ], + "score": 0.92, + "content": "x y = \\hat { z } _ { 0 } ^ { \\prime } + z _ { 1 } ^ { \\prime } + z _ { 2 } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 428, + 506, + 443 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 86, + 439, + 506, + 453 + ], + "spans": [ + { + "bbox": [ + 86, + 442, + 99, + 452 + ], + "score": 1.0, + "content": "101", + "type": "text" + }, + { + "bbox": [ + 105, + 439, + 271, + 453 + ], + "score": 1.0, + "content": "pseudo-random, and it is thus secure for", + "type": "text" + }, + { + "bbox": [ + 271, + 441, + 282, + 451 + ], + "score": 0.88, + "content": "P _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 439, + 315, + 453 + ], + "score": 1.0, + "content": "to send", + "type": "text" + }, + { + "bbox": [ + 315, + 441, + 324, + 452 + ], + "score": 0.88, + "content": "z _ { i } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 439, + 335, + 453 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 336, + 441, + 357, + 452 + ], + "score": 0.92, + "content": "P _ { i + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 439, + 506, + 453 + ], + "score": 1.0, + "content": "in order to create a replicated secret", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 87, + 451, + 271, + 464 + ], + "spans": [ + { + "bbox": [ + 87, + 453, + 99, + 462 + ], + "score": 1.0, + "content": "102", + "type": "text" + }, + { + "bbox": [ + 105, + 451, + 139, + 464 + ], + "score": 1.0, + "content": "sharing", + "type": "text" + }, + { + "bbox": [ + 139, + 451, + 267, + 464 + ], + "score": 0.9, + "content": "\\left( ( x y ) _ { i - 1 } , ( x y ) _ { i + 1 } \\right) = \\left( z _ { i } ^ { \\prime } , z _ { i - 1 } ^ { \\prime } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 451, + 271, + 464 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 27 + }, + { + "type": "title", + "bbox": [ + 87, + 477, + 315, + 491 + ], + "lines": [ + { + "bbox": [ + 84, + 475, + 316, + 495 + ], + "spans": [ + { + "bbox": [ + 84, + 475, + 316, + 495 + ], + "score": 1.0, + "content": "103 3 Secure Computation Building Blocks", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 87, + 502, + 505, + 546 + ], + "lines": [ + { + "bbox": [ + 87, + 501, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 87, + 504, + 99, + 514 + ], + "score": 1.0, + "content": "104", + "type": "text" + }, + { + "bbox": [ + 105, + 501, + 505, + 515 + ], + "score": 1.0, + "content": "In this section, we will discuss how to implement computation with MPC with a focus on how", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 86, + 512, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 86, + 515, + 100, + 524 + ], + "score": 1.0, + "content": "105", + "type": "text" + }, + { + "bbox": [ + 105, + 512, + 505, + 525 + ], + "score": 1.0, + "content": "it differs from computation on CPUs or GPUs. Most of the techniques below are already known", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 86, + 522, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 86, + 525, + 100, + 535 + ], + "score": 1.0, + "content": "106", + "type": "text" + }, + { + "bbox": [ + 105, + 522, + 505, + 537 + ], + "score": 1.0, + "content": "individually. To the best of our knowledge however, we are the first to put them together in an efficient", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 86, + 533, + 410, + 549 + ], + "spans": [ + { + "bbox": [ + 86, + 537, + 100, + 546 + ], + "score": 1.0, + "content": "107", + "type": "text" + }, + { + "bbox": [ + 104, + 533, + 410, + 549 + ], + "score": 1.0, + "content": "and extensible framework for secure computation of deep learning training.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 86, + 557, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 86, + 556, + 506, + 570 + ], + "spans": [ + { + "bbox": [ + 86, + 560, + 99, + 569 + ], + "score": 1.0, + "content": "108", + "type": "text" + }, + { + "bbox": [ + 105, + 556, + 377, + 570 + ], + "score": 1.0, + "content": "Domain conversion Recall we that we use computation modulo", + "type": "text" + }, + { + "bbox": [ + 377, + 557, + 388, + 568 + ], + "score": 0.87, + "content": "2 ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 556, + 404, + 570 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 404, + 558, + 411, + 568 + ], + "score": 0.8, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 556, + 506, + 570 + ], + "score": 1.0, + "content": "being a multiple of 64", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 86, + 568, + 506, + 582 + ], + "spans": [ + { + "bbox": [ + 86, + 571, + 99, + 580 + ], + "score": 1.0, + "content": "109", + "type": "text" + }, + { + "bbox": [ + 104, + 568, + 506, + 582 + ], + "score": 1.0, + "content": "as well as 1. Given that the main operations are just addition and multiplication in the respective", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 86, + 580, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 86, + 582, + 100, + 591 + ], + "score": 1.0, + "content": "110", + "type": "text" + }, + { + "bbox": [ + 105, + 580, + 506, + 592 + ], + "score": 1.0, + "content": "domain, it is desirable to compute integer arithmetic in the large domain but operations with a", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 86, + 588, + 506, + 605 + ], + "spans": [ + { + "bbox": [ + 86, + 592, + 100, + 603 + ], + "score": 1.0, + "content": "111", + "type": "text" + }, + { + "bbox": [ + 104, + 588, + 506, + 605 + ], + "score": 1.0, + "content": "straight-forward binary circuit modulo two. There has been a long-running interest in this going", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 86, + 601, + 506, + 615 + ], + "spans": [ + { + "bbox": [ + 86, + 603, + 100, + 614 + ], + "score": 1.0, + "content": "112", + "type": "text" + }, + { + "bbox": [ + 105, + 601, + 506, + 615 + ], + "score": 1.0, + "content": "back to least Kolesnikov et al. [2013]. We mainly rely on the approach proposed by Mohassel and", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 86, + 610, + 510, + 645 + ], + "spans": [ + { + "bbox": [ + 86, + 614, + 100, + 638 + ], + "score": 1.0, + "content": "113 114", + "type": "text" + }, + { + "bbox": [ + 107, + 623, + 148, + 639 + ], + "score": 0.92, + "content": "\\{ x _ { 0 } ^ { ( i ) } \\} _ { j = 0 } ^ { k - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 610, + 246, + 645 + ], + "score": 1.0, + "content": "018] and Araki et al. [20the bit decomposition of", + "type": "text" + }, + { + "bbox": [ + 246, + 626, + 257, + 636 + ], + "score": 0.85, + "content": "x _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 610, + 289, + 645 + ], + "score": 1.0, + "content": ". Recal, that is,", + "type": "text" + }, + { + "bbox": [ + 289, + 623, + 342, + 638 + ], + "score": 0.93, + "content": "x _ { 0 } ^ { ( i ) } \\in \\{ 0 , 1 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 612, + 339, + 622 + ], + "score": 0.9, + "content": "x \\in 2 ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 610, + 360, + 645 + ], + "score": 1.0, + "content": "is shand", + "type": "text" + }, + { + "bbox": [ + 360, + 623, + 435, + 638 + ], + "score": 0.94, + "content": "\\begin{array} { r } { x _ { 0 } = \\sum _ { i = 0 } ^ { k - 1 } x _ { 0 } ^ { ( i ) } 2 ^ { i } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 613, + 466, + 623 + ], + "score": 0.89, + "content": "x = x _ { 0 } + x _ { 1 } + x _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 466, + 610, + 510, + 645 + ], + "score": 1.0, + "content": ". Now letlf-evident", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 86, + 633, + 509, + 661 + ], + "spans": [ + { + "bbox": [ + 86, + 642, + 100, + 653 + ], + "score": 1.0, + "content": "115", + "type": "text" + }, + { + "bbox": [ + 103, + 633, + 124, + 661 + ], + "score": 1.0, + "content": "that", + "type": "text" + }, + { + "bbox": [ + 124, + 638, + 203, + 653 + ], + "score": 0.93, + "content": "x _ { 0 } ^ { ( i ) } = x _ { 0 } ^ { ( i ) } + 0 + 0", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 633, + 509, + 661 + ], + "score": 1.0, + "content": "is a valid secret sharing modulo two (albeit not a random one). Furthermore,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 86, + 649, + 510, + 673 + ], + "spans": [ + { + "bbox": [ + 86, + 657, + 100, + 666 + ], + "score": 1.0, + "content": "116", + "type": "text" + }, + { + "bbox": [ + 102, + 649, + 187, + 673 + ], + "score": 1.0, + "content": "every party holding", + "type": "text" + }, + { + "bbox": [ + 187, + 657, + 199, + 666 + ], + "score": 0.84, + "content": "x _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 649, + 253, + 673 + ], + "score": 1.0, + "content": "can generate", + "type": "text" + }, + { + "bbox": [ + 253, + 652, + 270, + 667 + ], + "score": 0.91, + "content": "x _ { 0 } ^ { ( i ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 649, + 510, + 673 + ], + "score": 1.0, + "content": ". It is therefore possible for the parties to generate a secret", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 86, + 666, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 86, + 669, + 99, + 678 + ], + "score": 1.0, + "content": "117", + "type": "text" + }, + { + "bbox": [ + 105, + 666, + 294, + 680 + ], + "score": 1.0, + "content": "sharing modulo two of a single share modulo", + "type": "text" + }, + { + "bbox": [ + 294, + 666, + 305, + 677 + ], + "score": 0.86, + "content": "2 ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 666, + 506, + 680 + ], + "score": 1.0, + "content": ". Repeating this for all shares and the computing", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 86, + 677, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 86, + 680, + 100, + 689 + ], + "score": 1.0, + "content": "118", + "type": "text" + }, + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "score": 1.0, + "content": "the addition as a binary circuit allows the parties to generate a secret sharing modulo two from a", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 86, + 687, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 86, + 691, + 100, + 700 + ], + "score": 1.0, + "content": "119", + "type": "text" + }, + { + "bbox": [ + 104, + 687, + 195, + 702 + ], + "score": 1.0, + "content": "secret sharing modulo", + "type": "text" + }, + { + "bbox": [ + 195, + 688, + 206, + 699 + ], + "score": 0.85, + "content": "\\bar { 2 } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 687, + 506, + 702 + ], + "score": 1.0, + "content": ". Conversion in the other direction can be achieved using a similar technique", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 86, + 699, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 86, + 702, + 100, + 711 + ], + "score": 1.0, + "content": "120", + "type": "text" + }, + { + "bbox": [ + 105, + 699, + 506, + 713 + ], + "score": 1.0, + "content": "or using daBits as described by Rotaru and Wood [2019]. In the following we will use the term", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 86, + 710, + 462, + 724 + ], + "spans": [ + { + "bbox": [ + 86, + 713, + 99, + 722 + ], + "score": 1.0, + "content": "121", + "type": "text" + }, + { + "bbox": [ + 104, + 710, + 462, + 724 + ], + "score": 1.0, + "content": "mixed-circuit computation for any technique that works over both computation domains.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 43 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "index", + "bbox": [ + 90, + 72, + 505, + 172 + ], + "lines": [ + { + "bbox": [ + 89, + 72, + 506, + 85 + ], + "spans": [ + { + "bbox": [ + 89, + 74, + 99, + 84 + ], + "score": 1.0, + "content": "74", + "type": "text" + }, + { + "bbox": [ + 105, + 72, + 506, + 85 + ], + "score": 1.0, + "content": "Secret sharing All intermediate values in our protocol are stored using replicated secret sharing. A", + "type": "text" + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 83, + 505, + 97 + ], + "spans": [ + { + "bbox": [ + 89, + 86, + 100, + 95 + ], + "score": 1.0, + "content": "75", + "type": "text" + }, + { + "bbox": [ + 105, + 83, + 156, + 97 + ], + "score": 1.0, + "content": "secret value", + "type": "text" + }, + { + "bbox": [ + 156, + 86, + 163, + 93 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 83, + 298, + 97 + ], + "score": 1.0, + "content": "is a represented as a random sum", + "type": "text" + }, + { + "bbox": [ + 299, + 84, + 373, + 95 + ], + "score": 0.92, + "content": "x = x _ { 0 } + x _ { 1 } + x _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 83, + 417, + 97 + ], + "score": 1.0, + "content": ", and party", + "type": "text" + }, + { + "bbox": [ + 417, + 84, + 428, + 95 + ], + "score": 0.88, + "content": "P _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 83, + 453, + 97 + ], + "score": 1.0, + "content": "holds", + "type": "text" + }, + { + "bbox": [ + 454, + 84, + 505, + 96 + ], + "score": 0.92, + "content": "( x _ { i - 1 } , x _ { i + 1 } )", + "type": "inline_equation" + } + ], + "index": 1, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 93, + 505, + 108 + ], + "spans": [ + { + "bbox": [ + 89, + 97, + 99, + 106 + ], + "score": 1.0, + "content": "76", + "type": "text" + }, + { + "bbox": [ + 105, + 93, + 505, + 108 + ], + "score": 1.0, + "content": "where the indices are computed modulo three. Clearly, each party is missing one value to compute", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 105, + 506, + 118 + ], + "spans": [ + { + "bbox": [ + 89, + 107, + 100, + 117 + ], + "score": 1.0, + "content": "77", + "type": "text" + }, + { + "bbox": [ + 105, + 105, + 506, + 118 + ], + "score": 1.0, + "content": "the sum. On the other hand, each pair of parties hold all necessary to reconstruct the secret. For a", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 116, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 89, + 118, + 100, + 128 + ], + "score": 1.0, + "content": "78", + "type": "text" + }, + { + "bbox": [ + 105, + 116, + 506, + 129 + ], + "score": 1.0, + "content": "uniformly random generation of shares, the computation domain has to be finite. Most commonly,", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 126, + 505, + 140 + ], + "spans": [ + { + "bbox": [ + 88, + 128, + 100, + 139 + ], + "score": 1.0, + "content": "79", + "type": "text" + }, + { + "bbox": [ + 105, + 126, + 392, + 140 + ], + "score": 1.0, + "content": "this domain is defined by integer computation modulo a number. We use", + "type": "text" + }, + { + "bbox": [ + 393, + 127, + 404, + 137 + ], + "score": 0.86, + "content": "2 ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 126, + 418, + 140 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 419, + 127, + 426, + 137 + ], + "score": 0.79, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 126, + 505, + 140 + ], + "score": 1.0, + "content": "being a multiple 64", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 138, + 505, + 150 + ], + "spans": [ + { + "bbox": [ + 88, + 140, + 100, + 149 + ], + "score": 1.0, + "content": "80", + "type": "text" + }, + { + "bbox": [ + 106, + 138, + 505, + 150 + ], + "score": 1.0, + "content": "and 2 as the moduli. The first case corresponds to an extension of 64-bit arithmetic found on most", + "type": "text" + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 149, + 505, + 162 + ], + "spans": [ + { + "bbox": [ + 88, + 151, + 99, + 161 + ], + "score": 1.0, + "content": "81", + "type": "text" + }, + { + "bbox": [ + 104, + 149, + 505, + 162 + ], + "score": 1.0, + "content": "processors. We will refer to the two settings as arithmetic and binary secret sharing throughout the", + "type": "text" + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 162, + 133, + 173 + ], + "spans": [ + { + "bbox": [ + 88, + 162, + 100, + 173 + ], + "score": 1.0, + "content": "82", + "type": "text" + }, + { + "bbox": [ + 104, + 162, + 133, + 173 + ], + "score": 1.0, + "content": "paper.", + "type": "text" + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 183, + 506, + 196 + ], + "spans": [ + { + "bbox": [ + 89, + 185, + 99, + 195 + ], + "score": 1.0, + "content": "83", + "type": "text" + }, + { + "bbox": [ + 105, + 183, + 506, + 196 + ], + "score": 1.0, + "content": "Input sharing The secret sharing scheme implies a protocol to share inputs where the inputting", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 194, + 506, + 207 + ], + "spans": [ + { + "bbox": [ + 89, + 196, + 100, + 205 + ], + "score": 1.0, + "content": "84", + "type": "text" + }, + { + "bbox": [ + 105, + 194, + 506, + 207 + ], + "score": 1.0, + "content": "party samples the shares and distributes them accordingly. Eerikson et al. [2020] have proposed a", + "type": "text" + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 204, + 506, + 217 + ], + "spans": [ + { + "bbox": [ + 88, + 206, + 100, + 216 + ], + "score": 1.0, + "content": "85", + "type": "text" + }, + { + "bbox": [ + 105, + 204, + 506, + 217 + ], + "score": 1.0, + "content": "more efficient protocol where the inputting party only needs to send one value instead of two pairs of", + "type": "text" + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 216, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 89, + 218, + 99, + 227 + ], + "score": 1.0, + "content": "86", + "type": "text" + }, + { + "bbox": [ + 105, + 216, + 147, + 228 + ], + "score": 1.0, + "content": "values. If", + "type": "text" + }, + { + "bbox": [ + 148, + 216, + 158, + 227 + ], + "score": 0.87, + "content": "P _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 216, + 240, + 228 + ], + "score": 1.0, + "content": "would like to input", + "type": "text" + }, + { + "bbox": [ + 241, + 218, + 247, + 226 + ], + "score": 0.57, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 216, + 251, + 228 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 251, + 217, + 262, + 227 + ], + "score": 0.79, + "content": "x _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 216, + 337, + 228 + ], + "score": 1.0, + "content": "is set to zero, and", + "type": "text" + }, + { + "bbox": [ + 338, + 217, + 358, + 227 + ], + "score": 0.88, + "content": "x _ { i - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 216, + 506, + 228 + ], + "score": 1.0, + "content": "is generated with a pseudo-random", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 226, + 506, + 240 + ], + "spans": [ + { + "bbox": [ + 89, + 229, + 99, + 238 + ], + "score": 1.0, + "content": "87", + "type": "text" + }, + { + "bbox": [ + 105, + 226, + 298, + 240 + ], + "score": 1.0, + "content": "generator using a key previously shared between", + "type": "text" + }, + { + "bbox": [ + 299, + 227, + 309, + 237 + ], + "score": 0.87, + "content": "P _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 310, + 226, + 327, + 240 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 327, + 227, + 348, + 238 + ], + "score": 0.87, + "content": "P _ { i + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 226, + 352, + 240 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 353, + 227, + 363, + 237 + ], + "score": 0.82, + "content": "P _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 364, + 226, + 417, + 240 + ], + "score": 1.0, + "content": "can compute", + "type": "text" + }, + { + "bbox": [ + 417, + 228, + 487, + 238 + ], + "score": 0.9, + "content": "x _ { i + 1 } = x - x _ { i - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 226, + 506, + 240 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 237, + 505, + 250 + ], + "spans": [ + { + "bbox": [ + 89, + 240, + 99, + 249 + ], + "score": 1.0, + "content": "88", + "type": "text" + }, + { + "bbox": [ + 105, + 237, + 147, + 250 + ], + "score": 1.0, + "content": "send it to", + "type": "text" + }, + { + "bbox": [ + 147, + 238, + 168, + 249 + ], + "score": 0.91, + "content": "P _ { i - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 237, + 460, + 250 + ], + "score": 1.0, + "content": ". While the resulting secret sharing is not entirely random, the fact that", + "type": "text" + }, + { + "bbox": [ + 461, + 238, + 471, + 248 + ], + "score": 0.87, + "content": "P _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 237, + 505, + 250 + ], + "score": 1.0, + "content": "already", + "type": "text" + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 248, + 273, + 261 + ], + "spans": [ + { + "bbox": [ + 89, + 250, + 99, + 260 + ], + "score": 1.0, + "content": "89", + "type": "text" + }, + { + "bbox": [ + 105, + 248, + 135, + 261 + ], + "score": 1.0, + "content": "knows", + "type": "text" + }, + { + "bbox": [ + 135, + 251, + 142, + 258 + ], + "score": 0.73, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 248, + 223, + 261 + ], + "score": 1.0, + "content": "makes randomizing", + "type": "text" + }, + { + "bbox": [ + 224, + 250, + 234, + 259 + ], + "score": 0.85, + "content": "x _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 248, + 273, + 261 + ], + "score": 1.0, + "content": "obsolete.", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 271, + 506, + 283 + ], + "spans": [ + { + "bbox": [ + 89, + 273, + 100, + 282 + ], + "score": 1.0, + "content": "90", + "type": "text" + }, + { + "bbox": [ + 105, + 271, + 506, + 283 + ], + "score": 1.0, + "content": "Addition The commutative nature of addition allows to add secret sharings without communication.", + "type": "text" + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 282, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 90, + 284, + 100, + 293 + ], + "score": 1.0, + "content": "91", + "type": "text" + }, + { + "bbox": [ + 104, + 282, + 239, + 295 + ], + "score": 1.0, + "content": "More concretely, secret sharings", + "type": "text" + }, + { + "bbox": [ + 239, + 283, + 314, + 293 + ], + "score": 0.91, + "content": "x = x _ { 0 } + x _ { 1 } + x _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 282, + 333, + 295 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 333, + 283, + 404, + 294 + ], + "score": 0.92, + "content": "y = y _ { 0 } + y _ { 1 } + y _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 282, + 505, + 295 + ], + "score": 1.0, + "content": "imply the secret sharing", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 293, + 291, + 306 + ], + "spans": [ + { + "bbox": [ + 88, + 293, + 106, + 306 + ], + "score": 1.0, + "content": "92", + "type": "text" + }, + { + "bbox": [ + 106, + 293, + 286, + 306 + ], + "score": 0.92, + "content": "x + y = ( x _ { 0 } + y _ { 0 } ) + ( x _ { 1 } + y _ { 1 } ) + ( x _ { 2 } + y _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 293, + 291, + 306 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 18, + "is_list_start_line": true + } + ], + "index": 4, + "bbox_fs": [ + 88, + 72, + 506, + 173 + ] + }, + { + "type": "index", + "bbox": [ + 89, + 182, + 505, + 260 + ], + "lines": [], + "index": 12, + "bbox_fs": [ + 88, + 183, + 506, + 261 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 90, + 271, + 505, + 306 + ], + "lines": [], + "index": 17, + "bbox_fs": [ + 88, + 271, + 506, + 306 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 90, + 316, + 415, + 328 + ], + "lines": [ + { + "bbox": [ + 86, + 314, + 416, + 330 + ], + "spans": [ + { + "bbox": [ + 86, + 314, + 239, + 330 + ], + "score": 1.0, + "content": "93 Multiplication The product of", + "type": "text" + }, + { + "bbox": [ + 239, + 317, + 314, + 327 + ], + "score": 0.91, + "content": "x = x _ { 0 } + x _ { 1 } + x _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 314, + 332, + 330 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 333, + 317, + 404, + 328 + ], + "score": 0.91, + "content": "y = y _ { 0 } + y _ { 1 } + y _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 314, + 416, + 330 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19, + "bbox_fs": [ + 86, + 314, + 416, + 330 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 138, + 331, + 472, + 361 + ], + "lines": [ + { + "bbox": [ + 138, + 331, + 472, + 361 + ], + "spans": [ + { + "bbox": [ + 138, + 331, + 472, + 361 + ], + "score": 0.89, + "content": "{ \\begin{array} { r l } & { x \\cdot y = ( x _ { 0 } + x _ { 1 } + x _ { 2 } ) \\cdot ( y _ { 0 } + y _ { 1 } + y _ { 2 } ) } \\\\ & { \\qquad = ( x _ { 0 } y _ { 0 } + x _ { 0 } y _ { 1 } + x _ { 1 } y _ { 0 } ) + ( x _ { 1 } y _ { 1 } + x _ { 1 } y _ { 2 } + x _ { 1 } y _ { 1 } ) + ( x _ { 2 } y _ { 2 } + x _ { 2 } y _ { 0 } + x _ { 0 } y _ { 2 } ) . } \\end{array} }", + "type": "interline_equation", + "image_path": "0e562fbeff8dbe326579bda5ca198bdb7384ec8bfd3048b083e4bf0b03a38815.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 138, + 331, + 472, + 341.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 138, + 341.0, + 472, + 351.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 138, + 351.0, + 472, + 361.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "index", + "bbox": [ + 88, + 363, + 505, + 464 + ], + "lines": [ + { + "bbox": [ + 89, + 363, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 89, + 366, + 100, + 375 + ], + "score": 1.0, + "content": "94", + "type": "text" + }, + { + "bbox": [ + 105, + 363, + 505, + 376 + ], + "score": 1.0, + "content": "Each of the brackets only contains shares known by one of the parties. They can thus compute an", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 374, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 89, + 376, + 100, + 386 + ], + "score": 1.0, + "content": "95", + "type": "text" + }, + { + "bbox": [ + 105, + 374, + 505, + 388 + ], + "score": 1.0, + "content": "additive secret sharing (one summand per party) of the product. However, every party only holding", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 386, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 89, + 388, + 100, + 397 + ], + "score": 1.0, + "content": "96", + "type": "text" + }, + { + "bbox": [ + 105, + 386, + 505, + 398 + ], + "score": 1.0, + "content": "one share does not satisfy the replication requirement for further multiplications. It is not secure", + "type": "text" + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 396, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 89, + 398, + 100, + 408 + ], + "score": 1.0, + "content": "97", + "type": "text" + }, + { + "bbox": [ + 105, + 396, + 505, + 408 + ], + "score": 1.0, + "content": "for every party to pass their value on to another party because the summands are not distributed", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 407, + 506, + 420 + ], + "spans": [ + { + "bbox": [ + 89, + 409, + 100, + 419 + ], + "score": 1.0, + "content": "98", + "type": "text" + }, + { + "bbox": [ + 105, + 407, + 326, + 420 + ], + "score": 1.0, + "content": "randomly. This can be fixed by rerandomization: Let", + "type": "text" + }, + { + "bbox": [ + 327, + 408, + 406, + 419 + ], + "score": 0.92, + "content": "x y = z _ { 0 } + z _ { 1 } + z _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 407, + 407, + 435, + 420 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 435, + 409, + 445, + 419 + ], + "score": 0.86, + "content": "z _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 407, + 492, + 420 + ], + "score": 1.0, + "content": "is know to", + "type": "text" + }, + { + "bbox": [ + 492, + 408, + 503, + 419 + ], + "score": 0.88, + "content": "P _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 407, + 506, + 420 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 418, + 506, + 432 + ], + "spans": [ + { + "bbox": [ + 88, + 420, + 100, + 431 + ], + "score": 1.0, + "content": "99", + "type": "text" + }, + { + "bbox": [ + 105, + 418, + 155, + 432 + ], + "score": 1.0, + "content": "Every party", + "type": "text" + }, + { + "bbox": [ + 155, + 419, + 167, + 429 + ], + "score": 0.87, + "content": "P _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 418, + 208, + 432 + ], + "score": 1.0, + "content": "computes", + "type": "text" + }, + { + "bbox": [ + 208, + 419, + 309, + 430 + ], + "score": 0.92, + "content": "z _ { i } ^ { \\prime } = z _ { i } + r _ { i , i + 1 } - r _ { i - 1 , i }", + "type": "inline_equation" + }, + { + "bbox": [ + 310, + 418, + 337, + 432 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 338, + 421, + 362, + 430 + ], + "score": 0.91, + "content": "r _ { i , i + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 418, + 506, + 432 + ], + "score": 1.0, + "content": "is generated with a pseudo-random", + "type": "text" + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 428, + 506, + 443 + ], + "spans": [ + { + "bbox": [ + 86, + 431, + 100, + 442 + ], + "score": 1.0, + "content": "100", + "type": "text" + }, + { + "bbox": [ + 105, + 428, + 280, + 443 + ], + "score": 1.0, + "content": "generator using a key pre-shared between", + "type": "text" + }, + { + "bbox": [ + 280, + 430, + 291, + 440 + ], + "score": 0.86, + "content": "P _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 428, + 310, + 443 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 311, + 430, + 331, + 441 + ], + "score": 0.9, + "content": "P _ { i + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 428, + 414, + 443 + ], + "score": 1.0, + "content": ". The resulting sum", + "type": "text" + }, + { + "bbox": [ + 414, + 429, + 494, + 441 + ], + "score": 0.92, + "content": "x y = \\hat { z } _ { 0 } ^ { \\prime } + z _ { 1 } ^ { \\prime } + z _ { 2 } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 428, + 506, + 443 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 439, + 506, + 453 + ], + "spans": [ + { + "bbox": [ + 86, + 442, + 99, + 452 + ], + "score": 1.0, + "content": "101", + "type": "text" + }, + { + "bbox": [ + 105, + 439, + 271, + 453 + ], + "score": 1.0, + "content": "pseudo-random, and it is thus secure for", + "type": "text" + }, + { + "bbox": [ + 271, + 441, + 282, + 451 + ], + "score": 0.88, + "content": "P _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 439, + 315, + 453 + ], + "score": 1.0, + "content": "to send", + "type": "text" + }, + { + "bbox": [ + 315, + 441, + 324, + 452 + ], + "score": 0.88, + "content": "z _ { i } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 439, + 335, + 453 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 336, + 441, + 357, + 452 + ], + "score": 0.92, + "content": "P _ { i + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 439, + 506, + 453 + ], + "score": 1.0, + "content": "in order to create a replicated secret", + "type": "text" + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 87, + 451, + 271, + 464 + ], + "spans": [ + { + "bbox": [ + 87, + 453, + 99, + 462 + ], + "score": 1.0, + "content": "102", + "type": "text" + }, + { + "bbox": [ + 105, + 451, + 139, + 464 + ], + "score": 1.0, + "content": "sharing", + "type": "text" + }, + { + "bbox": [ + 139, + 451, + 267, + 464 + ], + "score": 0.9, + "content": "\\left( ( x y ) _ { i - 1 } , ( x y ) _ { i + 1 } \\right) = \\left( z _ { i } ^ { \\prime } , z _ { i - 1 } ^ { \\prime } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 451, + 271, + 464 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true + } + ], + "index": 27, + "bbox_fs": [ + 86, + 363, + 506, + 464 + ] + }, + { + "type": "title", + "bbox": [ + 87, + 477, + 315, + 491 + ], + "lines": [ + { + "bbox": [ + 84, + 475, + 316, + 495 + ], + "spans": [ + { + "bbox": [ + 84, + 475, + 316, + 495 + ], + "score": 1.0, + "content": "103 3 Secure Computation Building Blocks", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "index", + "bbox": [ + 87, + 502, + 505, + 546 + ], + "lines": [ + { + "bbox": [ + 87, + 501, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 87, + 504, + 99, + 514 + ], + "score": 1.0, + "content": "104", + "type": "text" + }, + { + "bbox": [ + 105, + 501, + 505, + 515 + ], + "score": 1.0, + "content": "In this section, we will discuss how to implement computation with MPC with a focus on how", + "type": "text" + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 512, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 86, + 515, + 100, + 524 + ], + "score": 1.0, + "content": "105", + "type": "text" + }, + { + "bbox": [ + 105, + 512, + 505, + 525 + ], + "score": 1.0, + "content": "it differs from computation on CPUs or GPUs. Most of the techniques below are already known", + "type": "text" + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 522, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 86, + 525, + 100, + 535 + ], + "score": 1.0, + "content": "106", + "type": "text" + }, + { + "bbox": [ + 105, + 522, + 505, + 537 + ], + "score": 1.0, + "content": "individually. To the best of our knowledge however, we are the first to put them together in an efficient", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 533, + 410, + 549 + ], + "spans": [ + { + "bbox": [ + 86, + 537, + 100, + 546 + ], + "score": 1.0, + "content": "107", + "type": "text" + }, + { + "bbox": [ + 104, + 533, + 410, + 549 + ], + "score": 1.0, + "content": "and extensible framework for secure computation of deep learning training.", + "type": "text" + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 556, + 506, + 570 + ], + "spans": [ + { + "bbox": [ + 86, + 560, + 99, + 569 + ], + "score": 1.0, + "content": "108", + "type": "text" + }, + { + "bbox": [ + 105, + 556, + 377, + 570 + ], + "score": 1.0, + "content": "Domain conversion Recall we that we use computation modulo", + "type": "text" + }, + { + "bbox": [ + 377, + 557, + 388, + 568 + ], + "score": 0.87, + "content": "2 ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 556, + 404, + 570 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 404, + 558, + 411, + 568 + ], + "score": 0.8, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 556, + 506, + 570 + ], + "score": 1.0, + "content": "being a multiple of 64", + "type": "text" + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 568, + 506, + 582 + ], + "spans": [ + { + "bbox": [ + 86, + 571, + 99, + 580 + ], + "score": 1.0, + "content": "109", + "type": "text" + }, + { + "bbox": [ + 104, + 568, + 506, + 582 + ], + "score": 1.0, + "content": "as well as 1. Given that the main operations are just addition and multiplication in the respective", + "type": "text" + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 580, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 86, + 582, + 100, + 591 + ], + "score": 1.0, + "content": "110", + "type": "text" + }, + { + "bbox": [ + 105, + 580, + 506, + 592 + ], + "score": 1.0, + "content": "domain, it is desirable to compute integer arithmetic in the large domain but operations with a", + "type": "text" + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 588, + 506, + 605 + ], + "spans": [ + { + "bbox": [ + 86, + 592, + 100, + 603 + ], + "score": 1.0, + "content": "111", + "type": "text" + }, + { + "bbox": [ + 104, + 588, + 506, + 605 + ], + "score": 1.0, + "content": "straight-forward binary circuit modulo two. There has been a long-running interest in this going", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 601, + 506, + 615 + ], + "spans": [ + { + "bbox": [ + 86, + 603, + 100, + 614 + ], + "score": 1.0, + "content": "112", + "type": "text" + }, + { + "bbox": [ + 105, + 601, + 506, + 615 + ], + "score": 1.0, + "content": "back to least Kolesnikov et al. [2013]. We mainly rely on the approach proposed by Mohassel and", + "type": "text" + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 610, + 510, + 645 + ], + "spans": [ + { + "bbox": [ + 86, + 614, + 100, + 638 + ], + "score": 1.0, + "content": "113 114", + "type": "text" + }, + { + "bbox": [ + 107, + 623, + 148, + 639 + ], + "score": 0.92, + "content": "\\{ x _ { 0 } ^ { ( i ) } \\} _ { j = 0 } ^ { k - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 610, + 246, + 645 + ], + "score": 1.0, + "content": "018] and Araki et al. [20the bit decomposition of", + "type": "text" + }, + { + "bbox": [ + 246, + 626, + 257, + 636 + ], + "score": 0.85, + "content": "x _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 610, + 289, + 645 + ], + "score": 1.0, + "content": ". Recal, that is,", + "type": "text" + }, + { + "bbox": [ + 289, + 623, + 342, + 638 + ], + "score": 0.93, + "content": "x _ { 0 } ^ { ( i ) } \\in \\{ 0 , 1 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 612, + 339, + 622 + ], + "score": 0.9, + "content": "x \\in 2 ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 610, + 360, + 645 + ], + "score": 1.0, + "content": "is shand", + "type": "text" + }, + { + "bbox": [ + 360, + 623, + 435, + 638 + ], + "score": 0.94, + "content": "\\begin{array} { r } { x _ { 0 } = \\sum _ { i = 0 } ^ { k - 1 } x _ { 0 } ^ { ( i ) } 2 ^ { i } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 613, + 466, + 623 + ], + "score": 0.89, + "content": "x = x _ { 0 } + x _ { 1 } + x _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 466, + 610, + 510, + 645 + ], + "score": 1.0, + "content": ". Now letlf-evident", + "type": "text" + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 633, + 509, + 661 + ], + "spans": [ + { + "bbox": [ + 86, + 642, + 100, + 653 + ], + "score": 1.0, + "content": "115", + "type": "text" + }, + { + "bbox": [ + 103, + 633, + 124, + 661 + ], + "score": 1.0, + "content": "that", + "type": "text" + }, + { + "bbox": [ + 124, + 638, + 203, + 653 + ], + "score": 0.93, + "content": "x _ { 0 } ^ { ( i ) } = x _ { 0 } ^ { ( i ) } + 0 + 0", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 633, + 509, + 661 + ], + "score": 1.0, + "content": "is a valid secret sharing modulo two (albeit not a random one). Furthermore,", + "type": "text" + } + ], + "index": 43, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 649, + 510, + 673 + ], + "spans": [ + { + "bbox": [ + 86, + 657, + 100, + 666 + ], + "score": 1.0, + "content": "116", + "type": "text" + }, + { + "bbox": [ + 102, + 649, + 187, + 673 + ], + "score": 1.0, + "content": "every party holding", + "type": "text" + }, + { + "bbox": [ + 187, + 657, + 199, + 666 + ], + "score": 0.84, + "content": "x _ { 0 }", + "type": "inline_equation" + }, + { + "bbox": [ + 199, + 649, + 253, + 673 + ], + "score": 1.0, + "content": "can generate", + "type": "text" + }, + { + "bbox": [ + 253, + 652, + 270, + 667 + ], + "score": 0.91, + "content": "x _ { 0 } ^ { ( i ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 649, + 510, + 673 + ], + "score": 1.0, + "content": ". It is therefore possible for the parties to generate a secret", + "type": "text" + } + ], + "index": 44, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 666, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 86, + 669, + 99, + 678 + ], + "score": 1.0, + "content": "117", + "type": "text" + }, + { + "bbox": [ + 105, + 666, + 294, + 680 + ], + "score": 1.0, + "content": "sharing modulo two of a single share modulo", + "type": "text" + }, + { + "bbox": [ + 294, + 666, + 305, + 677 + ], + "score": 0.86, + "content": "2 ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 306, + 666, + 506, + 680 + ], + "score": 1.0, + "content": ". Repeating this for all shares and the computing", + "type": "text" + } + ], + "index": 45, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 677, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 86, + 680, + 100, + 689 + ], + "score": 1.0, + "content": "118", + "type": "text" + }, + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "score": 1.0, + "content": "the addition as a binary circuit allows the parties to generate a secret sharing modulo two from a", + "type": "text" + } + ], + "index": 46, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 687, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 86, + 691, + 100, + 700 + ], + "score": 1.0, + "content": "119", + "type": "text" + }, + { + "bbox": [ + 104, + 687, + 195, + 702 + ], + "score": 1.0, + "content": "secret sharing modulo", + "type": "text" + }, + { + "bbox": [ + 195, + 688, + 206, + 699 + ], + "score": 0.85, + "content": "\\bar { 2 } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 687, + 506, + 702 + ], + "score": 1.0, + "content": ". Conversion in the other direction can be achieved using a similar technique", + "type": "text" + } + ], + "index": 47, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 699, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 86, + 702, + 100, + 711 + ], + "score": 1.0, + "content": "120", + "type": "text" + }, + { + "bbox": [ + 105, + 699, + 506, + 713 + ], + "score": 1.0, + "content": "or using daBits as described by Rotaru and Wood [2019]. In the following we will use the term", + "type": "text" + } + ], + "index": 48, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 710, + 462, + 724 + ], + "spans": [ + { + "bbox": [ + 86, + 713, + 99, + 722 + ], + "score": 1.0, + "content": "121", + "type": "text" + }, + { + "bbox": [ + 104, + 710, + 462, + 724 + ], + "score": 1.0, + "content": "mixed-circuit computation for any technique that works over both computation domains.", + "type": "text" + } + ], + "index": 49, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 73, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 86, + 74, + 100, + 84 + ], + "score": 1.0, + "content": "122", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 73, + 505, + 86 + ], + "score": 1.0, + "content": "Quantization While Aliasgari et al. [2013] showed that it is possible to implement floating-point", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 84, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 86, + 86, + 100, + 96 + ], + "score": 1.0, + "content": "123", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 84, + 505, + 96 + ], + "score": 1.0, + "content": "computation, the cost is far higher than integer computation. It is therefore common to represent", + "type": "text", + "cross_page": true + } + ], + "index": 1, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 95, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 86, + 96, + 100, + 106 + ], + "score": 1.0, + "content": "124", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 95, + 506, + 107 + ], + "score": 1.0, + "content": "fractional numbers using quantization (also called fixed-point representation) as suggested by Catrina", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 103, + 506, + 120 + ], + "spans": [ + { + "bbox": [ + 86, + 108, + 100, + 117 + ], + "score": 1.0, + "content": "125", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 103, + 245, + 120 + ], + "score": 1.0, + "content": "and Saxena [2010]. A real number", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 246, + 108, + 253, + 115 + ], + "score": 0.76, + "content": "x", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 253, + 103, + 322, + 120 + ], + "score": 1.0, + "content": "is represented as", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 322, + 105, + 374, + 118 + ], + "score": 0.93, + "content": "\\bar { \\boldsymbol { x } } = \\lfloor \\boldsymbol { x } \\cdot \\mathbf { \\bar { 2 } } ^ { f } \\rceil", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 374, + 103, + 402, + 120 + ], + "score": 1.0, + "content": "where", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 402, + 106, + 409, + 117 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 410, + 103, + 506, + 120 + ], + "score": 1.0, + "content": "is an integer specifying", + "type": "text", + "cross_page": true + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 116, + 505, + 129 + ], + "spans": [ + { + "bbox": [ + 86, + 118, + 100, + 128 + ], + "score": 1.0, + "content": "126", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 116, + 505, + 129 + ], + "score": 1.0, + "content": "the precision. The linearity of the representation allows to compute addition by simply adding the", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 127, + 506, + 140 + ], + "spans": [ + { + "bbox": [ + 86, + 129, + 100, + 139 + ], + "score": 1.0, + "content": "127", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 127, + 506, + 140 + ], + "score": 1.0, + "content": "representing integers. Multiplication however requires to adjust the result because it will have twice", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 137, + 423, + 151 + ], + "spans": [ + { + "bbox": [ + 86, + 140, + 100, + 150 + ], + "score": 1.0, + "content": "128", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 137, + 164, + 151 + ], + "score": 1.0, + "content": "the precision:", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 164, + 137, + 279, + 150 + ], + "score": 0.93, + "content": "( x \\cdot 2 ^ { f } ) \\cdot ( y \\cdot \\dot { 2 } ^ { f } ) = x y \\cdot 2 ^ { 2 f }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 279, + 137, + 423, + 151 + ], + "score": 1.0, + "content": ". There are two ways to rectify this:", + "type": "text", + "cross_page": true + } + ], + "index": 6, + "is_list_start_line": true + } + ], + "index": 34.5, + "bbox_fs": [ + 86, + 501, + 505, + 549 + ] + }, + { + "type": "index", + "bbox": [ + 86, + 557, + 505, + 722 + ], + "lines": [], + "index": 43, + "bbox_fs": [ + 86, + 556, + 510, + 724 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 87, + 72, + 505, + 150 + ], + "lines": [ + { + "bbox": [ + 86, + 73, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 86, + 74, + 100, + 84 + ], + "score": 1.0, + "content": "122", + "type": "text" + }, + { + "bbox": [ + 106, + 73, + 505, + 86 + ], + "score": 1.0, + "content": "Quantization While Aliasgari et al. [2013] showed that it is possible to implement floating-point", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 86, + 84, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 86, + 86, + 100, + 96 + ], + "score": 1.0, + "content": "123", + "type": "text" + }, + { + "bbox": [ + 106, + 84, + 505, + 96 + ], + "score": 1.0, + "content": "computation, the cost is far higher than integer computation. It is therefore common to represent", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 86, + 95, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 86, + 96, + 100, + 106 + ], + "score": 1.0, + "content": "124", + "type": "text" + }, + { + "bbox": [ + 106, + 95, + 506, + 107 + ], + "score": 1.0, + "content": "fractional numbers using quantization (also called fixed-point representation) as suggested by Catrina", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 86, + 103, + 506, + 120 + ], + "spans": [ + { + "bbox": [ + 86, + 108, + 100, + 117 + ], + "score": 1.0, + "content": "125", + "type": "text" + }, + { + "bbox": [ + 104, + 103, + 245, + 120 + ], + "score": 1.0, + "content": "and Saxena [2010]. A real number", + "type": "text" + }, + { + "bbox": [ + 246, + 108, + 253, + 115 + ], + "score": 0.76, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 103, + 322, + 120 + ], + "score": 1.0, + "content": "is represented as", + "type": "text" + }, + { + "bbox": [ + 322, + 105, + 374, + 118 + ], + "score": 0.93, + "content": "\\bar { \\boldsymbol { x } } = \\lfloor \\boldsymbol { x } \\cdot \\mathbf { \\bar { 2 } } ^ { f } \\rceil", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 103, + 402, + 120 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 402, + 106, + 409, + 117 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 103, + 506, + 120 + ], + "score": 1.0, + "content": "is an integer specifying", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 86, + 116, + 505, + 129 + ], + "spans": [ + { + "bbox": [ + 86, + 118, + 100, + 128 + ], + "score": 1.0, + "content": "126", + "type": "text" + }, + { + "bbox": [ + 105, + 116, + 505, + 129 + ], + "score": 1.0, + "content": "the precision. The linearity of the representation allows to compute addition by simply adding the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 86, + 127, + 506, + 140 + ], + "spans": [ + { + "bbox": [ + 86, + 129, + 100, + 139 + ], + "score": 1.0, + "content": "127", + "type": "text" + }, + { + "bbox": [ + 104, + 127, + 506, + 140 + ], + "score": 1.0, + "content": "representing integers. Multiplication however requires to adjust the result because it will have twice", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 86, + 137, + 423, + 151 + ], + "spans": [ + { + "bbox": [ + 86, + 140, + 100, + 150 + ], + "score": 1.0, + "content": "128", + "type": "text" + }, + { + "bbox": [ + 105, + 137, + 164, + 151 + ], + "score": 1.0, + "content": "the precision:", + "type": "text" + }, + { + "bbox": [ + 164, + 137, + 279, + 150 + ], + "score": 0.93, + "content": "( x \\cdot 2 ^ { f } ) \\cdot ( y \\cdot \\dot { 2 } ^ { f } ) = x y \\cdot 2 ^ { 2 f }", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 137, + 423, + 151 + ], + "score": 1.0, + "content": ". There are two ways to rectify this:", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 134, + 159, + 505, + 204 + ], + "lines": [ + { + "bbox": [ + 133, + 158, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 133, + 158, + 353, + 173 + ], + "score": 1.0, + "content": "• An obvious correction would be to shift the result by", + "type": "text" + }, + { + "bbox": [ + 353, + 160, + 360, + 172 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 158, + 428, + 173 + ], + "score": 1.0, + "content": "bits after adding", + "type": "text" + }, + { + "bbox": [ + 428, + 159, + 450, + 170 + ], + "score": 0.91, + "content": "2 ^ { f - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 158, + 506, + 173 + ], + "score": 1.0, + "content": "to the integer", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 171, + 506, + 183 + ], + "spans": [ + { + "bbox": [ + 141, + 171, + 506, + 183 + ], + "score": 1.0, + "content": "representation. This ensures rounding to the nearest number possible in the representation,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 182, + 505, + 194 + ], + "spans": [ + { + "bbox": [ + 141, + 182, + 505, + 194 + ], + "score": 1.0, + "content": "with the tie being broken by rounding up. Dalskov et al. [2021] presented an efficient", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 142, + 193, + 410, + 205 + ], + "spans": [ + { + "bbox": [ + 142, + 193, + 410, + 205 + ], + "score": 1.0, + "content": "implementation of the truncation using mixed-circuit computation.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 133, + 208, + 505, + 285 + ], + "lines": [ + { + "bbox": [ + 132, + 208, + 506, + 221 + ], + "spans": [ + { + "bbox": [ + 132, + 208, + 506, + 221 + ], + "score": 1.0, + "content": "• However, Catrina and Saxena have found that in the context of secure computation it is more", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 219, + 504, + 231 + ], + "spans": [ + { + "bbox": [ + 141, + 219, + 504, + 231 + ], + "score": 1.0, + "content": "efficient to use probabilistic truncation. This method rounds up or down probabilistically", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 230, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 141, + 230, + 505, + 243 + ], + "score": 1.0, + "content": "depending on the input. For example, probabilistically rounding 0.75 to an integer would", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 142, + 241, + 505, + 254 + ], + "spans": [ + { + "bbox": [ + 142, + 241, + 505, + 254 + ], + "score": 1.0, + "content": "see it rounded of up with probability 0.75 and down with probability 0.25. The probabilistic", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 142, + 252, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 142, + 252, + 505, + 264 + ], + "score": 1.0, + "content": "truncation is an effect of the fact that the operation involves the truncation of a randomized", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 262, + 506, + 277 + ], + "spans": [ + { + "bbox": [ + 141, + 262, + 272, + 277 + ], + "score": 1.0, + "content": "value, that is the computation of", + "type": "text" + }, + { + "bbox": [ + 272, + 263, + 329, + 275 + ], + "score": 0.92, + "content": "\\left\\lfloor ( x + r ) / 2 ^ { m } \\right\\rfloor", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 262, + 383, + 277 + ], + "score": 1.0, + "content": "for a random", + "type": "text" + }, + { + "bbox": [ + 384, + 264, + 393, + 273 + ], + "score": 0.79, + "content": "m", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 262, + 432, + 277 + ], + "score": 1.0, + "content": "-bit value", + "type": "text" + }, + { + "bbox": [ + 432, + 265, + 438, + 273 + ], + "score": 0.74, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 262, + 506, + 277 + ], + "score": 1.0, + "content": ". It is easy to see", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 273, + 162, + 286 + ], + "spans": [ + { + "bbox": [ + 141, + 273, + 162, + 286 + ], + "score": 1.0, + "content": "that", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14 + }, + { + "type": "interline_equation", + "bbox": [ + 205, + 284, + 441, + 312 + ], + "lines": [ + { + "bbox": [ + 205, + 284, + 441, + 312 + ], + "spans": [ + { + "bbox": [ + 205, + 284, + 441, + 312 + ], + "score": 0.92, + "content": "\\lfloor ( x + r ) / 2 ^ { m } \\rfloor = \\left\\{ \\begin{array} { l l } { \\lfloor x / 2 ^ { m } \\rfloor } & { ( x \\bmod 2 ^ { m } + r ) < 2 ^ { m } } \\\\ { \\lfloor x / 2 ^ { m } \\rfloor + 1 } & { ( x \\bmod 2 ^ { m } + r ) \\ge 2 ^ { m } . } \\end{array} \\right.", + "type": "interline_equation", + "image_path": "8c5c9712adea6b5b528c5e11a746ad687533c04d65eda8941ed882721c5e70cc.jpg" + } + ] + } + ], + "index": 18.5, + "virtual_lines": [ + { + "bbox": [ + 205, + 284, + 441, + 298.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 205, + 298.0, + 441, + 312.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 141, + 316, + 504, + 339 + ], + "lines": [ + { + "bbox": [ + 142, + 316, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 142, + 316, + 231, + 329 + ], + "score": 1.0, + "content": "Therefore, the larger", + "type": "text" + }, + { + "bbox": [ + 231, + 316, + 277, + 327 + ], + "score": 0.44, + "content": "( x \\bmod 2 ^ { m } )", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 316, + 505, + 329 + ], + "score": 1.0, + "content": ") is, the more likely the latter condition is true. Dalskov", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 142, + 327, + 397, + 340 + ], + "spans": [ + { + "bbox": [ + 142, + 327, + 397, + 340 + ], + "score": 1.0, + "content": "et al. [2020] present an efficient protocol in our security model.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 87, + 347, + 506, + 381 + ], + "lines": [ + { + "bbox": [ + 86, + 347, + 506, + 361 + ], + "spans": [ + { + "bbox": [ + 86, + 350, + 100, + 360 + ], + "score": 1.0, + "content": "142", + "type": "text" + }, + { + "bbox": [ + 105, + 347, + 506, + 361 + ], + "score": 1.0, + "content": "Our quantization scheme is related to quantized neural networks (see e.g. Hubara et al. [2016]).", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 87, + 358, + 505, + 372 + ], + "spans": [ + { + "bbox": [ + 87, + 362, + 99, + 370 + ], + "score": 1.0, + "content": "143", + "type": "text" + }, + { + "bbox": [ + 104, + 358, + 505, + 372 + ], + "score": 1.0, + "content": "However, our consideration is not to compress the model, but to improve the computational speed", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 86, + 370, + 228, + 381 + ], + "spans": [ + { + "bbox": [ + 86, + 372, + 100, + 381 + ], + "score": 1.0, + "content": "144", + "type": "text" + }, + { + "bbox": [ + 105, + 370, + 228, + 381 + ], + "score": 1.0, + "content": "and save communication cost.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 86, + 393, + 505, + 475 + ], + "lines": [ + { + "bbox": [ + 86, + 393, + 506, + 406 + ], + "spans": [ + { + "bbox": [ + 86, + 396, + 100, + 405 + ], + "score": 1.0, + "content": "145", + "type": "text" + }, + { + "bbox": [ + 104, + 393, + 506, + 406 + ], + "score": 1.0, + "content": "Dot products Dot products are an essential building block of linear computation such as matrix", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 86, + 403, + 506, + 418 + ], + "spans": [ + { + "bbox": [ + 86, + 406, + 100, + 416 + ], + "score": 1.0, + "content": "146", + "type": "text" + }, + { + "bbox": [ + 105, + 403, + 506, + 418 + ], + "score": 1.0, + "content": "multiplication. In the light of quantization, it is possible to reduce the usage of truncation by deferring", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 86, + 416, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 86, + 418, + 99, + 426 + ], + "score": 1.0, + "content": "147", + "type": "text" + }, + { + "bbox": [ + 106, + 416, + 505, + 428 + ], + "score": 1.0, + "content": "after the summation. In other words, the dot product in the integer representations is computed before", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 86, + 427, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 86, + 429, + 100, + 438 + ], + "score": 1.0, + "content": "148", + "type": "text" + }, + { + "bbox": [ + 106, + 427, + 505, + 438 + ], + "score": 1.0, + "content": "truncating. This not only reduces the truncation error, it is also more efficient because the truncation", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 86, + 437, + 506, + 450 + ], + "spans": [ + { + "bbox": [ + 86, + 439, + 100, + 449 + ], + "score": 1.0, + "content": "149", + "type": "text" + }, + { + "bbox": [ + 104, + 437, + 506, + 450 + ], + "score": 1.0, + "content": "is the most expensive part in quantized secure multiplication. Similarly, our protocol allows to defer", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 86, + 448, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 86, + 450, + 100, + 460 + ], + "score": 1.0, + "content": "150", + "type": "text" + }, + { + "bbox": [ + 106, + 448, + 313, + 460 + ], + "score": 1.0, + "content": "the communication needed for multiplication. Let", + "type": "text" + }, + { + "bbox": [ + 313, + 449, + 320, + 458 + ], + "score": 0.81, + "content": "\\vec { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 448, + 338, + 460 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 339, + 448, + 346, + 460 + ], + "score": 0.85, + "content": "\\vec { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 448, + 505, + 460 + ], + "score": 1.0, + "content": "be two vectors where the elements are", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 86, + 458, + 496, + 477 + ], + "spans": [ + { + "bbox": [ + 86, + 464, + 99, + 473 + ], + "score": 1.0, + "content": "151", + "type": "text" + }, + { + "bbox": [ + 103, + 458, + 192, + 477 + ], + "score": 1.0, + "content": "secret shared, that is,", + "type": "text" + }, + { + "bbox": [ + 192, + 459, + 302, + 474 + ], + "score": 0.94, + "content": "\\{ x ^ { ( i ) } \\} = x _ { 0 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } + x _ { 2 } ^ { ( i ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 459, + 372, + 476 + ], + "score": 1.0, + "content": "and similarly for", + "type": "text" + }, + { + "bbox": [ + 372, + 461, + 388, + 474 + ], + "score": 0.91, + "content": "\\boldsymbol y ^ { ( i ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 459, + 496, + 476 + ], + "score": 1.0, + "content": ". The inner product then is", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28 + }, + { + "type": "interline_equation", + "bbox": [ + 127, + 479, + 483, + 561 + ], + "lines": [ + { + "bbox": [ + 127, + 479, + 483, + 561 + ], + "spans": [ + { + "bbox": [ + 127, + 479, + 483, + 561 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { \\displaystyle \\sum _ { i } x ^ { ( i ) } \\cdot y ^ { ( i ) } = \\sum _ { i } ( x _ { 0 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } + x _ { 2 } ^ { ( i ) } ) \\cdot ( y _ { 0 } ^ { ( i ) } + y _ { 1 } ^ { ( i ) } + y _ { 2 } ^ { ( i ) } ) } \\\\ & { \\quad \\quad \\quad \\quad = \\displaystyle \\sum _ { i } ( x _ { 0 } ^ { ( i ) } y _ { 0 } ^ { ( i ) } + x _ { 0 } ^ { ( i ) } y _ { 1 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } y _ { 0 } ^ { ( i ) } ) + \\sum _ { i } ( x _ { 1 } ^ { ( i ) } y _ { 1 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } y _ { 2 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } y _ { 1 } ^ { ( i ) } ) } \\\\ & { \\quad \\quad \\quad \\quad + \\displaystyle \\sum _ { i } ( x _ { 2 } ^ { ( i ) } y _ { 2 } ^ { ( i ) } + x _ { 2 } ^ { ( i ) } y _ { 0 } ^ { ( i ) } + x _ { 0 } ^ { ( i ) } y _ { 2 } ^ { ( i ) } ) . } \\end{array}", + "type": "interline_equation", + "image_path": "36540aeb1a5d2f89955c3e16fde0f520a4fd1af5a546d08dd4d4fe06ee6c50ec.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 127, + 479, + 483, + 506.3333333333333 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 127, + 506.3333333333333, + 483, + 533.6666666666666 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 127, + 533.6666666666666, + 483, + 561.0 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "text", + "bbox": [ + 91, + 566, + 504, + 589 + ], + "lines": [ + { + "bbox": [ + 89, + 566, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 89, + 569, + 101, + 578 + ], + "score": 1.0, + "content": "152", + "type": "text" + }, + { + "bbox": [ + 104, + 566, + 505, + 579 + ], + "score": 1.0, + "content": "The three sums in the last term can be compute locally by one party each before applying the same", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 87, + 578, + 261, + 590 + ], + "spans": [ + { + "bbox": [ + 87, + 578, + 261, + 590 + ], + "score": 1.0, + "content": "153 protocol as for a single multiplication.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 86, + 601, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 86, + 602, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 86, + 604, + 99, + 613 + ], + "score": 1.0, + "content": "154", + "type": "text" + }, + { + "bbox": [ + 106, + 602, + 505, + 614 + ], + "score": 1.0, + "content": "Comparisons Arithmetic secret sharing does not allow to access the individual bits directly. It", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 86, + 612, + 506, + 625 + ], + "spans": [ + { + "bbox": [ + 86, + 614, + 99, + 624 + ], + "score": 1.0, + "content": "155", + "type": "text" + }, + { + "bbox": [ + 105, + 612, + 506, + 625 + ], + "score": 1.0, + "content": "is therefore not straightforward to compute comparisons such as “less than”. There is a long line", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 86, + 622, + 506, + 637 + ], + "spans": [ + { + "bbox": [ + 86, + 626, + 99, + 635 + ], + "score": 1.0, + "content": "156", + "type": "text" + }, + { + "bbox": [ + 105, + 622, + 506, + 637 + ], + "score": 1.0, + "content": "of literature on how to achieve this going back to at least Damgård et al. [2006]. More recently,", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 86, + 633, + 506, + 648 + ], + "spans": [ + { + "bbox": [ + 86, + 637, + 99, + 646 + ], + "score": 1.0, + "content": "157", + "type": "text" + }, + { + "bbox": [ + 105, + 633, + 506, + 648 + ], + "score": 1.0, + "content": "most attention has been given to combine the power of arithmetic and binary secret sharing in", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 86, + 644, + 506, + 659 + ], + "spans": [ + { + "bbox": [ + 86, + 648, + 99, + 657 + ], + "score": 1.0, + "content": "158", + "type": "text" + }, + { + "bbox": [ + 105, + 644, + 506, + 659 + ], + "score": 1.0, + "content": "order to combine the best of worlds. One possibility to do so is to plainly convert to the binary", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 86, + 656, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 86, + 658, + 99, + 668 + ], + "score": 1.0, + "content": "159", + "type": "text" + }, + { + "bbox": [ + 105, + 656, + 505, + 669 + ], + "score": 1.0, + "content": "domain and compute the comparison circuit there. In our concrete implementation we use the more", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 86, + 667, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 86, + 669, + 99, + 679 + ], + "score": 1.0, + "content": "160", + "type": "text" + }, + { + "bbox": [ + 105, + 667, + 506, + 680 + ], + "score": 1.0, + "content": "efficient approach by Mohassel and Rindal [2018]. It starts by taking the difference of the two", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 86, + 678, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 86, + 680, + 99, + 690 + ], + "score": 1.0, + "content": "161", + "type": "text" + }, + { + "bbox": [ + 105, + 678, + 505, + 691 + ], + "score": 1.0, + "content": "inputs. Computing the comparison then reduces to comparing to zero, which in turn is equivalent", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 86, + 687, + 506, + 704 + ], + "spans": [ + { + "bbox": [ + 86, + 691, + 99, + 700 + ], + "score": 1.0, + "content": "162", + "type": "text" + }, + { + "bbox": [ + 104, + 687, + 506, + 704 + ], + "score": 1.0, + "content": "to extracting the most significant bit as it indicates the sign. The latter is achieved by converting", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 86, + 700, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 86, + 702, + 99, + 711 + ], + "score": 1.0, + "content": "163", + "type": "text" + }, + { + "bbox": [ + 105, + 700, + 505, + 712 + ], + "score": 1.0, + "content": "the shares locally to bit-wise sharing of the arithmetic shares, which sum up to the secret value. It", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 86, + 711, + 499, + 724 + ], + "spans": [ + { + "bbox": [ + 86, + 713, + 99, + 722 + ], + "score": 1.0, + "content": "164", + "type": "text" + }, + { + "bbox": [ + 105, + 711, + 499, + 724 + ], + "score": 1.0, + "content": "remains to compute the sum of the binary shares in order to come up with the most significant bit.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 42 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 9 + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 86, + 161, + 101, + 286 + ], + "lines": [ + { + "bbox": [ + 85, + 162, + 100, + 173 + ], + "spans": [ + { + "bbox": [ + 85, + 162, + 100, + 173 + ], + "score": 1.0, + "content": "129", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 173, + 100, + 184 + ], + "spans": [ + { + "bbox": [ + 85, + 173, + 100, + 184 + ], + "score": 1.0, + "content": "130", + "type": "text" + } + ] + }, + { + "bbox": [ + 86, + 184, + 99, + 194 + ], + "spans": [ + { + "bbox": [ + 86, + 184, + 99, + 194 + ], + "score": 1.0, + "content": "131", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 195, + 100, + 207 + ], + "spans": [ + { + "bbox": [ + 85, + 195, + 100, + 207 + ], + "score": 1.0, + "content": "132", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 210, + 100, + 222 + ], + "spans": [ + { + "bbox": [ + 85, + 210, + 100, + 222 + ], + "score": 1.0, + "content": "133", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 221, + 100, + 233 + ], + "spans": [ + { + "bbox": [ + 85, + 221, + 100, + 233 + ], + "score": 1.0, + "content": "134", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 232, + 100, + 244 + ], + "spans": [ + { + "bbox": [ + 85, + 232, + 100, + 244 + ], + "score": 1.0, + "content": "135", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 243, + 100, + 254 + ], + "spans": [ + { + "bbox": [ + 85, + 243, + 100, + 254 + ], + "score": 1.0, + "content": "136", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 254, + 100, + 265 + ], + "spans": [ + { + "bbox": [ + 85, + 254, + 100, + 265 + ], + "score": 1.0, + "content": "137", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 265, + 100, + 276 + ], + "spans": [ + { + "bbox": [ + 85, + 265, + 100, + 276 + ], + "score": 1.0, + "content": "138", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 276, + 100, + 287 + ], + "spans": [ + { + "bbox": [ + 85, + 276, + 100, + 287 + ], + "score": 1.0, + "content": "139", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "index", + "bbox": [ + 87, + 72, + 505, + 150 + ], + "lines": [], + "index": 3, + "bbox_fs": [ + 86, + 73, + 506, + 151 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 134, + 159, + 505, + 204 + ], + "lines": [ + { + "bbox": [ + 133, + 158, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 133, + 158, + 353, + 173 + ], + "score": 1.0, + "content": "• An obvious correction would be to shift the result by", + "type": "text" + }, + { + "bbox": [ + 353, + 160, + 360, + 172 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 158, + 428, + 173 + ], + "score": 1.0, + "content": "bits after adding", + "type": "text" + }, + { + "bbox": [ + 428, + 159, + 450, + 170 + ], + "score": 0.91, + "content": "2 ^ { f - 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 158, + 506, + 173 + ], + "score": 1.0, + "content": "to the integer", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 171, + 506, + 183 + ], + "spans": [ + { + "bbox": [ + 141, + 171, + 506, + 183 + ], + "score": 1.0, + "content": "representation. This ensures rounding to the nearest number possible in the representation,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 182, + 505, + 194 + ], + "spans": [ + { + "bbox": [ + 141, + 182, + 505, + 194 + ], + "score": 1.0, + "content": "with the tie being broken by rounding up. Dalskov et al. [2021] presented an efficient", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 142, + 193, + 410, + 205 + ], + "spans": [ + { + "bbox": [ + 142, + 193, + 410, + 205 + ], + "score": 1.0, + "content": "implementation of the truncation using mixed-circuit computation.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8.5, + "bbox_fs": [ + 133, + 158, + 506, + 205 + ] + }, + { + "type": "text", + "bbox": [ + 133, + 208, + 505, + 285 + ], + "lines": [ + { + "bbox": [ + 132, + 208, + 506, + 221 + ], + "spans": [ + { + "bbox": [ + 132, + 208, + 506, + 221 + ], + "score": 1.0, + "content": "• However, Catrina and Saxena have found that in the context of secure computation it is more", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 219, + 504, + 231 + ], + "spans": [ + { + "bbox": [ + 141, + 219, + 504, + 231 + ], + "score": 1.0, + "content": "efficient to use probabilistic truncation. This method rounds up or down probabilistically", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 230, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 141, + 230, + 505, + 243 + ], + "score": 1.0, + "content": "depending on the input. For example, probabilistically rounding 0.75 to an integer would", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 142, + 241, + 505, + 254 + ], + "spans": [ + { + "bbox": [ + 142, + 241, + 505, + 254 + ], + "score": 1.0, + "content": "see it rounded of up with probability 0.75 and down with probability 0.25. The probabilistic", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 142, + 252, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 142, + 252, + 505, + 264 + ], + "score": 1.0, + "content": "truncation is an effect of the fact that the operation involves the truncation of a randomized", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 262, + 506, + 277 + ], + "spans": [ + { + "bbox": [ + 141, + 262, + 272, + 277 + ], + "score": 1.0, + "content": "value, that is the computation of", + "type": "text" + }, + { + "bbox": [ + 272, + 263, + 329, + 275 + ], + "score": 0.92, + "content": "\\left\\lfloor ( x + r ) / 2 ^ { m } \\right\\rfloor", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 262, + 383, + 277 + ], + "score": 1.0, + "content": "for a random", + "type": "text" + }, + { + "bbox": [ + 384, + 264, + 393, + 273 + ], + "score": 0.79, + "content": "m", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 262, + 432, + 277 + ], + "score": 1.0, + "content": "-bit value", + "type": "text" + }, + { + "bbox": [ + 432, + 265, + 438, + 273 + ], + "score": 0.74, + "content": "r", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 262, + 506, + 277 + ], + "score": 1.0, + "content": ". It is easy to see", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 273, + 162, + 286 + ], + "spans": [ + { + "bbox": [ + 141, + 273, + 162, + 286 + ], + "score": 1.0, + "content": "that", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14, + "bbox_fs": [ + 132, + 208, + 506, + 286 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 205, + 284, + 441, + 312 + ], + "lines": [ + { + "bbox": [ + 205, + 284, + 441, + 312 + ], + "spans": [ + { + "bbox": [ + 205, + 284, + 441, + 312 + ], + "score": 0.92, + "content": "\\lfloor ( x + r ) / 2 ^ { m } \\rfloor = \\left\\{ \\begin{array} { l l } { \\lfloor x / 2 ^ { m } \\rfloor } & { ( x \\bmod 2 ^ { m } + r ) < 2 ^ { m } } \\\\ { \\lfloor x / 2 ^ { m } \\rfloor + 1 } & { ( x \\bmod 2 ^ { m } + r ) \\ge 2 ^ { m } . } \\end{array} \\right.", + "type": "interline_equation", + "image_path": "8c5c9712adea6b5b528c5e11a746ad687533c04d65eda8941ed882721c5e70cc.jpg" + } + ] + } + ], + "index": 18.5, + "virtual_lines": [ + { + "bbox": [ + 205, + 284, + 441, + 298.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 205, + 298.0, + 441, + 312.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 141, + 316, + 504, + 339 + ], + "lines": [ + { + "bbox": [ + 142, + 316, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 142, + 316, + 231, + 329 + ], + "score": 1.0, + "content": "Therefore, the larger", + "type": "text" + }, + { + "bbox": [ + 231, + 316, + 277, + 327 + ], + "score": 0.44, + "content": "( x \\bmod 2 ^ { m } )", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 316, + 505, + 329 + ], + "score": 1.0, + "content": ") is, the more likely the latter condition is true. Dalskov", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 142, + 327, + 397, + 340 + ], + "spans": [ + { + "bbox": [ + 142, + 327, + 397, + 340 + ], + "score": 1.0, + "content": "et al. [2020] present an efficient protocol in our security model.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5, + "bbox_fs": [ + 142, + 316, + 505, + 340 + ] + }, + { + "type": "index", + "bbox": [ + 87, + 347, + 506, + 381 + ], + "lines": [ + { + "bbox": [ + 86, + 347, + 506, + 361 + ], + "spans": [ + { + "bbox": [ + 86, + 350, + 100, + 360 + ], + "score": 1.0, + "content": "142", + "type": "text" + }, + { + "bbox": [ + 105, + 347, + 506, + 361 + ], + "score": 1.0, + "content": "Our quantization scheme is related to quantized neural networks (see e.g. Hubara et al. [2016]).", + "type": "text" + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 87, + 358, + 505, + 372 + ], + "spans": [ + { + "bbox": [ + 87, + 362, + 99, + 370 + ], + "score": 1.0, + "content": "143", + "type": "text" + }, + { + "bbox": [ + 104, + 358, + 505, + 372 + ], + "score": 1.0, + "content": "However, our consideration is not to compress the model, but to improve the computational speed", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 370, + 228, + 381 + ], + "spans": [ + { + "bbox": [ + 86, + 372, + 100, + 381 + ], + "score": 1.0, + "content": "144", + "type": "text" + }, + { + "bbox": [ + 105, + 370, + 228, + 381 + ], + "score": 1.0, + "content": "and save communication cost.", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 393, + 506, + 406 + ], + "spans": [ + { + "bbox": [ + 86, + 396, + 100, + 405 + ], + "score": 1.0, + "content": "145", + "type": "text" + }, + { + "bbox": [ + 104, + 393, + 506, + 406 + ], + "score": 1.0, + "content": "Dot products Dot products are an essential building block of linear computation such as matrix", + "type": "text" + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 403, + 506, + 418 + ], + "spans": [ + { + "bbox": [ + 86, + 406, + 100, + 416 + ], + "score": 1.0, + "content": "146", + "type": "text" + }, + { + "bbox": [ + 105, + 403, + 506, + 418 + ], + "score": 1.0, + "content": "multiplication. In the light of quantization, it is possible to reduce the usage of truncation by deferring", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 416, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 86, + 418, + 99, + 426 + ], + "score": 1.0, + "content": "147", + "type": "text" + }, + { + "bbox": [ + 106, + 416, + 505, + 428 + ], + "score": 1.0, + "content": "after the summation. In other words, the dot product in the integer representations is computed before", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 427, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 86, + 429, + 100, + 438 + ], + "score": 1.0, + "content": "148", + "type": "text" + }, + { + "bbox": [ + 106, + 427, + 505, + 438 + ], + "score": 1.0, + "content": "truncating. This not only reduces the truncation error, it is also more efficient because the truncation", + "type": "text" + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 437, + 506, + 450 + ], + "spans": [ + { + "bbox": [ + 86, + 439, + 100, + 449 + ], + "score": 1.0, + "content": "149", + "type": "text" + }, + { + "bbox": [ + 104, + 437, + 506, + 450 + ], + "score": 1.0, + "content": "is the most expensive part in quantized secure multiplication. Similarly, our protocol allows to defer", + "type": "text" + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 448, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 86, + 450, + 100, + 460 + ], + "score": 1.0, + "content": "150", + "type": "text" + }, + { + "bbox": [ + 106, + 448, + 313, + 460 + ], + "score": 1.0, + "content": "the communication needed for multiplication. Let", + "type": "text" + }, + { + "bbox": [ + 313, + 449, + 320, + 458 + ], + "score": 0.81, + "content": "\\vec { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 448, + 338, + 460 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 339, + 448, + 346, + 460 + ], + "score": 0.85, + "content": "\\vec { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 448, + 505, + 460 + ], + "score": 1.0, + "content": "be two vectors where the elements are", + "type": "text" + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 458, + 496, + 477 + ], + "spans": [ + { + "bbox": [ + 86, + 464, + 99, + 473 + ], + "score": 1.0, + "content": "151", + "type": "text" + }, + { + "bbox": [ + 103, + 458, + 192, + 477 + ], + "score": 1.0, + "content": "secret shared, that is,", + "type": "text" + }, + { + "bbox": [ + 192, + 459, + 302, + 474 + ], + "score": 0.94, + "content": "\\{ x ^ { ( i ) } \\} = x _ { 0 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } + x _ { 2 } ^ { ( i ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 459, + 372, + 476 + ], + "score": 1.0, + "content": "and similarly for", + "type": "text" + }, + { + "bbox": [ + 372, + 461, + 388, + 474 + ], + "score": 0.91, + "content": "\\boldsymbol y ^ { ( i ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 459, + 496, + 476 + ], + "score": 1.0, + "content": ". The inner product then is", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true + } + ], + "index": 23, + "bbox_fs": [ + 86, + 347, + 506, + 381 + ] + }, + { + "type": "index", + "bbox": [ + 86, + 393, + 505, + 475 + ], + "lines": [], + "index": 28, + "bbox_fs": [ + 86, + 393, + 506, + 477 + ], + "lines_deleted": true + }, + { + "type": "interline_equation", + "bbox": [ + 127, + 479, + 483, + 561 + ], + "lines": [ + { + "bbox": [ + 127, + 479, + 483, + 561 + ], + "spans": [ + { + "bbox": [ + 127, + 479, + 483, + 561 + ], + "score": 0.94, + "content": "\\begin{array} { r l } & { \\displaystyle \\sum _ { i } x ^ { ( i ) } \\cdot y ^ { ( i ) } = \\sum _ { i } ( x _ { 0 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } + x _ { 2 } ^ { ( i ) } ) \\cdot ( y _ { 0 } ^ { ( i ) } + y _ { 1 } ^ { ( i ) } + y _ { 2 } ^ { ( i ) } ) } \\\\ & { \\quad \\quad \\quad \\quad = \\displaystyle \\sum _ { i } ( x _ { 0 } ^ { ( i ) } y _ { 0 } ^ { ( i ) } + x _ { 0 } ^ { ( i ) } y _ { 1 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } y _ { 0 } ^ { ( i ) } ) + \\sum _ { i } ( x _ { 1 } ^ { ( i ) } y _ { 1 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } y _ { 2 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } y _ { 1 } ^ { ( i ) } ) } \\\\ & { \\quad \\quad \\quad \\quad + \\displaystyle \\sum _ { i } ( x _ { 2 } ^ { ( i ) } y _ { 2 } ^ { ( i ) } + x _ { 2 } ^ { ( i ) } y _ { 0 } ^ { ( i ) } + x _ { 0 } ^ { ( i ) } y _ { 2 } ^ { ( i ) } ) . } \\end{array}", + "type": "interline_equation", + "image_path": "36540aeb1a5d2f89955c3e16fde0f520a4fd1af5a546d08dd4d4fe06ee6c50ec.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 127, + 479, + 483, + 506.3333333333333 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 127, + 506.3333333333333, + 483, + 533.6666666666666 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 127, + 533.6666666666666, + 483, + 561.0 + ], + "spans": [], + "index": 34 + } + ] + }, + { + "type": "index", + "bbox": [ + 91, + 566, + 504, + 589 + ], + "lines": [ + { + "bbox": [ + 89, + 566, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 89, + 569, + 101, + 578 + ], + "score": 1.0, + "content": "152", + "type": "text" + }, + { + "bbox": [ + 104, + 566, + 505, + 579 + ], + "score": 1.0, + "content": "The three sums in the last term can be compute locally by one party each before applying the same", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 87, + 578, + 261, + 590 + ], + "spans": [ + { + "bbox": [ + 87, + 578, + 261, + 590 + ], + "score": 1.0, + "content": "153 protocol as for a single multiplication.", + "type": "text" + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 602, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 86, + 604, + 99, + 613 + ], + "score": 1.0, + "content": "154", + "type": "text" + }, + { + "bbox": [ + 106, + 602, + 505, + 614 + ], + "score": 1.0, + "content": "Comparisons Arithmetic secret sharing does not allow to access the individual bits directly. It", + "type": "text" + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 612, + 506, + 625 + ], + "spans": [ + { + "bbox": [ + 86, + 614, + 99, + 624 + ], + "score": 1.0, + "content": "155", + "type": "text" + }, + { + "bbox": [ + 105, + 612, + 506, + 625 + ], + "score": 1.0, + "content": "is therefore not straightforward to compute comparisons such as “less than”. There is a long line", + "type": "text" + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 622, + 506, + 637 + ], + "spans": [ + { + "bbox": [ + 86, + 626, + 99, + 635 + ], + "score": 1.0, + "content": "156", + "type": "text" + }, + { + "bbox": [ + 105, + 622, + 506, + 637 + ], + "score": 1.0, + "content": "of literature on how to achieve this going back to at least Damgård et al. [2006]. More recently,", + "type": "text" + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 633, + 506, + 648 + ], + "spans": [ + { + "bbox": [ + 86, + 637, + 99, + 646 + ], + "score": 1.0, + "content": "157", + "type": "text" + }, + { + "bbox": [ + 105, + 633, + 506, + 648 + ], + "score": 1.0, + "content": "most attention has been given to combine the power of arithmetic and binary secret sharing in", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 644, + 506, + 659 + ], + "spans": [ + { + "bbox": [ + 86, + 648, + 99, + 657 + ], + "score": 1.0, + "content": "158", + "type": "text" + }, + { + "bbox": [ + 105, + 644, + 506, + 659 + ], + "score": 1.0, + "content": "order to combine the best of worlds. One possibility to do so is to plainly convert to the binary", + "type": "text" + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 656, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 86, + 658, + 99, + 668 + ], + "score": 1.0, + "content": "159", + "type": "text" + }, + { + "bbox": [ + 105, + 656, + 505, + 669 + ], + "score": 1.0, + "content": "domain and compute the comparison circuit there. In our concrete implementation we use the more", + "type": "text" + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 667, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 86, + 669, + 99, + 679 + ], + "score": 1.0, + "content": "160", + "type": "text" + }, + { + "bbox": [ + 105, + 667, + 506, + 680 + ], + "score": 1.0, + "content": "efficient approach by Mohassel and Rindal [2018]. It starts by taking the difference of the two", + "type": "text" + } + ], + "index": 43, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 678, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 86, + 680, + 99, + 690 + ], + "score": 1.0, + "content": "161", + "type": "text" + }, + { + "bbox": [ + 105, + 678, + 505, + 691 + ], + "score": 1.0, + "content": "inputs. Computing the comparison then reduces to comparing to zero, which in turn is equivalent", + "type": "text" + } + ], + "index": 44, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 687, + 506, + 704 + ], + "spans": [ + { + "bbox": [ + 86, + 691, + 99, + 700 + ], + "score": 1.0, + "content": "162", + "type": "text" + }, + { + "bbox": [ + 104, + 687, + 506, + 704 + ], + "score": 1.0, + "content": "to extracting the most significant bit as it indicates the sign. The latter is achieved by converting", + "type": "text" + } + ], + "index": 45, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 700, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 86, + 702, + 99, + 711 + ], + "score": 1.0, + "content": "163", + "type": "text" + }, + { + "bbox": [ + 105, + 700, + 505, + 712 + ], + "score": 1.0, + "content": "the shares locally to bit-wise sharing of the arithmetic shares, which sum up to the secret value. It", + "type": "text" + } + ], + "index": 46, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 711, + 499, + 724 + ], + "spans": [ + { + "bbox": [ + 86, + 713, + 99, + 722 + ], + "score": 1.0, + "content": "164", + "type": "text" + }, + { + "bbox": [ + 105, + 711, + 499, + 724 + ], + "score": 1.0, + "content": "remains to compute the sum of the binary shares in order to come up with the most significant bit.", + "type": "text" + } + ], + "index": 47, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 72, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 86, + 74, + 99, + 84 + ], + "score": 1.0, + "content": "165", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 72, + 505, + 86 + ], + "score": 1.0, + "content": "Oblivious Selection Plain secure computation does not allow branching because the parties would", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 82, + 506, + 97 + ], + "spans": [ + { + "bbox": [ + 85, + 85, + 100, + 96 + ], + "score": 1.0, + "content": "166", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 82, + 506, + 97 + ], + "score": 1.0, + "content": "need to be aware which branch is followed. Conditional assignment can be implemented as follows", + "type": "text", + "cross_page": true + } + ], + "index": 1, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 94, + 505, + 107 + ], + "spans": [ + { + "bbox": [ + 86, + 96, + 99, + 106 + ], + "score": 1.0, + "content": "167", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 94, + 155, + 107 + ], + "score": 1.0, + "content": "however. If", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 155, + 94, + 197, + 106 + ], + "score": 0.94, + "content": "b \\in \\{ 0 , 1 \\}", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 197, + 94, + 290, + 107 + ], + "score": 1.0, + "content": "denotes the condition,", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 290, + 95, + 351, + 106 + ], + "score": 0.93, + "content": "x + b \\cdot ( y - x )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 351, + 94, + 387, + 107 + ], + "score": 1.0, + "content": "is either", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 387, + 96, + 394, + 104 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 394, + 94, + 406, + 107 + ], + "score": 1.0, + "content": "or", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 406, + 96, + 413, + 106 + ], + "score": 0.78, + "content": "y", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 414, + 94, + 471, + 107 + ], + "score": 1.0, + "content": "depending on", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 471, + 95, + 477, + 104 + ], + "score": 0.55, + "content": "b", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 477, + 94, + 505, + 107 + ], + "score": 1.0, + "content": ". If the", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 105, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 86, + 108, + 100, + 117 + ], + "score": 1.0, + "content": "168", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 105, + 309, + 118 + ], + "score": 1.0, + "content": "condition is available in binary secret sharing but", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 309, + 108, + 317, + 115 + ], + "score": 0.79, + "content": "x", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 317, + 105, + 335, + 118 + ], + "score": 1.0, + "content": "and", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 335, + 108, + 342, + 117 + ], + "score": 0.81, + "content": "y", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 342, + 105, + 459, + 118 + ], + "score": 1.0, + "content": "in arithmetic secret sharing,", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 459, + 106, + 465, + 115 + ], + "score": 0.67, + "content": "b", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 465, + 105, + 505, + 118 + ], + "score": 1.0, + "content": "has to be", + "type": "text", + "cross_page": true + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 116, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 85, + 118, + 100, + 129 + ], + "score": 1.0, + "content": "169", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 116, + 506, + 129 + ], + "score": 1.0, + "content": "converted to the latter. This can be done using a daBit as introduced by Rotaru and Wood [2019],", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 127, + 506, + 139 + ], + "spans": [ + { + "bbox": [ + 86, + 129, + 99, + 139 + ], + "score": 1.0, + "content": "170", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 127, + 506, + 139 + ], + "score": 1.0, + "content": "which is a secret random bit shared both in arithmetic and binary. It allows to mask a bit in one world", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 138, + 458, + 150 + ], + "spans": [ + { + "bbox": [ + 86, + 140, + 99, + 149 + ], + "score": 1.0, + "content": "171", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 138, + 458, + 150 + ], + "score": 1.0, + "content": "by XORing it. The result is then revealed and the masking is undone in the other world.", + "type": "text", + "cross_page": true + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 160, + 506, + 174 + ], + "spans": [ + { + "bbox": [ + 86, + 163, + 100, + 172 + ], + "score": 1.0, + "content": "172", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 160, + 506, + 174 + ], + "score": 1.0, + "content": "Division Catrina and Saxena [2010] have shown how to implement quantized division using the", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 172, + 506, + 185 + ], + "spans": [ + { + "bbox": [ + 86, + 174, + 100, + 184 + ], + "score": 1.0, + "content": "173", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 172, + 506, + 185 + ], + "score": 1.0, + "content": "algorithm by Goldschmidt [1964]. It mainly uses arithmetic and the probabilistic truncation already", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 183, + 507, + 196 + ], + "spans": [ + { + "bbox": [ + 86, + 186, + 100, + 194 + ], + "score": 1.0, + "content": "174", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 183, + 507, + 196 + ], + "score": 1.0, + "content": "explained. In addition, the initial approximation requires a full bit decomposition as described above.", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 193, + 507, + 207 + ], + "spans": [ + { + "bbox": [ + 86, + 196, + 100, + 206 + ], + "score": 1.0, + "content": "175", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 193, + 507, + 207 + ], + "score": 1.0, + "content": "The error of the output depends on the error in the multiplications used for Goldschmidt’s iteration,", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 205, + 506, + 218 + ], + "spans": [ + { + "bbox": [ + 86, + 207, + 100, + 216 + ], + "score": 1.0, + "content": "176", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 205, + 506, + 218 + ], + "score": 1.0, + "content": "which compounds in particular when using probabilistic truncation. Due to the nature of secure", + "type": "text", + "cross_page": true + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 216, + 505, + 229 + ], + "spans": [ + { + "bbox": [ + 85, + 218, + 100, + 228 + ], + "score": 1.0, + "content": "177", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 216, + 505, + 229 + ], + "score": 1.0, + "content": "computation, the result of division by zero is undefined. One could obtain a secret failure bit by", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 227, + 506, + 239 + ], + "spans": [ + { + "bbox": [ + 85, + 228, + 100, + 239 + ], + "score": 1.0, + "content": "178", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 227, + 506, + 239 + ], + "score": 1.0, + "content": "testing the divisor to zero. However, we found that not to be necessary in our algorithm. This is", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 237, + 506, + 250 + ], + "spans": [ + { + "bbox": [ + 85, + 239, + 100, + 250 + ], + "score": 1.0, + "content": "179", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 237, + 506, + 250 + ], + "score": 1.0, + "content": "because we only use division by secret value only for the softmax function where the it is guaranteed", + "type": "text", + "cross_page": true + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 249, + 183, + 261 + ], + "spans": [ + { + "bbox": [ + 86, + 251, + 100, + 260 + ], + "score": 1.0, + "content": "180", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 249, + 183, + 261 + ], + "score": 1.0, + "content": "to strictly positive.", + "type": "text", + "cross_page": true + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 271, + 506, + 284 + ], + "spans": [ + { + "bbox": [ + 86, + 273, + 99, + 283 + ], + "score": 1.0, + "content": "181", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 271, + 506, + 284 + ], + "score": 1.0, + "content": "Logarithm Computation logarithm with any public base can be reduced to logarithm to base two", + "type": "text", + "cross_page": true + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 282, + 506, + 295 + ], + "spans": [ + { + "bbox": [ + 86, + 284, + 100, + 294 + ], + "score": 1.0, + "content": "182", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 282, + 132, + 295 + ], + "score": 1.0, + "content": "using", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 132, + 283, + 228, + 294 + ], + "score": 0.85, + "content": "\\log _ { x } y = \\log _ { 2 } y \\cdot \\log _ { x } 2", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 228, + 282, + 433, + 295 + ], + "score": 1.0, + "content": ". Aly and Smart [2019] have proposed computing", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 433, + 282, + 476, + 294 + ], + "score": 0.92, + "content": "y = a \\cdot 2 ^ { b }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 477, + 282, + 506, + 295 + ], + "score": 1.0, + "content": "where", + "type": "text", + "cross_page": true + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 293, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 86, + 295, + 100, + 306 + ], + "score": 1.0, + "content": "183", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 294, + 153, + 305 + ], + "score": 0.88, + "content": "a \\in [ 0 . 5 , 1 )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 154, + 293, + 172, + 306 + ], + "score": 1.0, + "content": "and", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 172, + 294, + 196, + 304 + ], + "score": 0.89, + "content": "b \\in \\mathbb { Z }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 197, + 293, + 316, + 306 + ], + "score": 1.0, + "content": ". This then allows to compute", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 316, + 294, + 396, + 305 + ], + "score": 0.91, + "content": "\\log _ { 2 } y = \\log _ { 2 } a + b", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 396, + 293, + 505, + 306 + ], + "score": 1.0, + "content": ". Given the restricted range", + "type": "text", + "cross_page": true + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 303, + 505, + 317 + ], + "spans": [ + { + "bbox": [ + 86, + 306, + 100, + 316 + ], + "score": 1.0, + "content": "184", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 303, + 117, + 317 + ], + "score": 1.0, + "content": "of", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 117, + 307, + 123, + 315 + ], + "score": 0.53, + "content": "a", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 124, + 303, + 127, + 317 + ], + "score": 1.0, + "content": ",", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 127, + 306, + 153, + 316 + ], + "score": 0.77, + "content": "\\log _ { 2 } a", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 153, + 303, + 505, + 317 + ], + "score": 1.0, + "content": "can be approximated using a division of polynomials. Numerical stability and input range", + "type": "text", + "cross_page": true + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 315, + 506, + 328 + ], + "spans": [ + { + "bbox": [ + 86, + 317, + 100, + 326 + ], + "score": 1.0, + "content": "185", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 315, + 506, + 328 + ], + "score": 1.0, + "content": "control are less of an issue here because we only use logarithm for the loss computation, which does", + "type": "text", + "cross_page": true + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 326, + 211, + 339 + ], + "spans": [ + { + "bbox": [ + 86, + 328, + 100, + 338 + ], + "score": 1.0, + "content": "186", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 326, + 211, + 339 + ], + "score": 1.0, + "content": "not influence the training.", + "type": "text", + "cross_page": true + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 347, + 506, + 362 + ], + "spans": [ + { + "bbox": [ + 86, + 351, + 100, + 361 + ], + "score": 1.0, + "content": "187", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 347, + 222, + 362 + ], + "score": 1.0, + "content": "Exponentiation By using", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 222, + 348, + 280, + 360 + ], + "score": 0.92, + "content": "x ^ { y } = 2 ^ { y \\log _ { 2 } x }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 280, + 347, + 506, + 362 + ], + "score": 1.0, + "content": ", any exponentiation can be reduced to exponentiation", + "type": "text", + "cross_page": true + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 358, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 86, + 363, + 100, + 372 + ], + "score": 1.0, + "content": "188", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 358, + 370, + 375 + ], + "score": 1.0, + "content": "with base two. Aly and Smart [2019] have shown how to compute", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 371, + 360, + 446, + 371 + ], + "score": 0.9, + "content": "2 ^ { a } = 2 ^ { \\lfloor a \\rfloor } \\cdot 2 ^ { a - \\lfloor a \\rfloor }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 446, + 358, + 506, + 375 + ], + "score": 1.0, + "content": "by computing", + "type": "text", + "cross_page": true + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 372, + 506, + 385 + ], + "spans": [ + { + "bbox": [ + 86, + 374, + 100, + 384 + ], + "score": 1.0, + "content": "189", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 372, + 506, + 385 + ], + "score": 1.0, + "content": "the two exponents using bit decomposition and the second factor using a polynomial approximation.", + "type": "text", + "cross_page": true + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 382, + 402, + 397 + ], + "spans": [ + { + "bbox": [ + 86, + 385, + 100, + 395 + ], + "score": 1.0, + "content": "190", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 382, + 219, + 397 + ], + "score": 1.0, + "content": "Regarding the first factor, if", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 219, + 383, + 279, + 397 + ], + "score": 0.94, + "content": "\\begin{array} { r } { b = \\sum _ { i \\geq 0 } b _ { i } 2 ^ { i } } \\end{array}", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 280, + 382, + 352, + 397 + ], + "score": 1.0, + "content": "is an integer with", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 352, + 383, + 397, + 395 + ], + "score": 0.93, + "content": "b _ { i } \\in \\{ 0 , 1 \\}", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 397, + 382, + 402, + 397 + ], + "score": 1.0, + "content": ",", + "type": "text", + "cross_page": true + } + ], + "index": 25, + "is_list_start_line": true + } + ], + "index": 35.5, + "bbox_fs": [ + 87, + 566, + 505, + 590 + ] + }, + { + "type": "index", + "bbox": [ + 86, + 601, + 505, + 722 + ], + "lines": [], + "index": 42, + "bbox_fs": [ + 86, + 602, + 506, + 724 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 87, + 72, + 505, + 150 + ], + "lines": [ + { + "bbox": [ + 86, + 72, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 86, + 74, + 99, + 84 + ], + "score": 1.0, + "content": "165", + "type": "text" + }, + { + "bbox": [ + 106, + 72, + 505, + 86 + ], + "score": 1.0, + "content": "Oblivious Selection Plain secure computation does not allow branching because the parties would", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 85, + 82, + 506, + 97 + ], + "spans": [ + { + "bbox": [ + 85, + 85, + 100, + 96 + ], + "score": 1.0, + "content": "166", + "type": "text" + }, + { + "bbox": [ + 105, + 82, + 506, + 97 + ], + "score": 1.0, + "content": "need to be aware which branch is followed. Conditional assignment can be implemented as follows", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 86, + 94, + 505, + 107 + ], + "spans": [ + { + "bbox": [ + 86, + 96, + 99, + 106 + ], + "score": 1.0, + "content": "167", + "type": "text" + }, + { + "bbox": [ + 106, + 94, + 155, + 107 + ], + "score": 1.0, + "content": "however. If", + "type": "text" + }, + { + "bbox": [ + 155, + 94, + 197, + 106 + ], + "score": 0.94, + "content": "b \\in \\{ 0 , 1 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 94, + 290, + 107 + ], + "score": 1.0, + "content": "denotes the condition,", + "type": "text" + }, + { + "bbox": [ + 290, + 95, + 351, + 106 + ], + "score": 0.93, + "content": "x + b \\cdot ( y - x )", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 94, + 387, + 107 + ], + "score": 1.0, + "content": "is either", + "type": "text" + }, + { + "bbox": [ + 387, + 96, + 394, + 104 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 94, + 406, + 107 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 406, + 96, + 413, + 106 + ], + "score": 0.78, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 94, + 471, + 107 + ], + "score": 1.0, + "content": "depending on", + "type": "text" + }, + { + "bbox": [ + 471, + 95, + 477, + 104 + ], + "score": 0.55, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 94, + 505, + 107 + ], + "score": 1.0, + "content": ". If the", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 86, + 105, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 86, + 108, + 100, + 117 + ], + "score": 1.0, + "content": "168", + "type": "text" + }, + { + "bbox": [ + 106, + 105, + 309, + 118 + ], + "score": 1.0, + "content": "condition is available in binary secret sharing but", + "type": "text" + }, + { + "bbox": [ + 309, + 108, + 317, + 115 + ], + "score": 0.79, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 105, + 335, + 118 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 335, + 108, + 342, + 117 + ], + "score": 0.81, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 105, + 459, + 118 + ], + "score": 1.0, + "content": "in arithmetic secret sharing,", + "type": "text" + }, + { + "bbox": [ + 459, + 106, + 465, + 115 + ], + "score": 0.67, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 105, + 505, + 118 + ], + "score": 1.0, + "content": "has to be", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 85, + 116, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 85, + 118, + 100, + 129 + ], + "score": 1.0, + "content": "169", + "type": "text" + }, + { + "bbox": [ + 106, + 116, + 506, + 129 + ], + "score": 1.0, + "content": "converted to the latter. This can be done using a daBit as introduced by Rotaru and Wood [2019],", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 86, + 127, + 506, + 139 + ], + "spans": [ + { + "bbox": [ + 86, + 129, + 99, + 139 + ], + "score": 1.0, + "content": "170", + "type": "text" + }, + { + "bbox": [ + 105, + 127, + 506, + 139 + ], + "score": 1.0, + "content": "which is a secret random bit shared both in arithmetic and binary. It allows to mask a bit in one world", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 86, + 138, + 458, + 150 + ], + "spans": [ + { + "bbox": [ + 86, + 140, + 99, + 149 + ], + "score": 1.0, + "content": "171", + "type": "text" + }, + { + "bbox": [ + 106, + 138, + 458, + 150 + ], + "score": 1.0, + "content": "by XORing it. The result is then revealed and the masking is undone in the other world.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 86, + 161, + 505, + 260 + ], + "lines": [ + { + "bbox": [ + 86, + 160, + 506, + 174 + ], + "spans": [ + { + "bbox": [ + 86, + 163, + 100, + 172 + ], + "score": 1.0, + "content": "172", + "type": "text" + }, + { + "bbox": [ + 104, + 160, + 506, + 174 + ], + "score": 1.0, + "content": "Division Catrina and Saxena [2010] have shown how to implement quantized division using the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 86, + 172, + 506, + 185 + ], + "spans": [ + { + "bbox": [ + 86, + 174, + 100, + 184 + ], + "score": 1.0, + "content": "173", + "type": "text" + }, + { + "bbox": [ + 105, + 172, + 506, + 185 + ], + "score": 1.0, + "content": "algorithm by Goldschmidt [1964]. It mainly uses arithmetic and the probabilistic truncation already", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 86, + 183, + 507, + 196 + ], + "spans": [ + { + "bbox": [ + 86, + 186, + 100, + 194 + ], + "score": 1.0, + "content": "174", + "type": "text" + }, + { + "bbox": [ + 105, + 183, + 507, + 196 + ], + "score": 1.0, + "content": "explained. In addition, the initial approximation requires a full bit decomposition as described above.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 86, + 193, + 507, + 207 + ], + "spans": [ + { + "bbox": [ + 86, + 196, + 100, + 206 + ], + "score": 1.0, + "content": "175", + "type": "text" + }, + { + "bbox": [ + 105, + 193, + 507, + 207 + ], + "score": 1.0, + "content": "The error of the output depends on the error in the multiplications used for Goldschmidt’s iteration,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 86, + 205, + 506, + 218 + ], + "spans": [ + { + "bbox": [ + 86, + 207, + 100, + 216 + ], + "score": 1.0, + "content": "176", + "type": "text" + }, + { + "bbox": [ + 105, + 205, + 506, + 218 + ], + "score": 1.0, + "content": "which compounds in particular when using probabilistic truncation. Due to the nature of secure", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 85, + 216, + 505, + 229 + ], + "spans": [ + { + "bbox": [ + 85, + 218, + 100, + 228 + ], + "score": 1.0, + "content": "177", + "type": "text" + }, + { + "bbox": [ + 105, + 216, + 505, + 229 + ], + "score": 1.0, + "content": "computation, the result of division by zero is undefined. One could obtain a secret failure bit by", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 85, + 227, + 506, + 239 + ], + "spans": [ + { + "bbox": [ + 85, + 228, + 100, + 239 + ], + "score": 1.0, + "content": "178", + "type": "text" + }, + { + "bbox": [ + 106, + 227, + 506, + 239 + ], + "score": 1.0, + "content": "testing the divisor to zero. However, we found that not to be necessary in our algorithm. This is", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 85, + 237, + 506, + 250 + ], + "spans": [ + { + "bbox": [ + 85, + 239, + 100, + 250 + ], + "score": 1.0, + "content": "179", + "type": "text" + }, + { + "bbox": [ + 105, + 237, + 506, + 250 + ], + "score": 1.0, + "content": "because we only use division by secret value only for the softmax function where the it is guaranteed", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 86, + 249, + 183, + 261 + ], + "spans": [ + { + "bbox": [ + 86, + 251, + 100, + 260 + ], + "score": 1.0, + "content": "180", + "type": "text" + }, + { + "bbox": [ + 105, + 249, + 183, + 261 + ], + "score": 1.0, + "content": "to strictly positive.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 86, + 271, + 505, + 338 + ], + "lines": [ + { + "bbox": [ + 86, + 271, + 506, + 284 + ], + "spans": [ + { + "bbox": [ + 86, + 273, + 99, + 283 + ], + "score": 1.0, + "content": "181", + "type": "text" + }, + { + "bbox": [ + 105, + 271, + 506, + 284 + ], + "score": 1.0, + "content": "Logarithm Computation logarithm with any public base can be reduced to logarithm to base two", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 86, + 282, + 506, + 295 + ], + "spans": [ + { + "bbox": [ + 86, + 284, + 100, + 294 + ], + "score": 1.0, + "content": "182", + "type": "text" + }, + { + "bbox": [ + 104, + 282, + 132, + 295 + ], + "score": 1.0, + "content": "using", + "type": "text" + }, + { + "bbox": [ + 132, + 283, + 228, + 294 + ], + "score": 0.85, + "content": "\\log _ { x } y = \\log _ { 2 } y \\cdot \\log _ { x } 2", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 282, + 433, + 295 + ], + "score": 1.0, + "content": ". Aly and Smart [2019] have proposed computing", + "type": "text" + }, + { + "bbox": [ + 433, + 282, + 476, + 294 + ], + "score": 0.92, + "content": "y = a \\cdot 2 ^ { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 282, + 506, + 295 + ], + "score": 1.0, + "content": "where", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 86, + 293, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 86, + 295, + 100, + 306 + ], + "score": 1.0, + "content": "183", + "type": "text" + }, + { + "bbox": [ + 106, + 294, + 153, + 305 + ], + "score": 0.88, + "content": "a \\in [ 0 . 5 , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 293, + 172, + 306 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 172, + 294, + 196, + 304 + ], + "score": 0.89, + "content": "b \\in \\mathbb { Z }", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 293, + 316, + 306 + ], + "score": 1.0, + "content": ". This then allows to compute", + "type": "text" + }, + { + "bbox": [ + 316, + 294, + 396, + 305 + ], + "score": 0.91, + "content": "\\log _ { 2 } y = \\log _ { 2 } a + b", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 293, + 505, + 306 + ], + "score": 1.0, + "content": ". Given the restricted range", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 86, + 303, + 505, + 317 + ], + "spans": [ + { + "bbox": [ + 86, + 306, + 100, + 316 + ], + "score": 1.0, + "content": "184", + "type": "text" + }, + { + "bbox": [ + 105, + 303, + 117, + 317 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 117, + 307, + 123, + 315 + ], + "score": 0.53, + "content": "a", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 303, + 127, + 317 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 127, + 306, + 153, + 316 + ], + "score": 0.77, + "content": "\\log _ { 2 } a", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 303, + 505, + 317 + ], + "score": 1.0, + "content": "can be approximated using a division of polynomials. Numerical stability and input range", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 86, + 315, + 506, + 328 + ], + "spans": [ + { + "bbox": [ + 86, + 317, + 100, + 326 + ], + "score": 1.0, + "content": "185", + "type": "text" + }, + { + "bbox": [ + 105, + 315, + 506, + 328 + ], + "score": 1.0, + "content": "control are less of an issue here because we only use logarithm for the loss computation, which does", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 86, + 326, + 211, + 339 + ], + "spans": [ + { + "bbox": [ + 86, + 328, + 100, + 338 + ], + "score": 1.0, + "content": "186", + "type": "text" + }, + { + "bbox": [ + 105, + 326, + 211, + 339 + ], + "score": 1.0, + "content": "not influence the training.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 87, + 348, + 506, + 396 + ], + "lines": [ + { + "bbox": [ + 86, + 347, + 506, + 362 + ], + "spans": [ + { + "bbox": [ + 86, + 351, + 100, + 361 + ], + "score": 1.0, + "content": "187", + "type": "text" + }, + { + "bbox": [ + 105, + 347, + 222, + 362 + ], + "score": 1.0, + "content": "Exponentiation By using", + "type": "text" + }, + { + "bbox": [ + 222, + 348, + 280, + 360 + ], + "score": 0.92, + "content": "x ^ { y } = 2 ^ { y \\log _ { 2 } x }", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 347, + 506, + 362 + ], + "score": 1.0, + "content": ", any exponentiation can be reduced to exponentiation", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 86, + 358, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 86, + 363, + 100, + 372 + ], + "score": 1.0, + "content": "188", + "type": "text" + }, + { + "bbox": [ + 104, + 358, + 370, + 375 + ], + "score": 1.0, + "content": "with base two. Aly and Smart [2019] have shown how to compute", + "type": "text" + }, + { + "bbox": [ + 371, + 360, + 446, + 371 + ], + "score": 0.9, + "content": "2 ^ { a } = 2 ^ { \\lfloor a \\rfloor } \\cdot 2 ^ { a - \\lfloor a \\rfloor }", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 358, + 506, + 375 + ], + "score": 1.0, + "content": "by computing", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 86, + 372, + 506, + 385 + ], + "spans": [ + { + "bbox": [ + 86, + 374, + 100, + 384 + ], + "score": 1.0, + "content": "189", + "type": "text" + }, + { + "bbox": [ + 106, + 372, + 506, + 385 + ], + "score": 1.0, + "content": "the two exponents using bit decomposition and the second factor using a polynomial approximation.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 86, + 382, + 402, + 397 + ], + "spans": [ + { + "bbox": [ + 86, + 385, + 100, + 395 + ], + "score": 1.0, + "content": "190", + "type": "text" + }, + { + "bbox": [ + 104, + 382, + 219, + 397 + ], + "score": 1.0, + "content": "Regarding the first factor, if", + "type": "text" + }, + { + "bbox": [ + 219, + 383, + 279, + 397 + ], + "score": 0.94, + "content": "\\begin{array} { r } { b = \\sum _ { i \\geq 0 } b _ { i } 2 ^ { i } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 382, + 352, + 397 + ], + "score": 1.0, + "content": "is an integer with", + "type": "text" + }, + { + "bbox": [ + 352, + 383, + 397, + 395 + ], + "score": 0.93, + "content": "b _ { i } \\in \\{ 0 , 1 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 382, + 402, + 397 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23.5 + }, + { + "type": "interline_equation", + "bbox": [ + 196, + 401, + 415, + 430 + ], + "lines": [ + { + "bbox": [ + 196, + 401, + 415, + 430 + ], + "spans": [ + { + "bbox": [ + 196, + 401, + 415, + 430 + ], + "score": 0.91, + "content": "2 ^ { b } = 2 ^ { \\sum _ { i \\geq 0 } b _ { i } 2 ^ { i } } = \\prod _ { i \\geq 0 } 2 ^ { b _ { i } 2 ^ { i } } = \\prod _ { i \\geq 0 } ( 1 + b _ { i } \\cdot ( 2 ^ { 2 ^ { i } } - 1 ) ) .", + "type": "interline_equation", + "image_path": "8c23276368b1b8c7d81c896fefea584c2c7d3d49e6ccd58490eca1b2a9f47135.jpg" + } + ] + } + ], + "index": 26.5, + "virtual_lines": [ + { + "bbox": [ + 196, + 401, + 415, + 415.5 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 196, + 415.5, + 415, + 430.0 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 92, + 434, + 465, + 447 + ], + "lines": [ + { + "bbox": [ + 88, + 434, + 467, + 448 + ], + "spans": [ + { + "bbox": [ + 88, + 434, + 467, + 448 + ], + "score": 1.0, + "content": "191 As with division the numerical stability depends on the truncation used for multiplication.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 100, + 457, + 505, + 492 + ], + "lines": [ + { + "bbox": [ + 105, + 457, + 507, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 507, + 471 + ], + "score": 1.0, + "content": "Inverse square root Aly and Smart [2019] have proposed to compute square root using Gold-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 469, + 505, + 481 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 505, + 481 + ], + "score": 1.0, + "content": "schmidt and Raphson-Newton iterations. We could combine this with the division above. However, Lu", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 480, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 506, + 492 + ], + "score": 1.0, + "content": "et al. [2020] have proposed a more direct computation that avoids running two successive iterations.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 105, + 502, + 505, + 570 + ], + "lines": [ + { + "bbox": [ + 106, + 502, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 106, + 502, + 446, + 515 + ], + "score": 1.0, + "content": "Uniformly random fractional number Limiting ourselves to intervals of the form", + "type": "text" + }, + { + "bbox": [ + 446, + 503, + 490, + 515 + ], + "score": 0.92, + "content": "[ x , x + 2 ^ { e } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 490, + 502, + 506, + 515 + ], + "score": 1.0, + "content": "for", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 514, + 506, + 526 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 222, + 526 + ], + "score": 1.0, + "content": "a potentially negative integer", + "type": "text" + }, + { + "bbox": [ + 223, + 516, + 228, + 524 + ], + "score": 0.53, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 514, + 426, + 526 + ], + "score": 1.0, + "content": ", we can reduce the problem to generate a random", + "type": "text" + }, + { + "bbox": [ + 427, + 514, + 457, + 525 + ], + "score": 0.92, + "content": "( f + { \\dot { e } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 458, + 514, + 506, + 526 + ], + "score": 1.0, + "content": "-bit number", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 524, + 502, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 133, + 538 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 526, + 141, + 537 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 524, + 443, + 538 + ], + "score": 1.0, + "content": "is the fixed-point precision. Recall that we represent a fractional number", + "type": "text" + }, + { + "bbox": [ + 443, + 528, + 451, + 536 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 524, + 464, + 538 + ], + "score": 1.0, + "content": "as", + "type": "text" + }, + { + "bbox": [ + 464, + 525, + 502, + 538 + ], + "score": 0.93, + "content": "\\lfloor x \\cdot 2 ^ { - f } \\rceil", + "type": "inline_equation" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 537, + 505, + 549 + ], + "spans": [ + { + "bbox": [ + 106, + 537, + 193, + 549 + ], + "score": 1.0, + "content": "Generating a random", + "type": "text" + }, + { + "bbox": [ + 193, + 539, + 200, + 547 + ], + "score": 0.76, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 537, + 505, + 549 + ], + "score": 1.0, + "content": "-bit number is straight-forward using random bits, which in our protocol can", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 548, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 548, + 505, + 559 + ], + "score": 1.0, + "content": "be generated as presented by Damgård et al. [2019]. In the context of our protocol however, Dalskov", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 559, + 493, + 570 + ], + "spans": [ + { + "bbox": [ + 106, + 559, + 493, + 570 + ], + "score": 1.0, + "content": "et al. [2021] have a presented a more efficient approach that involves mixed-circuit computation.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 105, + 581, + 505, + 615 + ], + "lines": [ + { + "bbox": [ + 106, + 581, + 505, + 594 + ], + "spans": [ + { + "bbox": [ + 106, + 581, + 505, + 594 + ], + "score": 1.0, + "content": "Communication cost Table 1 show the total communication cost of some of the building blocks", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 104, + 590, + 507, + 606 + ], + "spans": [ + { + "bbox": [ + 104, + 590, + 184, + 606 + ], + "score": 1.0, + "content": "in our protocol for", + "type": "text" + }, + { + "bbox": [ + 184, + 593, + 215, + 604 + ], + "score": 0.9, + "content": "f = 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 590, + 363, + 606 + ], + "score": 1.0, + "content": ". This setting mandates the modulus", + "type": "text" + }, + { + "bbox": [ + 364, + 592, + 382, + 603 + ], + "score": 0.86, + "content": "2 ^ { 1 2 8 }", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 590, + 507, + 606 + ], + "score": 1.0, + "content": "because the division protocol", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 603, + 198, + 616 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 181, + 616 + ], + "score": 1.0, + "content": "requires bit length", + "type": "text" + }, + { + "bbox": [ + 182, + 604, + 194, + 615 + ], + "score": 0.86, + "content": "4 f", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 603, + 198, + 616 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39 + }, + { + "type": "title", + "bbox": [ + 101, + 630, + 304, + 644 + ], + "lines": [ + { + "bbox": [ + 105, + 628, + 306, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 306, + 647 + ], + "score": 1.0, + "content": "4 Machine Learning Building Blocks", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 107, + 654, + 504, + 677 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "score": 1.0, + "content": "In this section, we will use the building blocks in Section 3 to construct high-level computational", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 664, + 216, + 679 + ], + "spans": [ + { + "bbox": [ + 105, + 664, + 216, + 679 + ], + "score": 1.0, + "content": "modules for deep learning.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42.5 + }, + { + "type": "text", + "bbox": [ + 106, + 689, + 504, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "Fully connected layers Both forward and back-propagation of fully connected layers can be seen", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 700, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 106, + 700, + 505, + 713 + ], + "score": 1.0, + "content": "as matrix multiplications and thus can be implemented using dot products. A particular challenge in", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 711, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 506, + 723 + ], + "score": 1.0, + "content": "secure computation is to compute a number of outputs in parallel in order to save communication", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 740, + 309, + 753 + ], + "spans": [ + { + "bbox": [ + 302, + 740, + 309, + 753 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "index", + "bbox": [ + 87, + 72, + 505, + 150 + ], + "lines": [], + "index": 3, + "bbox_fs": [ + 85, + 72, + 506, + 150 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 86, + 161, + 505, + 260 + ], + "lines": [], + "index": 11, + "bbox_fs": [ + 85, + 160, + 507, + 261 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 86, + 271, + 505, + 338 + ], + "lines": [], + "index": 18.5, + "bbox_fs": [ + 86, + 271, + 506, + 339 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 87, + 348, + 506, + 396 + ], + "lines": [], + "index": 23.5, + "bbox_fs": [ + 86, + 347, + 506, + 397 + ], + "lines_deleted": true + }, + { + "type": "interline_equation", + "bbox": [ + 196, + 401, + 415, + 430 + ], + "lines": [ + { + "bbox": [ + 196, + 401, + 415, + 430 + ], + "spans": [ + { + "bbox": [ + 196, + 401, + 415, + 430 + ], + "score": 0.91, + "content": "2 ^ { b } = 2 ^ { \\sum _ { i \\geq 0 } b _ { i } 2 ^ { i } } = \\prod _ { i \\geq 0 } 2 ^ { b _ { i } 2 ^ { i } } = \\prod _ { i \\geq 0 } ( 1 + b _ { i } \\cdot ( 2 ^ { 2 ^ { i } } - 1 ) ) .", + "type": "interline_equation", + "image_path": "8c23276368b1b8c7d81c896fefea584c2c7d3d49e6ccd58490eca1b2a9f47135.jpg" + } + ] + } + ], + "index": 26.5, + "virtual_lines": [ + { + "bbox": [ + 196, + 401, + 415, + 415.5 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 196, + 415.5, + 415, + 430.0 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 92, + 434, + 465, + 447 + ], + "lines": [ + { + "bbox": [ + 88, + 434, + 467, + 448 + ], + "spans": [ + { + "bbox": [ + 88, + 434, + 467, + 448 + ], + "score": 1.0, + "content": "191 As with division the numerical stability depends on the truncation used for multiplication.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28, + "bbox_fs": [ + 88, + 434, + 467, + 448 + ] + }, + { + "type": "text", + "bbox": [ + 100, + 457, + 505, + 492 + ], + "lines": [ + { + "bbox": [ + 105, + 457, + 507, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 507, + 471 + ], + "score": 1.0, + "content": "Inverse square root Aly and Smart [2019] have proposed to compute square root using Gold-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 469, + 505, + 481 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 505, + 481 + ], + "score": 1.0, + "content": "schmidt and Raphson-Newton iterations. We could combine this with the division above. However, Lu", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 480, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 506, + 492 + ], + "score": 1.0, + "content": "et al. [2020] have proposed a more direct computation that avoids running two successive iterations.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 457, + 507, + 492 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 502, + 505, + 570 + ], + "lines": [ + { + "bbox": [ + 106, + 502, + 506, + 515 + ], + "spans": [ + { + "bbox": [ + 106, + 502, + 446, + 515 + ], + "score": 1.0, + "content": "Uniformly random fractional number Limiting ourselves to intervals of the form", + "type": "text" + }, + { + "bbox": [ + 446, + 503, + 490, + 515 + ], + "score": 0.92, + "content": "[ x , x + 2 ^ { e } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 490, + 502, + 506, + 515 + ], + "score": 1.0, + "content": "for", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 514, + 506, + 526 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 222, + 526 + ], + "score": 1.0, + "content": "a potentially negative integer", + "type": "text" + }, + { + "bbox": [ + 223, + 516, + 228, + 524 + ], + "score": 0.53, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 514, + 426, + 526 + ], + "score": 1.0, + "content": ", we can reduce the problem to generate a random", + "type": "text" + }, + { + "bbox": [ + 427, + 514, + 457, + 525 + ], + "score": 0.92, + "content": "( f + { \\dot { e } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 458, + 514, + 506, + 526 + ], + "score": 1.0, + "content": "-bit number", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 524, + 502, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 133, + 538 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 134, + 526, + 141, + 537 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 524, + 443, + 538 + ], + "score": 1.0, + "content": "is the fixed-point precision. Recall that we represent a fractional number", + "type": "text" + }, + { + "bbox": [ + 443, + 528, + 451, + 536 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 451, + 524, + 464, + 538 + ], + "score": 1.0, + "content": "as", + "type": "text" + }, + { + "bbox": [ + 464, + 525, + 502, + 538 + ], + "score": 0.93, + "content": "\\lfloor x \\cdot 2 ^ { - f } \\rceil", + "type": "inline_equation" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 537, + 505, + 549 + ], + "spans": [ + { + "bbox": [ + 106, + 537, + 193, + 549 + ], + "score": 1.0, + "content": "Generating a random", + "type": "text" + }, + { + "bbox": [ + 193, + 539, + 200, + 547 + ], + "score": 0.76, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 537, + 505, + 549 + ], + "score": 1.0, + "content": "-bit number is straight-forward using random bits, which in our protocol can", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 548, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 548, + 505, + 559 + ], + "score": 1.0, + "content": "be generated as presented by Damgård et al. [2019]. In the context of our protocol however, Dalskov", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 559, + 493, + 570 + ], + "spans": [ + { + "bbox": [ + 106, + 559, + 493, + 570 + ], + "score": 1.0, + "content": "et al. [2021] have a presented a more efficient approach that involves mixed-circuit computation.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 34.5, + "bbox_fs": [ + 105, + 502, + 506, + 570 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 581, + 505, + 615 + ], + "lines": [ + { + "bbox": [ + 106, + 581, + 505, + 594 + ], + "spans": [ + { + "bbox": [ + 106, + 581, + 505, + 594 + ], + "score": 1.0, + "content": "Communication cost Table 1 show the total communication cost of some of the building blocks", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 104, + 590, + 507, + 606 + ], + "spans": [ + { + "bbox": [ + 104, + 590, + 184, + 606 + ], + "score": 1.0, + "content": "in our protocol for", + "type": "text" + }, + { + "bbox": [ + 184, + 593, + 215, + 604 + ], + "score": 0.9, + "content": "f = 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 590, + 363, + 606 + ], + "score": 1.0, + "content": ". This setting mandates the modulus", + "type": "text" + }, + { + "bbox": [ + 364, + 592, + 382, + 603 + ], + "score": 0.86, + "content": "2 ^ { 1 2 8 }", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 590, + 507, + 606 + ], + "score": 1.0, + "content": "because the division protocol", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 603, + 198, + 616 + ], + "spans": [ + { + "bbox": [ + 105, + 603, + 181, + 616 + ], + "score": 1.0, + "content": "requires bit length", + "type": "text" + }, + { + "bbox": [ + 182, + 604, + 194, + 615 + ], + "score": 0.86, + "content": "4 f", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 603, + 198, + 616 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39, + "bbox_fs": [ + 104, + 581, + 507, + 616 + ] + }, + { + "type": "title", + "bbox": [ + 101, + 630, + 304, + 644 + ], + "lines": [ + { + "bbox": [ + 105, + 628, + 306, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 306, + 647 + ], + "score": 1.0, + "content": "4 Machine Learning Building Blocks", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 107, + 654, + 504, + 677 + ], + "lines": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 505, + 667 + ], + "score": 1.0, + "content": "In this section, we will use the building blocks in Section 3 to construct high-level computational", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 664, + 216, + 679 + ], + "spans": [ + { + "bbox": [ + 105, + 664, + 216, + 679 + ], + "score": 1.0, + "content": "modules for deep learning.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42.5, + "bbox_fs": [ + 105, + 654, + 505, + 679 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 689, + 504, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "Fully connected layers Both forward and back-propagation of fully connected layers can be seen", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 700, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 106, + 700, + 505, + 713 + ], + "score": 1.0, + "content": "as matrix multiplications and thus can be implemented using dot products. A particular challenge in", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 711, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 506, + 723 + ], + "score": 1.0, + "content": "secure computation is to compute a number of outputs in parallel in order to save communication", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45, + "bbox_fs": [ + 105, + 688, + 506, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 201, + 88, + 411, + 222 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 125, + 70, + 484, + 83 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 123, + 67, + 482, + 85 + ], + "spans": [ + { + "bbox": [ + 123, + 67, + 348, + 85 + ], + "score": 1.0, + "content": "Table 1: Communication cost of select computation for", + "type": "text" + }, + { + "bbox": [ + 348, + 71, + 379, + 83 + ], + "score": 0.91, + "content": "f = 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 67, + 464, + 85 + ], + "score": 1.0, + "content": "and integer modulus", + "type": "text" + }, + { + "bbox": [ + 464, + 70, + 482, + 81 + ], + "score": 0.84, + "content": "2 ^ { 1 2 8 }", + "type": "inline_equation" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 201, + 88, + 411, + 222 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 201, + 88, + 411, + 222 + ], + "spans": [ + { + "bbox": [ + 201, + 88, + 411, + 222 + ], + "score": 0.981, + "html": "
Bits
Integer multiplication384
Probabilistic truncation Nearest truncation1,536 4,462
Comparison1,369
Division (prob. truncation)29,866
Division (nearest truncation)57,798
Exponentiation (prob. truncation)77,684
Exponentiation (nearest truncation)
171,638
Invert square root (prob. truncation) Invert square root (nearest truncation)20,073 27,699
", + "type": "table", + "image_path": "9fd53d77e94058a5729b0569126beb6fc6cee5af420d0afe5b03e2771b4647cb.jpg" + } + ] + } + ], + "index": 5.5, + "virtual_lines": [ + { + "bbox": [ + 201, + 88, + 411, + 101.4 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 201, + 101.4, + 411, + 114.80000000000001 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 201, + 114.80000000000001, + 411, + 128.20000000000002 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 201, + 128.20000000000002, + 411, + 141.60000000000002 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 201, + 141.60000000000002, + 411, + 155.00000000000003 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 201, + 155.00000000000003, + 411, + 168.40000000000003 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 201, + 168.40000000000003, + 411, + 181.80000000000004 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 201, + 181.80000000000004, + 411, + 195.20000000000005 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 201, + 195.20000000000005, + 411, + 208.60000000000005 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 201, + 208.60000000000005, + 411, + 222.00000000000006 + ], + "spans": [], + "index": 10 + } + ] + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 91, + 241, + 505, + 264 + ], + "lines": [ + { + "bbox": [ + 90, + 241, + 505, + 254 + ], + "spans": [ + { + "bbox": [ + 90, + 241, + 505, + 254 + ], + "score": 1.0, + "content": "10 rounds. We solve this by having a dedicated infrastructure in our implementation that computes all dot", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 90, + 252, + 506, + 265 + ], + "spans": [ + { + "bbox": [ + 90, + 252, + 506, + 265 + ], + "score": 1.0, + "content": "11 products for a matrix multiplication in a single batch, thus combining all necessary communication.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 104, + 275, + 506, + 309 + ], + "lines": [ + { + "bbox": [ + 105, + 274, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 274, + 505, + 289 + ], + "score": 1.0, + "content": "2D convolution layers Similar to fully connected layers, 2D convolution and its corresponding", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 287, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 106, + 287, + 505, + 299 + ], + "score": 1.0, + "content": "gradient can be implemented using only dot products, and we again compute several output values in", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 297, + 142, + 309 + ], + "spans": [ + { + "bbox": [ + 104, + 297, + 142, + 309 + ], + "score": 1.0, + "content": "parallel.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 86, + 319, + 451, + 333 + ], + "lines": [ + { + "bbox": [ + 84, + 319, + 452, + 334 + ], + "spans": [ + { + "bbox": [ + 84, + 319, + 452, + 334 + ], + "score": 1.0, + "content": "215 Rectified Linear Unit (ReLU) ReLU Nair and Hinton [2010] is defined as follows:", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "interline_equation", + "bbox": [ + 244, + 336, + 364, + 365 + ], + "lines": [ + { + "bbox": [ + 244, + 336, + 364, + 365 + ], + "spans": [ + { + "bbox": [ + 244, + 336, + 364, + 365 + ], + "score": 0.94, + "content": "{ \\mathsf { R e L U } } ( x ) : = { \\left\\{ \\begin{array} { l l } { x , } & { { \\mathrm { i f ~ } } x > 0 } \\\\ { 0 . } & { { \\mathrm { o t h e r w i s e } } } \\end{array} \\right. }", + "type": "interline_equation", + "image_path": "6552a81b7732498e72666ed8446d188466cdfbd1913c006a8ad902a1062afcf8.jpg" + } + ] + } + ], + "index": 17.5, + "virtual_lines": [ + { + "bbox": [ + 244, + 336, + 364, + 350.5 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 244, + 350.5, + 364, + 365.0 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "text", + "bbox": [ + 86, + 369, + 505, + 414 + ], + "lines": [ + { + "bbox": [ + 85, + 369, + 506, + 382 + ], + "spans": [ + { + "bbox": [ + 85, + 371, + 100, + 381 + ], + "score": 1.0, + "content": "216", + "type": "text" + }, + { + "bbox": [ + 105, + 369, + 506, + 382 + ], + "score": 1.0, + "content": "It can thus be implemented as a comparison followed by an oblivious selection. For back-propagation,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 85, + 380, + 505, + 393 + ], + "spans": [ + { + "bbox": [ + 85, + 382, + 100, + 392 + ], + "score": 1.0, + "content": "217", + "type": "text" + }, + { + "bbox": [ + 105, + 380, + 505, + 393 + ], + "score": 1.0, + "content": "it is advantageous to reuse the comparison results from forward propagation due to the relatively high", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 85, + 391, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 85, + 392, + 100, + 403 + ], + "score": 1.0, + "content": "218", + "type": "text" + }, + { + "bbox": [ + 105, + 391, + 505, + 403 + ], + "score": 1.0, + "content": "cost in secure computation. Note that the comparison results are stored in secret-shared form and", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 85, + 401, + 255, + 415 + ], + "spans": [ + { + "bbox": [ + 85, + 403, + 100, + 414 + ], + "score": 1.0, + "content": "219", + "type": "text" + }, + { + "bbox": [ + 105, + 401, + 255, + 415 + ], + "score": 1.0, + "content": "thus there is no reduction in security.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 86, + 424, + 506, + 470 + ], + "lines": [ + { + "bbox": [ + 86, + 424, + 507, + 439 + ], + "spans": [ + { + "bbox": [ + 86, + 427, + 100, + 437 + ], + "score": 1.0, + "content": "220", + "type": "text" + }, + { + "bbox": [ + 105, + 424, + 507, + 439 + ], + "score": 1.0, + "content": "Max pooling Similar to ReLU, max pooling can be reduced to comparison and oblivious selection.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 85, + 435, + 505, + 448 + ], + "spans": [ + { + "bbox": [ + 85, + 437, + 100, + 448 + ], + "score": 1.0, + "content": "221", + "type": "text" + }, + { + "bbox": [ + 105, + 435, + 505, + 448 + ], + "score": 1.0, + "content": "In secure computation, it saves communication rounds if the process uses a balanced tree rather than", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 85, + 447, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 85, + 448, + 100, + 459 + ], + "score": 1.0, + "content": "222", + "type": "text" + }, + { + "bbox": [ + 105, + 447, + 505, + 460 + ], + "score": 1.0, + "content": "iterating over all input values of one maximum computation. For back-propagation it again pays off", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 86, + 457, + 456, + 470 + ], + "spans": [ + { + "bbox": [ + 86, + 460, + 100, + 469 + ], + "score": 1.0, + "content": "223", + "type": "text" + }, + { + "bbox": [ + 105, + 457, + 456, + 470 + ], + "score": 1.0, + "content": "to the store intermediate results from forward propagation, again in secret-shared form.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24.5 + }, + { + "type": "text", + "bbox": [ + 91, + 480, + 505, + 503 + ], + "lines": [ + { + "bbox": [ + 88, + 479, + 506, + 494 + ], + "spans": [ + { + "bbox": [ + 88, + 479, + 506, + 494 + ], + "score": 1.0, + "content": "224 Softmax and cross entropy loss This combination requires computing the following gradient for", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 87, + 492, + 182, + 504 + ], + "spans": [ + { + "bbox": [ + 87, + 492, + 182, + 504 + ], + "score": 1.0, + "content": "225 back-propagation:", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5 + }, + { + "type": "interline_equation", + "bbox": [ + 166, + 506, + 444, + 538 + ], + "lines": [ + { + "bbox": [ + 166, + 506, + 444, + 538 + ], + "spans": [ + { + "bbox": [ + 166, + 506, + 444, + 538 + ], + "score": 0.93, + "content": "\\nabla i : = \\frac { \\partial \\ell } { \\partial x _ { i } } = \\frac { \\partial } { \\partial x _ { i } } \\Big ( - \\sum _ { k } y _ { k } \\cdot x _ { k } + \\log \\sum _ { j } e ^ { x _ { j } } \\Big ) = - y _ { i } + \\frac { e ^ { x _ { i } } } { \\sum _ { j } e ^ { x _ { j } } } ,", + "type": "interline_equation", + "image_path": "eae777e4bc355b0a5c52d65a7001eadc3f0665c189d84dfb531be701ff493e40.jpg" + } + ] + } + ], + "index": 30, + "virtual_lines": [ + { + "bbox": [ + 166, + 506, + 444, + 516.6666666666666 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 166, + 516.6666666666666, + 444, + 527.3333333333333 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 166, + 527.3333333333333, + 444, + 537.9999999999999 + ], + "spans": [], + "index": 31 + } + ] + }, + { + "type": "text", + "bbox": [ + 88, + 542, + 507, + 565 + ], + "lines": [ + { + "bbox": [ + 84, + 541, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 84, + 541, + 132, + 555 + ], + "score": 1.0, + "content": "226 where", + "type": "text" + }, + { + "bbox": [ + 132, + 544, + 142, + 554 + ], + "score": 0.85, + "content": "y _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 541, + 393, + 555 + ], + "score": 1.0, + "content": "denotes an element of the ground truth as a one-hot vector, and", + "type": "text" + }, + { + "bbox": [ + 393, + 544, + 404, + 554 + ], + "score": 0.86, + "content": "x _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 541, + 506, + 555 + ], + "score": 1.0, + "content": "denotes the output of the", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 84, + 551, + 149, + 567 + ], + "spans": [ + { + "bbox": [ + 84, + 551, + 149, + 567 + ], + "score": 1.0, + "content": "227 last layer.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32.5 + }, + { + "type": "text", + "bbox": [ + 86, + 569, + 505, + 614 + ], + "lines": [ + { + "bbox": [ + 86, + 569, + 505, + 583 + ], + "spans": [ + { + "bbox": [ + 86, + 572, + 99, + 582 + ], + "score": 1.0, + "content": "228", + "type": "text" + }, + { + "bbox": [ + 105, + 569, + 505, + 583 + ], + "score": 1.0, + "content": "On the right hand side of eq. (1), the values in the denominator are potentially large due to the use", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 86, + 582, + 505, + 592 + ], + "spans": [ + { + "bbox": [ + 86, + 582, + 99, + 592 + ], + "score": 1.0, + "content": "229", + "type": "text" + }, + { + "bbox": [ + 106, + 582, + 505, + 592 + ], + "score": 1.0, + "content": "of the exponential. This is prone to numerical overflow in our quantized representation because the", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 86, + 590, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 86, + 594, + 99, + 603 + ], + "score": 1.0, + "content": "230", + "type": "text" + }, + { + "bbox": [ + 105, + 590, + 505, + 606 + ], + "score": 1.0, + "content": "latter puts relatively strict limits on the values. We therefore optimize the computation by computing", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 86, + 602, + 244, + 615 + ], + "spans": [ + { + "bbox": [ + 86, + 605, + 99, + 614 + ], + "score": 1.0, + "content": "231", + "type": "text" + }, + { + "bbox": [ + 105, + 602, + 244, + 615 + ], + "score": 1.0, + "content": "the maximum of the input values:", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35.5 + }, + { + "type": "interline_equation", + "bbox": [ + 268, + 612, + 342, + 631 + ], + "lines": [ + { + "bbox": [ + 268, + 612, + 342, + 631 + ], + "spans": [ + { + "bbox": [ + 268, + 612, + 342, + 631 + ], + "score": 0.92, + "content": "m = \\operatorname* { m a x } _ { j } ( \\{ x _ { j } \\} ) .", + "type": "interline_equation", + "image_path": "6c2446d51ceae645baff7cf2c22ccb1d25b3d2e64c166c57a212c0dae2a3f332.jpg" + } + ] + } + ], + "index": 38, + "virtual_lines": [ + { + "bbox": [ + 268, + 612, + 342, + 631 + ], + "spans": [], + "index": 38 + } + ] + }, + { + "type": "text", + "bbox": [ + 87, + 633, + 180, + 645 + ], + "lines": [ + { + "bbox": [ + 84, + 632, + 181, + 646 + ], + "spans": [ + { + "bbox": [ + 84, + 632, + 181, + 646 + ], + "score": 1.0, + "content": "232 Then we compute", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "interline_equation", + "bbox": [ + 222, + 643, + 389, + 673 + ], + "lines": [ + { + "bbox": [ + 222, + 643, + 389, + 673 + ], + "spans": [ + { + "bbox": [ + 222, + 643, + 389, + 673 + ], + "score": 0.94, + "content": "{ \\frac { e ^ { x _ { i } - m } } { \\sum _ { j } e ^ { x _ { j } - m } } } = { \\frac { e ^ { x _ { i } } e ^ { - m } } { ( \\sum _ { j } e ^ { x _ { j } } ) e ^ { - m } } } = { \\frac { e ^ { x _ { i } } } { \\sum _ { j } e ^ { x _ { j } } } } .", + "type": "interline_equation", + "image_path": "e97d4b3c6d76abf1abd9d00e5c9538cb1f5ba3ae1df09857309d21f1c004e343.jpg" + } + ] + } + ], + "index": 40.5, + "virtual_lines": [ + { + "bbox": [ + 222, + 643, + 389, + 658.0 + ], + "spans": [], + "index": 40 + }, + { + "bbox": [ + 222, + 658.0, + 389, + 673.0 + ], + "spans": [], + "index": 41 + } + ] + }, + { + "type": "text", + "bbox": [ + 86, + 674, + 506, + 723 + ], + "lines": [ + { + "bbox": [ + 86, + 674, + 505, + 685 + ], + "spans": [ + { + "bbox": [ + 86, + 675, + 100, + 685 + ], + "score": 1.0, + "content": "233", + "type": "text" + }, + { + "bbox": [ + 105, + 674, + 505, + 685 + ], + "score": 1.0, + "content": "All the exponents on the left-most term are at most zero, and thus the dividend is at most one and the", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 86, + 685, + 505, + 696 + ], + "spans": [ + { + "bbox": [ + 86, + 686, + 100, + 696 + ], + "score": 1.0, + "content": "234", + "type": "text" + }, + { + "bbox": [ + 106, + 685, + 505, + 696 + ], + "score": 1.0, + "content": "divisor is at most the number of labels (which is 10 in MNIST). The same technique can be used to", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 86, + 694, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 86, + 700, + 100, + 710 + ], + "score": 1.0, + "content": "235", + "type": "text" + }, + { + "bbox": [ + 103, + 694, + 393, + 714 + ], + "score": 1.0, + "content": "compute the sigmoid activation function, as sigmoid(x) = 11+exp(−x)", + "type": "text" + }, + { + "bbox": [ + 311, + 695, + 456, + 713 + ], + "score": 0.56, + "content": "\\begin{array} { r } { \\mathrm { \\begin{array} { r } { \\mathrm { i d } ( x ) = \\frac { 1 } { 1 + \\exp ( - x ) } = \\frac { \\exp ( 0 ) } { \\exp ( 0 ) + \\exp ( x ) } } } \\end{array} } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 694, + 506, + 714 + ], + "score": 1.0, + "content": "is a special", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 86, + 711, + 173, + 722 + ], + "spans": [ + { + "bbox": [ + 86, + 713, + 100, + 721 + ], + "score": 1.0, + "content": "236", + "type": "text" + }, + { + "bbox": [ + 105, + 711, + 173, + 722 + ], + "score": 1.0, + "content": "case of softmax.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 43.5 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 310, + 752 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 201, + 88, + 411, + 222 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 125, + 70, + 484, + 83 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 123, + 67, + 482, + 85 + ], + "spans": [ + { + "bbox": [ + 123, + 67, + 348, + 85 + ], + "score": 1.0, + "content": "Table 1: Communication cost of select computation for", + "type": "text" + }, + { + "bbox": [ + 348, + 71, + 379, + 83 + ], + "score": 0.91, + "content": "f = 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 67, + 464, + 85 + ], + "score": 1.0, + "content": "and integer modulus", + "type": "text" + }, + { + "bbox": [ + 464, + 70, + 482, + 81 + ], + "score": 0.84, + "content": "2 ^ { 1 2 8 }", + "type": "inline_equation" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 201, + 88, + 411, + 222 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 201, + 88, + 411, + 222 + ], + "spans": [ + { + "bbox": [ + 201, + 88, + 411, + 222 + ], + "score": 0.981, + "html": "
Bits
Integer multiplication384
Probabilistic truncation Nearest truncation1,536 4,462
Comparison1,369
Division (prob. truncation)29,866
Division (nearest truncation)57,798
Exponentiation (prob. truncation)77,684
Exponentiation (nearest truncation)
171,638
Invert square root (prob. truncation) Invert square root (nearest truncation)20,073 27,699
", + "type": "table", + "image_path": "9fd53d77e94058a5729b0569126beb6fc6cee5af420d0afe5b03e2771b4647cb.jpg" + } + ] + } + ], + "index": 5.5, + "virtual_lines": [ + { + "bbox": [ + 201, + 88, + 411, + 101.4 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 201, + 101.4, + 411, + 114.80000000000001 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 201, + 114.80000000000001, + 411, + 128.20000000000002 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 201, + 128.20000000000002, + 411, + 141.60000000000002 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 201, + 141.60000000000002, + 411, + 155.00000000000003 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 201, + 155.00000000000003, + 411, + 168.40000000000003 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 201, + 168.40000000000003, + 411, + 181.80000000000004 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 201, + 181.80000000000004, + 411, + 195.20000000000005 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 201, + 195.20000000000005, + 411, + 208.60000000000005 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 201, + 208.60000000000005, + 411, + 222.00000000000006 + ], + "spans": [], + "index": 10 + } + ] + } + ], + "index": 2.75 + }, + { + "type": "index", + "bbox": [ + 91, + 241, + 505, + 264 + ], + "lines": [ + { + "bbox": [ + 90, + 241, + 505, + 254 + ], + "spans": [ + { + "bbox": [ + 90, + 241, + 505, + 254 + ], + "score": 1.0, + "content": "10 rounds. We solve this by having a dedicated infrastructure in our implementation that computes all dot", + "type": "text" + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 252, + 506, + 265 + ], + "spans": [ + { + "bbox": [ + 90, + 252, + 506, + 265 + ], + "score": 1.0, + "content": "11 products for a matrix multiplication in a single batch, thus combining all necessary communication.", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true + } + ], + "index": 11.5, + "bbox_fs": [ + 90, + 241, + 506, + 265 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 275, + 506, + 309 + ], + "lines": [ + { + "bbox": [ + 105, + 274, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 274, + 505, + 289 + ], + "score": 1.0, + "content": "2D convolution layers Similar to fully connected layers, 2D convolution and its corresponding", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 287, + 505, + 299 + ], + "spans": [ + { + "bbox": [ + 106, + 287, + 505, + 299 + ], + "score": 1.0, + "content": "gradient can be implemented using only dot products, and we again compute several output values in", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 297, + 142, + 309 + ], + "spans": [ + { + "bbox": [ + 104, + 297, + 142, + 309 + ], + "score": 1.0, + "content": "parallel.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14, + "bbox_fs": [ + 104, + 274, + 505, + 309 + ] + }, + { + "type": "text", + "bbox": [ + 86, + 319, + 451, + 333 + ], + "lines": [ + { + "bbox": [ + 84, + 319, + 452, + 334 + ], + "spans": [ + { + "bbox": [ + 84, + 319, + 452, + 334 + ], + "score": 1.0, + "content": "215 Rectified Linear Unit (ReLU) ReLU Nair and Hinton [2010] is defined as follows:", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16, + "bbox_fs": [ + 84, + 319, + 452, + 334 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 244, + 336, + 364, + 365 + ], + "lines": [ + { + "bbox": [ + 244, + 336, + 364, + 365 + ], + "spans": [ + { + "bbox": [ + 244, + 336, + 364, + 365 + ], + "score": 0.94, + "content": "{ \\mathsf { R e L U } } ( x ) : = { \\left\\{ \\begin{array} { l l } { x , } & { { \\mathrm { i f ~ } } x > 0 } \\\\ { 0 . } & { { \\mathrm { o t h e r w i s e } } } \\end{array} \\right. }", + "type": "interline_equation", + "image_path": "6552a81b7732498e72666ed8446d188466cdfbd1913c006a8ad902a1062afcf8.jpg" + } + ] + } + ], + "index": 17.5, + "virtual_lines": [ + { + "bbox": [ + 244, + 336, + 364, + 350.5 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 244, + 350.5, + 364, + 365.0 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "index", + "bbox": [ + 86, + 369, + 505, + 414 + ], + "lines": [ + { + "bbox": [ + 85, + 369, + 506, + 382 + ], + "spans": [ + { + "bbox": [ + 85, + 371, + 100, + 381 + ], + "score": 1.0, + "content": "216", + "type": "text" + }, + { + "bbox": [ + 105, + 369, + 506, + 382 + ], + "score": 1.0, + "content": "It can thus be implemented as a comparison followed by an oblivious selection. For back-propagation,", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 380, + 505, + 393 + ], + "spans": [ + { + "bbox": [ + 85, + 382, + 100, + 392 + ], + "score": 1.0, + "content": "217", + "type": "text" + }, + { + "bbox": [ + 105, + 380, + 505, + 393 + ], + "score": 1.0, + "content": "it is advantageous to reuse the comparison results from forward propagation due to the relatively high", + "type": "text" + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 391, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 85, + 392, + 100, + 403 + ], + "score": 1.0, + "content": "218", + "type": "text" + }, + { + "bbox": [ + 105, + 391, + 505, + 403 + ], + "score": 1.0, + "content": "cost in secure computation. Note that the comparison results are stored in secret-shared form and", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 401, + 255, + 415 + ], + "spans": [ + { + "bbox": [ + 85, + 403, + 100, + 414 + ], + "score": 1.0, + "content": "219", + "type": "text" + }, + { + "bbox": [ + 105, + 401, + 255, + 415 + ], + "score": 1.0, + "content": "thus there is no reduction in security.", + "type": "text" + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 424, + 507, + 439 + ], + "spans": [ + { + "bbox": [ + 86, + 427, + 100, + 437 + ], + "score": 1.0, + "content": "220", + "type": "text" + }, + { + "bbox": [ + 105, + 424, + 507, + 439 + ], + "score": 1.0, + "content": "Max pooling Similar to ReLU, max pooling can be reduced to comparison and oblivious selection.", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 435, + 505, + 448 + ], + "spans": [ + { + "bbox": [ + 85, + 437, + 100, + 448 + ], + "score": 1.0, + "content": "221", + "type": "text" + }, + { + "bbox": [ + 105, + 435, + 505, + 448 + ], + "score": 1.0, + "content": "In secure computation, it saves communication rounds if the process uses a balanced tree rather than", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 447, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 85, + 448, + 100, + 459 + ], + "score": 1.0, + "content": "222", + "type": "text" + }, + { + "bbox": [ + 105, + 447, + 505, + 460 + ], + "score": 1.0, + "content": "iterating over all input values of one maximum computation. For back-propagation it again pays off", + "type": "text" + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 457, + 456, + 470 + ], + "spans": [ + { + "bbox": [ + 86, + 460, + 100, + 469 + ], + "score": 1.0, + "content": "223", + "type": "text" + }, + { + "bbox": [ + 105, + 457, + 456, + 470 + ], + "score": 1.0, + "content": "to the store intermediate results from forward propagation, again in secret-shared form.", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 479, + 506, + 494 + ], + "spans": [ + { + "bbox": [ + 88, + 479, + 506, + 494 + ], + "score": 1.0, + "content": "224 Softmax and cross entropy loss This combination requires computing the following gradient for", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 87, + 492, + 182, + 504 + ], + "spans": [ + { + "bbox": [ + 87, + 492, + 182, + 504 + ], + "score": 1.0, + "content": "225 back-propagation:", + "type": "text" + } + ], + "index": 28, + "is_list_start_line": true + } + ], + "index": 20.5, + "bbox_fs": [ + 85, + 369, + 506, + 415 + ] + }, + { + "type": "index", + "bbox": [ + 86, + 424, + 506, + 470 + ], + "lines": [], + "index": 24.5, + "bbox_fs": [ + 85, + 424, + 507, + 470 + ], + "lines_deleted": true + }, + { + "type": "index", + "bbox": [ + 91, + 480, + 505, + 503 + ], + "lines": [], + "index": 27.5, + "bbox_fs": [ + 87, + 479, + 506, + 504 + ], + "lines_deleted": true + }, + { + "type": "interline_equation", + "bbox": [ + 166, + 506, + 444, + 538 + ], + "lines": [ + { + "bbox": [ + 166, + 506, + 444, + 538 + ], + "spans": [ + { + "bbox": [ + 166, + 506, + 444, + 538 + ], + "score": 0.93, + "content": "\\nabla i : = \\frac { \\partial \\ell } { \\partial x _ { i } } = \\frac { \\partial } { \\partial x _ { i } } \\Big ( - \\sum _ { k } y _ { k } \\cdot x _ { k } + \\log \\sum _ { j } e ^ { x _ { j } } \\Big ) = - y _ { i } + \\frac { e ^ { x _ { i } } } { \\sum _ { j } e ^ { x _ { j } } } ,", + "type": "interline_equation", + "image_path": "eae777e4bc355b0a5c52d65a7001eadc3f0665c189d84dfb531be701ff493e40.jpg" + } + ] + } + ], + "index": 30, + "virtual_lines": [ + { + "bbox": [ + 166, + 506, + 444, + 516.6666666666666 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 166, + 516.6666666666666, + 444, + 527.3333333333333 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 166, + 527.3333333333333, + 444, + 537.9999999999999 + ], + "spans": [], + "index": 31 + } + ] + }, + { + "type": "index", + "bbox": [ + 88, + 542, + 507, + 565 + ], + "lines": [ + { + "bbox": [ + 84, + 541, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 84, + 541, + 132, + 555 + ], + "score": 1.0, + "content": "226 where", + "type": "text" + }, + { + "bbox": [ + 132, + 544, + 142, + 554 + ], + "score": 0.85, + "content": "y _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 541, + 393, + 555 + ], + "score": 1.0, + "content": "denotes an element of the ground truth as a one-hot vector, and", + "type": "text" + }, + { + "bbox": [ + 393, + 544, + 404, + 554 + ], + "score": 0.86, + "content": "x _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 541, + 506, + 555 + ], + "score": 1.0, + "content": "denotes the output of the", + "type": "text" + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 551, + 149, + 567 + ], + "spans": [ + { + "bbox": [ + 84, + 551, + 149, + 567 + ], + "score": 1.0, + "content": "227 last layer.", + "type": "text" + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 569, + 505, + 583 + ], + "spans": [ + { + "bbox": [ + 86, + 572, + 99, + 582 + ], + "score": 1.0, + "content": "228", + "type": "text" + }, + { + "bbox": [ + 105, + 569, + 505, + 583 + ], + "score": 1.0, + "content": "On the right hand side of eq. (1), the values in the denominator are potentially large due to the use", + "type": "text" + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 582, + 505, + 592 + ], + "spans": [ + { + "bbox": [ + 86, + 582, + 99, + 592 + ], + "score": 1.0, + "content": "229", + "type": "text" + }, + { + "bbox": [ + 106, + 582, + 505, + 592 + ], + "score": 1.0, + "content": "of the exponential. This is prone to numerical overflow in our quantized representation because the", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 590, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 86, + 594, + 99, + 603 + ], + "score": 1.0, + "content": "230", + "type": "text" + }, + { + "bbox": [ + 105, + 590, + 505, + 606 + ], + "score": 1.0, + "content": "latter puts relatively strict limits on the values. We therefore optimize the computation by computing", + "type": "text" + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 602, + 244, + 615 + ], + "spans": [ + { + "bbox": [ + 86, + 605, + 99, + 614 + ], + "score": 1.0, + "content": "231", + "type": "text" + }, + { + "bbox": [ + 105, + 602, + 244, + 615 + ], + "score": 1.0, + "content": "the maximum of the input values:", + "type": "text" + } + ], + "index": 37, + "is_list_start_line": true + } + ], + "index": 32.5, + "bbox_fs": [ + 84, + 541, + 506, + 567 + ] + }, + { + "type": "index", + "bbox": [ + 86, + 569, + 505, + 614 + ], + "lines": [], + "index": 35.5, + "bbox_fs": [ + 86, + 569, + 505, + 615 + ], + "lines_deleted": true + }, + { + "type": "interline_equation", + "bbox": [ + 268, + 612, + 342, + 631 + ], + "lines": [ + { + "bbox": [ + 268, + 612, + 342, + 631 + ], + "spans": [ + { + "bbox": [ + 268, + 612, + 342, + 631 + ], + "score": 0.92, + "content": "m = \\operatorname* { m a x } _ { j } ( \\{ x _ { j } \\} ) .", + "type": "interline_equation", + "image_path": "6c2446d51ceae645baff7cf2c22ccb1d25b3d2e64c166c57a212c0dae2a3f332.jpg" + } + ] + } + ], + "index": 38, + "virtual_lines": [ + { + "bbox": [ + 268, + 612, + 342, + 631 + ], + "spans": [], + "index": 38 + } + ] + }, + { + "type": "text", + "bbox": [ + 87, + 633, + 180, + 645 + ], + "lines": [ + { + "bbox": [ + 84, + 632, + 181, + 646 + ], + "spans": [ + { + "bbox": [ + 84, + 632, + 181, + 646 + ], + "score": 1.0, + "content": "232 Then we compute", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39, + "bbox_fs": [ + 84, + 632, + 181, + 646 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 222, + 643, + 389, + 673 + ], + "lines": [ + { + "bbox": [ + 222, + 643, + 389, + 673 + ], + "spans": [ + { + "bbox": [ + 222, + 643, + 389, + 673 + ], + "score": 0.94, + "content": "{ \\frac { e ^ { x _ { i } - m } } { \\sum _ { j } e ^ { x _ { j } - m } } } = { \\frac { e ^ { x _ { i } } e ^ { - m } } { ( \\sum _ { j } e ^ { x _ { j } } ) e ^ { - m } } } = { \\frac { e ^ { x _ { i } } } { \\sum _ { j } e ^ { x _ { j } } } } .", + "type": "interline_equation", + "image_path": "e97d4b3c6d76abf1abd9d00e5c9538cb1f5ba3ae1df09857309d21f1c004e343.jpg" + } + ] + } + ], + "index": 40.5, + "virtual_lines": [ + { + "bbox": [ + 222, + 643, + 389, + 658.0 + ], + "spans": [], + "index": 40 + }, + { + "bbox": [ + 222, + 658.0, + 389, + 673.0 + ], + "spans": [], + "index": 41 + } + ] + }, + { + "type": "index", + "bbox": [ + 86, + 674, + 506, + 723 + ], + "lines": [ + { + "bbox": [ + 86, + 674, + 505, + 685 + ], + "spans": [ + { + "bbox": [ + 86, + 675, + 100, + 685 + ], + "score": 1.0, + "content": "233", + "type": "text" + }, + { + "bbox": [ + 105, + 674, + 505, + 685 + ], + "score": 1.0, + "content": "All the exponents on the left-most term are at most zero, and thus the dividend is at most one and the", + "type": "text" + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 685, + 505, + 696 + ], + "spans": [ + { + "bbox": [ + 86, + 686, + 100, + 696 + ], + "score": 1.0, + "content": "234", + "type": "text" + }, + { + "bbox": [ + 106, + 685, + 505, + 696 + ], + "score": 1.0, + "content": "divisor is at most the number of labels (which is 10 in MNIST). The same technique can be used to", + "type": "text" + } + ], + "index": 43, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 694, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 86, + 700, + 100, + 710 + ], + "score": 1.0, + "content": "235", + "type": "text" + }, + { + "bbox": [ + 103, + 694, + 393, + 714 + ], + "score": 1.0, + "content": "compute the sigmoid activation function, as sigmoid(x) = 11+exp(−x)", + "type": "text" + }, + { + "bbox": [ + 311, + 695, + 456, + 713 + ], + "score": 0.56, + "content": "\\begin{array} { r } { \\mathrm { \\begin{array} { r } { \\mathrm { i d } ( x ) = \\frac { 1 } { 1 + \\exp ( - x ) } = \\frac { \\exp ( 0 ) } { \\exp ( 0 ) + \\exp ( x ) } } } \\end{array} } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 694, + 506, + 714 + ], + "score": 1.0, + "content": "is a special", + "type": "text" + } + ], + "index": 44, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 711, + 173, + 722 + ], + "spans": [ + { + "bbox": [ + 86, + 713, + 100, + 721 + ], + "score": 1.0, + "content": "236", + "type": "text" + }, + { + "bbox": [ + 105, + 711, + 173, + 722 + ], + "score": 1.0, + "content": "case of softmax.", + "type": "text" + } + ], + "index": 45, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 72, + 507, + 86 + ], + "spans": [ + { + "bbox": [ + 86, + 75, + 99, + 84 + ], + "score": 1.0, + "content": "237", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 72, + 507, + 86 + ], + "score": 1.0, + "content": "Stochastic gradient descent The model parameter update in SGD only involves basic arithmetic:", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 79, + 509, + 104 + ], + "spans": [ + { + "bbox": [ + 86, + 88, + 99, + 97 + ], + "score": 1.0, + "content": "238", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 84, + 205, + 99 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\theta _ { j } \\theta _ { j } - \\frac { \\gamma } { B } \\sum _ { i = 1 } ^ { B } \\bigtriangledown _ { i j } } \\end{array}", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 205, + 79, + 233, + 104 + ], + "score": 1.0, + "content": "where", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 233, + 86, + 243, + 98 + ], + "score": 0.87, + "content": "\\theta _ { j }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 244, + 79, + 357, + 104 + ], + "score": 1.0, + "content": "is the parameter indexed by", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 358, + 86, + 378, + 97 + ], + "score": 0.3, + "content": "j , B", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 378, + 79, + 469, + 104 + ], + "score": 1.0, + "content": "is the mini-batch size,", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 469, + 86, + 495, + 97 + ], + "score": 0.91, + "content": "\\gamma > 0", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 495, + 79, + 509, + 104 + ], + "score": 1.0, + "content": "is", + "type": "text", + "cross_page": true + } + ], + "index": 1, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 96, + 505, + 110 + ], + "spans": [ + { + "bbox": [ + 85, + 98, + 100, + 109 + ], + "score": 1.0, + "content": "239", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 96, + 192, + 110 + ], + "score": 1.0, + "content": "the learning rate, and", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 193, + 99, + 209, + 109 + ], + "score": 0.89, + "content": "\\nabla i j", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 210, + 96, + 387, + 110 + ], + "score": 1.0, + "content": "is the gradient of the loss with respect to the", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 394, + 96, + 505, + 110 + ], + "score": 1.0, + "content": "’th sample in the mini-batch", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 106, + 505, + 121 + ], + "spans": [ + { + "bbox": [ + 85, + 109, + 100, + 119 + ], + "score": 1.0, + "content": "240", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 106, + 179, + 121 + ], + "score": 1.0, + "content": "and the parameter", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 180, + 108, + 189, + 119 + ], + "score": 0.87, + "content": "\\theta _ { j }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 190, + 106, + 505, + 121 + ], + "score": 1.0, + "content": ". In order to tackle the limited precision with quantization, we defer dividing by", + "type": "text", + "cross_page": true + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 118, + 505, + 130 + ], + "spans": [ + { + "bbox": [ + 86, + 120, + 99, + 129 + ], + "score": 1.0, + "content": "241", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 118, + 505, + 130 + ], + "score": 1.0, + "content": "the batch size to the model update. This means that we do not divide the gradient value by the batch", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 129, + 505, + 142 + ], + "spans": [ + { + "bbox": [ + 85, + 131, + 100, + 141 + ], + "score": 1.0, + "content": "242", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 129, + 505, + 142 + ], + "score": 1.0, + "content": "size when computing them as described in the previous paragraph. Instead, we divide the model", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 140, + 506, + 153 + ], + "spans": [ + { + "bbox": [ + 85, + 142, + 100, + 153 + ], + "score": 1.0, + "content": "243", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 140, + 506, + 153 + ], + "score": 1.0, + "content": "update by the batch size. Since we use a batch size that is a power of two (128), it is sufficient to use", + "type": "text", + "cross_page": true + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 151, + 505, + 163 + ], + "spans": [ + { + "bbox": [ + 85, + 153, + 100, + 163 + ], + "score": 1.0, + "content": "244", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 151, + 505, + 163 + ], + "score": 1.0, + "content": "probabilistic truncation instead of full-blown division. This saves both time and decreases the error.", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_start_line": true + } + ], + "index": 43.5, + "bbox_fs": [ + 86, + 674, + 506, + 722 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 86, + 72, + 505, + 163 + ], + "lines": [ + { + "bbox": [ + 86, + 72, + 507, + 86 + ], + "spans": [ + { + "bbox": [ + 86, + 75, + 99, + 84 + ], + "score": 1.0, + "content": "237", + "type": "text" + }, + { + "bbox": [ + 105, + 72, + 507, + 86 + ], + "score": 1.0, + "content": "Stochastic gradient descent The model parameter update in SGD only involves basic arithmetic:", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 86, + 79, + 509, + 104 + ], + "spans": [ + { + "bbox": [ + 86, + 88, + 99, + 97 + ], + "score": 1.0, + "content": "238", + "type": "text" + }, + { + "bbox": [ + 106, + 84, + 205, + 99 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\theta _ { j } \\theta _ { j } - \\frac { \\gamma } { B } \\sum _ { i = 1 } ^ { B } \\bigtriangledown _ { i j } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 79, + 233, + 104 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 233, + 86, + 243, + 98 + ], + "score": 0.87, + "content": "\\theta _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 79, + 357, + 104 + ], + "score": 1.0, + "content": "is the parameter indexed by", + "type": "text" + }, + { + "bbox": [ + 358, + 86, + 378, + 97 + ], + "score": 0.3, + "content": "j , B", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 79, + 469, + 104 + ], + "score": 1.0, + "content": "is the mini-batch size,", + "type": "text" + }, + { + "bbox": [ + 469, + 86, + 495, + 97 + ], + "score": 0.91, + "content": "\\gamma > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 495, + 79, + 509, + 104 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 85, + 96, + 505, + 110 + ], + "spans": [ + { + "bbox": [ + 85, + 98, + 100, + 109 + ], + "score": 1.0, + "content": "239", + "type": "text" + }, + { + "bbox": [ + 105, + 96, + 192, + 110 + ], + "score": 1.0, + "content": "the learning rate, and", + "type": "text" + }, + { + "bbox": [ + 193, + 99, + 209, + 109 + ], + "score": 0.89, + "content": "\\nabla i j", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 96, + 387, + 110 + ], + "score": 1.0, + "content": "is the gradient of the loss with respect to the", + "type": "text" + }, + { + "bbox": [ + 394, + 96, + 505, + 110 + ], + "score": 1.0, + "content": "’th sample in the mini-batch", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 85, + 106, + 505, + 121 + ], + "spans": [ + { + "bbox": [ + 85, + 109, + 100, + 119 + ], + "score": 1.0, + "content": "240", + "type": "text" + }, + { + "bbox": [ + 104, + 106, + 179, + 121 + ], + "score": 1.0, + "content": "and the parameter", + "type": "text" + }, + { + "bbox": [ + 180, + 108, + 189, + 119 + ], + "score": 0.87, + "content": "\\theta _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 106, + 505, + 121 + ], + "score": 1.0, + "content": ". In order to tackle the limited precision with quantization, we defer dividing by", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 86, + 118, + 505, + 130 + ], + "spans": [ + { + "bbox": [ + 86, + 120, + 99, + 129 + ], + "score": 1.0, + "content": "241", + "type": "text" + }, + { + "bbox": [ + 106, + 118, + 505, + 130 + ], + "score": 1.0, + "content": "the batch size to the model update. This means that we do not divide the gradient value by the batch", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 85, + 129, + 505, + 142 + ], + "spans": [ + { + "bbox": [ + 85, + 131, + 100, + 141 + ], + "score": 1.0, + "content": "242", + "type": "text" + }, + { + "bbox": [ + 106, + 129, + 505, + 142 + ], + "score": 1.0, + "content": "size when computing them as described in the previous paragraph. Instead, we divide the model", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 85, + 140, + 506, + 153 + ], + "spans": [ + { + "bbox": [ + 85, + 142, + 100, + 153 + ], + "score": 1.0, + "content": "243", + "type": "text" + }, + { + "bbox": [ + 105, + 140, + 506, + 153 + ], + "score": 1.0, + "content": "update by the batch size. Since we use a batch size that is a power of two (128), it is sufficient to use", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 85, + 151, + 505, + 163 + ], + "spans": [ + { + "bbox": [ + 85, + 153, + 100, + 163 + ], + "score": 1.0, + "content": "244", + "type": "text" + }, + { + "bbox": [ + 105, + 151, + 505, + 163 + ], + "score": 1.0, + "content": "probabilistic truncation instead of full-blown division. This saves both time and decreases the error.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 99, + 179, + 505, + 213 + ], + "lines": [ + { + "bbox": [ + 106, + 179, + 505, + 191 + ], + "spans": [ + { + "bbox": [ + 106, + 179, + 505, + 191 + ], + "score": 1.0, + "content": "Adam The main difference to SGD in terms of basic computational operations is the additional", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 191, + 505, + 203 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 505, + 203 + ], + "score": 1.0, + "content": "use of an inverse square root. We again defer the division by the batch size to just before the model", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 200, + 140, + 215 + ], + "spans": [ + { + "bbox": [ + 105, + 200, + 140, + 215 + ], + "score": 1.0, + "content": "update.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 99, + 230, + 505, + 253 + ], + "lines": [ + { + "bbox": [ + 105, + 229, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 105, + 229, + 505, + 243 + ], + "score": 1.0, + "content": "Parameter initialization We use the Glorot initialization by Glorot and Bengio [2010]. Besides", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 241, + 506, + 253 + ], + "spans": [ + { + "bbox": [ + 106, + 241, + 506, + 253 + ], + "score": 1.0, + "content": "basic operations, it mainly involves generating a uniformly random fractional value in a given interval.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + }, + { + "type": "title", + "bbox": [ + 93, + 273, + 207, + 286 + ], + "lines": [ + { + "bbox": [ + 89, + 271, + 209, + 290 + ], + "spans": [ + { + "bbox": [ + 89, + 271, + 209, + 290 + ], + "score": 1.0, + "content": "250 5 Implementation", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 92, + 300, + 505, + 367 + ], + "lines": [ + { + "bbox": [ + 90, + 301, + 506, + 314 + ], + "spans": [ + { + "bbox": [ + 90, + 304, + 100, + 313 + ], + "score": 1.0, + "content": "51", + "type": "text" + }, + { + "bbox": [ + 105, + 301, + 506, + 314 + ], + "score": 1.0, + "content": "We built our implementation on MP-SPDZ by Keller [2020]. MP-SPDZ not only implements a", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 90, + 312, + 506, + 325 + ], + "spans": [ + { + "bbox": [ + 90, + 315, + 99, + 323 + ], + "score": 1.0, + "content": "52", + "type": "text" + }, + { + "bbox": [ + 105, + 312, + 506, + 325 + ], + "score": 1.0, + "content": "range of MPC protocols, it also comes with a high-level library containing the building blocks in", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 89, + 322, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 89, + 326, + 100, + 335 + ], + "score": 1.0, + "content": "253", + "type": "text" + }, + { + "bbox": [ + 105, + 322, + 506, + 336 + ], + "score": 1.0, + "content": "Section 3. MP-SPDZ already featured capabilities to train dense neural networks as well as inference", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 89, + 332, + 506, + 348 + ], + "spans": [ + { + "bbox": [ + 89, + 336, + 100, + 345 + ], + "score": 1.0, + "content": "254", + "type": "text" + }, + { + "bbox": [ + 105, + 332, + 506, + 348 + ], + "score": 1.0, + "content": "for convolutional neural networks. We have added backward propagation for a number of layer types,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 89, + 344, + 506, + 358 + ], + "spans": [ + { + "bbox": [ + 89, + 348, + 100, + 357 + ], + "score": 1.0, + "content": "255", + "type": "text" + }, + { + "bbox": [ + 105, + 344, + 506, + 358 + ], + "score": 1.0, + "content": "including 2D convolution. Furthermore, we have corrected a bug in the backward propagation for", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 89, + 355, + 162, + 370 + ], + "spans": [ + { + "bbox": [ + 89, + 358, + 100, + 367 + ], + "score": 1.0, + "content": "256", + "type": "text" + }, + { + "bbox": [ + 104, + 355, + 162, + 370 + ], + "score": 1.0, + "content": "dense layers.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 16.5 + }, + { + "type": "text", + "bbox": [ + 106, + 372, + 506, + 405 + ], + "lines": [ + { + "bbox": [ + 105, + 372, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 505, + 385 + ], + "score": 1.0, + "content": "MP-SPDZ allows implementing the computation in Python code, which is then compiled a specific", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 384, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 506, + 396 + ], + "score": 1.0, + "content": "bytecode. This code can be execute by a virtual machine executing the actual secure computation.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 394, + 393, + 406 + ], + "spans": [ + { + "bbox": [ + 106, + 394, + 393, + 406 + ], + "score": 1.0, + "content": "The process allows to optimize the computation in the context of MPC.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 99, + 410, + 505, + 433 + ], + "lines": [ + { + "bbox": [ + 106, + 411, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 106, + 411, + 505, + 422 + ], + "score": 1.0, + "content": "The framework also features an emulator that executes the exact computation that could be done", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 421, + 507, + 434 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 507, + 434 + ], + "score": 1.0, + "content": "securely in the clear. This allowed us to collect the accuracy figures in the next section at a lower cost.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 90, + 437, + 401, + 449 + ], + "lines": [ + { + "bbox": [ + 86, + 437, + 401, + 450 + ], + "spans": [ + { + "bbox": [ + 86, + 437, + 401, + 450 + ], + "score": 1.0, + "content": "262 It is licensed under a BSD-style license, which allows to extend the code.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "title", + "bbox": [ + 95, + 470, + 164, + 483 + ], + "lines": [ + { + "bbox": [ + 90, + 469, + 166, + 486 + ], + "spans": [ + { + "bbox": [ + 90, + 469, + 166, + 486 + ], + "score": 1.0, + "content": "63 6 MNIST", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 102, + 498, + 505, + 575 + ], + "lines": [ + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "score": 1.0, + "content": "For a concrete measurement of accuracy and running times, we have implemented training for the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 509, + 506, + 521 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 506, + 521 + ], + "score": 1.0, + "content": "well-known MNIST dataset by LeCun et al. [2010]. We work mainly with the models that have been", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 519, + 506, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 506, + 532 + ], + "score": 1.0, + "content": "used by Wagh et al. [2019] with secure computation, and we will reuse their numbering (A–D). The", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 531, + 506, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 506, + 543 + ], + "score": 1.0, + "content": "models contain up to four linear layers. Network C is a convolutional neural network going back", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 541, + 506, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 541, + 506, + 554 + ], + "score": 1.0, + "content": "to the seminal work by LeCun et al. [1998] whereas the others are simpler networks that have been", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 552, + 506, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 506, + 564 + ], + "score": 1.0, + "content": "proposed by works on secure computation such as Mohassel and Zhang [2017], Liu et al. [2017], and", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 564, + 460, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 460, + 576 + ], + "score": 1.0, + "content": "Riazi et al. [2018]. We present the networks as Keras code in the supplemental material.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 105, + 579, + 505, + 646 + ], + "lines": [ + { + "bbox": [ + 105, + 579, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 579, + 506, + 592 + ], + "score": 1.0, + "content": "Figure 2 shows the results for various quantization precisions and and the two rounding options.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 590, + 507, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 507, + 603 + ], + "score": 1.0, + "content": "We have used SGD with learning rate 0.01, batch size 128, and the usual MNIST training/test split.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 601, + 503, + 614 + ], + "spans": [ + { + "bbox": [ + 106, + 602, + 140, + 613 + ], + "score": 0.91, + "content": "f = 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 601, + 415, + 614 + ], + "score": 1.0, + "content": "is the best option with probabilistic rounding, improving on both", + "type": "text" + }, + { + "bbox": [ + 416, + 602, + 450, + 613 + ], + "score": 0.91, + "content": "f = 1 6", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 601, + 469, + 614 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 470, + 602, + 503, + 613 + ], + "score": 0.91, + "content": "f = 3 2", + "type": "inline_equation" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 612, + 506, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 378, + 625 + ], + "score": 1.0, + "content": "Furthermore, nearest rounding performs worse that probabilistic for", + "type": "text" + }, + { + "bbox": [ + 378, + 613, + 408, + 624 + ], + "score": 0.92, + "content": "f = 1 6", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 612, + 426, + 625 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 426, + 613, + 457, + 624 + ], + "score": 0.9, + "content": "f = 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 612, + 506, + 625 + ], + "score": 1.0, + "content": ". Due to the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 623, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 105, + 623, + 196, + 636 + ], + "score": 1.0, + "content": "high cost, we only ran", + "type": "text" + }, + { + "bbox": [ + 196, + 623, + 227, + 635 + ], + "score": 0.91, + "content": "f = 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 623, + 505, + 636 + ], + "score": 1.0, + "content": "with probabilistic rounding several times. The range is indicate by the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 633, + 404, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 210, + 647 + ], + "score": 1.0, + "content": "shaded area. We focus on", + "type": "text" + }, + { + "bbox": [ + 211, + 635, + 241, + 646 + ], + "score": 0.91, + "content": "f = 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 633, + 404, + 647 + ], + "score": 1.0, + "content": "because it offers the faster convergence.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 103, + 650, + 505, + 684 + ], + "lines": [ + { + "bbox": [ + 105, + 649, + 505, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 505, + 663 + ], + "score": 1.0, + "content": "Figure 3 then shows the result with a variety of optimizers. While increasing the learning rate for", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 661, + 505, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 505, + 675 + ], + "score": 1.0, + "content": "SGD leads to a lower stability, Adam exposes a smoother learning learning curve albeit not a faster", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 674, + 141, + 685 + ], + "spans": [ + { + "bbox": [ + 105, + 674, + 141, + 685 + ], + "score": 1.0, + "content": "process.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 89, + 689, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 88, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 88, + 691, + 99, + 700 + ], + "score": 1.0, + "content": "280", + "type": "text" + }, + { + "bbox": [ + 104, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "Finally, Figures 4 shows our results for all networks used by Wagh et al. [2019]. As one would expect,", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 88, + 700, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 88, + 703, + 99, + 711 + ], + "score": 1.0, + "content": "281", + "type": "text" + }, + { + "bbox": [ + 105, + 700, + 506, + 712 + ], + "score": 1.0, + "content": "the most sophisticated network performs best. Somewhat surprisingly, however, Network A (without", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 88, + 711, + 492, + 723 + ], + "spans": [ + { + "bbox": [ + 88, + 714, + 99, + 722 + ], + "score": 1.0, + "content": "282", + "type": "text" + }, + { + "bbox": [ + 105, + 711, + 492, + 723 + ], + "score": 1.0, + "content": "convolutional layers) performs better than the simpler networks containing convolutional layers.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "index", + "bbox": [ + 86, + 72, + 505, + 163 + ], + "lines": [], + "index": 3.5, + "bbox_fs": [ + 85, + 72, + 509, + 163 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 99, + 179, + 505, + 213 + ], + "lines": [ + { + "bbox": [ + 106, + 179, + 505, + 191 + ], + "spans": [ + { + "bbox": [ + 106, + 179, + 505, + 191 + ], + "score": 1.0, + "content": "Adam The main difference to SGD in terms of basic computational operations is the additional", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 191, + 505, + 203 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 505, + 203 + ], + "score": 1.0, + "content": "use of an inverse square root. We again defer the division by the batch size to just before the model", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 200, + 140, + 215 + ], + "spans": [ + { + "bbox": [ + 105, + 200, + 140, + 215 + ], + "score": 1.0, + "content": "update.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 179, + 505, + 215 + ] + }, + { + "type": "text", + "bbox": [ + 99, + 230, + 505, + 253 + ], + "lines": [ + { + "bbox": [ + 105, + 229, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 105, + 229, + 505, + 243 + ], + "score": 1.0, + "content": "Parameter initialization We use the Glorot initialization by Glorot and Bengio [2010]. Besides", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 241, + 506, + 253 + ], + "spans": [ + { + "bbox": [ + 106, + 241, + 506, + 253 + ], + "score": 1.0, + "content": "basic operations, it mainly involves generating a uniformly random fractional value in a given interval.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 229, + 506, + 253 + ] + }, + { + "type": "title", + "bbox": [ + 93, + 273, + 207, + 286 + ], + "lines": [ + { + "bbox": [ + 89, + 271, + 209, + 290 + ], + "spans": [ + { + "bbox": [ + 89, + 271, + 209, + 290 + ], + "score": 1.0, + "content": "250 5 Implementation", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "index", + "bbox": [ + 92, + 300, + 505, + 367 + ], + "lines": [ + { + "bbox": [ + 90, + 301, + 506, + 314 + ], + "spans": [ + { + "bbox": [ + 90, + 304, + 100, + 313 + ], + "score": 1.0, + "content": "51", + "type": "text" + }, + { + "bbox": [ + 105, + 301, + 506, + 314 + ], + "score": 1.0, + "content": "We built our implementation on MP-SPDZ by Keller [2020]. MP-SPDZ not only implements a", + "type": "text" + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 90, + 312, + 506, + 325 + ], + "spans": [ + { + "bbox": [ + 90, + 315, + 99, + 323 + ], + "score": 1.0, + "content": "52", + "type": "text" + }, + { + "bbox": [ + 105, + 312, + 506, + 325 + ], + "score": 1.0, + "content": "range of MPC protocols, it also comes with a high-level library containing the building blocks in", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 322, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 89, + 326, + 100, + 335 + ], + "score": 1.0, + "content": "253", + "type": "text" + }, + { + "bbox": [ + 105, + 322, + 506, + 336 + ], + "score": 1.0, + "content": "Section 3. MP-SPDZ already featured capabilities to train dense neural networks as well as inference", + "type": "text" + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 332, + 506, + 348 + ], + "spans": [ + { + "bbox": [ + 89, + 336, + 100, + 345 + ], + "score": 1.0, + "content": "254", + "type": "text" + }, + { + "bbox": [ + 105, + 332, + 506, + 348 + ], + "score": 1.0, + "content": "for convolutional neural networks. We have added backward propagation for a number of layer types,", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 344, + 506, + 358 + ], + "spans": [ + { + "bbox": [ + 89, + 348, + 100, + 357 + ], + "score": 1.0, + "content": "255", + "type": "text" + }, + { + "bbox": [ + 105, + 344, + 506, + 358 + ], + "score": 1.0, + "content": "including 2D convolution. Furthermore, we have corrected a bug in the backward propagation for", + "type": "text" + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 355, + 162, + 370 + ], + "spans": [ + { + "bbox": [ + 89, + 358, + 100, + 367 + ], + "score": 1.0, + "content": "256", + "type": "text" + }, + { + "bbox": [ + 104, + 355, + 162, + 370 + ], + "score": 1.0, + "content": "dense layers.", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true + } + ], + "index": 16.5, + "bbox_fs": [ + 89, + 301, + 506, + 370 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 372, + 506, + 405 + ], + "lines": [ + { + "bbox": [ + 105, + 372, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 505, + 385 + ], + "score": 1.0, + "content": "MP-SPDZ allows implementing the computation in Python code, which is then compiled a specific", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 384, + 506, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 506, + 396 + ], + "score": 1.0, + "content": "bytecode. This code can be execute by a virtual machine executing the actual secure computation.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 394, + 393, + 406 + ], + "spans": [ + { + "bbox": [ + 106, + 394, + 393, + 406 + ], + "score": 1.0, + "content": "The process allows to optimize the computation in the context of MPC.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21, + "bbox_fs": [ + 105, + 372, + 506, + 406 + ] + }, + { + "type": "text", + "bbox": [ + 99, + 410, + 505, + 433 + ], + "lines": [ + { + "bbox": [ + 106, + 411, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 106, + 411, + 505, + 422 + ], + "score": 1.0, + "content": "The framework also features an emulator that executes the exact computation that could be done", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 421, + 507, + 434 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 507, + 434 + ], + "score": 1.0, + "content": "securely in the clear. This allowed us to collect the accuracy figures in the next section at a lower cost.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 105, + 411, + 507, + 434 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 437, + 401, + 449 + ], + "lines": [ + { + "bbox": [ + 86, + 437, + 401, + 450 + ], + "spans": [ + { + "bbox": [ + 86, + 437, + 401, + 450 + ], + "score": 1.0, + "content": "262 It is licensed under a BSD-style license, which allows to extend the code.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25, + "bbox_fs": [ + 86, + 437, + 401, + 450 + ] + }, + { + "type": "title", + "bbox": [ + 95, + 470, + 164, + 483 + ], + "lines": [ + { + "bbox": [ + 90, + 469, + 166, + 486 + ], + "spans": [ + { + "bbox": [ + 90, + 469, + 166, + 486 + ], + "score": 1.0, + "content": "63 6 MNIST", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 102, + 498, + 505, + 575 + ], + "lines": [ + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "score": 1.0, + "content": "For a concrete measurement of accuracy and running times, we have implemented training for the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 509, + 506, + 521 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 506, + 521 + ], + "score": 1.0, + "content": "well-known MNIST dataset by LeCun et al. [2010]. We work mainly with the models that have been", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 519, + 506, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 506, + 532 + ], + "score": 1.0, + "content": "used by Wagh et al. [2019] with secure computation, and we will reuse their numbering (A–D). The", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 531, + 506, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 506, + 543 + ], + "score": 1.0, + "content": "models contain up to four linear layers. Network C is a convolutional neural network going back", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 541, + 506, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 541, + 506, + 554 + ], + "score": 1.0, + "content": "to the seminal work by LeCun et al. [1998] whereas the others are simpler networks that have been", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 552, + 506, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 506, + 564 + ], + "score": 1.0, + "content": "proposed by works on secure computation such as Mohassel and Zhang [2017], Liu et al. [2017], and", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 564, + 460, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 460, + 576 + ], + "score": 1.0, + "content": "Riazi et al. [2018]. We present the networks as Keras code in the supplemental material.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 498, + 506, + 576 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 579, + 505, + 646 + ], + "lines": [ + { + "bbox": [ + 105, + 579, + 506, + 592 + ], + "spans": [ + { + "bbox": [ + 105, + 579, + 506, + 592 + ], + "score": 1.0, + "content": "Figure 2 shows the results for various quantization precisions and and the two rounding options.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 590, + 507, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 507, + 603 + ], + "score": 1.0, + "content": "We have used SGD with learning rate 0.01, batch size 128, and the usual MNIST training/test split.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 601, + 503, + 614 + ], + "spans": [ + { + "bbox": [ + 106, + 602, + 140, + 613 + ], + "score": 0.91, + "content": "f = 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 601, + 415, + 614 + ], + "score": 1.0, + "content": "is the best option with probabilistic rounding, improving on both", + "type": "text" + }, + { + "bbox": [ + 416, + 602, + 450, + 613 + ], + "score": 0.91, + "content": "f = 1 6", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 601, + 469, + 614 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 470, + 602, + 503, + 613 + ], + "score": 0.91, + "content": "f = 3 2", + "type": "inline_equation" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 612, + 506, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 612, + 378, + 625 + ], + "score": 1.0, + "content": "Furthermore, nearest rounding performs worse that probabilistic for", + "type": "text" + }, + { + "bbox": [ + 378, + 613, + 408, + 624 + ], + "score": 0.92, + "content": "f = 1 6", + "type": "inline_equation" + }, + { + "bbox": [ + 409, + 612, + 426, + 625 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 426, + 613, + 457, + 624 + ], + "score": 0.9, + "content": "f = 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 612, + 506, + 625 + ], + "score": 1.0, + "content": ". Due to the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 623, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 105, + 623, + 196, + 636 + ], + "score": 1.0, + "content": "high cost, we only ran", + "type": "text" + }, + { + "bbox": [ + 196, + 623, + 227, + 635 + ], + "score": 0.91, + "content": "f = 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 623, + 505, + 636 + ], + "score": 1.0, + "content": "with probabilistic rounding several times. The range is indicate by the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 633, + 404, + 647 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 210, + 647 + ], + "score": 1.0, + "content": "shaded area. We focus on", + "type": "text" + }, + { + "bbox": [ + 211, + 635, + 241, + 646 + ], + "score": 0.91, + "content": "f = 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 633, + 404, + 647 + ], + "score": 1.0, + "content": "because it offers the faster convergence.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 36.5, + "bbox_fs": [ + 105, + 579, + 507, + 647 + ] + }, + { + "type": "text", + "bbox": [ + 103, + 650, + 505, + 684 + ], + "lines": [ + { + "bbox": [ + 105, + 649, + 505, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 505, + 663 + ], + "score": 1.0, + "content": "Figure 3 then shows the result with a variety of optimizers. While increasing the learning rate for", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 661, + 505, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 505, + 675 + ], + "score": 1.0, + "content": "SGD leads to a lower stability, Adam exposes a smoother learning learning curve albeit not a faster", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 674, + 141, + 685 + ], + "spans": [ + { + "bbox": [ + 105, + 674, + 141, + 685 + ], + "score": 1.0, + "content": "process.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 41, + "bbox_fs": [ + 105, + 649, + 505, + 685 + ] + }, + { + "type": "index", + "bbox": [ + 89, + 689, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 88, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 88, + 691, + 99, + 700 + ], + "score": 1.0, + "content": "280", + "type": "text" + }, + { + "bbox": [ + 104, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "Finally, Figures 4 shows our results for all networks used by Wagh et al. [2019]. As one would expect,", + "type": "text" + } + ], + "index": 43, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 700, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 88, + 703, + 99, + 711 + ], + "score": 1.0, + "content": "281", + "type": "text" + }, + { + "bbox": [ + 105, + 700, + 506, + 712 + ], + "score": 1.0, + "content": "the most sophisticated network performs best. Somewhat surprisingly, however, Network A (without", + "type": "text" + } + ], + "index": 44, + "is_list_start_line": true + }, + { + "bbox": [ + 88, + 711, + 492, + 723 + ], + "spans": [ + { + "bbox": [ + 88, + 714, + 99, + 722 + ], + "score": 1.0, + "content": "282", + "type": "text" + }, + { + "bbox": [ + 105, + 711, + 492, + 723 + ], + "score": 1.0, + "content": "convolutional layers) performs better than the simpler networks containing convolutional layers.", + "type": "text" + } + ], + "index": 45, + "is_list_start_line": true + } + ], + "index": 44, + "bbox_fs": [ + 88, + 688, + 506, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 124, + 69, + 488, + 261 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 124, + 69, + 488, + 261 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 124, + 69, + 488, + 261 + ], + "spans": [ + { + "bbox": [ + 124, + 69, + 488, + 261 + ], + "score": 0.956, + "type": "image", + "image_path": "3a38cf90fbe641b15f4921c674ee686f614e2a42472767f16d74facf9f248ace.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 124, + 69, + 488, + 133.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 124, + 133.0, + 488, + 197.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 124, + 197.0, + 488, + 261.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 272, + 505, + 285 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 271, + 507, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 507, + 286 + ], + "score": 1.0, + "content": "Figure 2: Loss and accuracy for network C and precision options when running SGD with rate 0.01.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "image", + "bbox": [ + 121, + 297, + 473, + 489 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 121, + 297, + 473, + 489 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 121, + 297, + 473, + 489 + ], + "spans": [ + { + "bbox": [ + 121, + 297, + 473, + 489 + ], + "score": 0.695, + "type": "image", + "image_path": "ca865aaa95abbc18690e028b3ca52b226046ffa5e278339ad08eb557c5491c91.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 121, + 297, + 473, + 361.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 121, + 361.0, + 473, + 425.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 121, + 425.0, + 473, + 489.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 501, + 504, + 523 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 500, + 506, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 402, + 514 + ], + "score": 1.0, + "content": "Figure 3: Loss and accuracy for network C with various optimizer options,", + "type": "text" + }, + { + "bbox": [ + 403, + 501, + 433, + 513 + ], + "score": 0.91, + "content": "f = 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 500, + 506, + 514 + ], + "score": 1.0, + "content": ", and probabilistic", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 511, + 151, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 151, + 523 + ], + "score": 1.0, + "content": "truncation.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + } + ], + "index": 6.25 + }, + { + "type": "text", + "bbox": [ + 102, + 546, + 505, + 579 + ], + "lines": [ + { + "bbox": [ + 106, + 545, + 506, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 506, + 559 + ], + "score": 1.0, + "content": "Resources We ran the emulator on AWS c5.9xlarge instances. One epoch takes a few second to", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 557, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 557, + 505, + 569 + ], + "score": 1.0, + "content": "several minutes depending on the model. Overall, we estimate that we have used a few weeks worth", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 568, + 460, + 580 + ], + "spans": [ + { + "bbox": [ + 106, + 568, + 460, + 580 + ], + "score": 1.0, + "content": "of computing time including experiments not included here because of bugs in the code.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "title", + "bbox": [ + 107, + 596, + 215, + 608 + ], + "lines": [ + { + "bbox": [ + 106, + 596, + 216, + 610 + ], + "spans": [ + { + "bbox": [ + 106, + 596, + 216, + 610 + ], + "score": 1.0, + "content": "6.1 Secure computation", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 618, + 505, + 695 + ], + "lines": [ + { + "bbox": [ + 106, + 618, + 505, + 631 + ], + "spans": [ + { + "bbox": [ + 106, + 618, + 449, + 631 + ], + "score": 1.0, + "content": "In order to verify our emulation results, we have run Network C with precision", + "type": "text" + }, + { + "bbox": [ + 449, + 619, + 486, + 630 + ], + "score": 0.91, + "content": "f \\ = \\ 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 618, + 505, + 631 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 630, + 506, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 630, + 506, + 642 + ], + "score": 1.0, + "content": "probabilistic rounding in our actual multi-party computation protocol. We could verify that it", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 640, + 505, + 653 + ], + "spans": [ + { + "bbox": [ + 106, + 640, + 159, + 653 + ], + "score": 1.0, + "content": "converges on", + "type": "text" + }, + { + "bbox": [ + 160, + 640, + 187, + 651 + ], + "score": 0.87, + "content": "9 8 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 640, + 505, + 653 + ], + "score": 1.0, + "content": "accuracy at 150 epochs, taking 20 hours. Table 2 compares our result to previous", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 650, + 506, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 506, + 664 + ], + "score": 1.0, + "content": "works in a LAN setting. Note that Wagh et al. [2019] and Wagh et al. [2021] give accuracy figures.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 661, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 506, + 675 + ], + "score": 1.0, + "content": "From personal communication with the authors and the fact that the source repository for the latter", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 672, + 506, + 686 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 506, + 686 + ], + "score": 1.0, + "content": "work2 says that their “code has not run end-to-end training”, we derive our assessment that their", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 684, + 288, + 696 + ], + "spans": [ + { + "bbox": [ + 106, + 684, + 288, + 696 + ], + "score": 1.0, + "content": "figures do not reflect the secure computation.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 16 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 118, + 712, + 308, + 722 + ], + "lines": [ + { + "bbox": [ + 118, + 709, + 309, + 725 + ], + "spans": [ + { + "bbox": [ + 118, + 709, + 309, + 725 + ], + "score": 1.0, + "content": "2https://github.com/snwagh/falcon-public", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 742, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 9 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 124, + 69, + 488, + 261 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 124, + 69, + 488, + 261 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 124, + 69, + 488, + 261 + ], + "spans": [ + { + "bbox": [ + 124, + 69, + 488, + 261 + ], + "score": 0.956, + "type": "image", + "image_path": "3a38cf90fbe641b15f4921c674ee686f614e2a42472767f16d74facf9f248ace.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 124, + 69, + 488, + 133.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 124, + 133.0, + 488, + 197.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 124, + 197.0, + 488, + 261.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 272, + 505, + 285 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 271, + 507, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 271, + 507, + 286 + ], + "score": 1.0, + "content": "Figure 2: Loss and accuracy for network C and precision options when running SGD with rate 0.01.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "image", + "bbox": [ + 121, + 297, + 473, + 489 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 121, + 297, + 473, + 489 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 121, + 297, + 473, + 489 + ], + "spans": [ + { + "bbox": [ + 121, + 297, + 473, + 489 + ], + "score": 0.695, + "type": "image", + "image_path": "ca865aaa95abbc18690e028b3ca52b226046ffa5e278339ad08eb557c5491c91.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 121, + 297, + 473, + 361.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 121, + 361.0, + 473, + 425.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 121, + 425.0, + 473, + 489.0 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 501, + 504, + 523 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 500, + 506, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 402, + 514 + ], + "score": 1.0, + "content": "Figure 3: Loss and accuracy for network C with various optimizer options,", + "type": "text" + }, + { + "bbox": [ + 403, + 501, + 433, + 513 + ], + "score": 0.91, + "content": "f = 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 500, + 506, + 514 + ], + "score": 1.0, + "content": ", and probabilistic", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 511, + 151, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 151, + 523 + ], + "score": 1.0, + "content": "truncation.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + } + ], + "index": 6.25 + }, + { + "type": "text", + "bbox": [ + 102, + 546, + 505, + 579 + ], + "lines": [ + { + "bbox": [ + 106, + 545, + 506, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 545, + 506, + 559 + ], + "score": 1.0, + "content": "Resources We ran the emulator on AWS c5.9xlarge instances. One epoch takes a few second to", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 557, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 557, + 505, + 569 + ], + "score": 1.0, + "content": "several minutes depending on the model. Overall, we estimate that we have used a few weeks worth", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 568, + 460, + 580 + ], + "spans": [ + { + "bbox": [ + 106, + 568, + 460, + 580 + ], + "score": 1.0, + "content": "of computing time including experiments not included here because of bugs in the code.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 106, + 545, + 506, + 580 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 596, + 215, + 608 + ], + "lines": [ + { + "bbox": [ + 106, + 596, + 216, + 610 + ], + "spans": [ + { + "bbox": [ + 106, + 596, + 216, + 610 + ], + "score": 1.0, + "content": "6.1 Secure computation", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 618, + 505, + 695 + ], + "lines": [ + { + "bbox": [ + 106, + 618, + 505, + 631 + ], + "spans": [ + { + "bbox": [ + 106, + 618, + 449, + 631 + ], + "score": 1.0, + "content": "In order to verify our emulation results, we have run Network C with precision", + "type": "text" + }, + { + "bbox": [ + 449, + 619, + 486, + 630 + ], + "score": 0.91, + "content": "f \\ = \\ 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 618, + 505, + 631 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 630, + 506, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 630, + 506, + 642 + ], + "score": 1.0, + "content": "probabilistic rounding in our actual multi-party computation protocol. We could verify that it", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 640, + 505, + 653 + ], + "spans": [ + { + "bbox": [ + 106, + 640, + 159, + 653 + ], + "score": 1.0, + "content": "converges on", + "type": "text" + }, + { + "bbox": [ + 160, + 640, + 187, + 651 + ], + "score": 0.87, + "content": "9 8 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 640, + 505, + 653 + ], + "score": 1.0, + "content": "accuracy at 150 epochs, taking 20 hours. Table 2 compares our result to previous", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 650, + 506, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 506, + 664 + ], + "score": 1.0, + "content": "works in a LAN setting. Note that Wagh et al. [2019] and Wagh et al. [2021] give accuracy figures.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 661, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 506, + 675 + ], + "score": 1.0, + "content": "From personal communication with the authors and the fact that the source repository for the latter", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 672, + 506, + 686 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 506, + 686 + ], + "score": 1.0, + "content": "work2 says that their “code has not run end-to-end training”, we derive our assessment that their", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 684, + 288, + 696 + ], + "spans": [ + { + "bbox": [ + 106, + 684, + 288, + 696 + ], + "score": 1.0, + "content": "figures do not reflect the secure computation.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 618, + 506, + 696 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 133, + 69, + 480, + 261 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 133, + 69, + 480, + 261 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 133, + 69, + 480, + 261 + ], + "spans": [ + { + "bbox": [ + 133, + 69, + 480, + 261 + ], + "score": 0.965, + "type": "image", + "image_path": "700a37281e56f21023ee263453c7785f933f7176bea7c10046675acc845cf598.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 133, + 69, + 480, + 133.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 133, + 133.0, + 480, + 197.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 133, + 197.0, + 480, + 261.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 130, + 272, + 480, + 285 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 129, + 272, + 481, + 286 + ], + "spans": [ + { + "bbox": [ + 129, + 272, + 332, + 286 + ], + "score": 1.0, + "content": "Figure 4: Loss and accuracy for various networks,", + "type": "text" + }, + { + "bbox": [ + 333, + 273, + 363, + 285 + ], + "score": 0.91, + "content": "f = 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 272, + 481, + 286 + ], + "score": 1.0, + "content": ", and probabilistic truncation.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "table", + "bbox": [ + 112, + 356, + 499, + 604 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 305, + 505, + 350 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 305, + 506, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 305, + 354, + 318 + ], + "score": 1.0, + "content": "Table 2: Comparison to previous work in the LAN setting.", + "type": "text" + }, + { + "bbox": [ + 354, + 306, + 367, + 317 + ], + "score": 0.7, + "content": "( ^ { * } )", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 305, + 506, + 318 + ], + "score": 1.0, + "content": "Mohassel and Zhang [2017] and", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 317, + 506, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 506, + 329 + ], + "score": 1.0, + "content": "Agrawal et al. [2019] use a different security model and are thus incomparable. We include them for", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 327, + 506, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 506, + 341 + ], + "score": 1.0, + "content": "completeness. Two numbers refer to online and offline time. Accuracy N/A means that the accuracy", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 339, + 464, + 351 + ], + "spans": [ + { + "bbox": [ + 106, + 339, + 464, + 351 + ], + "score": 1.0, + "content": "figures were not given or computed in a way that does not reflect the secure computation.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5.5 + }, + { + "type": "table_body", + "bbox": [ + 112, + 356, + 499, + 604 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 356, + 499, + 604 + ], + "spans": [ + { + "bbox": [ + 112, + 356, + 499, + 604 + ], + "score": 0.986, + "html": "
NetworkEpoch time (s)Acc. (# epochs)Precision(f)
AMohassel and Zhang [2017]283/19333*93.4% (15)13
Mohassel and Rindal [2018]18094.0% (15)N/A
Agrawal et al. [2019]31392*95.0% (10)N/A
Wagh et al. [2019]247N/A13
Wagh et al. [2021]41N/A13
Ours3197.9% (15)16
Ours5097.7% (15)32
BWagh et al. [2019]4176N/A13
Wagh et al. [2021]101N/A13
Ours14493.6% (15)16
Ours24994.7% (15)32
CWagh et al. [2019]7188N/A13
Wagh et al. [2021]891N/A13
Tan et al. [2021]103694.0% (5)20
Ours34494.9% (5)16
Ours64393.8% (5)32
DMohassel and Rindal [2018]234N/AN/A
Ours4196.8% (15)16
Ours6896.8% (15)32
", + "type": "table", + "image_path": "97abf13640224b2e7ab8627c268f2fab07076453ba8f2fc10f4cc1bc2f920aae.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 112, + 356, + 499, + 438.6666666666667 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 112, + 438.6666666666667, + 499, + 521.3333333333334 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 112, + 521.3333333333334, + 499, + 604.0 + ], + "spans": [], + "index": 10 + } + ] + } + ], + "index": 7.25 + }, + { + "type": "title", + "bbox": [ + 86, + 634, + 188, + 648 + ], + "lines": [ + { + "bbox": [ + 84, + 632, + 189, + 650 + ], + "spans": [ + { + "bbox": [ + 84, + 632, + 189, + 650 + ], + "score": 1.0, + "content": "294 7 Conclusions", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 86, + 667, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 86, + 666, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 86, + 669, + 100, + 678 + ], + "score": 1.0, + "content": "295", + "type": "text" + }, + { + "bbox": [ + 105, + 666, + 505, + 680 + ], + "score": 1.0, + "content": "We have presented an implementation of deep learning training purely in multi-party computation", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 86, + 678, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 86, + 680, + 100, + 689 + ], + "score": 1.0, + "content": "296", + "type": "text" + }, + { + "bbox": [ + 106, + 678, + 505, + 690 + ], + "score": 1.0, + "content": "with extensive results on the accuracy. We have found that the lower precision of MPC computation", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 86, + 688, + 505, + 702 + ], + "spans": [ + { + "bbox": [ + 86, + 691, + 100, + 700 + ], + "score": 1.0, + "content": "297", + "type": "text" + }, + { + "bbox": [ + 105, + 688, + 505, + 702 + ], + "score": 1.0, + "content": "increases the error considerably. We only have considered one particular implementation of more", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 86, + 699, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 86, + 702, + 100, + 711 + ], + "score": 1.0, + "content": "298", + "type": "text" + }, + { + "bbox": [ + 105, + 699, + 505, + 713 + ], + "score": 1.0, + "content": "complex computation such as division and exponentiation, which are crucial to the learning process", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 86, + 710, + 498, + 723 + ], + "spans": [ + { + "bbox": [ + 86, + 713, + 100, + 722 + ], + "score": 1.0, + "content": "299", + "type": "text" + }, + { + "bbox": [ + 104, + 710, + 498, + 723 + ], + "score": 1.0, + "content": "as part of softmax. Future work might consider different approximations of these building blocks.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 133, + 69, + 480, + 261 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 133, + 69, + 480, + 261 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 133, + 69, + 480, + 261 + ], + "spans": [ + { + "bbox": [ + 133, + 69, + 480, + 261 + ], + "score": 0.965, + "type": "image", + "image_path": "700a37281e56f21023ee263453c7785f933f7176bea7c10046675acc845cf598.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 133, + 69, + 480, + 133.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 133, + 133.0, + 480, + 197.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 133, + 197.0, + 480, + 261.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 130, + 272, + 480, + 285 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 129, + 272, + 481, + 286 + ], + "spans": [ + { + "bbox": [ + 129, + 272, + 332, + 286 + ], + "score": 1.0, + "content": "Figure 4: Loss and accuracy for various networks,", + "type": "text" + }, + { + "bbox": [ + 333, + 273, + 363, + 285 + ], + "score": 0.91, + "content": "f = 3 2", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 272, + 481, + 286 + ], + "score": 1.0, + "content": ", and probabilistic truncation.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "table", + "bbox": [ + 112, + 356, + 499, + 604 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 305, + 505, + 350 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 305, + 506, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 305, + 354, + 318 + ], + "score": 1.0, + "content": "Table 2: Comparison to previous work in the LAN setting.", + "type": "text" + }, + { + "bbox": [ + 354, + 306, + 367, + 317 + ], + "score": 0.7, + "content": "( ^ { * } )", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 305, + 506, + 318 + ], + "score": 1.0, + "content": "Mohassel and Zhang [2017] and", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 317, + 506, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 317, + 506, + 329 + ], + "score": 1.0, + "content": "Agrawal et al. [2019] use a different security model and are thus incomparable. We include them for", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 327, + 506, + 341 + ], + "spans": [ + { + "bbox": [ + 105, + 327, + 506, + 341 + ], + "score": 1.0, + "content": "completeness. Two numbers refer to online and offline time. Accuracy N/A means that the accuracy", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 339, + 464, + 351 + ], + "spans": [ + { + "bbox": [ + 106, + 339, + 464, + 351 + ], + "score": 1.0, + "content": "figures were not given or computed in a way that does not reflect the secure computation.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5.5 + }, + { + "type": "table_body", + "bbox": [ + 112, + 356, + 499, + 604 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 356, + 499, + 604 + ], + "spans": [ + { + "bbox": [ + 112, + 356, + 499, + 604 + ], + "score": 0.986, + "html": "
NetworkEpoch time (s)Acc. (# epochs)Precision(f)
AMohassel and Zhang [2017]283/19333*93.4% (15)13
Mohassel and Rindal [2018]18094.0% (15)N/A
Agrawal et al. [2019]31392*95.0% (10)N/A
Wagh et al. [2019]247N/A13
Wagh et al. [2021]41N/A13
Ours3197.9% (15)16
Ours5097.7% (15)32
BWagh et al. [2019]4176N/A13
Wagh et al. [2021]101N/A13
Ours14493.6% (15)16
Ours24994.7% (15)32
CWagh et al. [2019]7188N/A13
Wagh et al. [2021]891N/A13
Tan et al. [2021]103694.0% (5)20
Ours34494.9% (5)16
Ours64393.8% (5)32
DMohassel and Rindal [2018]234N/AN/A
Ours4196.8% (15)16
Ours6896.8% (15)32
", + "type": "table", + "image_path": "97abf13640224b2e7ab8627c268f2fab07076453ba8f2fc10f4cc1bc2f920aae.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 112, + 356, + 499, + 438.6666666666667 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 112, + 438.6666666666667, + 499, + 521.3333333333334 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 112, + 521.3333333333334, + 499, + 604.0 + ], + "spans": [], + "index": 10 + } + ] + } + ], + "index": 7.25 + }, + { + "type": "title", + "bbox": [ + 86, + 634, + 188, + 648 + ], + "lines": [ + { + "bbox": [ + 84, + 632, + 189, + 650 + ], + "spans": [ + { + "bbox": [ + 84, + 632, + 189, + 650 + ], + "score": 1.0, + "content": "294 7 Conclusions", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "index", + "bbox": [ + 86, + 667, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 86, + 666, + 505, + 680 + ], + "spans": [ + { + "bbox": [ + 86, + 669, + 100, + 678 + ], + "score": 1.0, + "content": "295", + "type": "text" + }, + { + "bbox": [ + 105, + 666, + 505, + 680 + ], + "score": 1.0, + "content": "We have presented an implementation of deep learning training purely in multi-party computation", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 678, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 86, + 680, + 100, + 689 + ], + "score": 1.0, + "content": "296", + "type": "text" + }, + { + "bbox": [ + 106, + 678, + 505, + 690 + ], + "score": 1.0, + "content": "with extensive results on the accuracy. We have found that the lower precision of MPC computation", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 688, + 505, + 702 + ], + "spans": [ + { + "bbox": [ + 86, + 691, + 100, + 700 + ], + "score": 1.0, + "content": "297", + "type": "text" + }, + { + "bbox": [ + 105, + 688, + 505, + 702 + ], + "score": 1.0, + "content": "increases the error considerably. We only have considered one particular implementation of more", + "type": "text" + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 699, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 86, + 702, + 100, + 711 + ], + "score": 1.0, + "content": "298", + "type": "text" + }, + { + "bbox": [ + 105, + 699, + 505, + 713 + ], + "score": 1.0, + "content": "complex computation such as division and exponentiation, which are crucial to the learning process", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 710, + 498, + 723 + ], + "spans": [ + { + "bbox": [ + 86, + 713, + 100, + 722 + ], + "score": 1.0, + "content": "299", + "type": "text" + }, + { + "bbox": [ + 104, + 710, + 498, + 723 + ], + "score": 1.0, + "content": "as part of softmax. Future work might consider different approximations of these building blocks.", + "type": "text" + } + ], + "index": 16, + "is_list_start_line": true + } + ], + "index": 14, + "bbox_fs": [ + 86, + 666, + 505, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 92, + 72, + 164, + 84 + ], + "lines": [ + { + "bbox": [ + 92, + 72, + 164, + 84 + ], + "spans": [], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 86, + 75, + 508, + 728 + ], + "lines": [ + { + "bbox": [ + 90, + 71, + 165, + 86 + ], + "spans": [ + { + "bbox": [ + 90, + 71, + 165, + 86 + ], + "score": 1.0, + "content": "00 References", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 84, + 91, + 506, + 104 + ], + "spans": [ + { + "bbox": [ + 84, + 91, + 101, + 104 + ], + "score": 1.0, + "content": "301", + "type": "text" + }, + { + "bbox": [ + 104, + 91, + 506, + 104 + ], + "score": 1.0, + "content": "N. Agrawal, A. S. Shamsabadi, M. J. Kusner, and A. Gascón. QUOTIENT: Two-party secure neural", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 84, + 101, + 506, + 115 + ], + "spans": [ + { + "bbox": [ + 84, + 103, + 101, + 115 + ], + "score": 1.0, + "content": "302", + "type": "text" + }, + { + "bbox": [ + 114, + 101, + 506, + 115 + ], + "score": 1.0, + "content": "network training and prediction. In L. Cavallaro, J. Kinder, X. Wang, and J. Katz, editors, ACM", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 85, + 113, + 466, + 126 + ], + "spans": [ + { + "bbox": [ + 85, + 114, + 100, + 126 + ], + "score": 1.0, + "content": "303", + "type": "text" + }, + { + "bbox": [ + 115, + 113, + 466, + 126 + ], + "score": 1.0, + "content": "CCS 2019, pages 1231–1247. ACM Press, Nov. 2019. doi: 10.1145/3319535.3339819.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 85, + 132, + 507, + 148 + ], + "spans": [ + { + "bbox": [ + 85, + 135, + 101, + 146 + ], + "score": 1.0, + "content": "304", + "type": "text" + }, + { + "bbox": [ + 104, + 132, + 507, + 148 + ], + "score": 1.0, + "content": "M. Aliasgari, M. Blanton, Y. Zhang, and A. Steele. Secure computation on floating point numbers.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 85, + 144, + 309, + 157 + ], + "spans": [ + { + "bbox": [ + 85, + 146, + 100, + 157 + ], + "score": 1.0, + "content": "305", + "type": "text" + }, + { + "bbox": [ + 114, + 144, + 309, + 157 + ], + "score": 1.0, + "content": "In NDSS 2013. The Internet Society, Feb. 2013.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 85, + 162, + 507, + 180 + ], + "spans": [ + { + "bbox": [ + 85, + 166, + 101, + 177 + ], + "score": 1.0, + "content": "306", + "type": "text" + }, + { + "bbox": [ + 102, + 162, + 507, + 180 + ], + "score": 1.0, + "content": "A. Aly and N. P. Smart. Benchmarking privacy preserving scientific operations. In R. H. Deng,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 85, + 174, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 85, + 177, + 100, + 188 + ], + "score": 1.0, + "content": "307", + "type": "text" + }, + { + "bbox": [ + 114, + 174, + 506, + 190 + ], + "score": 1.0, + "content": "V. Gauthier-Umaña, M. Ochoa, and M. Yung, editors, ACNS 19, volume 11464 of LNCS, pages", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 85, + 185, + 446, + 201 + ], + "spans": [ + { + "bbox": [ + 85, + 187, + 100, + 199 + ], + "score": 1.0, + "content": "308", + "type": "text" + }, + { + "bbox": [ + 115, + 185, + 446, + 201 + ], + "score": 1.0, + "content": "509–529. Springer, Heidelberg, June 2019. doi: 10.1007/978-3-030-21568-2_25.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 85, + 205, + 506, + 220 + ], + "spans": [ + { + "bbox": [ + 85, + 208, + 101, + 218 + ], + "score": 1.0, + "content": "309", + "type": "text" + }, + { + "bbox": [ + 104, + 205, + 506, + 220 + ], + "score": 1.0, + "content": "T. Araki, J. Furukawa, Y. Lindell, A. Nof, and K. Ohara. High-throughput semi-honest secure", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 85, + 216, + 507, + 232 + ], + "spans": [ + { + "bbox": [ + 85, + 218, + 100, + 230 + ], + "score": 1.0, + "content": "310", + "type": "text" + }, + { + "bbox": [ + 114, + 216, + 507, + 232 + ], + "score": 1.0, + "content": "three-party computation with an honest majority. In E. R. Weippl, S. Katzenbeisser, C. Kruegel,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 85, + 227, + 507, + 242 + ], + "spans": [ + { + "bbox": [ + 85, + 230, + 100, + 241 + ], + "score": 1.0, + "content": "311", + "type": "text" + }, + { + "bbox": [ + 113, + 227, + 507, + 242 + ], + "score": 1.0, + "content": "A. C. Myers, and S. Halevi, editors, ACM CCS 2016, pages 805–817. ACM Press, Oct. 2016. doi:", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 85, + 238, + 229, + 252 + ], + "spans": [ + { + "bbox": [ + 85, + 240, + 100, + 252 + ], + "score": 1.0, + "content": "312", + "type": "text" + }, + { + "bbox": [ + 115, + 238, + 229, + 252 + ], + "score": 1.0, + "content": "10.1145/2976749.2978331.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 84, + 255, + 506, + 275 + ], + "spans": [ + { + "bbox": [ + 84, + 260, + 101, + 273 + ], + "score": 1.0, + "content": "313", + "type": "text" + }, + { + "bbox": [ + 102, + 255, + 506, + 275 + ], + "score": 1.0, + "content": "T. Araki, A. Barak, J. Furukawa, M. Keller, Y. Lindell, K. Ohara, and H. Tsuchida. Generalizing", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 84, + 268, + 507, + 285 + ], + "spans": [ + { + "bbox": [ + 84, + 271, + 101, + 284 + ], + "score": 1.0, + "content": "314", + "type": "text" + }, + { + "bbox": [ + 114, + 268, + 507, + 285 + ], + "score": 1.0, + "content": "the SPDZ compiler for other protocols. In D. Lie, M. Mannan, M. Backes, and X. Wang, editors,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 85, + 280, + 478, + 295 + ], + "spans": [ + { + "bbox": [ + 85, + 283, + 100, + 294 + ], + "score": 1.0, + "content": "315", + "type": "text" + }, + { + "bbox": [ + 112, + 280, + 478, + 295 + ], + "score": 1.0, + "content": "ACM CCS 2018, pages 880–895. ACM Press, Oct. 2018. doi: 10.1145/3243734.3243854.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 85, + 300, + 506, + 316 + ], + "spans": [ + { + "bbox": [ + 85, + 303, + 101, + 314 + ], + "score": 1.0, + "content": "316", + "type": "text" + }, + { + "bbox": [ + 104, + 300, + 506, + 316 + ], + "score": 1.0, + "content": "M. Barni, C. Orlandi, and A. Piva. A privacy-preserving protocol for neural-network-based computa-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 85, + 312, + 487, + 326 + ], + "spans": [ + { + "bbox": [ + 85, + 314, + 100, + 325 + ], + "score": 1.0, + "content": "317", + "type": "text" + }, + { + "bbox": [ + 115, + 312, + 487, + 326 + ], + "score": 1.0, + "content": "tion. In Proceedings of the 8th workshop on Multimedia and security, pages 146–151, 2006.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 85, + 332, + 506, + 346 + ], + "spans": [ + { + "bbox": [ + 85, + 334, + 101, + 344 + ], + "score": 1.0, + "content": "318", + "type": "text" + }, + { + "bbox": [ + 103, + 332, + 506, + 346 + ], + "score": 1.0, + "content": "J. C. Benaloh and J. Leichter. Generalized secret sharing and monotone functions. In S. Goldwasser,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 85, + 343, + 506, + 357 + ], + "spans": [ + { + "bbox": [ + 85, + 344, + 100, + 356 + ], + "score": 1.0, + "content": "319", + "type": "text" + }, + { + "bbox": [ + 115, + 343, + 506, + 357 + ], + "score": 1.0, + "content": "editor, CRYPTO’88, volume 403 of LNCS, pages 27–35. Springer, Heidelberg, Aug. 1990. doi:", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 85, + 354, + 227, + 367 + ], + "spans": [ + { + "bbox": [ + 85, + 355, + 100, + 367 + ], + "score": 1.0, + "content": "320", + "type": "text" + }, + { + "bbox": [ + 114, + 354, + 227, + 367 + ], + "score": 1.0, + "content": "10.1007/0-387-34799-2_3.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 85, + 374, + 506, + 388 + ], + "spans": [ + { + "bbox": [ + 85, + 376, + 101, + 387 + ], + "score": 1.0, + "content": "321", + "type": "text" + }, + { + "bbox": [ + 105, + 374, + 506, + 388 + ], + "score": 1.0, + "content": "O. Catrina and A. Saxena. Secure computation with fixed-point numbers. In R. Sion, editor, FC", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 85, + 385, + 422, + 399 + ], + "spans": [ + { + "bbox": [ + 85, + 387, + 100, + 398 + ], + "score": 1.0, + "content": "322", + "type": "text" + }, + { + "bbox": [ + 115, + 385, + 422, + 399 + ], + "score": 1.0, + "content": "2010, volume 6052 of LNCS, pages 35–50. Springer, Heidelberg, Jan. 2010.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 84, + 406, + 506, + 419 + ], + "spans": [ + { + "bbox": [ + 84, + 406, + 101, + 419 + ], + "score": 1.0, + "content": "323", + "type": "text" + }, + { + "bbox": [ + 105, + 406, + 506, + 419 + ], + "score": 1.0, + "content": "M. Dahl, J. Mancuso, Y. Dupis, B. Decoste, M. Giraud, I. Livingstone, J. Patriquin, and G. Uhma.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 85, + 416, + 508, + 432 + ], + "spans": [ + { + "bbox": [ + 85, + 418, + 100, + 429 + ], + "score": 1.0, + "content": "324", + "type": "text" + }, + { + "bbox": [ + 114, + 416, + 508, + 432 + ], + "score": 1.0, + "content": "Private machine learning in tensorflow using secure computation. CoRR, abs/1810.08130, 2018.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 85, + 428, + 307, + 442 + ], + "spans": [ + { + "bbox": [ + 85, + 429, + 100, + 441 + ], + "score": 1.0, + "content": "325", + "type": "text" + }, + { + "bbox": [ + 116, + 428, + 307, + 442 + ], + "score": 1.0, + "content": "URL http://arxiv.org/abs/1810.08130.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 85, + 447, + 506, + 462 + ], + "spans": [ + { + "bbox": [ + 85, + 450, + 101, + 460 + ], + "score": 1.0, + "content": "326", + "type": "text" + }, + { + "bbox": [ + 104, + 447, + 506, + 462 + ], + "score": 1.0, + "content": "A. Dalskov, D. Escudero, and M. Keller. Fantastic four: Honest-majority four-party secure com-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 85, + 457, + 507, + 474 + ], + "spans": [ + { + "bbox": [ + 85, + 460, + 100, + 471 + ], + "score": 1.0, + "content": "327", + "type": "text" + }, + { + "bbox": [ + 114, + 457, + 507, + 474 + ], + "score": 1.0, + "content": "putation with malicious security. In 30th USENIX Security Symposium (USENIX Security 21),", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 85, + 468, + 142, + 483 + ], + "spans": [ + { + "bbox": [ + 85, + 471, + 100, + 482 + ], + "score": 1.0, + "content": "328", + "type": "text" + }, + { + "bbox": [ + 114, + 468, + 142, + 483 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 84, + 489, + 506, + 503 + ], + "spans": [ + { + "bbox": [ + 84, + 490, + 101, + 503 + ], + "score": 1.0, + "content": "329", + "type": "text" + }, + { + "bbox": [ + 104, + 489, + 506, + 502 + ], + "score": 1.0, + "content": "A. P. K. Dalskov, D. Escudero, and M. Keller. Secure evaluation of quantized neural networks.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 85, + 499, + 406, + 515 + ], + "spans": [ + { + "bbox": [ + 85, + 502, + 100, + 513 + ], + "score": 1.0, + "content": "330", + "type": "text" + }, + { + "bbox": [ + 114, + 499, + 406, + 515 + ], + "score": 1.0, + "content": "PoPETs, 2020(4):355–375, Oct. 2020. doi: 10.2478/popets-2020-0077.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 85, + 520, + 506, + 536 + ], + "spans": [ + { + "bbox": [ + 85, + 523, + 101, + 533 + ], + "score": 1.0, + "content": "331", + "type": "text" + }, + { + "bbox": [ + 104, + 520, + 506, + 536 + ], + "score": 1.0, + "content": "I. Damgård, M. Fitzi, E. Kiltz, J. B. Nielsen, and T. Toft. Unconditionally secure constant-rounds", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 85, + 531, + 506, + 547 + ], + "spans": [ + { + "bbox": [ + 85, + 533, + 100, + 545 + ], + "score": 1.0, + "content": "332", + "type": "text" + }, + { + "bbox": [ + 114, + 531, + 506, + 547 + ], + "score": 1.0, + "content": "multi-party computation for equality, comparison, bits and exponentiation. In S. Halevi and", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 85, + 542, + 507, + 558 + ], + "spans": [ + { + "bbox": [ + 85, + 544, + 100, + 556 + ], + "score": 1.0, + "content": "333", + "type": "text" + }, + { + "bbox": [ + 114, + 542, + 507, + 558 + ], + "score": 1.0, + "content": "T. Rabin, editors, TCC 2006, volume 3876 of LNCS, pages 285–304. Springer, Heidelberg, Mar.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 85, + 554, + 258, + 567 + ], + "spans": [ + { + "bbox": [ + 85, + 555, + 100, + 567 + ], + "score": 1.0, + "content": "334", + "type": "text" + }, + { + "bbox": [ + 115, + 554, + 258, + 567 + ], + "score": 1.0, + "content": "2006. doi: 10.1007/11681878_15.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 85, + 574, + 506, + 587 + ], + "spans": [ + { + "bbox": [ + 85, + 575, + 101, + 586 + ], + "score": 1.0, + "content": "335", + "type": "text" + }, + { + "bbox": [ + 104, + 574, + 506, + 587 + ], + "score": 1.0, + "content": "I. Damgård, D. Escudero, T. K. Frederiksen, M. Keller, P. Scholl, and N. Volgushev. New primitives", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 85, + 585, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 85, + 586, + 100, + 598 + ], + "score": 1.0, + "content": "336", + "type": "text" + }, + { + "bbox": [ + 115, + 585, + 505, + 599 + ], + "score": 1.0, + "content": "for actively-secure MPC over rings with applications to private machine learning. In 2019 IEEE", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 85, + 595, + 506, + 611 + ], + "spans": [ + { + "bbox": [ + 85, + 597, + 100, + 609 + ], + "score": 1.0, + "content": "337", + "type": "text" + }, + { + "bbox": [ + 114, + 595, + 506, + 611 + ], + "score": 1.0, + "content": "Symposium on Security and Privacy, pages 1102–1120. IEEE Computer Society Press, May 2019.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 85, + 606, + 235, + 619 + ], + "spans": [ + { + "bbox": [ + 85, + 608, + 100, + 619 + ], + "score": 1.0, + "content": "338", + "type": "text" + }, + { + "bbox": [ + 115, + 606, + 235, + 619 + ], + "score": 1.0, + "content": "doi: 10.1109/SP.2019.00078.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 84, + 625, + 506, + 641 + ], + "spans": [ + { + "bbox": [ + 84, + 627, + 101, + 640 + ], + "score": 1.0, + "content": "339", + "type": "text" + }, + { + "bbox": [ + 104, + 625, + 506, + 641 + ], + "score": 1.0, + "content": "H. Eerikson, M. Keller, C. Orlandi, P. Pullonen, J. Puura, and M. Simkin. Use your brain! Arithmetic", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 84, + 637, + 506, + 653 + ], + "spans": [ + { + "bbox": [ + 84, + 639, + 101, + 652 + ], + "score": 1.0, + "content": "340", + "type": "text" + }, + { + "bbox": [ + 114, + 637, + 506, + 653 + ], + "score": 1.0, + "content": "3PC for any modulus with active security. In Y. T. Kalai, A. D. Smith, and D. Wichs, editors, ITC", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 85, + 648, + 461, + 663 + ], + "spans": [ + { + "bbox": [ + 85, + 650, + 100, + 661 + ], + "score": 1.0, + "content": "341", + "type": "text" + }, + { + "bbox": [ + 114, + 648, + 461, + 663 + ], + "score": 1.0, + "content": "2020, pages 5:1–5:24. Schloss Dagstuhl, June 2020. doi: 10.4230/LIPIcs.ITC.2020.5.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 85, + 668, + 507, + 684 + ], + "spans": [ + { + "bbox": [ + 85, + 671, + 101, + 681 + ], + "score": 1.0, + "content": "342", + "type": "text" + }, + { + "bbox": [ + 104, + 668, + 507, + 684 + ], + "score": 1.0, + "content": "X. Glorot and Y. Bengio. Understanding the difficulty of training deep feedforward neural networks.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 85, + 678, + 508, + 695 + ], + "spans": [ + { + "bbox": [ + 85, + 681, + 100, + 693 + ], + "score": 1.0, + "content": "343", + "type": "text" + }, + { + "bbox": [ + 114, + 678, + 508, + 695 + ], + "score": 1.0, + "content": "In Proceedings of the thirteenth international conference on artificial intelligence and statistics,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 85, + 689, + 398, + 705 + ], + "spans": [ + { + "bbox": [ + 85, + 693, + 100, + 704 + ], + "score": 1.0, + "content": "344", + "type": "text" + }, + { + "bbox": [ + 113, + 689, + 398, + 705 + ], + "score": 1.0, + "content": "pages 249–256. JMLR Workshop and Conference Proceedings, 2010.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 86, + 711, + 465, + 724 + ], + "spans": [ + { + "bbox": [ + 86, + 711, + 465, + 724 + ], + "score": 1.0, + "content": "345 R. E. Goldschmidt. Applications of division by convergence. Master’s thesis, MIT, 1964.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 23.5 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "10", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 92, + 72, + 164, + 84 + ], + "lines": [ + { + "bbox": [ + 92, + 72, + 164, + 84 + ], + "spans": [], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "index", + "bbox": [ + 86, + 75, + 508, + 728 + ], + "lines": [ + { + "bbox": [ + 90, + 71, + 165, + 86 + ], + "spans": [ + { + "bbox": [ + 90, + 71, + 165, + 86 + ], + "score": 1.0, + "content": "00 References", + "type": "text" + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 91, + 506, + 104 + ], + "spans": [ + { + "bbox": [ + 84, + 91, + 101, + 104 + ], + "score": 1.0, + "content": "301", + "type": "text" + }, + { + "bbox": [ + 104, + 91, + 506, + 104 + ], + "score": 1.0, + "content": "N. Agrawal, A. S. Shamsabadi, M. J. Kusner, and A. Gascón. QUOTIENT: Two-party secure neural", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 101, + 506, + 115 + ], + "spans": [ + { + "bbox": [ + 84, + 103, + 101, + 115 + ], + "score": 1.0, + "content": "302", + "type": "text" + }, + { + "bbox": [ + 114, + 101, + 506, + 115 + ], + "score": 1.0, + "content": "network training and prediction. In L. Cavallaro, J. Kinder, X. Wang, and J. Katz, editors, ACM", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 113, + 466, + 126 + ], + "spans": [ + { + "bbox": [ + 85, + 114, + 100, + 126 + ], + "score": 1.0, + "content": "303", + "type": "text" + }, + { + "bbox": [ + 115, + 113, + 466, + 126 + ], + "score": 1.0, + "content": "CCS 2019, pages 1231–1247. ACM Press, Nov. 2019. doi: 10.1145/3319535.3339819.", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 132, + 507, + 148 + ], + "spans": [ + { + "bbox": [ + 85, + 135, + 101, + 146 + ], + "score": 1.0, + "content": "304", + "type": "text" + }, + { + "bbox": [ + 104, + 132, + 507, + 148 + ], + "score": 1.0, + "content": "M. Aliasgari, M. Blanton, Y. Zhang, and A. Steele. Secure computation on floating point numbers.", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 144, + 309, + 157 + ], + "spans": [ + { + "bbox": [ + 85, + 146, + 100, + 157 + ], + "score": 1.0, + "content": "305", + "type": "text" + }, + { + "bbox": [ + 114, + 144, + 309, + 157 + ], + "score": 1.0, + "content": "In NDSS 2013. The Internet Society, Feb. 2013.", + "type": "text" + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 162, + 507, + 180 + ], + "spans": [ + { + "bbox": [ + 85, + 166, + 101, + 177 + ], + "score": 1.0, + "content": "306", + "type": "text" + }, + { + "bbox": [ + 102, + 162, + 507, + 180 + ], + "score": 1.0, + "content": "A. Aly and N. P. Smart. Benchmarking privacy preserving scientific operations. In R. H. Deng,", + "type": "text" + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 174, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 85, + 177, + 100, + 188 + ], + "score": 1.0, + "content": "307", + "type": "text" + }, + { + "bbox": [ + 114, + 174, + 506, + 190 + ], + "score": 1.0, + "content": "V. Gauthier-Umaña, M. Ochoa, and M. Yung, editors, ACNS 19, volume 11464 of LNCS, pages", + "type": "text" + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 185, + 446, + 201 + ], + "spans": [ + { + "bbox": [ + 85, + 187, + 100, + 199 + ], + "score": 1.0, + "content": "308", + "type": "text" + }, + { + "bbox": [ + 115, + 185, + 446, + 201 + ], + "score": 1.0, + "content": "509–529. Springer, Heidelberg, June 2019. doi: 10.1007/978-3-030-21568-2_25.", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 205, + 506, + 220 + ], + "spans": [ + { + "bbox": [ + 85, + 208, + 101, + 218 + ], + "score": 1.0, + "content": "309", + "type": "text" + }, + { + "bbox": [ + 104, + 205, + 506, + 220 + ], + "score": 1.0, + "content": "T. Araki, J. Furukawa, Y. Lindell, A. Nof, and K. Ohara. High-throughput semi-honest secure", + "type": "text" + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 216, + 507, + 232 + ], + "spans": [ + { + "bbox": [ + 85, + 218, + 100, + 230 + ], + "score": 1.0, + "content": "310", + "type": "text" + }, + { + "bbox": [ + 114, + 216, + 507, + 232 + ], + "score": 1.0, + "content": "three-party computation with an honest majority. In E. R. Weippl, S. Katzenbeisser, C. Kruegel,", + "type": "text" + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 227, + 507, + 242 + ], + "spans": [ + { + "bbox": [ + 85, + 230, + 100, + 241 + ], + "score": 1.0, + "content": "311", + "type": "text" + }, + { + "bbox": [ + 113, + 227, + 507, + 242 + ], + "score": 1.0, + "content": "A. C. Myers, and S. Halevi, editors, ACM CCS 2016, pages 805–817. ACM Press, Oct. 2016. doi:", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 238, + 229, + 252 + ], + "spans": [ + { + "bbox": [ + 85, + 240, + 100, + 252 + ], + "score": 1.0, + "content": "312", + "type": "text" + }, + { + "bbox": [ + 115, + 238, + 229, + 252 + ], + "score": 1.0, + "content": "10.1145/2976749.2978331.", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 255, + 506, + 275 + ], + "spans": [ + { + "bbox": [ + 84, + 260, + 101, + 273 + ], + "score": 1.0, + "content": "313", + "type": "text" + }, + { + "bbox": [ + 102, + 255, + 506, + 275 + ], + "score": 1.0, + "content": "T. Araki, A. Barak, J. Furukawa, M. Keller, Y. Lindell, K. Ohara, and H. Tsuchida. Generalizing", + "type": "text" + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 268, + 507, + 285 + ], + "spans": [ + { + "bbox": [ + 84, + 271, + 101, + 284 + ], + "score": 1.0, + "content": "314", + "type": "text" + }, + { + "bbox": [ + 114, + 268, + 507, + 285 + ], + "score": 1.0, + "content": "the SPDZ compiler for other protocols. In D. Lie, M. Mannan, M. Backes, and X. Wang, editors,", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 280, + 478, + 295 + ], + "spans": [ + { + "bbox": [ + 85, + 283, + 100, + 294 + ], + "score": 1.0, + "content": "315", + "type": "text" + }, + { + "bbox": [ + 112, + 280, + 478, + 295 + ], + "score": 1.0, + "content": "ACM CCS 2018, pages 880–895. ACM Press, Oct. 2018. doi: 10.1145/3243734.3243854.", + "type": "text" + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 300, + 506, + 316 + ], + "spans": [ + { + "bbox": [ + 85, + 303, + 101, + 314 + ], + "score": 1.0, + "content": "316", + "type": "text" + }, + { + "bbox": [ + 104, + 300, + 506, + 316 + ], + "score": 1.0, + "content": "M. Barni, C. Orlandi, and A. Piva. A privacy-preserving protocol for neural-network-based computa-", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 312, + 487, + 326 + ], + "spans": [ + { + "bbox": [ + 85, + 314, + 100, + 325 + ], + "score": 1.0, + "content": "317", + "type": "text" + }, + { + "bbox": [ + 115, + 312, + 487, + 326 + ], + "score": 1.0, + "content": "tion. In Proceedings of the 8th workshop on Multimedia and security, pages 146–151, 2006.", + "type": "text" + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 332, + 506, + 346 + ], + "spans": [ + { + "bbox": [ + 85, + 334, + 101, + 344 + ], + "score": 1.0, + "content": "318", + "type": "text" + }, + { + "bbox": [ + 103, + 332, + 506, + 346 + ], + "score": 1.0, + "content": "J. C. Benaloh and J. Leichter. Generalized secret sharing and monotone functions. In S. Goldwasser,", + "type": "text" + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 343, + 506, + 357 + ], + "spans": [ + { + "bbox": [ + 85, + 344, + 100, + 356 + ], + "score": 1.0, + "content": "319", + "type": "text" + }, + { + "bbox": [ + 115, + 343, + 506, + 357 + ], + "score": 1.0, + "content": "editor, CRYPTO’88, volume 403 of LNCS, pages 27–35. Springer, Heidelberg, Aug. 1990. doi:", + "type": "text" + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 354, + 227, + 367 + ], + "spans": [ + { + "bbox": [ + 85, + 355, + 100, + 367 + ], + "score": 1.0, + "content": "320", + "type": "text" + }, + { + "bbox": [ + 114, + 354, + 227, + 367 + ], + "score": 1.0, + "content": "10.1007/0-387-34799-2_3.", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 374, + 506, + 388 + ], + "spans": [ + { + "bbox": [ + 85, + 376, + 101, + 387 + ], + "score": 1.0, + "content": "321", + "type": "text" + }, + { + "bbox": [ + 105, + 374, + 506, + 388 + ], + "score": 1.0, + "content": "O. Catrina and A. Saxena. Secure computation with fixed-point numbers. In R. Sion, editor, FC", + "type": "text" + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 385, + 422, + 399 + ], + "spans": [ + { + "bbox": [ + 85, + 387, + 100, + 398 + ], + "score": 1.0, + "content": "322", + "type": "text" + }, + { + "bbox": [ + 115, + 385, + 422, + 399 + ], + "score": 1.0, + "content": "2010, volume 6052 of LNCS, pages 35–50. Springer, Heidelberg, Jan. 2010.", + "type": "text" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 406, + 506, + 419 + ], + "spans": [ + { + "bbox": [ + 84, + 406, + 101, + 419 + ], + "score": 1.0, + "content": "323", + "type": "text" + }, + { + "bbox": [ + 105, + 406, + 506, + 419 + ], + "score": 1.0, + "content": "M. Dahl, J. Mancuso, Y. Dupis, B. Decoste, M. Giraud, I. Livingstone, J. Patriquin, and G. Uhma.", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 416, + 508, + 432 + ], + "spans": [ + { + "bbox": [ + 85, + 418, + 100, + 429 + ], + "score": 1.0, + "content": "324", + "type": "text" + }, + { + "bbox": [ + 114, + 416, + 508, + 432 + ], + "score": 1.0, + "content": "Private machine learning in tensorflow using secure computation. CoRR, abs/1810.08130, 2018.", + "type": "text" + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 428, + 307, + 442 + ], + "spans": [ + { + "bbox": [ + 85, + 429, + 100, + 441 + ], + "score": 1.0, + "content": "325", + "type": "text" + }, + { + "bbox": [ + 116, + 428, + 307, + 442 + ], + "score": 1.0, + "content": "URL http://arxiv.org/abs/1810.08130.", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 447, + 506, + 462 + ], + "spans": [ + { + "bbox": [ + 85, + 450, + 101, + 460 + ], + "score": 1.0, + "content": "326", + "type": "text" + }, + { + "bbox": [ + 104, + 447, + 506, + 462 + ], + "score": 1.0, + "content": "A. Dalskov, D. Escudero, and M. Keller. Fantastic four: Honest-majority four-party secure com-", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 457, + 507, + 474 + ], + "spans": [ + { + "bbox": [ + 85, + 460, + 100, + 471 + ], + "score": 1.0, + "content": "327", + "type": "text" + }, + { + "bbox": [ + 114, + 457, + 507, + 474 + ], + "score": 1.0, + "content": "putation with malicious security. In 30th USENIX Security Symposium (USENIX Security 21),", + "type": "text" + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 468, + 142, + 483 + ], + "spans": [ + { + "bbox": [ + 85, + 471, + 100, + 482 + ], + "score": 1.0, + "content": "328", + "type": "text" + }, + { + "bbox": [ + 114, + 468, + 142, + 483 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 489, + 506, + 503 + ], + "spans": [ + { + "bbox": [ + 84, + 490, + 101, + 503 + ], + "score": 1.0, + "content": "329", + "type": "text" + }, + { + "bbox": [ + 104, + 489, + 506, + 502 + ], + "score": 1.0, + "content": "A. P. K. Dalskov, D. Escudero, and M. Keller. Secure evaluation of quantized neural networks.", + "type": "text" + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 499, + 406, + 515 + ], + "spans": [ + { + "bbox": [ + 85, + 502, + 100, + 513 + ], + "score": 1.0, + "content": "330", + "type": "text" + }, + { + "bbox": [ + 114, + 499, + 406, + 515 + ], + "score": 1.0, + "content": "PoPETs, 2020(4):355–375, Oct. 2020. doi: 10.2478/popets-2020-0077.", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 520, + 506, + 536 + ], + "spans": [ + { + "bbox": [ + 85, + 523, + 101, + 533 + ], + "score": 1.0, + "content": "331", + "type": "text" + }, + { + "bbox": [ + 104, + 520, + 506, + 536 + ], + "score": 1.0, + "content": "I. Damgård, M. Fitzi, E. Kiltz, J. B. Nielsen, and T. Toft. Unconditionally secure constant-rounds", + "type": "text" + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 531, + 506, + 547 + ], + "spans": [ + { + "bbox": [ + 85, + 533, + 100, + 545 + ], + "score": 1.0, + "content": "332", + "type": "text" + }, + { + "bbox": [ + 114, + 531, + 506, + 547 + ], + "score": 1.0, + "content": "multi-party computation for equality, comparison, bits and exponentiation. In S. Halevi and", + "type": "text" + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 542, + 507, + 558 + ], + "spans": [ + { + "bbox": [ + 85, + 544, + 100, + 556 + ], + "score": 1.0, + "content": "333", + "type": "text" + }, + { + "bbox": [ + 114, + 542, + 507, + 558 + ], + "score": 1.0, + "content": "T. Rabin, editors, TCC 2006, volume 3876 of LNCS, pages 285–304. Springer, Heidelberg, Mar.", + "type": "text" + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 554, + 258, + 567 + ], + "spans": [ + { + "bbox": [ + 85, + 555, + 100, + 567 + ], + "score": 1.0, + "content": "334", + "type": "text" + }, + { + "bbox": [ + 115, + 554, + 258, + 567 + ], + "score": 1.0, + "content": "2006. doi: 10.1007/11681878_15.", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 574, + 506, + 587 + ], + "spans": [ + { + "bbox": [ + 85, + 575, + 101, + 586 + ], + "score": 1.0, + "content": "335", + "type": "text" + }, + { + "bbox": [ + 104, + 574, + 506, + 587 + ], + "score": 1.0, + "content": "I. Damgård, D. Escudero, T. K. Frederiksen, M. Keller, P. Scholl, and N. Volgushev. New primitives", + "type": "text" + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 585, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 85, + 586, + 100, + 598 + ], + "score": 1.0, + "content": "336", + "type": "text" + }, + { + "bbox": [ + 115, + 585, + 505, + 599 + ], + "score": 1.0, + "content": "for actively-secure MPC over rings with applications to private machine learning. In 2019 IEEE", + "type": "text" + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 595, + 506, + 611 + ], + "spans": [ + { + "bbox": [ + 85, + 597, + 100, + 609 + ], + "score": 1.0, + "content": "337", + "type": "text" + }, + { + "bbox": [ + 114, + 595, + 506, + 611 + ], + "score": 1.0, + "content": "Symposium on Security and Privacy, pages 1102–1120. IEEE Computer Society Press, May 2019.", + "type": "text" + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 606, + 235, + 619 + ], + "spans": [ + { + "bbox": [ + 85, + 608, + 100, + 619 + ], + "score": 1.0, + "content": "338", + "type": "text" + }, + { + "bbox": [ + 115, + 606, + 235, + 619 + ], + "score": 1.0, + "content": "doi: 10.1109/SP.2019.00078.", + "type": "text" + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 625, + 506, + 641 + ], + "spans": [ + { + "bbox": [ + 84, + 627, + 101, + 640 + ], + "score": 1.0, + "content": "339", + "type": "text" + }, + { + "bbox": [ + 104, + 625, + 506, + 641 + ], + "score": 1.0, + "content": "H. Eerikson, M. Keller, C. Orlandi, P. Pullonen, J. Puura, and M. Simkin. Use your brain! Arithmetic", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 637, + 506, + 653 + ], + "spans": [ + { + "bbox": [ + 84, + 639, + 101, + 652 + ], + "score": 1.0, + "content": "340", + "type": "text" + }, + { + "bbox": [ + 114, + 637, + 506, + 653 + ], + "score": 1.0, + "content": "3PC for any modulus with active security. In Y. T. Kalai, A. D. Smith, and D. Wichs, editors, ITC", + "type": "text" + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 648, + 461, + 663 + ], + "spans": [ + { + "bbox": [ + 85, + 650, + 100, + 661 + ], + "score": 1.0, + "content": "341", + "type": "text" + }, + { + "bbox": [ + 114, + 648, + 461, + 663 + ], + "score": 1.0, + "content": "2020, pages 5:1–5:24. Schloss Dagstuhl, June 2020. doi: 10.4230/LIPIcs.ITC.2020.5.", + "type": "text" + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 668, + 507, + 684 + ], + "spans": [ + { + "bbox": [ + 85, + 671, + 101, + 681 + ], + "score": 1.0, + "content": "342", + "type": "text" + }, + { + "bbox": [ + 104, + 668, + 507, + 684 + ], + "score": 1.0, + "content": "X. Glorot and Y. Bengio. Understanding the difficulty of training deep feedforward neural networks.", + "type": "text" + } + ], + "index": 43, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 678, + 508, + 695 + ], + "spans": [ + { + "bbox": [ + 85, + 681, + 100, + 693 + ], + "score": 1.0, + "content": "343", + "type": "text" + }, + { + "bbox": [ + 114, + 678, + 508, + 695 + ], + "score": 1.0, + "content": "In Proceedings of the thirteenth international conference on artificial intelligence and statistics,", + "type": "text" + } + ], + "index": 44, + "is_list_start_line": true + }, + { + "bbox": [ + 85, + 689, + 398, + 705 + ], + "spans": [ + { + "bbox": [ + 85, + 693, + 100, + 704 + ], + "score": 1.0, + "content": "344", + "type": "text" + }, + { + "bbox": [ + 113, + 689, + 398, + 705 + ], + "score": 1.0, + "content": "pages 249–256. JMLR Workshop and Conference Proceedings, 2010.", + "type": "text" + } + ], + "index": 45, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 711, + 465, + 724 + ], + "spans": [ + { + "bbox": [ + 86, + 711, + 465, + 724 + ], + "score": 1.0, + "content": "345 R. E. Goldschmidt. Applications of division by convergence. Master’s thesis, MIT, 1964.", + "type": "text" + } + ], + "index": 46, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 72, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 86, + 74, + 101, + 86 + ], + "score": 1.0, + "content": "346", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 102, + 72, + 506, + 86 + ], + "score": 1.0, + "content": "D. Gunning, A. Hannun, M. Ibrahim, B. Knott, L. van der Maaten, V. Reis, S. Sengupta, S. Venkatara-", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 83, + 506, + 97 + ], + "spans": [ + { + "bbox": [ + 84, + 86, + 101, + 97 + ], + "score": 1.0, + "content": "347", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 83, + 506, + 97 + ], + "score": 1.0, + "content": "man, and X. Zhou. Crypten: A new research tool for secure machine learning with pytorch,", + "type": "text", + "cross_page": true + } + ], + "index": 1, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 93, + 144, + 108 + ], + "spans": [ + { + "bbox": [ + 84, + 96, + 101, + 107 + ], + "score": 1.0, + "content": "348", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 114, + 93, + 144, + 108 + ], + "score": 1.0, + "content": "2019.", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 82, + 113, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 82, + 113, + 506, + 129 + ], + "score": 1.0, + "content": "349 I. Hubara, M. Courbariaux, D. Soudry, R. El-Yaniv, and Y. Bengio. Binarized neural networks. In", + "type": "text", + "cross_page": true + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 124, + 501, + 138 + ], + "spans": [ + { + "bbox": [ + 86, + 127, + 101, + 138 + ], + "score": 1.0, + "content": "350", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 113, + 124, + 501, + 138 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems, volume 29. Curran Associates, Inc., 2016.", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 146, + 506, + 158 + ], + "spans": [ + { + "bbox": [ + 84, + 147, + 101, + 158 + ], + "score": 1.0, + "content": "351", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 146, + 506, + 158 + ], + "score": 1.0, + "content": "M. Keller. MP-SPDZ: A versatile framework for multi-party computation. In J. Ligatti, X. Ou,", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 157, + 506, + 169 + ], + "spans": [ + { + "bbox": [ + 84, + 158, + 101, + 169 + ], + "score": 1.0, + "content": "352", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 114, + 157, + 506, + 169 + ], + "score": 1.0, + "content": "J. Katz, and G. Vigna, editors, ACM CCS 20, pages 1575–1590. ACM Press, Nov. 2020. doi:", + "type": "text", + "cross_page": true + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 166, + 228, + 180 + ], + "spans": [ + { + "bbox": [ + 84, + 169, + 101, + 180 + ], + "score": 1.0, + "content": "353", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 166, + 228, + 180 + ], + "score": 1.0, + "content": "10.1145/3372297.3417872.", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 188, + 435, + 200 + ], + "spans": [ + { + "bbox": [ + 86, + 189, + 101, + 200 + ], + "score": 1.0, + "content": "354", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 188, + 435, + 200 + ], + "score": 1.0, + "content": "M. Keller and K. Sun. Effectiveness of MPC-friendly softmax replacement, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 206, + 506, + 221 + ], + "spans": [ + { + "bbox": [ + 84, + 206, + 506, + 221 + ], + "score": 1.0, + "content": "355 V. Kolesnikov, A.-R. Sadeghi, and T. Schneider. A systematic approach to practically efficient general", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 218, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 84, + 220, + 101, + 231 + ], + "score": 1.0, + "content": "356", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 218, + 505, + 232 + ], + "score": 1.0, + "content": "two-party secure function evaluation protocols and their modular design. Journal of Computer", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 229, + 243, + 243 + ], + "spans": [ + { + "bbox": [ + 84, + 230, + 100, + 242 + ], + "score": 1.0, + "content": "357", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 114, + 229, + 243, + 243 + ], + "score": 1.0, + "content": "Security, 21(2):283–315, 2013.", + "type": "text", + "cross_page": true + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 248, + 506, + 262 + ], + "spans": [ + { + "bbox": [ + 86, + 250, + 101, + 262 + ], + "score": 1.0, + "content": "358", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 248, + 506, + 262 + ], + "score": 1.0, + "content": "Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner. Gradient-based learning applied to document", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 260, + 376, + 273 + ], + "spans": [ + { + "bbox": [ + 84, + 262, + 101, + 273 + ], + "score": 1.0, + "content": "359", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 113, + 260, + 376, + 273 + ], + "score": 1.0, + "content": "recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998.", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 280, + 506, + 292 + ], + "spans": [ + { + "bbox": [ + 84, + 281, + 101, + 292 + ], + "score": 1.0, + "content": "360", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 280, + 506, + 292 + ], + "score": 1.0, + "content": "Y. LeCun, C. Cortes, and C. Burges. Mnist handwritten digit database. ATT Labs [Online]. Avail-", + "type": "text", + "cross_page": true + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 291, + 506, + 305 + ], + "spans": [ + { + "bbox": [ + 84, + 292, + 101, + 303 + ], + "score": 1.0, + "content": "361", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 291, + 506, + 305 + ], + "score": 1.0, + "content": "able: http: // yann. lecun. com/ exdb/ mnist , 2, 2010. Creative Commons Attribution-Share", + "type": "text", + "cross_page": true + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 301, + 440, + 315 + ], + "spans": [ + { + "bbox": [ + 84, + 303, + 101, + 315 + ], + "score": 1.0, + "content": "362", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 301, + 440, + 315 + ], + "score": 1.0, + "content": "Alike 3.0 license, https://creativecommons.org/licenses/by-sa/3.0/.", + "type": "text", + "cross_page": true + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 322, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 86, + 322, + 505, + 334 + ], + "score": 1.0, + "content": "363 J. Liu, M. Juuti, Y. Lu, and N. Asokan. Oblivious neural network predictions via MiniONN", + "type": "text", + "cross_page": true + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 333, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 84, + 334, + 101, + 345 + ], + "score": 1.0, + "content": "364", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 116, + 333, + 505, + 345 + ], + "score": 1.0, + "content": "transformations. In B. M. Thuraisingham, D. Evans, T. Malkin, and D. Xu, editors, ACM CCS", + "type": "text", + "cross_page": true + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 343, + 460, + 357 + ], + "spans": [ + { + "bbox": [ + 84, + 345, + 101, + 356 + ], + "score": 1.0, + "content": "365", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 343, + 460, + 357 + ], + "score": 1.0, + "content": "2017, pages 619–631. ACM Press, Oct. / Nov. 2017. doi: 10.1145/3133956.3134056.", + "type": "text", + "cross_page": true + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 363, + 506, + 378 + ], + "spans": [ + { + "bbox": [ + 84, + 365, + 101, + 376 + ], + "score": 1.0, + "content": "366", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 105, + 363, + 506, + 378 + ], + "score": 1.0, + "content": "W.-j. Lu, Y. Fang, Z. Huang, C. Hong, C. Chen, H. Qu, Y. Zhou, and K. Ren. Faster secure multiparty", + "type": "text", + "cross_page": true + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 374, + 506, + 389 + ], + "spans": [ + { + "bbox": [ + 84, + 375, + 101, + 387 + ], + "score": 1.0, + "content": "367", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 374, + 506, + 389 + ], + "score": 1.0, + "content": "computation of adaptive gradient descent. In Proceedings of the 2020 Workshop on Privacy-", + "type": "text", + "cross_page": true + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 385, + 506, + 399 + ], + "spans": [ + { + "bbox": [ + 84, + 386, + 101, + 398 + ], + "score": 1.0, + "content": "368", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 385, + 506, + 399 + ], + "score": 1.0, + "content": "Preserving Machine Learning in Practice, PPMLP’20, page 47–49, New York, NY, USA, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 395, + 507, + 410 + ], + "spans": [ + { + "bbox": [ + 84, + 397, + 99, + 409 + ], + "score": 1.0, + "content": "369", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 395, + 507, + 410 + ], + "score": 1.0, + "content": "Association for Computing Machinery. ISBN 9781450380881. doi: 10.1145/3411501.3419427.", + "type": "text", + "cross_page": true + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 407, + 347, + 421 + ], + "spans": [ + { + "bbox": [ + 84, + 409, + 101, + 421 + ], + "score": 1.0, + "content": "370", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 117, + 407, + 347, + 419 + ], + "score": 1.0, + "content": "URL https://doi.org/10.1145/3411501.3419427.", + "type": "text", + "cross_page": true + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 86, + 424, + 506, + 442 + ], + "spans": [ + { + "bbox": [ + 86, + 428, + 101, + 439 + ], + "score": 1.0, + "content": "371", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 424, + 506, + 442 + ], + "score": 1.0, + "content": "P. Mohassel and P. Rindal. ABY3: A mixed protocol framework for machine learning. In D. Lie,", + "type": "text", + "cross_page": true + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 438, + 507, + 452 + ], + "spans": [ + { + "bbox": [ + 84, + 439, + 101, + 451 + ], + "score": 1.0, + "content": "372", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 438, + 507, + 452 + ], + "score": 1.0, + "content": "M. Mannan, M. Backes, and X. Wang, editors, ACM CCS 2018, pages 35–52. ACM Press, Oct.", + "type": "text", + "cross_page": true + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 450, + 275, + 462 + ], + "spans": [ + { + "bbox": [ + 84, + 451, + 101, + 462 + ], + "score": 1.0, + "content": "373", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 116, + 450, + 275, + 462 + ], + "score": 1.0, + "content": "2018. doi: 10.1145/3243734.3243760.", + "type": "text", + "cross_page": true + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 466, + 507, + 484 + ], + "spans": [ + { + "bbox": [ + 84, + 466, + 507, + 484 + ], + "score": 1.0, + "content": "374 P. Mohassel and Y. Zhang. SecureML: A system for scalable privacy-preserving machine learning.", + "type": "text", + "cross_page": true + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 480, + 506, + 494 + ], + "spans": [ + { + "bbox": [ + 84, + 481, + 101, + 493 + ], + "score": 1.0, + "content": "375", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 480, + 506, + 494 + ], + "score": 1.0, + "content": "In 2017 IEEE Symposium on Security and Privacy, pages 19–38. IEEE Computer Society Press,", + "type": "text", + "cross_page": true + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 492, + 267, + 504 + ], + "spans": [ + { + "bbox": [ + 84, + 492, + 101, + 504 + ], + "score": 1.0, + "content": "376", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 116, + 492, + 267, + 504 + ], + "score": 1.0, + "content": "May 2017. doi: 10.1109/SP.2017.12.", + "type": "text", + "cross_page": true + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 510, + 506, + 524 + ], + "spans": [ + { + "bbox": [ + 84, + 513, + 101, + 524 + ], + "score": 1.0, + "content": "377", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 106, + 510, + 506, + 524 + ], + "score": 1.0, + "content": "V. Nair and G. E. Hinton. Rectified linear units improve Restricted Boltzmann machines. In Pro-", + "type": "text", + "cross_page": true + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 520, + 507, + 537 + ], + "spans": [ + { + "bbox": [ + 84, + 523, + 101, + 535 + ], + "score": 1.0, + "content": "378", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 113, + 520, + 507, + 537 + ], + "score": 1.0, + "content": "ceedings of the 27th International Conference on International Conference on Machine Learning,", + "type": "text", + "cross_page": true + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 531, + 250, + 546 + ], + "spans": [ + { + "bbox": [ + 84, + 534, + 101, + 545 + ], + "score": 1.0, + "content": "379", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 531, + 250, + 546 + ], + "score": 1.0, + "content": "ICML’10, pages 807–814, 2010.", + "type": "text", + "cross_page": true + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 552, + 506, + 566 + ], + "spans": [ + { + "bbox": [ + 84, + 552, + 506, + 566 + ], + "score": 1.0, + "content": "380 D. L. Quoc, F. Gregor, S. Arnautov, R. Kunkel, P. Bhatotia, and C. Fetzer. securetf: A secure", + "type": "text", + "cross_page": true + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 564, + 506, + 577 + ], + "spans": [ + { + "bbox": [ + 84, + 565, + 100, + 577 + ], + "score": 1.0, + "content": "381", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 116, + 564, + 506, + 576 + ], + "score": 1.0, + "content": "tensorflow framework. CoRR, abs/2101.08204, 2021. URL https://arxiv.org/abs/2101.", + "type": "text", + "cross_page": true + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 574, + 149, + 587 + ], + "spans": [ + { + "bbox": [ + 84, + 576, + 101, + 587 + ], + "score": 1.0, + "content": "382", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 116, + 574, + 149, + 587 + ], + "score": 1.0, + "content": "08204.", + "type": "text", + "cross_page": true + } + ], + "index": 36, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 593, + 506, + 608 + ], + "spans": [ + { + "bbox": [ + 84, + 596, + 101, + 607 + ], + "score": 1.0, + "content": "383", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 103, + 593, + 506, + 608 + ], + "score": 1.0, + "content": "M. S. Riazi, C. Weinert, O. Tkachenko, E. M. Songhori, T. Schneider, and F. Koushanfar. Chameleon:", + "type": "text", + "cross_page": true + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 605, + 506, + 619 + ], + "spans": [ + { + "bbox": [ + 84, + 606, + 101, + 618 + ], + "score": 1.0, + "content": "384", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 605, + 506, + 619 + ], + "score": 1.0, + "content": "A hybrid secure computation framework for machine learning applications. In J. Kim, G.-J. Ahn,", + "type": "text", + "cross_page": true + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 614, + 506, + 631 + ], + "spans": [ + { + "bbox": [ + 84, + 617, + 101, + 629 + ], + "score": 1.0, + "content": "385", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 113, + 614, + 506, + 631 + ], + "score": 1.0, + "content": "S. Kim, Y. Kim, J. López, and T. Kim, editors, ASIACCS 18, pages 707–721. ACM Press, Apr.", + "type": "text", + "cross_page": true + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 626, + 144, + 641 + ], + "spans": [ + { + "bbox": [ + 84, + 629, + 101, + 640 + ], + "score": 1.0, + "content": "386", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 626, + 144, + 641 + ], + "score": 1.0, + "content": "2018.", + "type": "text", + "cross_page": true + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 645, + 506, + 660 + ], + "spans": [ + { + "bbox": [ + 84, + 648, + 101, + 659 + ], + "score": 1.0, + "content": "387", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 103, + 645, + 506, + 660 + ], + "score": 1.0, + "content": "D. Rotaru and T. Wood. MArBled circuits: Mixing arithmetic and Boolean circuits with active", + "type": "text", + "cross_page": true + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 657, + 506, + 672 + ], + "spans": [ + { + "bbox": [ + 84, + 659, + 101, + 671 + ], + "score": 1.0, + "content": "388", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 116, + 657, + 506, + 672 + ], + "score": 1.0, + "content": "security. In F. Hao, S. Ruj, and S. Sen Gupta, editors, INDOCRYPT 2019, volume 11898 of LNCS,", + "type": "text", + "cross_page": true + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 668, + 470, + 683 + ], + "spans": [ + { + "bbox": [ + 84, + 671, + 101, + 682 + ], + "score": 1.0, + "content": "389", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 113, + 668, + 470, + 683 + ], + "score": 1.0, + "content": "pages 227–249. Springer, Heidelberg, Dec. 2019. doi: 10.1007/978-3-030-35423-7_12.", + "type": "text", + "cross_page": true + } + ], + "index": 43, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 84, + 690, + 101, + 702 + ], + "score": 1.0, + "content": "390", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 104, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "T. Ryffel, A. Trask, M. Dahl, B. Wagner, J. Mancuso, D. Rueckert, and J. Passerat-Palmbach. A", + "type": "text", + "cross_page": true + } + ], + "index": 44, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 700, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 84, + 702, + 100, + 713 + ], + "score": 1.0, + "content": "391", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 700, + 506, + 714 + ], + "score": 1.0, + "content": "generic framework for privacy preserving deep learning. CoRR, abs/1811.04017, 2018. URL", + "type": "text", + "cross_page": true + } + ], + "index": 45, + "is_list_start_line": true + }, + { + "bbox": [ + 84, + 711, + 286, + 725 + ], + "spans": [ + { + "bbox": [ + 84, + 713, + 101, + 724 + ], + "score": 1.0, + "content": "392", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 115, + 711, + 286, + 725 + ], + "score": 1.0, + "content": "http://arxiv.org/abs/1811.04017.", + "type": "text", + "cross_page": true + } + ], + "index": 46, + "is_list_start_line": true + } + ], + "index": 23.5, + "bbox_fs": [ + 84, + 71, + 508, + 724 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 84, + 47, + 508, + 727 + ], + "lines": [ + { + "bbox": [ + 86, + 72, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 86, + 74, + 101, + 86 + ], + "score": 1.0, + "content": "346", + "type": "text" + }, + { + "bbox": [ + 102, + 72, + 506, + 86 + ], + "score": 1.0, + "content": "D. Gunning, A. Hannun, M. Ibrahim, B. Knott, L. van der Maaten, V. Reis, S. Sengupta, S. Venkatara-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 84, + 83, + 506, + 97 + ], + "spans": [ + { + "bbox": [ + 84, + 86, + 101, + 97 + ], + "score": 1.0, + "content": "347", + "type": "text" + }, + { + "bbox": [ + 115, + 83, + 506, + 97 + ], + "score": 1.0, + "content": "man, and X. Zhou. Crypten: A new research tool for secure machine learning with pytorch,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 84, + 93, + 144, + 108 + ], + "spans": [ + { + "bbox": [ + 84, + 96, + 101, + 107 + ], + "score": 1.0, + "content": "348", + "type": "text" + }, + { + "bbox": [ + 114, + 93, + 144, + 108 + ], + "score": 1.0, + "content": "2019.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 82, + 113, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 82, + 113, + 506, + 129 + ], + "score": 1.0, + "content": "349 I. Hubara, M. Courbariaux, D. Soudry, R. El-Yaniv, and Y. Bengio. Binarized neural networks. In", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 86, + 124, + 501, + 138 + ], + "spans": [ + { + "bbox": [ + 86, + 127, + 101, + 138 + ], + "score": 1.0, + "content": "350", + "type": "text" + }, + { + "bbox": [ + 113, + 124, + 501, + 138 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems, volume 29. Curran Associates, Inc., 2016.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 84, + 146, + 506, + 158 + ], + "spans": [ + { + "bbox": [ + 84, + 147, + 101, + 158 + ], + "score": 1.0, + "content": "351", + "type": "text" + }, + { + "bbox": [ + 105, + 146, + 506, + 158 + ], + "score": 1.0, + "content": "M. Keller. MP-SPDZ: A versatile framework for multi-party computation. In J. Ligatti, X. Ou,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 84, + 157, + 506, + 169 + ], + "spans": [ + { + "bbox": [ + 84, + 158, + 101, + 169 + ], + "score": 1.0, + "content": "352", + "type": "text" + }, + { + "bbox": [ + 114, + 157, + 506, + 169 + ], + "score": 1.0, + "content": "J. Katz, and G. Vigna, editors, ACM CCS 20, pages 1575–1590. ACM Press, Nov. 2020. doi:", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 84, + 166, + 228, + 180 + ], + "spans": [ + { + "bbox": [ + 84, + 169, + 101, + 180 + ], + "score": 1.0, + "content": "353", + "type": "text" + }, + { + "bbox": [ + 115, + 166, + 228, + 180 + ], + "score": 1.0, + "content": "10.1145/3372297.3417872.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 86, + 188, + 435, + 200 + ], + "spans": [ + { + "bbox": [ + 86, + 189, + 101, + 200 + ], + "score": 1.0, + "content": "354", + "type": "text" + }, + { + "bbox": [ + 104, + 188, + 435, + 200 + ], + "score": 1.0, + "content": "M. Keller and K. Sun. Effectiveness of MPC-friendly softmax replacement, 2020.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 84, + 206, + 506, + 221 + ], + "spans": [ + { + "bbox": [ + 84, + 206, + 506, + 221 + ], + "score": 1.0, + "content": "355 V. Kolesnikov, A.-R. Sadeghi, and T. Schneider. A systematic approach to practically efficient general", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 84, + 218, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 84, + 220, + 101, + 231 + ], + "score": 1.0, + "content": "356", + "type": "text" + }, + { + "bbox": [ + 115, + 218, + 505, + 232 + ], + "score": 1.0, + "content": "two-party secure function evaluation protocols and their modular design. Journal of Computer", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 84, + 229, + 243, + 243 + ], + "spans": [ + { + "bbox": [ + 84, + 230, + 100, + 242 + ], + "score": 1.0, + "content": "357", + "type": "text" + }, + { + "bbox": [ + 114, + 229, + 243, + 243 + ], + "score": 1.0, + "content": "Security, 21(2):283–315, 2013.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 86, + 248, + 506, + 262 + ], + "spans": [ + { + "bbox": [ + 86, + 250, + 101, + 262 + ], + "score": 1.0, + "content": "358", + "type": "text" + }, + { + "bbox": [ + 105, + 248, + 506, + 262 + ], + "score": 1.0, + "content": "Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner. Gradient-based learning applied to document", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 84, + 260, + 376, + 273 + ], + "spans": [ + { + "bbox": [ + 84, + 262, + 101, + 273 + ], + "score": 1.0, + "content": "359", + "type": "text" + }, + { + "bbox": [ + 113, + 260, + 376, + 273 + ], + "score": 1.0, + "content": "recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 84, + 280, + 506, + 292 + ], + "spans": [ + { + "bbox": [ + 84, + 281, + 101, + 292 + ], + "score": 1.0, + "content": "360", + "type": "text" + }, + { + "bbox": [ + 106, + 280, + 506, + 292 + ], + "score": 1.0, + "content": "Y. LeCun, C. Cortes, and C. Burges. Mnist handwritten digit database. ATT Labs [Online]. Avail-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 84, + 291, + 506, + 305 + ], + "spans": [ + { + "bbox": [ + 84, + 292, + 101, + 303 + ], + "score": 1.0, + "content": "361", + "type": "text" + }, + { + "bbox": [ + 115, + 291, + 506, + 305 + ], + "score": 1.0, + "content": "able: http: // yann. lecun. com/ exdb/ mnist , 2, 2010. Creative Commons Attribution-Share", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 84, + 301, + 440, + 315 + ], + "spans": [ + { + "bbox": [ + 84, + 303, + 101, + 315 + ], + "score": 1.0, + "content": "362", + "type": "text" + }, + { + "bbox": [ + 115, + 301, + 440, + 315 + ], + "score": 1.0, + "content": "Alike 3.0 license, https://creativecommons.org/licenses/by-sa/3.0/.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 86, + 322, + 505, + 334 + ], + "spans": [ + { + "bbox": [ + 86, + 322, + 505, + 334 + ], + "score": 1.0, + "content": "363 J. Liu, M. Juuti, Y. Lu, and N. Asokan. Oblivious neural network predictions via MiniONN", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 84, + 333, + 505, + 345 + ], + "spans": [ + { + "bbox": [ + 84, + 334, + 101, + 345 + ], + "score": 1.0, + "content": "364", + "type": "text" + }, + { + "bbox": [ + 116, + 333, + 505, + 345 + ], + "score": 1.0, + "content": "transformations. In B. M. Thuraisingham, D. Evans, T. Malkin, and D. Xu, editors, ACM CCS", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 84, + 343, + 460, + 357 + ], + "spans": [ + { + "bbox": [ + 84, + 345, + 101, + 356 + ], + "score": 1.0, + "content": "365", + "type": "text" + }, + { + "bbox": [ + 115, + 343, + 460, + 357 + ], + "score": 1.0, + "content": "2017, pages 619–631. ACM Press, Oct. / Nov. 2017. doi: 10.1145/3133956.3134056.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 84, + 363, + 506, + 378 + ], + "spans": [ + { + "bbox": [ + 84, + 365, + 101, + 376 + ], + "score": 1.0, + "content": "366", + "type": "text" + }, + { + "bbox": [ + 105, + 363, + 506, + 378 + ], + "score": 1.0, + "content": "W.-j. Lu, Y. Fang, Z. Huang, C. Hong, C. Chen, H. Qu, Y. Zhou, and K. Ren. Faster secure multiparty", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 84, + 374, + 506, + 389 + ], + "spans": [ + { + "bbox": [ + 84, + 375, + 101, + 387 + ], + "score": 1.0, + "content": "367", + "type": "text" + }, + { + "bbox": [ + 115, + 374, + 506, + 389 + ], + "score": 1.0, + "content": "computation of adaptive gradient descent. In Proceedings of the 2020 Workshop on Privacy-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 84, + 385, + 506, + 399 + ], + "spans": [ + { + "bbox": [ + 84, + 386, + 101, + 398 + ], + "score": 1.0, + "content": "368", + "type": "text" + }, + { + "bbox": [ + 115, + 385, + 506, + 399 + ], + "score": 1.0, + "content": "Preserving Machine Learning in Practice, PPMLP’20, page 47–49, New York, NY, USA, 2020.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 84, + 395, + 507, + 410 + ], + "spans": [ + { + "bbox": [ + 84, + 397, + 99, + 409 + ], + "score": 1.0, + "content": "369", + "type": "text" + }, + { + "bbox": [ + 115, + 395, + 507, + 410 + ], + "score": 1.0, + "content": "Association for Computing Machinery. ISBN 9781450380881. doi: 10.1145/3411501.3419427.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 84, + 407, + 347, + 421 + ], + "spans": [ + { + "bbox": [ + 84, + 409, + 101, + 421 + ], + "score": 1.0, + "content": "370", + "type": "text" + }, + { + "bbox": [ + 117, + 407, + 347, + 419 + ], + "score": 1.0, + "content": "URL https://doi.org/10.1145/3411501.3419427.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 86, + 424, + 506, + 442 + ], + "spans": [ + { + "bbox": [ + 86, + 428, + 101, + 439 + ], + "score": 1.0, + "content": "371", + "type": "text" + }, + { + "bbox": [ + 104, + 424, + 506, + 442 + ], + "score": 1.0, + "content": "P. Mohassel and P. Rindal. ABY3: A mixed protocol framework for machine learning. In D. Lie,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 84, + 438, + 507, + 452 + ], + "spans": [ + { + "bbox": [ + 84, + 439, + 101, + 451 + ], + "score": 1.0, + "content": "372", + "type": "text" + }, + { + "bbox": [ + 115, + 438, + 507, + 452 + ], + "score": 1.0, + "content": "M. Mannan, M. Backes, and X. Wang, editors, ACM CCS 2018, pages 35–52. ACM Press, Oct.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 84, + 450, + 275, + 462 + ], + "spans": [ + { + "bbox": [ + 84, + 451, + 101, + 462 + ], + "score": 1.0, + "content": "373", + "type": "text" + }, + { + "bbox": [ + 116, + 450, + 275, + 462 + ], + "score": 1.0, + "content": "2018. doi: 10.1145/3243734.3243760.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 84, + 466, + 507, + 484 + ], + "spans": [ + { + "bbox": [ + 84, + 466, + 507, + 484 + ], + "score": 1.0, + "content": "374 P. Mohassel and Y. Zhang. SecureML: A system for scalable privacy-preserving machine learning.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 84, + 480, + 506, + 494 + ], + "spans": [ + { + "bbox": [ + 84, + 481, + 101, + 493 + ], + "score": 1.0, + "content": "375", + "type": "text" + }, + { + "bbox": [ + 115, + 480, + 506, + 494 + ], + "score": 1.0, + "content": "In 2017 IEEE Symposium on Security and Privacy, pages 19–38. IEEE Computer Society Press,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 84, + 492, + 267, + 504 + ], + "spans": [ + { + "bbox": [ + 84, + 492, + 101, + 504 + ], + "score": 1.0, + "content": "376", + "type": "text" + }, + { + "bbox": [ + 116, + 492, + 267, + 504 + ], + "score": 1.0, + "content": "May 2017. doi: 10.1109/SP.2017.12.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 84, + 510, + 506, + 524 + ], + "spans": [ + { + "bbox": [ + 84, + 513, + 101, + 524 + ], + "score": 1.0, + "content": "377", + "type": "text" + }, + { + "bbox": [ + 106, + 510, + 506, + 524 + ], + "score": 1.0, + "content": "V. Nair and G. E. Hinton. Rectified linear units improve Restricted Boltzmann machines. In Pro-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 84, + 520, + 507, + 537 + ], + "spans": [ + { + "bbox": [ + 84, + 523, + 101, + 535 + ], + "score": 1.0, + "content": "378", + "type": "text" + }, + { + "bbox": [ + 113, + 520, + 507, + 537 + ], + "score": 1.0, + "content": "ceedings of the 27th International Conference on International Conference on Machine Learning,", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 84, + 531, + 250, + 546 + ], + "spans": [ + { + "bbox": [ + 84, + 534, + 101, + 545 + ], + "score": 1.0, + "content": "379", + "type": "text" + }, + { + "bbox": [ + 115, + 531, + 250, + 546 + ], + "score": 1.0, + "content": "ICML’10, pages 807–814, 2010.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 84, + 552, + 506, + 566 + ], + "spans": [ + { + "bbox": [ + 84, + 552, + 506, + 566 + ], + "score": 1.0, + "content": "380 D. L. Quoc, F. Gregor, S. Arnautov, R. Kunkel, P. Bhatotia, and C. Fetzer. securetf: A secure", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 84, + 564, + 506, + 577 + ], + "spans": [ + { + "bbox": [ + 84, + 565, + 100, + 577 + ], + "score": 1.0, + "content": "381", + "type": "text" + }, + { + "bbox": [ + 116, + 564, + 506, + 576 + ], + "score": 1.0, + "content": "tensorflow framework. CoRR, abs/2101.08204, 2021. URL https://arxiv.org/abs/2101.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 84, + 574, + 149, + 587 + ], + "spans": [ + { + "bbox": [ + 84, + 576, + 101, + 587 + ], + "score": 1.0, + "content": "382", + "type": "text" + }, + { + "bbox": [ + 116, + 574, + 149, + 587 + ], + "score": 1.0, + "content": "08204.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 84, + 593, + 506, + 608 + ], + "spans": [ + { + "bbox": [ + 84, + 596, + 101, + 607 + ], + "score": 1.0, + "content": "383", + "type": "text" + }, + { + "bbox": [ + 103, + 593, + 506, + 608 + ], + "score": 1.0, + "content": "M. S. Riazi, C. Weinert, O. Tkachenko, E. M. Songhori, T. Schneider, and F. Koushanfar. Chameleon:", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 84, + 605, + 506, + 619 + ], + "spans": [ + { + "bbox": [ + 84, + 606, + 101, + 618 + ], + "score": 1.0, + "content": "384", + "type": "text" + }, + { + "bbox": [ + 115, + 605, + 506, + 619 + ], + "score": 1.0, + "content": "A hybrid secure computation framework for machine learning applications. In J. Kim, G.-J. Ahn,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 84, + 614, + 506, + 631 + ], + "spans": [ + { + "bbox": [ + 84, + 617, + 101, + 629 + ], + "score": 1.0, + "content": "385", + "type": "text" + }, + { + "bbox": [ + 113, + 614, + 506, + 631 + ], + "score": 1.0, + "content": "S. Kim, Y. Kim, J. López, and T. Kim, editors, ASIACCS 18, pages 707–721. ACM Press, Apr.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 84, + 626, + 144, + 641 + ], + "spans": [ + { + "bbox": [ + 84, + 629, + 101, + 640 + ], + "score": 1.0, + "content": "386", + "type": "text" + }, + { + "bbox": [ + 115, + 626, + 144, + 641 + ], + "score": 1.0, + "content": "2018.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 84, + 645, + 506, + 660 + ], + "spans": [ + { + "bbox": [ + 84, + 648, + 101, + 659 + ], + "score": 1.0, + "content": "387", + "type": "text" + }, + { + "bbox": [ + 103, + 645, + 506, + 660 + ], + "score": 1.0, + "content": "D. Rotaru and T. Wood. MArBled circuits: Mixing arithmetic and Boolean circuits with active", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 84, + 657, + 506, + 672 + ], + "spans": [ + { + "bbox": [ + 84, + 659, + 101, + 671 + ], + "score": 1.0, + "content": "388", + "type": "text" + }, + { + "bbox": [ + 116, + 657, + 506, + 672 + ], + "score": 1.0, + "content": "security. In F. Hao, S. Ruj, and S. Sen Gupta, editors, INDOCRYPT 2019, volume 11898 of LNCS,", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 84, + 668, + 470, + 683 + ], + "spans": [ + { + "bbox": [ + 84, + 671, + 101, + 682 + ], + "score": 1.0, + "content": "389", + "type": "text" + }, + { + "bbox": [ + 113, + 668, + 470, + 683 + ], + "score": 1.0, + "content": "pages 227–249. Springer, Heidelberg, Dec. 2019. doi: 10.1007/978-3-030-35423-7_12.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 84, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 84, + 690, + 101, + 702 + ], + "score": 1.0, + "content": "390", + "type": "text" + }, + { + "bbox": [ + 104, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "T. Ryffel, A. Trask, M. Dahl, B. Wagner, J. Mancuso, D. Rueckert, and J. Passerat-Palmbach. A", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 84, + 700, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 84, + 702, + 100, + 713 + ], + "score": 1.0, + "content": "391", + "type": "text" + }, + { + "bbox": [ + 115, + 700, + 506, + 714 + ], + "score": 1.0, + "content": "generic framework for privacy preserving deep learning. CoRR, abs/1811.04017, 2018. URL", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 84, + 711, + 286, + 725 + ], + "spans": [ + { + "bbox": [ + 84, + 713, + 101, + 724 + ], + "score": 1.0, + "content": "392", + "type": "text" + }, + { + "bbox": [ + 115, + 711, + 286, + 725 + ], + "score": 1.0, + "content": "http://arxiv.org/abs/1811.04017.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 23 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 751 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 312, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 312, + 754 + ], + "score": 1.0, + "content": "11", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "index", + "bbox": [ + 84, + 47, + 508, + 727 + ], + "lines": [], + "index": 23, + "bbox_fs": [ + 82, + 72, + 507, + 725 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 95, + 72, + 506, + 166 + ], + "lines": [ + { + "bbox": [ + 96, + 72, + 505, + 87 + ], + "spans": [ + { + "bbox": [ + 96, + 72, + 505, + 87 + ], + "score": 1.0, + "content": "3 S. Tan, B. Knott, Y. Tian, and D. J. Wu. CryptGPU: Fast privacy-preserving machine learning on the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 82, + 167, + 97 + ], + "spans": [ + { + "bbox": [ + 115, + 82, + 167, + 97 + ], + "score": 1.0, + "content": "GPU, 2021.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 103, + 102, + 506, + 115 + ], + "spans": [ + { + "bbox": [ + 103, + 102, + 506, + 115 + ], + "score": 1.0, + "content": "S. Wagh, D. Gupta, and N. Chandran. SecureNN: 3-party secure computation for neural network", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 113, + 432, + 127 + ], + "spans": [ + { + "bbox": [ + 115, + 113, + 432, + 127 + ], + "score": 1.0, + "content": "training. PoPETs, 2019(3):26–49, July 2019. doi: 10.2478/popets-2019-0035.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 103, + 131, + 505, + 146 + ], + "spans": [ + { + "bbox": [ + 103, + 131, + 505, + 146 + ], + "score": 1.0, + "content": "S. Wagh, S. Tople, F. Benhamouda, E. Kushilevitz, P. Mittal, and T. Rabin. Falcon: Honest-majority", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 142, + 508, + 158 + ], + "spans": [ + { + "bbox": [ + 115, + 142, + 508, + 158 + ], + "score": 1.0, + "content": "maliciously secure framework for private deep learning. PoPETs, 2021(1):188–208, Jan. 2021.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 154, + 250, + 168 + ], + "spans": [ + { + "bbox": [ + 115, + 154, + 250, + 168 + ], + "score": 1.0, + "content": "doi: 10.2478/popets-2021-0011.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3 + }, + { + "type": "title", + "bbox": [ + 88, + 186, + 156, + 199 + ], + "lines": [ + { + "bbox": [ + 104, + 185, + 158, + 200 + ], + "spans": [ + { + "bbox": [ + 104, + 185, + 158, + 200 + ], + "score": 1.0, + "content": "Checklist", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 131, + 207, + 208, + 218 + ], + "lines": [ + { + "bbox": [ + 129, + 205, + 210, + 220 + ], + "spans": [ + { + "bbox": [ + 129, + 205, + 210, + 220 + ], + "score": 1.0, + "content": "1. For all authors...", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 146, + 222, + 505, + 316 + ], + "lines": [ + { + "bbox": [ + 146, + 222, + 504, + 234 + ], + "spans": [ + { + "bbox": [ + 146, + 222, + 504, + 234 + ], + "score": 1.0, + "content": "(a) Do the main claims made in the abstract and introduction accurately reflect the paper’s", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 162, + 233, + 288, + 245 + ], + "spans": [ + { + "bbox": [ + 162, + 233, + 288, + 245 + ], + "score": 1.0, + "content": "contributions and scope? [Yes]", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 144, + 245, + 436, + 259 + ], + "spans": [ + { + "bbox": [ + 144, + 245, + 436, + 259 + ], + "score": 1.0, + "content": "(b) Did you describe the limitations of your work? [Yes] See Section 7.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 145, + 258, + 505, + 271 + ], + "spans": [ + { + "bbox": [ + 145, + 258, + 505, + 271 + ], + "score": 1.0, + "content": "(c) Did you discuss any potential negative societal impacts of your work? [No] We only", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 160, + 269, + 506, + 282 + ], + "spans": [ + { + "bbox": [ + 160, + 269, + 506, + 282 + ], + "score": 1.0, + "content": "work with MNIST, and we believe that the potential negative impacts associated to it", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 162, + 280, + 217, + 293 + ], + "spans": [ + { + "bbox": [ + 162, + 280, + 217, + 293 + ], + "score": 1.0, + "content": "are very low.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 144, + 292, + 505, + 307 + ], + "spans": [ + { + "bbox": [ + 144, + 292, + 505, + 307 + ], + "score": 1.0, + "content": "(d) Have you read the ethics review guidelines and ensured that your paper conforms to", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 162, + 304, + 214, + 317 + ], + "spans": [ + { + "bbox": [ + 162, + 304, + 214, + 317 + ], + "score": 1.0, + "content": "them? [Yes]", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 131, + 319, + 302, + 331 + ], + "lines": [ + { + "bbox": [ + 129, + 318, + 304, + 333 + ], + "spans": [ + { + "bbox": [ + 129, + 318, + 304, + 333 + ], + "score": 1.0, + "content": "2. If you are including theoretical results...", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 146, + 334, + 452, + 359 + ], + "lines": [ + { + "bbox": [ + 145, + 334, + 453, + 347 + ], + "spans": [ + { + "bbox": [ + 145, + 334, + 453, + 347 + ], + "score": 1.0, + "content": "(a) Did you state the full set of assumptions of all theoretical results? [N/A]", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 145, + 347, + 424, + 360 + ], + "spans": [ + { + "bbox": [ + 145, + 347, + 424, + 360 + ], + "score": 1.0, + "content": "(b) Did you include complete proofs of all theoretical results? [N/A]", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 130, + 362, + 241, + 374 + ], + "lines": [ + { + "bbox": [ + 128, + 360, + 243, + 376 + ], + "spans": [ + { + "bbox": [ + 128, + 360, + 243, + 376 + ], + "score": 1.0, + "content": "3. If you ran experiments...", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 146, + 377, + 505, + 482 + ], + "lines": [ + { + "bbox": [ + 145, + 376, + 506, + 389 + ], + "spans": [ + { + "bbox": [ + 145, + 376, + 506, + 389 + ], + "score": 1.0, + "content": "(a) Did you include the code, data, and instructions needed to reproduce the main experi-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 162, + 388, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 162, + 388, + 505, + 399 + ], + "score": 1.0, + "content": "mental results (either in the supplemental material or as a URL)? [Yes] We include all", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 162, + 399, + 495, + 410 + ], + "spans": [ + { + "bbox": [ + 162, + 399, + 495, + 410 + ], + "score": 1.0, + "content": "code in the supplemental material, and the references include the URL for MNIST.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 146, + 411, + 505, + 424 + ], + "spans": [ + { + "bbox": [ + 146, + 411, + 505, + 424 + ], + "score": 1.0, + "content": "(b) Did you specify all the training details (e.g., data splits, hyperparameters, how they", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 161, + 423, + 305, + 434 + ], + "spans": [ + { + "bbox": [ + 161, + 423, + 305, + 434 + ], + "score": 1.0, + "content": "were chosen)? [Yes] See Section 6.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 146, + 434, + 507, + 449 + ], + "spans": [ + { + "bbox": [ + 146, + 434, + 507, + 449 + ], + "score": 1.0, + "content": "(c) Did you report error bars (e.g., with respect to the random seed after running experi-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 162, + 446, + 351, + 459 + ], + "spans": [ + { + "bbox": [ + 162, + 446, + 351, + 459 + ], + "score": 1.0, + "content": "ments multiple times)? [Yes] Only in Figure 2.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 146, + 457, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 146, + 457, + 505, + 473 + ], + "score": 1.0, + "content": "(d) Did you include the total amount of compute and the type of resources used (e.g., type", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 162, + 470, + 429, + 482 + ], + "spans": [ + { + "bbox": [ + 162, + 470, + 429, + 482 + ], + "score": 1.0, + "content": "of GPUs, internal cluster, or cloud provider)? [Yes] See Section 6.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 132, + 485, + 504, + 497 + ], + "lines": [ + { + "bbox": [ + 129, + 483, + 507, + 499 + ], + "spans": [ + { + "bbox": [ + 129, + 483, + 507, + 499 + ], + "score": 1.0, + "content": "4. If you are using existing assets (e.g., code, data, models) or curating/releasing new assets...", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 146, + 500, + 505, + 596 + ], + "lines": [ + { + "bbox": [ + 145, + 499, + 483, + 512 + ], + "spans": [ + { + "bbox": [ + 145, + 499, + 483, + 512 + ], + "score": 1.0, + "content": "(a) If your work uses existing assets, did you cite the creators? [Yes] See Section 6.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 145, + 513, + 506, + 525 + ], + "spans": [ + { + "bbox": [ + 145, + 513, + 506, + 525 + ], + "score": 1.0, + "content": "(b) Did you mention the license of the assets? [Yes] See the references and and Section 5.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 146, + 525, + 505, + 538 + ], + "spans": [ + { + "bbox": [ + 146, + 525, + 505, + 538 + ], + "score": 1.0, + "content": "(c) Did you include any new assets either in the supplemental material or as a URL? [Yes]", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 162, + 537, + 355, + 549 + ], + "spans": [ + { + "bbox": [ + 162, + 537, + 355, + 549 + ], + "score": 1.0, + "content": "Our code is included the supplemental material.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 146, + 549, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 146, + 549, + 505, + 563 + ], + "score": 1.0, + "content": "(d) Did you discuss whether and how consent was obtained from people whose data you’re", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 162, + 561, + 254, + 573 + ], + "spans": [ + { + "bbox": [ + 162, + 561, + 254, + 573 + ], + "score": 1.0, + "content": "using/curating? [N/A]", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 146, + 573, + 505, + 586 + ], + "spans": [ + { + "bbox": [ + 146, + 573, + 505, + 586 + ], + "score": 1.0, + "content": "(e) Did you discuss whether the data you are using/curating contains personally identifiable", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 162, + 584, + 325, + 596 + ], + "spans": [ + { + "bbox": [ + 162, + 584, + 325, + 596 + ], + "score": 1.0, + "content": "information or offensive content? [N/A]", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 131, + 599, + 432, + 611 + ], + "lines": [ + { + "bbox": [ + 128, + 597, + 433, + 613 + ], + "spans": [ + { + "bbox": [ + 128, + 597, + 433, + 613 + ], + "score": 1.0, + "content": "5. If you used crowdsourcing or conducted research with human subjects...", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 146, + 614, + 505, + 685 + ], + "lines": [ + { + "bbox": [ + 145, + 613, + 506, + 627 + ], + "spans": [ + { + "bbox": [ + 145, + 613, + 506, + 627 + ], + "score": 1.0, + "content": "(a) Did you include the full text of instructions given to participants and screenshots, if", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 162, + 625, + 237, + 637 + ], + "spans": [ + { + "bbox": [ + 162, + 625, + 237, + 637 + ], + "score": 1.0, + "content": "applicable? [N/A]", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 145, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 145, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "(b) Did you describe any potential participant risks, with links to Institutional Review", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 161, + 649, + 342, + 662 + ], + "spans": [ + { + "bbox": [ + 161, + 649, + 342, + 662 + ], + "score": 1.0, + "content": "Board (IRB) approvals, if applicable? [N/A]", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 147, + 661, + 505, + 674 + ], + "spans": [ + { + "bbox": [ + 147, + 661, + 505, + 674 + ], + "score": 1.0, + "content": "(c) Did you include the estimated hourly wage paid to participants and the total amount", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 162, + 673, + 333, + 685 + ], + "spans": [ + { + "bbox": [ + 162, + 673, + 333, + 685 + ], + "score": 1.0, + "content": "spent on participant compensation? [N/A]", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 42.5 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 85, + 73, + 102, + 699 + ], + "lines": [ + { + "bbox": [ + 85, + 74, + 100, + 85 + ], + "spans": [ + { + "bbox": [ + 85, + 74, + 100, + 85 + ], + "score": 1.0, + "content": "39", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 85, + 102, + 98 + ], + "spans": [ + { + "bbox": [ + 84, + 85, + 102, + 98 + ], + "score": 1.0, + "content": "394", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 104, + 100, + 115 + ], + "spans": [ + { + "bbox": [ + 84, + 104, + 100, + 115 + ], + "score": 1.0, + "content": "395", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 115, + 101, + 130 + ], + "spans": [ + { + "bbox": [ + 84, + 115, + 101, + 130 + ], + "score": 1.0, + "content": "396", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 133, + 100, + 145 + ], + "spans": [ + { + "bbox": [ + 84, + 133, + 100, + 145 + ], + "score": 1.0, + "content": "397", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 145, + 100, + 156 + ], + "spans": [ + { + "bbox": [ + 84, + 145, + 100, + 156 + ], + "score": 1.0, + "content": "398", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 155, + 101, + 168 + ], + "spans": [ + { + "bbox": [ + 84, + 155, + 101, + 168 + ], + "score": 1.0, + "content": "399", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 190, + 100, + 201 + ], + "spans": [ + { + "bbox": [ + 84, + 190, + 100, + 201 + ], + "score": 1.0, + "content": "400", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 208, + 101, + 222 + ], + "spans": [ + { + "bbox": [ + 84, + 208, + 101, + 222 + ], + "score": 1.0, + "content": "401", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 224, + 100, + 235 + ], + "spans": [ + { + "bbox": [ + 84, + 224, + 100, + 235 + ], + "score": 1.0, + "content": "402", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 235, + 100, + 246 + ], + "spans": [ + { + "bbox": [ + 84, + 235, + 100, + 246 + ], + "score": 1.0, + "content": "403", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 247, + 101, + 260 + ], + "spans": [ + { + "bbox": [ + 84, + 247, + 101, + 260 + ], + "score": 1.0, + "content": "404", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 259, + 102, + 272 + ], + "spans": [ + { + "bbox": [ + 84, + 259, + 102, + 272 + ], + "score": 1.0, + "content": "405", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 271, + 102, + 284 + ], + "spans": [ + { + "bbox": [ + 84, + 271, + 102, + 284 + ], + "score": 1.0, + "content": "406", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 282, + 100, + 294 + ], + "spans": [ + { + "bbox": [ + 85, + 282, + 100, + 294 + ], + "score": 1.0, + "content": "407", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 295, + 101, + 306 + ], + "spans": [ + { + "bbox": [ + 84, + 295, + 101, + 306 + ], + "score": 1.0, + "content": "408", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 306, + 101, + 319 + ], + "spans": [ + { + "bbox": [ + 84, + 306, + 101, + 319 + ], + "score": 1.0, + "content": "409", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 320, + 101, + 335 + ], + "spans": [ + { + "bbox": [ + 84, + 320, + 101, + 335 + ], + "score": 1.0, + "content": "410", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 336, + 100, + 347 + ], + "spans": [ + { + "bbox": [ + 84, + 336, + 100, + 347 + ], + "score": 1.0, + "content": "411", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 349, + 102, + 361 + ], + "spans": [ + { + "bbox": [ + 84, + 349, + 102, + 361 + ], + "score": 1.0, + "content": "412", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 363, + 101, + 378 + ], + "spans": [ + { + "bbox": [ + 84, + 363, + 101, + 378 + ], + "score": 1.0, + "content": "413", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 379, + 101, + 390 + ], + "spans": [ + { + "bbox": [ + 84, + 379, + 101, + 390 + ], + "score": 1.0, + "content": "414", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 390, + 101, + 401 + ], + "spans": [ + { + "bbox": [ + 84, + 390, + 101, + 401 + ], + "score": 1.0, + "content": "415", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 401, + 102, + 413 + ], + "spans": [ + { + "bbox": [ + 84, + 401, + 102, + 413 + ], + "score": 1.0, + "content": "416", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 413, + 100, + 425 + ], + "spans": [ + { + "bbox": [ + 84, + 413, + 100, + 425 + ], + "score": 1.0, + "content": "417", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 425, + 101, + 435 + ], + "spans": [ + { + "bbox": [ + 84, + 425, + 101, + 435 + ], + "score": 1.0, + "content": "418", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 438, + 101, + 448 + ], + "spans": [ + { + "bbox": [ + 84, + 438, + 101, + 448 + ], + "score": 1.0, + "content": "419", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 448, + 100, + 460 + ], + "spans": [ + { + "bbox": [ + 84, + 448, + 100, + 460 + ], + "score": 1.0, + "content": "420", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 461, + 100, + 472 + ], + "spans": [ + { + "bbox": [ + 84, + 461, + 100, + 472 + ], + "score": 1.0, + "content": "421", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 471, + 101, + 486 + ], + "spans": [ + { + "bbox": [ + 84, + 471, + 101, + 486 + ], + "score": 1.0, + "content": "422", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 486, + 101, + 500 + ], + "spans": [ + { + "bbox": [ + 84, + 486, + 101, + 500 + ], + "score": 1.0, + "content": "423", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 501, + 101, + 514 + ], + "spans": [ + { + "bbox": [ + 84, + 501, + 101, + 514 + ], + "score": 1.0, + "content": "424", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 514, + 101, + 527 + ], + "spans": [ + { + "bbox": [ + 84, + 514, + 101, + 527 + ], + "score": 1.0, + "content": "425", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 527, + 102, + 540 + ], + "spans": [ + { + "bbox": [ + 84, + 527, + 102, + 540 + ], + "score": 1.0, + "content": "426", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 537, + 101, + 551 + ], + "spans": [ + { + "bbox": [ + 84, + 537, + 101, + 551 + ], + "score": 1.0, + "content": "427", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 551, + 102, + 564 + ], + "spans": [ + { + "bbox": [ + 84, + 551, + 102, + 564 + ], + "score": 1.0, + "content": "428", + "type": "text" + } + ] + }, + { + "bbox": [ + 85, + 562, + 101, + 575 + ], + "spans": [ + { + "bbox": [ + 85, + 562, + 101, + 575 + ], + "score": 1.0, + "content": "429", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 575, + 100, + 586 + ], + "spans": [ + { + "bbox": [ + 84, + 575, + 100, + 586 + ], + "score": 1.0, + "content": "430", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 586, + 100, + 599 + ], + "spans": [ + { + "bbox": [ + 84, + 586, + 100, + 599 + ], + "score": 1.0, + "content": "431", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 600, + 101, + 614 + ], + "spans": [ + { + "bbox": [ + 84, + 600, + 101, + 614 + ], + "score": 1.0, + "content": "432", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 615, + 100, + 627 + ], + "spans": [ + { + "bbox": [ + 84, + 615, + 100, + 627 + ], + "score": 1.0, + "content": "433", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 626, + 102, + 640 + ], + "spans": [ + { + "bbox": [ + 84, + 626, + 102, + 640 + ], + "score": 1.0, + "content": "434", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 639, + 100, + 651 + ], + "spans": [ + { + "bbox": [ + 84, + 639, + 100, + 651 + ], + "score": 1.0, + "content": "435", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 650, + 102, + 663 + ], + "spans": [ + { + "bbox": [ + 84, + 650, + 102, + 663 + ], + "score": 1.0, + "content": "436", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 663, + 100, + 675 + ], + "spans": [ + { + "bbox": [ + 84, + 663, + 100, + 675 + ], + "score": 1.0, + "content": "437", + "type": "text" + } + ] + }, + { + "bbox": [ + 84, + 674, + 100, + 686 + ], + "spans": [ + { + "bbox": [ + 84, + 674, + 100, + 686 + ], + "score": 1.0, + "content": "438", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 95, + 72, + 506, + 166 + ], + "lines": [ + { + "bbox": [ + 96, + 72, + 505, + 87 + ], + "spans": [ + { + "bbox": [ + 96, + 72, + 505, + 87 + ], + "score": 1.0, + "content": "3 S. Tan, B. Knott, Y. Tian, and D. J. Wu. CryptGPU: Fast privacy-preserving machine learning on the", + "type": "text" + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 82, + 167, + 97 + ], + "spans": [ + { + "bbox": [ + 115, + 82, + 167, + 97 + ], + "score": 1.0, + "content": "GPU, 2021.", + "type": "text" + } + ], + "index": 1, + "is_list_end_line": true + }, + { + "bbox": [ + 103, + 102, + 506, + 115 + ], + "spans": [ + { + "bbox": [ + 103, + 102, + 506, + 115 + ], + "score": 1.0, + "content": "S. Wagh, D. Gupta, and N. Chandran. SecureNN: 3-party secure computation for neural network", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 113, + 432, + 127 + ], + "spans": [ + { + "bbox": [ + 115, + 113, + 432, + 127 + ], + "score": 1.0, + "content": "training. PoPETs, 2019(3):26–49, July 2019. doi: 10.2478/popets-2019-0035.", + "type": "text" + } + ], + "index": 3, + "is_list_end_line": true + }, + { + "bbox": [ + 103, + 131, + 505, + 146 + ], + "spans": [ + { + "bbox": [ + 103, + 131, + 505, + 146 + ], + "score": 1.0, + "content": "S. Wagh, S. Tople, F. Benhamouda, E. Kushilevitz, P. Mittal, and T. Rabin. Falcon: Honest-majority", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 115, + 142, + 508, + 158 + ], + "spans": [ + { + "bbox": [ + 115, + 142, + 508, + 158 + ], + "score": 1.0, + "content": "maliciously secure framework for private deep learning. PoPETs, 2021(1):188–208, Jan. 2021.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 154, + 250, + 168 + ], + "spans": [ + { + "bbox": [ + 115, + 154, + 250, + 168 + ], + "score": 1.0, + "content": "doi: 10.2478/popets-2021-0011.", + "type": "text" + } + ], + "index": 6, + "is_list_end_line": true + } + ], + "index": 3, + "bbox_fs": [ + 96, + 72, + 508, + 168 + ] + }, + { + "type": "title", + "bbox": [ + 88, + 186, + 156, + 199 + ], + "lines": [ + { + "bbox": [ + 104, + 185, + 158, + 200 + ], + "spans": [ + { + "bbox": [ + 104, + 185, + 158, + 200 + ], + "score": 1.0, + "content": "Checklist", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 131, + 207, + 208, + 218 + ], + "lines": [ + { + "bbox": [ + 129, + 205, + 210, + 220 + ], + "spans": [ + { + "bbox": [ + 129, + 205, + 210, + 220 + ], + "score": 1.0, + "content": "1. For all authors...", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 129, + 205, + 210, + 220 + ] + }, + { + "type": "list", + "bbox": [ + 146, + 222, + 505, + 316 + ], + "lines": [ + { + "bbox": [ + 146, + 222, + 504, + 234 + ], + "spans": [ + { + "bbox": [ + 146, + 222, + 504, + 234 + ], + "score": 1.0, + "content": "(a) Do the main claims made in the abstract and introduction accurately reflect the paper’s", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 162, + 233, + 288, + 245 + ], + "spans": [ + { + "bbox": [ + 162, + 233, + 288, + 245 + ], + "score": 1.0, + "content": "contributions and scope? [Yes]", + "type": "text" + } + ], + "index": 10, + "is_list_end_line": true + }, + { + "bbox": [ + 144, + 245, + 436, + 259 + ], + "spans": [ + { + "bbox": [ + 144, + 245, + 436, + 259 + ], + "score": 1.0, + "content": "(b) Did you describe the limitations of your work? [Yes] See Section 7.", + "type": "text" + } + ], + "index": 11, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 145, + 258, + 505, + 271 + ], + "spans": [ + { + "bbox": [ + 145, + 258, + 505, + 271 + ], + "score": 1.0, + "content": "(c) Did you discuss any potential negative societal impacts of your work? [No] We only", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 160, + 269, + 506, + 282 + ], + "spans": [ + { + "bbox": [ + 160, + 269, + 506, + 282 + ], + "score": 1.0, + "content": "work with MNIST, and we believe that the potential negative impacts associated to it", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 162, + 280, + 217, + 293 + ], + "spans": [ + { + "bbox": [ + 162, + 280, + 217, + 293 + ], + "score": 1.0, + "content": "are very low.", + "type": "text" + } + ], + "index": 14, + "is_list_end_line": true + }, + { + "bbox": [ + 144, + 292, + 505, + 307 + ], + "spans": [ + { + "bbox": [ + 144, + 292, + 505, + 307 + ], + "score": 1.0, + "content": "(d) Have you read the ethics review guidelines and ensured that your paper conforms to", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 162, + 304, + 214, + 317 + ], + "spans": [ + { + "bbox": [ + 162, + 304, + 214, + 317 + ], + "score": 1.0, + "content": "them? [Yes]", + "type": "text" + } + ], + "index": 16, + "is_list_end_line": true + } + ], + "index": 12.5, + "bbox_fs": [ + 144, + 222, + 506, + 317 + ] + }, + { + "type": "text", + "bbox": [ + 131, + 319, + 302, + 331 + ], + "lines": [ + { + "bbox": [ + 129, + 318, + 304, + 333 + ], + "spans": [ + { + "bbox": [ + 129, + 318, + 304, + 333 + ], + "score": 1.0, + "content": "2. If you are including theoretical results...", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17, + "bbox_fs": [ + 129, + 318, + 304, + 333 + ] + }, + { + "type": "text", + "bbox": [ + 146, + 334, + 452, + 359 + ], + "lines": [ + { + "bbox": [ + 145, + 334, + 453, + 347 + ], + "spans": [ + { + "bbox": [ + 145, + 334, + 453, + 347 + ], + "score": 1.0, + "content": "(a) Did you state the full set of assumptions of all theoretical results? [N/A]", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 145, + 347, + 424, + 360 + ], + "spans": [ + { + "bbox": [ + 145, + 347, + 424, + 360 + ], + "score": 1.0, + "content": "(b) Did you include complete proofs of all theoretical results? [N/A]", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 145, + 334, + 453, + 360 + ] + }, + { + "type": "text", + "bbox": [ + 130, + 362, + 241, + 374 + ], + "lines": [ + { + "bbox": [ + 128, + 360, + 243, + 376 + ], + "spans": [ + { + "bbox": [ + 128, + 360, + 243, + 376 + ], + "score": 1.0, + "content": "3. If you ran experiments...", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20, + "bbox_fs": [ + 128, + 360, + 243, + 376 + ] + }, + { + "type": "list", + "bbox": [ + 146, + 377, + 505, + 482 + ], + "lines": [ + { + "bbox": [ + 145, + 376, + 506, + 389 + ], + "spans": [ + { + "bbox": [ + 145, + 376, + 506, + 389 + ], + "score": 1.0, + "content": "(a) Did you include the code, data, and instructions needed to reproduce the main experi-", + "type": "text" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 162, + 388, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 162, + 388, + 505, + 399 + ], + "score": 1.0, + "content": "mental results (either in the supplemental material or as a URL)? [Yes] We include all", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 162, + 399, + 495, + 410 + ], + "spans": [ + { + "bbox": [ + 162, + 399, + 495, + 410 + ], + "score": 1.0, + "content": "code in the supplemental material, and the references include the URL for MNIST.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 146, + 411, + 505, + 424 + ], + "spans": [ + { + "bbox": [ + 146, + 411, + 505, + 424 + ], + "score": 1.0, + "content": "(b) Did you specify all the training details (e.g., data splits, hyperparameters, how they", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 161, + 423, + 305, + 434 + ], + "spans": [ + { + "bbox": [ + 161, + 423, + 305, + 434 + ], + "score": 1.0, + "content": "were chosen)? [Yes] See Section 6.", + "type": "text" + } + ], + "index": 25, + "is_list_end_line": true + }, + { + "bbox": [ + 146, + 434, + 507, + 449 + ], + "spans": [ + { + "bbox": [ + 146, + 434, + 507, + 449 + ], + "score": 1.0, + "content": "(c) Did you report error bars (e.g., with respect to the random seed after running experi-", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 162, + 446, + 351, + 459 + ], + "spans": [ + { + "bbox": [ + 162, + 446, + 351, + 459 + ], + "score": 1.0, + "content": "ments multiple times)? [Yes] Only in Figure 2.", + "type": "text" + } + ], + "index": 27, + "is_list_end_line": true + }, + { + "bbox": [ + 146, + 457, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 146, + 457, + 505, + 473 + ], + "score": 1.0, + "content": "(d) Did you include the total amount of compute and the type of resources used (e.g., type", + "type": "text" + } + ], + "index": 28, + "is_list_start_line": true + }, + { + "bbox": [ + 162, + 470, + 429, + 482 + ], + "spans": [ + { + "bbox": [ + 162, + 470, + 429, + 482 + ], + "score": 1.0, + "content": "of GPUs, internal cluster, or cloud provider)? [Yes] See Section 6.", + "type": "text" + } + ], + "index": 29, + "is_list_end_line": true + } + ], + "index": 25, + "bbox_fs": [ + 145, + 376, + 507, + 482 + ] + }, + { + "type": "text", + "bbox": [ + 132, + 485, + 504, + 497 + ], + "lines": [ + { + "bbox": [ + 129, + 483, + 507, + 499 + ], + "spans": [ + { + "bbox": [ + 129, + 483, + 507, + 499 + ], + "score": 1.0, + "content": "4. If you are using existing assets (e.g., code, data, models) or curating/releasing new assets...", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30, + "bbox_fs": [ + 129, + 483, + 507, + 499 + ] + }, + { + "type": "list", + "bbox": [ + 146, + 500, + 505, + 596 + ], + "lines": [ + { + "bbox": [ + 145, + 499, + 483, + 512 + ], + "spans": [ + { + "bbox": [ + 145, + 499, + 483, + 512 + ], + "score": 1.0, + "content": "(a) If your work uses existing assets, did you cite the creators? [Yes] See Section 6.", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 145, + 513, + 506, + 525 + ], + "spans": [ + { + "bbox": [ + 145, + 513, + 506, + 525 + ], + "score": 1.0, + "content": "(b) Did you mention the license of the assets? [Yes] See the references and and Section 5.", + "type": "text" + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 146, + 525, + 505, + 538 + ], + "spans": [ + { + "bbox": [ + 146, + 525, + 505, + 538 + ], + "score": 1.0, + "content": "(c) Did you include any new assets either in the supplemental material or as a URL? [Yes]", + "type": "text" + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 162, + 537, + 355, + 549 + ], + "spans": [ + { + "bbox": [ + 162, + 537, + 355, + 549 + ], + "score": 1.0, + "content": "Our code is included the supplemental material.", + "type": "text" + } + ], + "index": 34, + "is_list_end_line": true + }, + { + "bbox": [ + 146, + 549, + 505, + 563 + ], + "spans": [ + { + "bbox": [ + 146, + 549, + 505, + 563 + ], + "score": 1.0, + "content": "(d) Did you discuss whether and how consent was obtained from people whose data you’re", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 162, + 561, + 254, + 573 + ], + "spans": [ + { + "bbox": [ + 162, + 561, + 254, + 573 + ], + "score": 1.0, + "content": "using/curating? [N/A]", + "type": "text" + } + ], + "index": 36, + "is_list_end_line": true + }, + { + "bbox": [ + 146, + 573, + 505, + 586 + ], + "spans": [ + { + "bbox": [ + 146, + 573, + 505, + 586 + ], + "score": 1.0, + "content": "(e) Did you discuss whether the data you are using/curating contains personally identifiable", + "type": "text" + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 162, + 584, + 325, + 596 + ], + "spans": [ + { + "bbox": [ + 162, + 584, + 325, + 596 + ], + "score": 1.0, + "content": "information or offensive content? [N/A]", + "type": "text" + } + ], + "index": 38, + "is_list_end_line": true + } + ], + "index": 34.5, + "bbox_fs": [ + 145, + 499, + 506, + 596 + ] + }, + { + "type": "text", + "bbox": [ + 131, + 599, + 432, + 611 + ], + "lines": [ + { + "bbox": [ + 128, + 597, + 433, + 613 + ], + "spans": [ + { + "bbox": [ + 128, + 597, + 433, + 613 + ], + "score": 1.0, + "content": "5. If you used crowdsourcing or conducted research with human subjects...", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39, + "bbox_fs": [ + 128, + 597, + 433, + 613 + ] + }, + { + "type": "list", + "bbox": [ + 146, + 614, + 505, + 685 + ], + "lines": [ + { + "bbox": [ + 145, + 613, + 506, + 627 + ], + "spans": [ + { + "bbox": [ + 145, + 613, + 506, + 627 + ], + "score": 1.0, + "content": "(a) Did you include the full text of instructions given to participants and screenshots, if", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 162, + 625, + 237, + 637 + ], + "spans": [ + { + "bbox": [ + 162, + 625, + 237, + 637 + ], + "score": 1.0, + "content": "applicable? [N/A]", + "type": "text" + } + ], + "index": 41, + "is_list_end_line": true + }, + { + "bbox": [ + 145, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 145, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "(b) Did you describe any potential participant risks, with links to Institutional Review", + "type": "text" + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 161, + 649, + 342, + 662 + ], + "spans": [ + { + "bbox": [ + 161, + 649, + 342, + 662 + ], + "score": 1.0, + "content": "Board (IRB) approvals, if applicable? [N/A]", + "type": "text" + } + ], + "index": 43, + "is_list_end_line": true + }, + { + "bbox": [ + 147, + 661, + 505, + 674 + ], + "spans": [ + { + "bbox": [ + 147, + 661, + 505, + 674 + ], + "score": 1.0, + "content": "(c) Did you include the estimated hourly wage paid to participants and the total amount", + "type": "text" + } + ], + "index": 44, + "is_list_start_line": true + }, + { + "bbox": [ + 162, + 673, + 333, + 685 + ], + "spans": [ + { + "bbox": [ + 162, + 673, + 333, + 685 + ], + "score": 1.0, + "content": "spent on participant compensation? [N/A]", + "type": "text" + } + ], + "index": 45, + "is_list_end_line": true + } + ], + "index": 42.5, + "bbox_fs": [ + 145, + 613, + 506, + 685 + ] + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/train/NiM9Q7Z95z/NiM9Q7Z95z_model.json b/parse/train/NiM9Q7Z95z/NiM9Q7Z95z_model.json new file mode 100644 index 0000000000000000000000000000000000000000..7d91131d0b3e7c5e8da0255b3978f09f3f94c5cf --- /dev/null +++ b/parse/train/NiM9Q7Z95z/NiM9Q7Z95z_model.json @@ -0,0 +1,19750 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 257, + 711, + 1307, + 711, + 1307, + 957, + 257, + 957 + ], + "score": 0.965 + }, + { + "category_id": 1, + "poly": [ + 250, + 1486, + 1406, + 1486, + 1406, + 1881, + 250, + 1881 + ], + "score": 0.937 + }, + { + "category_id": 1, + "poly": [ + 253, + 1894, + 1402, + 1894, + 1402, + 1988, + 253, + 1988 + ], + "score": 0.931 + }, + { + "category_id": 0, + "poly": [ + 373, + 270, + 1325, + 270, + 1325, + 328, + 373, + 328 + ], + "score": 0.916 + }, + { + "category_id": 1, + "poly": [ + 255, + 1380, + 1401, + 1380, + 1401, + 1473, + 255, + 1473 + ], + "score": 0.899 + }, + { + "category_id": 1, + "poly": [ + 720, + 444, + 987, + 444, + 987, + 565, + 720, + 565 + ], + "score": 0.896 + }, + { + "category_id": 0, + "poly": [ + 788, + 645, + 913, + 645, + 913, + 681, + 788, + 681 + ], + "score": 0.891 + }, + { + "category_id": 0, + "poly": [ + 261, + 1009, + 531, + 1009, + 531, + 1047, + 261, + 1047 + ], + "score": 0.878 + }, + { + "category_id": 1, + "poly": [ + 253, + 1077, + 1403, + 1077, + 1403, + 1199, + 253, + 1199 + ], + "score": 0.865 + }, + { + "category_id": 1, + "poly": [ + 253, + 1213, + 1404, + 1213, + 1404, + 1366, + 253, + 1366 + ], + "score": 0.848 + }, + { + "category_id": 2, + "poly": [ + 284, + 2033, + 1388, + 2033, + 1388, + 2062, + 284, + 2062 + ], + "score": 0.711 + }, + { + "category_id": 1, + "poly": [ + 284, + 2033, + 1388, + 2033, + 1388, + 2062, + 284, + 2062 + ], + "score": 0.128 + }, + { + "category_id": 13, + "poly": [ + 464, + 1728, + 570, + 1728, + 570, + 1763, + 464, + 1763 + ], + "score": 0.93, + "latex": "\\lfloor x \\cdot 2 ^ { - f } \\rceil" + }, + { + "category_id": 13, + "poly": [ + 722, + 1548, + 776, + 1548, + 776, + 1577, + 722, + 1577 + ], + "score": 0.88, + "latex": "9 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 1348, + 1927, + 1403, + 1927, + 1403, + 1956, + 1348, + 1956 + ], + "score": 0.85, + "latex": "94 \\%" + }, + { + "category_id": 13, + "poly": [ + 1227, + 895, + 1304, + 895, + 1304, + 924, + 1227, + 924 + ], + "score": 0.84, + "latex": "9 8 . 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 999, + 1957, + 1076, + 1957, + 1076, + 1986, + 999, + 1986 + ], + "score": 0.84, + "latex": "9 8 . 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 1355, + 1704, + 1375, + 1704, + 1375, + 1726, + 1355, + 1726 + ], + "score": 0.77, + "latex": "x" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 262.0, + 1329.0, + 262.0, + 1329.0, + 341.0, + 371.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 782.0, + 642.0, + 920.0, + 642.0, + 920.0, + 686.0, + 782.0, + 686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 1006.0, + 535.0, + 1006.0, + 535.0, + 1054.0, + 252.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2029.0, + 1390.0, + 2029.0, + 1390.0, + 2066.0, + 295.0, + 2066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 259.0, + 721.0, + 276.0, + 721.0, + 276.0, + 743.0, + 259.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 711.0, + 1308.0, + 711.0, + 1308.0, + 748.0, + 395.0, + 748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 261.0, + 752.0, + 275.0, + 752.0, + 275.0, + 772.0, + 261.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 744.0, + 1307.0, + 744.0, + 1307.0, + 778.0, + 395.0, + 778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 259.0, + 781.0, + 276.0, + 781.0, + 276.0, + 805.0, + 259.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 774.0, + 1305.0, + 774.0, + 1305.0, + 808.0, + 395.0, + 808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 259.0, + 811.0, + 278.0, + 811.0, + 278.0, + 834.0, + 259.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 803.0, + 1308.0, + 803.0, + 1308.0, + 839.0, + 393.0, + 839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 259.0, + 841.0, + 278.0, + 841.0, + 278.0, + 866.0, + 259.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 834.0, + 1307.0, + 834.0, + 1307.0, + 868.0, + 395.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 259.0, + 871.0, + 278.0, + 871.0, + 278.0, + 896.0, + 259.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 860.0, + 1308.0, + 860.0, + 1308.0, + 901.0, + 393.0, + 901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 259.0, + 902.0, + 276.0, + 902.0, + 276.0, + 925.0, + 259.0, + 925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 895.0, + 1226.0, + 895.0, + 1226.0, + 929.0, + 395.0, + 929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 259.0, + 933.0, + 278.0, + 933.0, + 278.0, + 955.0, + 259.0, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 924.0, + 1175.0, + 924.0, + 1175.0, + 960.0, + 393.0, + 960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1491.0, + 277.0, + 1491.0, + 277.0, + 1519.0, + 248.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1484.0, + 1409.0, + 1484.0, + 1409.0, + 1521.0, + 292.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1522.0, + 277.0, + 1522.0, + 277.0, + 1551.0, + 248.0, + 1551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1515.0, + 1410.0, + 1515.0, + 1410.0, + 1556.0, + 293.0, + 1556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1553.0, + 277.0, + 1553.0, + 277.0, + 1579.0, + 248.0, + 1579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1544.0, + 721.0, + 1544.0, + 721.0, + 1587.0, + 290.0, + 1587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 1544.0, + 1410.0, + 1544.0, + 1410.0, + 1587.0, + 777.0, + 1587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1583.0, + 278.0, + 1583.0, + 278.0, + 1610.0, + 248.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1576.0, + 1405.0, + 1576.0, + 1405.0, + 1611.0, + 293.0, + 1611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1614.0, + 277.0, + 1614.0, + 277.0, + 1640.0, + 247.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1607.0, + 1406.0, + 1607.0, + 1406.0, + 1644.0, + 293.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1645.0, + 278.0, + 1645.0, + 278.0, + 1671.0, + 248.0, + 1671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1638.0, + 1406.0, + 1638.0, + 1406.0, + 1673.0, + 293.0, + 1673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1676.0, + 277.0, + 1676.0, + 277.0, + 1700.0, + 247.0, + 1700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1671.0, + 1405.0, + 1671.0, + 1405.0, + 1703.0, + 294.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1704.0, + 278.0, + 1704.0, + 278.0, + 1731.0, + 248.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1700.0, + 1354.0, + 1700.0, + 1354.0, + 1732.0, + 296.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1376.0, + 1700.0, + 1406.0, + 1700.0, + 1406.0, + 1732.0, + 1376.0, + 1732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1736.0, + 278.0, + 1736.0, + 278.0, + 1762.0, + 248.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1726.0, + 463.0, + 1726.0, + 463.0, + 1769.0, + 290.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 571.0, + 1726.0, + 1406.0, + 1726.0, + 1406.0, + 1769.0, + 571.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1765.0, + 277.0, + 1765.0, + 277.0, + 1792.0, + 247.0, + 1792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1761.0, + 1405.0, + 1761.0, + 1405.0, + 1793.0, + 294.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1796.0, + 278.0, + 1796.0, + 278.0, + 1823.0, + 248.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1789.0, + 1406.0, + 1789.0, + 1406.0, + 1824.0, + 293.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1827.0, + 278.0, + 1827.0, + 278.0, + 1852.0, + 248.0, + 1852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1820.0, + 1406.0, + 1820.0, + 1406.0, + 1856.0, + 293.0, + 1856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1858.0, + 278.0, + 1858.0, + 278.0, + 1883.0, + 248.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1851.0, + 675.0, + 1851.0, + 675.0, + 1886.0, + 293.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1902.0, + 277.0, + 1902.0, + 277.0, + 1926.0, + 249.0, + 1926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1892.0, + 1404.0, + 1892.0, + 1404.0, + 1933.0, + 292.0, + 1933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1936.0, + 277.0, + 1936.0, + 277.0, + 1956.0, + 251.0, + 1956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1926.0, + 1347.0, + 1926.0, + 1347.0, + 1960.0, + 291.0, + 1960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1965.0, + 278.0, + 1965.0, + 278.0, + 1990.0, + 250.0, + 1990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1957.0, + 998.0, + 1957.0, + 998.0, + 1991.0, + 293.0, + 1991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 1957.0, + 1405.0, + 1957.0, + 1405.0, + 1991.0, + 1077.0, + 1991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 253.0, + 1388.0, + 278.0, + 1388.0, + 278.0, + 1411.0, + 253.0, + 1411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1379.0, + 1406.0, + 1379.0, + 1406.0, + 1417.0, + 288.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1418.0, + 279.0, + 1418.0, + 279.0, + 1441.0, + 250.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1408.0, + 1406.0, + 1408.0, + 1406.0, + 1447.0, + 288.0, + 1447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1448.0, + 280.0, + 1448.0, + 280.0, + 1471.0, + 250.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1437.0, + 1081.0, + 1437.0, + 1081.0, + 1480.0, + 288.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 444.0, + 989.0, + 444.0, + 989.0, + 476.0, + 718.0, + 476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 471.0, + 913.0, + 471.0, + 913.0, + 507.0, + 787.0, + 507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 798.0, + 500.0, + 902.0, + 500.0, + 902.0, + 538.0, + 798.0, + 538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 538.0, + 891.0, + 538.0, + 891.0, + 564.0, + 807.0, + 564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1083.0, + 279.0, + 1083.0, + 279.0, + 1110.0, + 250.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1078.0, + 1405.0, + 1078.0, + 1405.0, + 1113.0, + 295.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1116.0, + 276.0, + 1116.0, + 276.0, + 1138.0, + 251.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1106.0, + 1410.0, + 1106.0, + 1410.0, + 1142.0, + 292.0, + 1142.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1144.0, + 279.0, + 1144.0, + 279.0, + 1171.0, + 250.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1137.0, + 1404.0, + 1137.0, + 1404.0, + 1174.0, + 294.0, + 1174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1174.0, + 279.0, + 1174.0, + 279.0, + 1200.0, + 250.0, + 1200.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1169.0, + 894.0, + 1169.0, + 894.0, + 1203.0, + 294.0, + 1203.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1221.0, + 277.0, + 1221.0, + 277.0, + 1242.0, + 251.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1214.0, + 1407.0, + 1214.0, + 1407.0, + 1247.0, + 296.0, + 1247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1250.0, + 277.0, + 1250.0, + 277.0, + 1278.0, + 250.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1242.0, + 1404.0, + 1242.0, + 1404.0, + 1279.0, + 293.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1282.0, + 279.0, + 1282.0, + 279.0, + 1307.0, + 250.0, + 1307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1274.0, + 1406.0, + 1274.0, + 1406.0, + 1309.0, + 293.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1311.0, + 279.0, + 1311.0, + 279.0, + 1338.0, + 250.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1303.0, + 1408.0, + 1303.0, + 1408.0, + 1340.0, + 294.0, + 1340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1342.0, + 279.0, + 1342.0, + 279.0, + 1367.0, + 250.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1334.0, + 1311.0, + 1334.0, + 1311.0, + 1370.0, + 293.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2029.0, + 1390.0, + 2029.0, + 1390.0, + 2066.0, + 295.0, + 2066.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 295, + 1677, + 1405, + 1677, + 1405, + 1953, + 295, + 1953 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 250, + 881, + 1403, + 881, + 1403, + 1095, + 250, + 1095 + ], + "score": 0.964 + }, + { + "category_id": 1, + "poly": [ + 252, + 1442, + 1404, + 1442, + 1404, + 1564, + 252, + 1564 + ], + "score": 0.964 + }, + { + "category_id": 1, + "poly": [ + 250, + 1108, + 1404, + 1108, + 1404, + 1322, + 250, + 1322 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 252, + 713, + 1404, + 713, + 1404, + 866, + 252, + 866 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 252, + 1336, + 1404, + 1336, + 1404, + 1429, + 252, + 1429 + ], + "score": 0.958 + }, + { + "category_id": 4, + "poly": [ + 296, + 533, + 1405, + 533, + 1405, + 656, + 296, + 656 + ], + "score": 0.953 + }, + { + "category_id": 0, + "poly": [ + 254, + 1607, + 1126, + 1607, + 1126, + 1647, + 254, + 1647 + ], + "score": 0.931 + }, + { + "category_id": 3, + "poly": [ + 434, + 206, + 1262, + 206, + 1262, + 499, + 434, + 499 + ], + "score": 0.927 + }, + { + "category_id": 2, + "poly": [ + 841, + 2062, + 859, + 2062, + 859, + 2085, + 841, + 2085 + ], + "score": 0.745 + }, + { + "category_id": 2, + "poly": [ + 334, + 1978, + 917, + 1978, + 917, + 2007, + 334, + 2007 + ], + "score": 0.647 + }, + { + "category_id": 2, + "poly": [ + 334, + 1978, + 917, + 1978, + 917, + 2007, + 334, + 2007 + ], + "score": 0.238 + }, + { + "category_id": 13, + "poly": [ + 680, + 745, + 736, + 745, + 736, + 774, + 680, + 774 + ], + "score": 0.87, + "latex": "40 \\%" + }, + { + "category_id": 13, + "poly": [ + 728, + 1201, + 785, + 1201, + 785, + 1230, + 728, + 1230 + ], + "score": 0.84, + "latex": "\\mathrm { C } { + + }" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 532.0, + 1408.0, + 532.0, + 1408.0, + 567.0, + 293.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 563.0, + 1406.0, + 563.0, + 1406.0, + 598.0, + 293.0, + 598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 595.0, + 1405.0, + 595.0, + 1405.0, + 631.0, + 293.0, + 631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 626.0, + 646.0, + 626.0, + 646.0, + 658.0, + 294.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 243.0, + 1606.0, + 1131.0, + 1606.0, + 1131.0, + 1654.0, + 243.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 371.0, + 526.0, + 371.0, + 526.0, + 380.0, + 516.0, + 380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 382.0, + 510.0, + 382.0, + 510.0, + 390.0, + 498.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 480.0, + 392.0, + 490.0, + 392.0, + 490.0, + 400.0, + 480.0, + 400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1069.0, + 273.0, + 1124.0, + 273.0, + 1124.0, + 304.0, + 1069.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1208.0, + 328.5, + 1266.0, + 328.5, + 1266.0, + 360.5, + 1208.0, + 360.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 423.0, + 1150.0, + 423.0, + 1150.0, + 455.0, + 1086.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2059.0, + 862.0, + 2059.0, + 862.0, + 2094.0, + 838.0, + 2094.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 1972.0, + 921.0, + 1972.0, + 921.0, + 2013.0, + 334.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 1972.0, + 921.0, + 1972.0, + 921.0, + 2013.0, + 334.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1678.0, + 1406.0, + 1678.0, + 1406.0, + 1714.0, + 295.0, + 1714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1708.0, + 1407.0, + 1708.0, + 1407.0, + 1745.0, + 294.0, + 1745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1735.0, + 1407.0, + 1735.0, + 1407.0, + 1777.0, + 292.0, + 1777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1767.0, + 1405.0, + 1767.0, + 1405.0, + 1806.0, + 293.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1799.0, + 1406.0, + 1799.0, + 1406.0, + 1836.0, + 293.0, + 1836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1829.0, + 1405.0, + 1829.0, + 1405.0, + 1866.0, + 293.0, + 1866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1859.0, + 1408.0, + 1859.0, + 1408.0, + 1895.0, + 293.0, + 1895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1888.0, + 1405.0, + 1888.0, + 1405.0, + 1926.0, + 293.0, + 1926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1922.0, + 1171.0, + 1922.0, + 1171.0, + 1955.0, + 297.0, + 1955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 887.0, + 277.0, + 887.0, + 277.0, + 913.0, + 250.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 880.0, + 1406.0, + 880.0, + 1406.0, + 918.0, + 291.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 918.0, + 277.0, + 918.0, + 277.0, + 944.0, + 248.0, + 944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 911.0, + 1403.0, + 911.0, + 1403.0, + 945.0, + 294.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 949.0, + 278.0, + 949.0, + 278.0, + 975.0, + 248.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 943.0, + 1405.0, + 943.0, + 1405.0, + 977.0, + 294.0, + 977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 979.0, + 277.0, + 979.0, + 277.0, + 1004.0, + 248.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 973.0, + 1405.0, + 973.0, + 1405.0, + 1008.0, + 294.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1009.0, + 278.0, + 1009.0, + 278.0, + 1035.0, + 248.0, + 1035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 999.0, + 1406.0, + 999.0, + 1406.0, + 1040.0, + 293.0, + 1040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1040.0, + 277.0, + 1040.0, + 277.0, + 1064.0, + 248.0, + 1064.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1033.0, + 1405.0, + 1033.0, + 1405.0, + 1068.0, + 294.0, + 1068.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1071.0, + 277.0, + 1071.0, + 277.0, + 1096.0, + 248.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1064.0, + 1312.0, + 1064.0, + 1312.0, + 1098.0, + 293.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1448.0, + 279.0, + 1448.0, + 279.0, + 1475.0, + 249.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1442.0, + 1407.0, + 1442.0, + 1407.0, + 1477.0, + 296.0, + 1477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1478.0, + 278.0, + 1478.0, + 278.0, + 1504.0, + 249.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1468.0, + 1406.0, + 1468.0, + 1406.0, + 1510.0, + 292.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1507.0, + 279.0, + 1507.0, + 279.0, + 1535.0, + 249.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1502.0, + 1404.0, + 1502.0, + 1404.0, + 1536.0, + 295.0, + 1536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1539.0, + 279.0, + 1539.0, + 279.0, + 1565.0, + 249.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1534.0, + 656.0, + 1534.0, + 656.0, + 1568.0, + 292.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1114.0, + 277.0, + 1114.0, + 277.0, + 1140.0, + 248.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1107.0, + 1406.0, + 1107.0, + 1406.0, + 1145.0, + 293.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1145.0, + 277.0, + 1145.0, + 277.0, + 1172.0, + 248.0, + 1172.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1140.0, + 1404.0, + 1140.0, + 1404.0, + 1175.0, + 295.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1176.0, + 278.0, + 1176.0, + 278.0, + 1202.0, + 248.0, + 1202.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1170.0, + 1406.0, + 1170.0, + 1406.0, + 1204.0, + 294.0, + 1204.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1206.0, + 277.0, + 1206.0, + 277.0, + 1231.0, + 248.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1197.0, + 727.0, + 1197.0, + 727.0, + 1237.0, + 293.0, + 1237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 1197.0, + 1407.0, + 1197.0, + 1407.0, + 1237.0, + 786.0, + 1237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1236.0, + 278.0, + 1236.0, + 278.0, + 1262.0, + 248.0, + 1262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1228.0, + 1407.0, + 1228.0, + 1407.0, + 1265.0, + 294.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1268.0, + 278.0, + 1268.0, + 278.0, + 1292.0, + 248.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1262.0, + 1406.0, + 1262.0, + 1406.0, + 1296.0, + 293.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1298.0, + 278.0, + 1298.0, + 278.0, + 1323.0, + 248.0, + 1323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1291.0, + 1026.0, + 1291.0, + 1026.0, + 1325.0, + 293.0, + 1325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 720.0, + 279.0, + 720.0, + 279.0, + 746.0, + 249.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 714.0, + 1404.0, + 714.0, + 1404.0, + 747.0, + 296.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 750.0, + 279.0, + 750.0, + 279.0, + 776.0, + 249.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 745.0, + 679.0, + 745.0, + 679.0, + 778.0, + 296.0, + 778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 737.0, + 745.0, + 1404.0, + 745.0, + 1404.0, + 778.0, + 737.0, + 778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 782.0, + 278.0, + 782.0, + 278.0, + 807.0, + 249.0, + 807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 771.0, + 1406.0, + 771.0, + 1406.0, + 812.0, + 292.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 812.0, + 276.0, + 812.0, + 276.0, + 838.0, + 249.0, + 838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 801.0, + 1410.0, + 801.0, + 1410.0, + 842.0, + 292.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 842.0, + 279.0, + 842.0, + 279.0, + 867.0, + 249.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 834.0, + 1073.0, + 834.0, + 1073.0, + 871.0, + 291.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1343.0, + 278.0, + 1343.0, + 278.0, + 1367.0, + 250.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1335.0, + 1406.0, + 1335.0, + 1406.0, + 1373.0, + 291.0, + 1373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1373.0, + 279.0, + 1373.0, + 279.0, + 1397.0, + 250.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1364.0, + 1406.0, + 1364.0, + 1406.0, + 1403.0, + 289.0, + 1403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1404.0, + 279.0, + 1404.0, + 279.0, + 1428.0, + 249.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1397.0, + 1193.0, + 1397.0, + 1193.0, + 1431.0, + 293.0, + 1431.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 240, + 1549, + 1405, + 1549, + 1405, + 2008, + 240, + 2008 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 243, + 1395, + 1404, + 1395, + 1404, + 1519, + 243, + 1519 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 250, + 202, + 1405, + 202, + 1405, + 479, + 250, + 479 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 247, + 1011, + 1405, + 1011, + 1405, + 1291, + 247, + 1291 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 249, + 508, + 1404, + 508, + 1404, + 724, + 249, + 724 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 252, + 754, + 1405, + 754, + 1405, + 851, + 252, + 851 + ], + "score": 0.965 + }, + { + "category_id": 0, + "poly": [ + 244, + 1326, + 876, + 1326, + 876, + 1366, + 244, + 1366 + ], + "score": 0.933 + }, + { + "category_id": 1, + "poly": [ + 252, + 879, + 1154, + 879, + 1154, + 913, + 252, + 913 + ], + "score": 0.922 + }, + { + "category_id": 8, + "poly": [ + 385, + 925, + 1310, + 925, + 1310, + 1001, + 385, + 1001 + ], + "score": 0.9 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.635 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.435 + }, + { + "category_id": 13, + "poly": [ + 1002, + 1733, + 1210, + 1733, + 1210, + 1774, + 1002, + 1774 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { x _ { 0 } = \\sum _ { i = 0 } ^ { k - 1 } x _ { 0 } ^ { ( i ) } 2 ^ { i } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 346, + 1774, + 564, + 1774, + 564, + 1816, + 346, + 1816 + ], + "score": 0.93, + "latex": "x _ { 0 } ^ { ( i ) } = x _ { 0 } ^ { ( i ) } + 0 + 0" + }, + { + "category_id": 13, + "poly": [ + 805, + 1732, + 952, + 1732, + 952, + 1773, + 805, + 1773 + ], + "score": 0.93, + "latex": "x _ { 0 } ^ { ( i ) } \\in \\{ 0 , 1 \\}" + }, + { + "category_id": 13, + "poly": [ + 580, + 1164, + 861, + 1164, + 861, + 1197, + 580, + 1197 + ], + "score": 0.92, + "latex": "z _ { i } ^ { \\prime } = z _ { i } + r _ { i , i + 1 } - r _ { i - 1 , i }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1732, + 412, + 1732, + 412, + 1776, + 298, + 1776 + ], + "score": 0.92, + "latex": "\\{ x _ { 0 } ^ { ( i ) } \\} _ { j = 0 } ^ { k - 1 }" + }, + { + "category_id": 13, + "poly": [ + 1152, + 1193, + 1373, + 1193, + 1373, + 1227, + 1152, + 1227 + ], + "score": 0.92, + "latex": "x y = \\hat { z } _ { 0 } ^ { \\prime } + z _ { 1 } ^ { \\prime } + z _ { 2 } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 926, + 788, + 1124, + 788, + 1124, + 818, + 926, + 818 + ], + "score": 0.92, + "latex": "y = y _ { 0 } + y _ { 1 } + y _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 934, + 1225, + 992, + 1225, + 992, + 1256, + 934, + 1256 + ], + "score": 0.92, + "latex": "P _ { i + 1 }" + }, + { + "category_id": 13, + "poly": [ + 909, + 1135, + 1130, + 1135, + 1130, + 1165, + 909, + 1165 + ], + "score": 0.92, + "latex": "x y = z _ { 0 } + z _ { 1 } + z _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1262, + 234, + 1403, + 234, + 1403, + 267, + 1262, + 267 + ], + "score": 0.92, + "latex": "( x _ { i - 1 } , x _ { i + 1 } )" + }, + { + "category_id": 13, + "poly": [ + 831, + 236, + 1038, + 236, + 1038, + 265, + 831, + 265 + ], + "score": 0.92, + "latex": "x = x _ { 0 } + x _ { 1 } + x _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 296, + 815, + 797, + 815, + 797, + 850, + 296, + 850 + ], + "score": 0.92, + "latex": "x + y = ( x _ { 0 } + y _ { 0 } ) + ( x _ { 1 } + y _ { 1 } ) + ( x _ { 2 } + y _ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 705, + 1813, + 750, + 1813, + 750, + 1855, + 705, + 1855 + ], + "score": 0.91, + "latex": "x _ { 0 } ^ { ( i ) }" + }, + { + "category_id": 13, + "poly": [ + 925, + 883, + 1124, + 883, + 1124, + 912, + 925, + 912 + ], + "score": 0.91, + "latex": "y = y _ { 0 } + y _ { 1 } + y _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 666, + 883, + 873, + 883, + 873, + 911, + 666, + 911 + ], + "score": 0.91, + "latex": "x = x _ { 0 } + x _ { 1 } + x _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 410, + 662, + 467, + 662, + 467, + 692, + 410, + 692 + ], + "score": 0.91, + "latex": "P _ { i - 1 }" + }, + { + "category_id": 13, + "poly": [ + 666, + 788, + 874, + 788, + 874, + 816, + 666, + 816 + ], + "score": 0.91, + "latex": "x = x _ { 0 } + x _ { 1 } + x _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 939, + 1170, + 1006, + 1170, + 1006, + 1197, + 939, + 1197 + ], + "score": 0.91, + "latex": "r _ { i , i + 1 }" + }, + { + "category_id": 13, + "poly": [ + 388, + 1253, + 744, + 1253, + 744, + 1290, + 388, + 1290 + ], + "score": 0.9, + "latex": "\\left( ( x y ) _ { i - 1 } , ( x y ) _ { i + 1 } \\right) = \\left( z _ { i } ^ { \\prime } , z _ { i - 1 } ^ { \\prime } \\right)" + }, + { + "category_id": 13, + "poly": [ + 1160, + 634, + 1354, + 634, + 1354, + 663, + 1160, + 663 + ], + "score": 0.9, + "latex": "x _ { i + 1 } = x - x _ { i - 1 }" + }, + { + "category_id": 13, + "poly": [ + 864, + 1701, + 944, + 1701, + 944, + 1730, + 864, + 1730 + ], + "score": 0.9, + "latex": "x \\in 2 ^ { k }" + }, + { + "category_id": 13, + "poly": [ + 864, + 1195, + 921, + 1195, + 921, + 1225, + 864, + 1225 + ], + "score": 0.9, + "latex": "P _ { i + 1 }" + }, + { + "category_id": 13, + "poly": [ + 1085, + 1705, + 1295, + 1705, + 1295, + 1733, + 1085, + 1733 + ], + "score": 0.89, + "latex": "x = x _ { 0 } + x _ { 1 } + x _ { 2 }" + }, + { + "category_id": 14, + "poly": [ + 385, + 922, + 1312, + 922, + 1312, + 1003, + 385, + 1003 + ], + "score": 0.89, + "latex": "{ \\begin{array} { r l } & { x \\cdot y = ( x _ { 0 } + x _ { 1 } + x _ { 2 } ) \\cdot ( y _ { 0 } + y _ { 1 } + y _ { 2 } ) } \\\\ & { \\qquad = ( x _ { 0 } y _ { 0 } + x _ { 0 } y _ { 1 } + x _ { 1 } y _ { 0 } ) + ( x _ { 1 } y _ { 1 } + x _ { 1 } y _ { 2 } + x _ { 1 } y _ { 1 } ) + ( x _ { 2 } y _ { 2 } + x _ { 2 } y _ { 0 } + x _ { 0 } y _ { 2 } ) . } \\end{array} }" + }, + { + "category_id": 13, + "poly": [ + 939, + 604, + 995, + 604, + 995, + 632, + 939, + 632 + ], + "score": 0.88, + "latex": "x _ { i - 1 }" + }, + { + "category_id": 13, + "poly": [ + 1160, + 234, + 1190, + 234, + 1190, + 264, + 1160, + 264 + ], + "score": 0.88, + "latex": "P _ { i }" + }, + { + "category_id": 13, + "poly": [ + 755, + 1225, + 785, + 1225, + 785, + 1254, + 755, + 1254 + ], + "score": 0.88, + "latex": "P _ { i }" + }, + { + "category_id": 13, + "poly": [ + 876, + 1226, + 902, + 1226, + 902, + 1256, + 876, + 1256 + ], + "score": 0.88, + "latex": "z _ { i } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 1368, + 1134, + 1398, + 1134, + 1398, + 1164, + 1368, + 1164 + ], + "score": 0.88, + "latex": "P _ { i }" + }, + { + "category_id": 13, + "poly": [ + 911, + 632, + 968, + 632, + 968, + 663, + 911, + 663 + ], + "score": 0.87, + "latex": "P _ { i + 1 }" + }, + { + "category_id": 13, + "poly": [ + 433, + 1164, + 464, + 1164, + 464, + 1194, + 433, + 1194 + ], + "score": 0.87, + "latex": "P _ { i }" + }, + { + "category_id": 13, + "poly": [ + 1281, + 663, + 1310, + 663, + 1310, + 691, + 1281, + 691 + ], + "score": 0.87, + "latex": "P _ { i }" + }, + { + "category_id": 13, + "poly": [ + 831, + 632, + 861, + 632, + 861, + 661, + 831, + 661 + ], + "score": 0.87, + "latex": "P _ { i }" + }, + { + "category_id": 13, + "poly": [ + 412, + 601, + 441, + 601, + 441, + 631, + 412, + 631 + ], + "score": 0.87, + "latex": "P _ { i }" + }, + { + "category_id": 13, + "poly": [ + 1049, + 1548, + 1080, + 1548, + 1080, + 1579, + 1049, + 1579 + ], + "score": 0.87, + "latex": "2 ^ { k }" + }, + { + "category_id": 13, + "poly": [ + 1092, + 353, + 1123, + 353, + 1123, + 382, + 1092, + 382 + ], + "score": 0.86, + "latex": "2 ^ { k }" + }, + { + "category_id": 13, + "poly": [ + 780, + 1196, + 810, + 1196, + 810, + 1224, + 780, + 1224 + ], + "score": 0.86, + "latex": "P _ { i }" + }, + { + "category_id": 13, + "poly": [ + 819, + 1852, + 849, + 1852, + 849, + 1883, + 819, + 1883 + ], + "score": 0.86, + "latex": "2 ^ { k }" + }, + { + "category_id": 13, + "poly": [ + 1211, + 1137, + 1237, + 1137, + 1237, + 1164, + 1211, + 1164 + ], + "score": 0.86, + "latex": "z _ { i }" + }, + { + "category_id": 13, + "poly": [ + 543, + 1913, + 573, + 1913, + 573, + 1942, + 543, + 1942 + ], + "score": 0.85, + "latex": "\\bar { 2 } ^ { k }" + }, + { + "category_id": 13, + "poly": [ + 685, + 1741, + 716, + 1741, + 716, + 1769, + 685, + 1769 + ], + "score": 0.85, + "latex": "x _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 623, + 696, + 651, + 696, + 651, + 722, + 623, + 722 + ], + "score": 0.85, + "latex": "x _ { i }" + }, + { + "category_id": 13, + "poly": [ + 522, + 1826, + 554, + 1826, + 554, + 1852, + 522, + 1852 + ], + "score": 0.84, + "latex": "x _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 981, + 633, + 1011, + 633, + 1011, + 661, + 981, + 661 + ], + "score": 0.82, + "latex": "P _ { i }" + }, + { + "category_id": 13, + "poly": [ + 1124, + 1552, + 1143, + 1552, + 1143, + 1578, + 1124, + 1578 + ], + "score": 0.8, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 1164, + 355, + 1184, + 355, + 1184, + 382, + 1164, + 382 + ], + "score": 0.79, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 699, + 605, + 728, + 605, + 728, + 631, + 699, + 631 + ], + "score": 0.79, + "latex": "x _ { i }" + }, + { + "category_id": 13, + "poly": [ + 435, + 239, + 454, + 239, + 454, + 261, + 435, + 261 + ], + "score": 0.75, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 376, + 698, + 395, + 698, + 395, + 719, + 376, + 719 + ], + "score": 0.73, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 670, + 607, + 688, + 607, + 688, + 628, + 670, + 628 + ], + "score": 0.57, + "latex": "x" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1322.0, + 880.0, + 1322.0, + 880.0, + 1376.0, + 234.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2091.0, + 838.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2091.0, + 838.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1557.0, + 277.0, + 1557.0, + 277.0, + 1581.0, + 241.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1547.0, + 1048.0, + 1547.0, + 1048.0, + 1585.0, + 293.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1081.0, + 1547.0, + 1123.0, + 1547.0, + 1123.0, + 1585.0, + 1081.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1144.0, + 1547.0, + 1406.0, + 1547.0, + 1406.0, + 1585.0, + 1144.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1588.0, + 277.0, + 1588.0, + 277.0, + 1613.0, + 240.0, + 1613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1578.0, + 1406.0, + 1578.0, + 1406.0, + 1617.0, + 291.0, + 1617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1617.0, + 278.0, + 1617.0, + 278.0, + 1643.0, + 240.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1612.0, + 1406.0, + 1612.0, + 1406.0, + 1645.0, + 294.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1645.0, + 278.0, + 1645.0, + 278.0, + 1675.0, + 239.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1636.0, + 1408.0, + 1636.0, + 1408.0, + 1682.0, + 290.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1676.0, + 280.0, + 1676.0, + 280.0, + 1706.0, + 239.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1671.0, + 1406.0, + 1671.0, + 1406.0, + 1709.0, + 293.0, + 1709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1707.0, + 280.0, + 1707.0, + 280.0, + 1773.0, + 239.0, + 1773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 1697.0, + 297.0, + 1697.0, + 297.0, + 1793.0, + 284.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 413.0, + 1697.0, + 684.0, + 1697.0, + 684.0, + 1793.0, + 413.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1697.0, + 804.0, + 1697.0, + 804.0, + 1793.0, + 717.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 953.0, + 1697.0, + 1001.0, + 1697.0, + 1001.0, + 1793.0, + 953.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1296.0, + 1697.0, + 1419.0, + 1697.0, + 1419.0, + 1793.0, + 1296.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1785.0, + 278.0, + 1785.0, + 278.0, + 1815.0, + 240.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 1761.0, + 345.0, + 1761.0, + 345.0, + 1838.0, + 287.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 565.0, + 1761.0, + 1416.0, + 1761.0, + 1416.0, + 1838.0, + 565.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1827.0, + 278.0, + 1827.0, + 278.0, + 1852.0, + 240.0, + 1852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 1803.0, + 521.0, + 1803.0, + 521.0, + 1870.0, + 286.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 1803.0, + 704.0, + 1803.0, + 704.0, + 1870.0, + 555.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 1803.0, + 1417.0, + 1803.0, + 1417.0, + 1870.0, + 751.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1860.0, + 277.0, + 1860.0, + 277.0, + 1886.0, + 240.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1851.0, + 818.0, + 1851.0, + 818.0, + 1891.0, + 293.0, + 1891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 850.0, + 1851.0, + 1406.0, + 1851.0, + 1406.0, + 1891.0, + 850.0, + 1891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1890.0, + 278.0, + 1890.0, + 278.0, + 1916.0, + 240.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1882.0, + 1408.0, + 1882.0, + 1408.0, + 1921.0, + 294.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1921.0, + 278.0, + 1921.0, + 278.0, + 1946.0, + 240.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1910.0, + 542.0, + 1910.0, + 542.0, + 1952.0, + 291.0, + 1952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 1910.0, + 1408.0, + 1910.0, + 1408.0, + 1952.0, + 574.0, + 1952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1952.0, + 278.0, + 1952.0, + 278.0, + 1977.0, + 240.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1944.0, + 1406.0, + 1944.0, + 1406.0, + 1981.0, + 293.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1981.0, + 277.0, + 1981.0, + 277.0, + 2007.0, + 241.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1973.0, + 1285.0, + 1973.0, + 1285.0, + 2012.0, + 291.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 242.0, + 1401.0, + 277.0, + 1401.0, + 277.0, + 1428.0, + 242.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1393.0, + 1404.0, + 1393.0, + 1404.0, + 1432.0, + 293.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1432.0, + 278.0, + 1432.0, + 278.0, + 1458.0, + 240.0, + 1458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1423.0, + 1404.0, + 1423.0, + 1404.0, + 1461.0, + 292.0, + 1461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1461.0, + 278.0, + 1461.0, + 278.0, + 1488.0, + 240.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1451.0, + 1405.0, + 1451.0, + 1405.0, + 1493.0, + 292.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1493.0, + 278.0, + 1493.0, + 278.0, + 1519.0, + 240.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1482.0, + 1140.0, + 1482.0, + 1140.0, + 1527.0, + 290.0, + 1527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 208.0, + 277.0, + 208.0, + 277.0, + 234.0, + 248.0, + 234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 200.0, + 1407.0, + 200.0, + 1407.0, + 238.0, + 293.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 240.0, + 278.0, + 240.0, + 278.0, + 266.0, + 248.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 232.0, + 434.0, + 232.0, + 434.0, + 270.0, + 292.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 232.0, + 830.0, + 232.0, + 830.0, + 270.0, + 455.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1039.0, + 232.0, + 1159.0, + 232.0, + 1159.0, + 270.0, + 1039.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 232.0, + 1261.0, + 232.0, + 1261.0, + 270.0, + 1191.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 232.0, + 1407.0, + 232.0, + 1407.0, + 270.0, + 1404.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 270.0, + 277.0, + 270.0, + 277.0, + 295.0, + 248.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 261.0, + 1405.0, + 261.0, + 1405.0, + 300.0, + 293.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 299.0, + 278.0, + 299.0, + 278.0, + 326.0, + 248.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 292.0, + 1408.0, + 292.0, + 1408.0, + 330.0, + 293.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 330.0, + 278.0, + 330.0, + 278.0, + 358.0, + 248.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 323.0, + 1408.0, + 323.0, + 1408.0, + 360.0, + 293.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 358.0, + 280.0, + 358.0, + 280.0, + 388.0, + 246.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 352.0, + 1091.0, + 352.0, + 1091.0, + 389.0, + 293.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 352.0, + 1163.0, + 352.0, + 1163.0, + 389.0, + 1124.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 352.0, + 1405.0, + 352.0, + 1405.0, + 389.0, + 1185.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 390.0, + 278.0, + 390.0, + 278.0, + 416.0, + 247.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 385.0, + 1405.0, + 385.0, + 1405.0, + 418.0, + 295.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 422.0, + 277.0, + 422.0, + 277.0, + 448.0, + 247.0, + 448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 415.0, + 1405.0, + 415.0, + 1405.0, + 450.0, + 291.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 450.0, + 280.0, + 450.0, + 280.0, + 482.0, + 246.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 450.0, + 371.0, + 450.0, + 371.0, + 483.0, + 291.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1018.0, + 278.0, + 1018.0, + 278.0, + 1043.0, + 248.0, + 1043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1010.0, + 1405.0, + 1010.0, + 1405.0, + 1045.0, + 293.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1047.0, + 278.0, + 1047.0, + 278.0, + 1073.0, + 248.0, + 1073.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1039.0, + 1405.0, + 1039.0, + 1405.0, + 1078.0, + 293.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1078.0, + 278.0, + 1078.0, + 278.0, + 1105.0, + 248.0, + 1105.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1073.0, + 1405.0, + 1073.0, + 1405.0, + 1106.0, + 294.0, + 1106.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1107.0, + 278.0, + 1107.0, + 278.0, + 1135.0, + 248.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1101.0, + 1405.0, + 1101.0, + 1405.0, + 1136.0, + 293.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1138.0, + 278.0, + 1138.0, + 278.0, + 1166.0, + 248.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1131.0, + 908.0, + 1131.0, + 908.0, + 1168.0, + 293.0, + 1168.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 1131.0, + 1210.0, + 1131.0, + 1210.0, + 1168.0, + 1131.0, + 1168.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 1131.0, + 1367.0, + 1131.0, + 1367.0, + 1168.0, + 1238.0, + 1168.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 1131.0, + 1408.0, + 1131.0, + 1408.0, + 1168.0, + 1399.0, + 1168.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 245.0, + 1168.0, + 280.0, + 1168.0, + 280.0, + 1199.0, + 245.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1162.0, + 432.0, + 1162.0, + 432.0, + 1202.0, + 293.0, + 1202.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 465.0, + 1162.0, + 579.0, + 1162.0, + 579.0, + 1202.0, + 465.0, + 1202.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 862.0, + 1162.0, + 938.0, + 1162.0, + 938.0, + 1202.0, + 862.0, + 1202.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1007.0, + 1162.0, + 1408.0, + 1162.0, + 1408.0, + 1202.0, + 1007.0, + 1202.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1199.0, + 278.0, + 1199.0, + 278.0, + 1228.0, + 240.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1191.0, + 779.0, + 1191.0, + 779.0, + 1231.0, + 293.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 811.0, + 1191.0, + 863.0, + 1191.0, + 863.0, + 1231.0, + 811.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 922.0, + 1191.0, + 1151.0, + 1191.0, + 1151.0, + 1231.0, + 922.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1374.0, + 1191.0, + 1408.0, + 1191.0, + 1408.0, + 1231.0, + 1374.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1228.0, + 277.0, + 1228.0, + 277.0, + 1257.0, + 240.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1222.0, + 754.0, + 1222.0, + 754.0, + 1259.0, + 293.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 1222.0, + 875.0, + 1222.0, + 875.0, + 1259.0, + 786.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 903.0, + 1222.0, + 933.0, + 1222.0, + 933.0, + 1259.0, + 903.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 993.0, + 1222.0, + 1408.0, + 1222.0, + 1408.0, + 1259.0, + 993.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 243.0, + 1261.0, + 277.0, + 1261.0, + 277.0, + 1286.0, + 243.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1254.0, + 387.0, + 1254.0, + 387.0, + 1291.0, + 293.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1254.0, + 755.0, + 1254.0, + 755.0, + 1291.0, + 745.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 516.0, + 277.0, + 516.0, + 277.0, + 542.0, + 249.0, + 542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 509.0, + 1406.0, + 509.0, + 1406.0, + 547.0, + 293.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 546.0, + 279.0, + 546.0, + 279.0, + 572.0, + 249.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 541.0, + 1407.0, + 541.0, + 1407.0, + 575.0, + 294.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 574.0, + 280.0, + 574.0, + 280.0, + 602.0, + 246.0, + 602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 569.0, + 1407.0, + 569.0, + 1407.0, + 604.0, + 293.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 607.0, + 277.0, + 607.0, + 277.0, + 632.0, + 249.0, + 632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 601.0, + 411.0, + 601.0, + 411.0, + 636.0, + 293.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 442.0, + 601.0, + 669.0, + 601.0, + 669.0, + 636.0, + 442.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 689.0, + 601.0, + 698.0, + 601.0, + 698.0, + 636.0, + 689.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 729.0, + 601.0, + 938.0, + 601.0, + 938.0, + 636.0, + 729.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 601.0, + 1406.0, + 601.0, + 1406.0, + 636.0, + 996.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 637.0, + 277.0, + 637.0, + 277.0, + 663.0, + 249.0, + 663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 630.0, + 830.0, + 630.0, + 830.0, + 668.0, + 293.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 862.0, + 630.0, + 910.0, + 630.0, + 910.0, + 668.0, + 862.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 969.0, + 630.0, + 980.0, + 630.0, + 980.0, + 668.0, + 969.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 630.0, + 1159.0, + 630.0, + 1159.0, + 668.0, + 1012.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1355.0, + 630.0, + 1407.0, + 630.0, + 1407.0, + 668.0, + 1355.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 668.0, + 277.0, + 668.0, + 277.0, + 693.0, + 249.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 660.0, + 409.0, + 660.0, + 409.0, + 695.0, + 293.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 468.0, + 660.0, + 1280.0, + 660.0, + 1280.0, + 695.0, + 468.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1311.0, + 660.0, + 1404.0, + 660.0, + 1404.0, + 695.0, + 1311.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 697.0, + 277.0, + 697.0, + 277.0, + 723.0, + 249.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 691.0, + 375.0, + 691.0, + 375.0, + 725.0, + 293.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 691.0, + 622.0, + 691.0, + 622.0, + 725.0, + 396.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 691.0, + 759.0, + 691.0, + 759.0, + 725.0, + 652.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 761.0, + 279.0, + 761.0, + 279.0, + 784.0, + 249.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 753.0, + 1407.0, + 753.0, + 1407.0, + 788.0, + 292.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 791.0, + 280.0, + 791.0, + 280.0, + 816.0, + 250.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 784.0, + 665.0, + 784.0, + 665.0, + 821.0, + 289.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 784.0, + 925.0, + 784.0, + 925.0, + 821.0, + 875.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.0, + 784.0, + 1403.0, + 784.0, + 1403.0, + 821.0, + 1125.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 245.0, + 816.0, + 295.0, + 816.0, + 295.0, + 851.0, + 245.0, + 851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 798.0, + 816.0, + 810.0, + 816.0, + 810.0, + 851.0, + 798.0, + 851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 874.0, + 665.0, + 874.0, + 665.0, + 919.0, + 241.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 874.0, + 874.0, + 924.0, + 874.0, + 924.0, + 919.0, + 874.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.0, + 874.0, + 1157.0, + 874.0, + 1157.0, + 919.0, + 1125.0, + 919.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 239, + 1671, + 1404, + 1671, + 1404, + 2008, + 239, + 2008 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 240, + 1093, + 1405, + 1093, + 1405, + 1320, + 240, + 1320 + ], + "score": 0.967 + }, + { + "category_id": 8, + "poly": [ + 354, + 1332, + 1343, + 1332, + 1343, + 1558, + 354, + 1558 + ], + "score": 0.966 + }, + { + "category_id": 1, + "poly": [ + 242, + 202, + 1405, + 202, + 1405, + 419, + 242, + 419 + ], + "score": 0.964 + }, + { + "category_id": 1, + "poly": [ + 243, + 966, + 1406, + 966, + 1406, + 1060, + 243, + 1060 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 393, + 879, + 1400, + 879, + 1400, + 943, + 393, + 943 + ], + "score": 0.952 + }, + { + "category_id": 1, + "poly": [ + 254, + 1574, + 1402, + 1574, + 1402, + 1638, + 254, + 1638 + ], + "score": 0.949 + }, + { + "category_id": 8, + "poly": [ + 571, + 793, + 1220, + 793, + 1220, + 868, + 571, + 868 + ], + "score": 0.934 + }, + { + "category_id": 2, + "poly": [ + 840, + 2061, + 858, + 2061, + 858, + 2084, + 840, + 2084 + ], + "score": 0.747 + }, + { + "category_id": 1, + "poly": [ + 372, + 579, + 1404, + 579, + 1404, + 792, + 372, + 792 + ], + "score": 0.639 + }, + { + "category_id": 1, + "poly": [ + 373, + 443, + 1404, + 443, + 1404, + 569, + 373, + 569 + ], + "score": 0.599 + }, + { + "category_id": 2, + "poly": [ + 240, + 448, + 281, + 448, + 281, + 796, + 240, + 796 + ], + "score": 0.277 + }, + { + "category_id": 14, + "poly": [ + 354, + 1332, + 1343, + 1332, + 1343, + 1561, + 354, + 1561 + ], + "score": 0.94, + "latex": "\\begin{array} { r l } & { \\displaystyle \\sum _ { i } x ^ { ( i ) } \\cdot y ^ { ( i ) } = \\sum _ { i } ( x _ { 0 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } + x _ { 2 } ^ { ( i ) } ) \\cdot ( y _ { 0 } ^ { ( i ) } + y _ { 1 } ^ { ( i ) } + y _ { 2 } ^ { ( i ) } ) } \\\\ & { \\quad \\quad \\quad \\quad = \\displaystyle \\sum _ { i } ( x _ { 0 } ^ { ( i ) } y _ { 0 } ^ { ( i ) } + x _ { 0 } ^ { ( i ) } y _ { 1 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } y _ { 0 } ^ { ( i ) } ) + \\sum _ { i } ( x _ { 1 } ^ { ( i ) } y _ { 1 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } y _ { 2 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } y _ { 1 } ^ { ( i ) } ) } \\\\ & { \\quad \\quad \\quad \\quad + \\displaystyle \\sum _ { i } ( x _ { 2 } ^ { ( i ) } y _ { 2 } ^ { ( i ) } + x _ { 2 } ^ { ( i ) } y _ { 0 } ^ { ( i ) } + x _ { 0 } ^ { ( i ) } y _ { 2 } ^ { ( i ) } ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 536, + 1277, + 840, + 1277, + 840, + 1319, + 536, + 1319 + ], + "score": 0.94, + "latex": "\\{ x ^ { ( i ) } \\} = x _ { 0 } ^ { ( i ) } + x _ { 1 } ^ { ( i ) } + x _ { 2 } ^ { ( i ) }" + }, + { + "category_id": 13, + "poly": [ + 457, + 382, + 776, + 382, + 776, + 418, + 457, + 418 + ], + "score": 0.93, + "latex": "( x \\cdot 2 ^ { f } ) \\cdot ( y \\cdot \\dot { 2 } ^ { f } ) = x y \\cdot 2 ^ { 2 f }" + }, + { + "category_id": 13, + "poly": [ + 896, + 293, + 1039, + 293, + 1039, + 328, + 896, + 328 + ], + "score": 0.93, + "latex": "\\bar { \\boldsymbol { x } } = \\lfloor \\boldsymbol { x } \\cdot \\mathbf { \\bar { 2 } } ^ { f } \\rceil" + }, + { + "category_id": 13, + "poly": [ + 757, + 731, + 914, + 731, + 914, + 766, + 757, + 766 + ], + "score": 0.92, + "latex": "\\left\\lfloor ( x + r ) / 2 ^ { m } \\right\\rfloor" + }, + { + "category_id": 14, + "poly": [ + 571, + 789, + 1225, + 789, + 1225, + 867, + 571, + 867 + ], + "score": 0.92, + "latex": "\\lfloor ( x + r ) / 2 ^ { m } \\rfloor = \\left\\{ \\begin{array} { l l } { \\lfloor x / 2 ^ { m } \\rfloor } & { ( x \\bmod 2 ^ { m } + r ) < 2 ^ { m } } \\\\ { \\lfloor x / 2 ^ { m } \\rfloor + 1 } & { ( x \\bmod 2 ^ { m } + r ) \\ge 2 ^ { m } . } \\end{array} \\right." + }, + { + "category_id": 13, + "poly": [ + 1190, + 443, + 1250, + 443, + 1250, + 474, + 1190, + 474 + ], + "score": 0.91, + "latex": "2 ^ { f - 1 }" + }, + { + "category_id": 13, + "poly": [ + 1036, + 1281, + 1078, + 1281, + 1078, + 1317, + 1036, + 1317 + ], + "score": 0.91, + "latex": "\\boldsymbol y ^ { ( i ) }" + }, + { + "category_id": 13, + "poly": [ + 982, + 447, + 1002, + 447, + 1002, + 478, + 982, + 478 + ], + "score": 0.85, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 1118, + 296, + 1138, + 296, + 1138, + 327, + 1118, + 327 + ], + "score": 0.85, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 942, + 1247, + 963, + 1247, + 963, + 1279, + 942, + 1279 + ], + "score": 0.85, + "latex": "\\vec { y }" + }, + { + "category_id": 13, + "poly": [ + 871, + 1248, + 891, + 1248, + 891, + 1274, + 871, + 1274 + ], + "score": 0.81, + "latex": "\\vec { x }" + }, + { + "category_id": 13, + "poly": [ + 1067, + 736, + 1094, + 736, + 1094, + 759, + 1067, + 759 + ], + "score": 0.79, + "latex": "m" + }, + { + "category_id": 13, + "poly": [ + 684, + 300, + 704, + 300, + 704, + 322, + 684, + 322 + ], + "score": 0.76, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 1201, + 738, + 1218, + 738, + 1218, + 759, + 1201, + 759 + ], + "score": 0.74, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 644, + 879, + 772, + 879, + 772, + 911, + 644, + 911 + ], + "score": 0.44, + "latex": "( x \\bmod 2 ^ { m } )" + }, + { + "category_id": 13, + "poly": [ + 732, + 880, + 769, + 880, + 769, + 908, + 732, + 908 + ], + "score": 0.39, + "latex": "2 ^ { m }" + }, + { + "category_id": 13, + "poly": [ + 648, + 887, + 666, + 887, + 666, + 907, + 648, + 907 + ], + "score": 0.35, + "latex": "x" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 863.0, + 2058.0, + 863.0, + 2091.0, + 838.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 451.0, + 279.0, + 451.0, + 279.0, + 481.0, + 238.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 481.0, + 278.0, + 481.0, + 278.0, + 513.0, + 238.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 513.0, + 277.0, + 513.0, + 277.0, + 541.0, + 240.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 543.0, + 279.0, + 543.0, + 279.0, + 576.0, + 238.0, + 576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 586.0, + 279.0, + 586.0, + 279.0, + 618.0, + 238.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 615.0, + 279.0, + 615.0, + 279.0, + 648.0, + 238.0, + 648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 646.0, + 279.0, + 646.0, + 279.0, + 678.0, + 238.0, + 678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 677.0, + 279.0, + 677.0, + 279.0, + 708.0, + 238.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 707.0, + 279.0, + 707.0, + 279.0, + 738.0, + 238.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 737.0, + 279.0, + 737.0, + 279.0, + 769.0, + 238.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 768.0, + 279.0, + 768.0, + 279.0, + 799.0, + 238.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1678.0, + 277.0, + 1678.0, + 277.0, + 1704.0, + 240.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1673.0, + 1405.0, + 1673.0, + 1405.0, + 1707.0, + 296.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1708.0, + 277.0, + 1708.0, + 277.0, + 1734.0, + 240.0, + 1734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1700.0, + 1407.0, + 1700.0, + 1407.0, + 1738.0, + 292.0, + 1738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1740.0, + 277.0, + 1740.0, + 277.0, + 1766.0, + 240.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1730.0, + 1408.0, + 1730.0, + 1408.0, + 1771.0, + 292.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1770.0, + 277.0, + 1770.0, + 277.0, + 1796.0, + 240.0, + 1796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1760.0, + 1408.0, + 1760.0, + 1408.0, + 1801.0, + 292.0, + 1801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1800.0, + 277.0, + 1800.0, + 277.0, + 1826.0, + 240.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1790.0, + 1407.0, + 1790.0, + 1407.0, + 1831.0, + 292.0, + 1831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1830.0, + 277.0, + 1830.0, + 277.0, + 1856.0, + 240.0, + 1856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1824.0, + 1405.0, + 1824.0, + 1405.0, + 1859.0, + 294.0, + 1859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1861.0, + 277.0, + 1861.0, + 277.0, + 1887.0, + 240.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1853.0, + 1407.0, + 1853.0, + 1407.0, + 1890.0, + 292.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1891.0, + 276.0, + 1891.0, + 276.0, + 1917.0, + 240.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1886.0, + 1405.0, + 1886.0, + 1405.0, + 1920.0, + 294.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1921.0, + 277.0, + 1921.0, + 277.0, + 1947.0, + 240.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1910.0, + 1408.0, + 1910.0, + 1408.0, + 1956.0, + 291.0, + 1956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1951.0, + 277.0, + 1951.0, + 277.0, + 1977.0, + 240.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1946.0, + 1404.0, + 1946.0, + 1404.0, + 1980.0, + 294.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1982.0, + 277.0, + 1982.0, + 277.0, + 2007.0, + 240.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1975.0, + 1388.0, + 1975.0, + 1388.0, + 2013.0, + 292.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1102.0, + 278.0, + 1102.0, + 278.0, + 1126.0, + 241.0, + 1126.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1093.0, + 1406.0, + 1093.0, + 1406.0, + 1130.0, + 291.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1130.0, + 280.0, + 1130.0, + 280.0, + 1158.0, + 239.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1121.0, + 1406.0, + 1121.0, + 1406.0, + 1163.0, + 292.0, + 1163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1162.0, + 277.0, + 1162.0, + 277.0, + 1186.0, + 240.0, + 1186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1157.0, + 1404.0, + 1157.0, + 1404.0, + 1189.0, + 295.0, + 1189.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1194.0, + 278.0, + 1194.0, + 278.0, + 1218.0, + 240.0, + 1218.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1187.0, + 1405.0, + 1187.0, + 1405.0, + 1219.0, + 295.0, + 1219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1221.0, + 280.0, + 1221.0, + 280.0, + 1249.0, + 239.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1215.0, + 1406.0, + 1215.0, + 1406.0, + 1250.0, + 291.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1251.0, + 280.0, + 1251.0, + 280.0, + 1279.0, + 239.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1247.0, + 870.0, + 1247.0, + 870.0, + 1279.0, + 295.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 1247.0, + 941.0, + 1247.0, + 941.0, + 1279.0, + 892.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 1247.0, + 1405.0, + 1247.0, + 1405.0, + 1279.0, + 964.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1291.0, + 277.0, + 1291.0, + 277.0, + 1315.0, + 240.0, + 1315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1273.0, + 535.0, + 1273.0, + 535.0, + 1327.0, + 288.0, + 1327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 841.0, + 1275.0, + 1035.0, + 1275.0, + 1035.0, + 1324.0, + 841.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1079.0, + 1275.0, + 1380.0, + 1275.0, + 1380.0, + 1324.0, + 1079.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.75, + 1285.0, + 754.75, + 1285.0, + 754.75, + 1320.5, + 685.75, + 1320.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 208.0, + 279.0, + 208.0, + 279.0, + 236.0, + 239.0, + 236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 204.0, + 1405.0, + 204.0, + 1405.0, + 239.0, + 296.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 239.0, + 279.0, + 239.0, + 279.0, + 267.0, + 239.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 234.0, + 1405.0, + 234.0, + 1405.0, + 268.0, + 295.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 268.0, + 279.0, + 268.0, + 279.0, + 296.0, + 239.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 265.0, + 1406.0, + 265.0, + 1406.0, + 299.0, + 295.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 302.0, + 279.0, + 302.0, + 279.0, + 325.0, + 241.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 288.0, + 683.0, + 288.0, + 683.0, + 334.0, + 291.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 288.0, + 895.0, + 288.0, + 895.0, + 334.0, + 705.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1040.0, + 288.0, + 1117.0, + 288.0, + 1117.0, + 334.0, + 1040.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1139.0, + 288.0, + 1408.0, + 288.0, + 1408.0, + 334.0, + 1139.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 329.0, + 279.0, + 329.0, + 279.0, + 358.0, + 239.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 323.0, + 1405.0, + 323.0, + 1405.0, + 360.0, + 293.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 360.0, + 279.0, + 360.0, + 279.0, + 387.0, + 239.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 354.0, + 1408.0, + 354.0, + 1408.0, + 390.0, + 291.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 390.0, + 279.0, + 390.0, + 279.0, + 417.0, + 241.0, + 417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 382.0, + 456.0, + 382.0, + 456.0, + 421.0, + 292.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 382.0, + 1175.0, + 382.0, + 1175.0, + 421.0, + 777.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 973.0, + 279.0, + 973.0, + 279.0, + 1000.0, + 240.0, + 1000.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 966.0, + 1407.0, + 966.0, + 1407.0, + 1003.0, + 293.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 242.0, + 1006.0, + 277.0, + 1006.0, + 277.0, + 1028.0, + 242.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 997.0, + 1405.0, + 997.0, + 1405.0, + 1034.0, + 290.0, + 1034.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1034.0, + 280.0, + 1034.0, + 280.0, + 1060.0, + 240.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1030.0, + 636.0, + 1030.0, + 636.0, + 1060.0, + 292.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 878.0, + 643.0, + 878.0, + 643.0, + 914.0, + 396.0, + 914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 773.0, + 878.0, + 1404.0, + 878.0, + 1404.0, + 914.0, + 773.0, + 914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 910.0, + 1104.0, + 910.0, + 1104.0, + 946.0, + 395.0, + 946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1582.0, + 282.0, + 1582.0, + 282.0, + 1606.0, + 248.0, + 1606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1573.0, + 1403.0, + 1573.0, + 1403.0, + 1611.0, + 289.0, + 1611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 242.0, + 1606.0, + 727.0, + 1606.0, + 727.0, + 1641.0, + 242.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 578.0, + 1406.0, + 578.0, + 1406.0, + 616.0, + 369.0, + 616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 610.0, + 1402.0, + 610.0, + 1402.0, + 644.0, + 394.0, + 644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 639.0, + 1404.0, + 639.0, + 1404.0, + 676.0, + 394.0, + 676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 672.0, + 1404.0, + 672.0, + 1404.0, + 706.0, + 395.0, + 706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 700.0, + 1404.0, + 700.0, + 1404.0, + 734.0, + 395.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 728.0, + 756.0, + 728.0, + 756.0, + 770.0, + 392.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 728.0, + 1066.0, + 728.0, + 1066.0, + 770.0, + 915.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 728.0, + 1200.0, + 728.0, + 1200.0, + 770.0, + 1095.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 728.0, + 1408.0, + 728.0, + 1408.0, + 770.0, + 1219.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 759.0, + 450.0, + 759.0, + 450.0, + 797.0, + 393.0, + 797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 439.0, + 981.0, + 439.0, + 981.0, + 482.0, + 370.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1003.0, + 439.0, + 1189.0, + 439.0, + 1189.0, + 482.0, + 1003.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 439.0, + 1407.0, + 439.0, + 1407.0, + 482.0, + 1251.0, + 482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 477.0, + 1407.0, + 477.0, + 1407.0, + 510.0, + 394.0, + 510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 506.0, + 1405.0, + 506.0, + 1405.0, + 541.0, + 394.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 538.0, + 1141.0, + 538.0, + 1141.0, + 571.0, + 396.0, + 571.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 241, + 448, + 1405, + 448, + 1405, + 724, + 241, + 724 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 242, + 202, + 1404, + 202, + 1404, + 417, + 242, + 417 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 240, + 754, + 1404, + 754, + 1404, + 940, + 240, + 940 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 293, + 1397, + 1404, + 1397, + 1404, + 1585, + 293, + 1585 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 242, + 968, + 1406, + 968, + 1406, + 1100, + 242, + 1100 + ], + "score": 0.969 + }, + { + "category_id": 1, + "poly": [ + 297, + 1914, + 1402, + 1914, + 1402, + 2007, + 297, + 2007 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 293, + 1616, + 1403, + 1616, + 1403, + 1711, + 293, + 1711 + ], + "score": 0.967 + }, + { + "category_id": 1, + "poly": [ + 279, + 1272, + 1405, + 1272, + 1405, + 1367, + 279, + 1367 + ], + "score": 0.961 + }, + { + "category_id": 8, + "poly": [ + 542, + 1116, + 1154, + 1116, + 1154, + 1194, + 542, + 1194 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 299, + 1819, + 1401, + 1819, + 1401, + 1883, + 299, + 1883 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 257, + 1208, + 1293, + 1208, + 1293, + 1242, + 257, + 1242 + ], + "score": 0.932 + }, + { + "category_id": 0, + "poly": [ + 281, + 1751, + 846, + 1751, + 846, + 1790, + 281, + 1790 + ], + "score": 0.921 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 858, + 2061, + 858, + 2085, + 841, + 2085 + ], + "score": 0.716 + }, + { + "category_id": 13, + "poly": [ + 611, + 1064, + 777, + 1064, + 777, + 1104, + 611, + 1104 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { b = \\sum _ { i \\geq 0 } b _ { i } 2 ^ { i } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 433, + 263, + 548, + 263, + 548, + 297, + 433, + 297 + ], + "score": 0.94, + "latex": "b \\in \\{ 0 , 1 \\}" + }, + { + "category_id": 13, + "poly": [ + 980, + 1066, + 1104, + 1066, + 1104, + 1099, + 980, + 1099 + ], + "score": 0.93, + "latex": "b _ { i } \\in \\{ 0 , 1 \\}" + }, + { + "category_id": 13, + "poly": [ + 807, + 264, + 976, + 264, + 976, + 296, + 807, + 296 + ], + "score": 0.93, + "latex": "x + b \\cdot ( y - x )" + }, + { + "category_id": 13, + "poly": [ + 1290, + 1459, + 1397, + 1459, + 1397, + 1495, + 1290, + 1495 + ], + "score": 0.93, + "latex": "\\lfloor x \\cdot 2 ^ { - f } \\rceil" + }, + { + "category_id": 13, + "poly": [ + 1205, + 785, + 1324, + 785, + 1324, + 818, + 1205, + 818 + ], + "score": 0.92, + "latex": "y = a \\cdot 2 ^ { b }" + }, + { + "category_id": 13, + "poly": [ + 618, + 969, + 779, + 969, + 779, + 1000, + 618, + 1000 + ], + "score": 0.92, + "latex": "x ^ { y } = 2 ^ { y \\log _ { 2 } x }" + }, + { + "category_id": 13, + "poly": [ + 1241, + 1398, + 1362, + 1398, + 1362, + 1431, + 1241, + 1431 + ], + "score": 0.92, + "latex": "[ x , x + 2 ^ { e } ]" + }, + { + "category_id": 13, + "poly": [ + 1187, + 1429, + 1272, + 1429, + 1272, + 1461, + 1187, + 1461 + ], + "score": 0.92, + "latex": "( f + { \\dot { e } } )" + }, + { + "category_id": 14, + "poly": [ + 545, + 1115, + 1153, + 1115, + 1153, + 1196, + 545, + 1196 + ], + "score": 0.91, + "latex": "2 ^ { b } = 2 ^ { \\sum _ { i \\geq 0 } b _ { i } 2 ^ { i } } = \\prod _ { i \\geq 0 } 2 ^ { b _ { i } 2 ^ { i } } = \\prod _ { i \\geq 0 } ( 1 + b _ { i } \\cdot ( 2 ^ { 2 ^ { i } } - 1 ) ) ." + }, + { + "category_id": 13, + "poly": [ + 880, + 818, + 1101, + 818, + 1101, + 849, + 880, + 849 + ], + "score": 0.91, + "latex": "\\log _ { 2 } y = \\log _ { 2 } a + b" + }, + { + "category_id": 13, + "poly": [ + 513, + 1648, + 598, + 1648, + 598, + 1679, + 513, + 1679 + ], + "score": 0.9, + "latex": "f = 3 2" + }, + { + "category_id": 13, + "poly": [ + 1031, + 1001, + 1240, + 1001, + 1240, + 1033, + 1031, + 1033 + ], + "score": 0.9, + "latex": "2 ^ { a } = 2 ^ { \\lfloor a \\rfloor } \\cdot 2 ^ { a - \\lfloor a \\rfloor }" + }, + { + "category_id": 13, + "poly": [ + 479, + 819, + 547, + 819, + 547, + 845, + 479, + 845 + ], + "score": 0.89, + "latex": "b \\in \\mathbb { Z }" + }, + { + "category_id": 13, + "poly": [ + 297, + 818, + 427, + 818, + 427, + 849, + 297, + 849 + ], + "score": 0.88, + "latex": "a \\in [ 0 . 5 , 1 )" + }, + { + "category_id": 13, + "poly": [ + 506, + 1680, + 539, + 1680, + 539, + 1710, + 506, + 1710 + ], + "score": 0.86, + "latex": "4 f" + }, + { + "category_id": 13, + "poly": [ + 1012, + 1645, + 1063, + 1645, + 1063, + 1675, + 1012, + 1675 + ], + "score": 0.86, + "latex": "2 ^ { 1 2 8 }" + }, + { + "category_id": 13, + "poly": [ + 373, + 1462, + 394, + 1462, + 394, + 1494, + 373, + 1494 + ], + "score": 0.85, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 368, + 788, + 634, + 788, + 634, + 818, + 368, + 818 + ], + "score": 0.85, + "latex": "\\log _ { x } y = \\log _ { 2 } y \\cdot \\log _ { x } 2" + }, + { + "category_id": 13, + "poly": [ + 933, + 300, + 951, + 300, + 951, + 326, + 933, + 326 + ], + "score": 0.81, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 861, + 301, + 881, + 301, + 881, + 322, + 861, + 322 + ], + "score": 0.79, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 1129, + 269, + 1149, + 269, + 1149, + 295, + 1129, + 295 + ], + "score": 0.78, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 354, + 850, + 426, + 850, + 426, + 880, + 354, + 880 + ], + "score": 0.77, + "latex": "\\log _ { 2 } a" + }, + { + "category_id": 13, + "poly": [ + 538, + 1498, + 557, + 1498, + 557, + 1520, + 538, + 1520 + ], + "score": 0.76, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 1076, + 269, + 1096, + 269, + 1096, + 291, + 1076, + 291 + ], + "score": 0.75, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 1232, + 1467, + 1253, + 1467, + 1253, + 1490, + 1232, + 1490 + ], + "score": 0.75, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 1277, + 296, + 1293, + 296, + 1293, + 322, + 1277, + 322 + ], + "score": 0.67, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 1310, + 265, + 1325, + 265, + 1325, + 291, + 1310, + 291 + ], + "score": 0.55, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 620, + 1435, + 636, + 1435, + 636, + 1456, + 620, + 1456 + ], + "score": 0.53, + "latex": "e" + }, + { + "category_id": 13, + "poly": [ + 326, + 853, + 344, + 853, + 344, + 875, + 326, + 875 + ], + "score": 0.53, + "latex": "a" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1745.0, + 850.0, + 1745.0, + 850.0, + 1799.0, + 292.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2058.0, + 861.0, + 2058.0, + 861.0, + 2093.0, + 839.0, + 2093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 455.0, + 278.0, + 455.0, + 278.0, + 480.0, + 241.0, + 480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 446.0, + 1406.0, + 446.0, + 1406.0, + 484.0, + 291.0, + 484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 484.0, + 278.0, + 484.0, + 278.0, + 513.0, + 240.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 478.0, + 1406.0, + 478.0, + 1406.0, + 516.0, + 292.0, + 516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 517.0, + 278.0, + 517.0, + 278.0, + 540.0, + 241.0, + 540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 509.0, + 1410.0, + 509.0, + 1410.0, + 546.0, + 294.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 546.0, + 278.0, + 546.0, + 278.0, + 574.0, + 240.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 538.0, + 1409.0, + 538.0, + 1409.0, + 576.0, + 292.0, + 576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 577.0, + 278.0, + 577.0, + 278.0, + 602.0, + 240.0, + 602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 570.0, + 1408.0, + 570.0, + 1408.0, + 607.0, + 294.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 606.0, + 279.0, + 606.0, + 279.0, + 634.0, + 238.0, + 634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 600.0, + 1405.0, + 600.0, + 1405.0, + 637.0, + 294.0, + 637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 636.0, + 278.0, + 636.0, + 278.0, + 664.0, + 238.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 632.0, + 1406.0, + 632.0, + 1406.0, + 664.0, + 295.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 666.0, + 278.0, + 666.0, + 278.0, + 696.0, + 238.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 660.0, + 1406.0, + 660.0, + 1406.0, + 697.0, + 294.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 698.0, + 278.0, + 698.0, + 278.0, + 723.0, + 240.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 693.0, + 510.0, + 693.0, + 510.0, + 727.0, + 294.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 208.0, + 277.0, + 208.0, + 277.0, + 235.0, + 241.0, + 235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 201.0, + 1405.0, + 201.0, + 1405.0, + 241.0, + 295.0, + 241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 238.0, + 280.0, + 238.0, + 280.0, + 267.0, + 238.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 229.0, + 1407.0, + 229.0, + 1407.0, + 270.0, + 292.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 269.0, + 277.0, + 269.0, + 277.0, + 295.0, + 241.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 263.0, + 432.0, + 263.0, + 432.0, + 298.0, + 295.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 549.0, + 263.0, + 806.0, + 263.0, + 806.0, + 298.0, + 549.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 977.0, + 263.0, + 1075.0, + 263.0, + 1075.0, + 298.0, + 977.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 263.0, + 1128.0, + 263.0, + 1128.0, + 298.0, + 1097.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1150.0, + 263.0, + 1309.0, + 263.0, + 1309.0, + 298.0, + 1150.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1326.0, + 263.0, + 1405.0, + 263.0, + 1405.0, + 298.0, + 1326.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 300.0, + 279.0, + 300.0, + 279.0, + 326.0, + 241.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 294.0, + 860.0, + 294.0, + 860.0, + 329.0, + 295.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 882.0, + 294.0, + 932.0, + 294.0, + 932.0, + 329.0, + 882.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 952.0, + 294.0, + 1276.0, + 294.0, + 1276.0, + 329.0, + 952.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1294.0, + 294.0, + 1405.0, + 294.0, + 1405.0, + 329.0, + 1294.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 328.0, + 279.0, + 328.0, + 279.0, + 359.0, + 238.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 324.0, + 1408.0, + 324.0, + 1408.0, + 359.0, + 295.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 360.0, + 277.0, + 360.0, + 277.0, + 387.0, + 241.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 353.0, + 1407.0, + 353.0, + 1407.0, + 387.0, + 293.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 390.0, + 276.0, + 390.0, + 276.0, + 416.0, + 241.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 384.0, + 1273.0, + 384.0, + 1273.0, + 419.0, + 295.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 759.0, + 277.0, + 759.0, + 277.0, + 787.0, + 240.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 754.0, + 1407.0, + 754.0, + 1407.0, + 790.0, + 293.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 791.0, + 278.0, + 791.0, + 278.0, + 819.0, + 240.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 784.0, + 367.0, + 784.0, + 367.0, + 821.0, + 291.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 635.0, + 784.0, + 1204.0, + 784.0, + 1204.0, + 821.0, + 635.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 784.0, + 1407.0, + 784.0, + 1407.0, + 821.0, + 1325.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 821.0, + 280.0, + 821.0, + 280.0, + 850.0, + 240.0, + 850.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 814.0, + 296.0, + 814.0, + 296.0, + 852.0, + 291.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 428.0, + 814.0, + 478.0, + 814.0, + 478.0, + 852.0, + 428.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 814.0, + 879.0, + 814.0, + 879.0, + 852.0, + 548.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1102.0, + 814.0, + 1405.0, + 814.0, + 1405.0, + 852.0, + 1102.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 851.0, + 280.0, + 851.0, + 280.0, + 879.0, + 240.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 844.0, + 325.0, + 844.0, + 325.0, + 883.0, + 293.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 844.0, + 353.0, + 844.0, + 353.0, + 883.0, + 345.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 427.0, + 844.0, + 1405.0, + 844.0, + 1405.0, + 883.0, + 427.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 883.0, + 278.0, + 883.0, + 278.0, + 907.0, + 240.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 876.0, + 1407.0, + 876.0, + 1407.0, + 912.0, + 293.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 912.0, + 280.0, + 912.0, + 280.0, + 940.0, + 240.0, + 940.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 906.0, + 588.0, + 906.0, + 588.0, + 943.0, + 293.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1397.0, + 1240.0, + 1397.0, + 1240.0, + 1433.0, + 296.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1363.0, + 1397.0, + 1407.0, + 1397.0, + 1407.0, + 1433.0, + 1363.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1429.0, + 619.0, + 1429.0, + 619.0, + 1463.0, + 293.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 637.0, + 1429.0, + 1186.0, + 1429.0, + 1186.0, + 1463.0, + 637.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 1429.0, + 1407.0, + 1429.0, + 1407.0, + 1463.0, + 1273.0, + 1463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1458.0, + 372.0, + 1458.0, + 372.0, + 1496.0, + 293.0, + 1496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1458.0, + 1231.0, + 1458.0, + 1231.0, + 1496.0, + 395.0, + 1496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1254.0, + 1458.0, + 1289.0, + 1458.0, + 1289.0, + 1496.0, + 1254.0, + 1496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 1458.0, + 1409.0, + 1458.0, + 1409.0, + 1496.0, + 1398.0, + 1496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1494.0, + 537.0, + 1494.0, + 537.0, + 1526.0, + 296.0, + 1526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 558.0, + 1494.0, + 1405.0, + 1494.0, + 1405.0, + 1526.0, + 558.0, + 1526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1523.0, + 1405.0, + 1523.0, + 1405.0, + 1555.0, + 295.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1554.0, + 1371.0, + 1554.0, + 1371.0, + 1586.0, + 295.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 976.0, + 278.0, + 976.0, + 278.0, + 1004.0, + 241.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 966.0, + 617.0, + 966.0, + 617.0, + 1007.0, + 293.0, + 1007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 966.0, + 1406.0, + 966.0, + 1406.0, + 1007.0, + 780.0, + 1007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1010.0, + 278.0, + 1010.0, + 278.0, + 1036.0, + 241.0, + 1036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 995.0, + 1030.0, + 995.0, + 1030.0, + 1044.0, + 290.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1241.0, + 995.0, + 1407.0, + 995.0, + 1407.0, + 1044.0, + 1241.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1041.0, + 279.0, + 1041.0, + 279.0, + 1068.0, + 239.0, + 1068.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1036.0, + 1407.0, + 1036.0, + 1407.0, + 1070.0, + 295.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1071.0, + 279.0, + 1071.0, + 279.0, + 1098.0, + 241.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1063.0, + 610.0, + 1063.0, + 610.0, + 1104.0, + 289.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 778.0, + 1063.0, + 979.0, + 1063.0, + 979.0, + 1104.0, + 778.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1105.0, + 1063.0, + 1119.0, + 1063.0, + 1119.0, + 1104.0, + 1105.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1913.0, + 1406.0, + 1913.0, + 1406.0, + 1951.0, + 293.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1947.0, + 1404.0, + 1947.0, + 1404.0, + 1981.0, + 295.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1976.0, + 1406.0, + 1976.0, + 1406.0, + 2010.0, + 293.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1614.0, + 1404.0, + 1614.0, + 1404.0, + 1650.0, + 296.0, + 1650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1640.0, + 512.0, + 1640.0, + 512.0, + 1685.0, + 291.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 599.0, + 1640.0, + 1011.0, + 1640.0, + 1011.0, + 1685.0, + 599.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 1640.0, + 1409.0, + 1640.0, + 1409.0, + 1685.0, + 1064.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1677.0, + 505.0, + 1677.0, + 505.0, + 1712.0, + 293.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 1677.0, + 550.0, + 1677.0, + 550.0, + 1712.0, + 540.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1272.0, + 1409.0, + 1272.0, + 1409.0, + 1309.0, + 293.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1304.0, + 1405.0, + 1304.0, + 1405.0, + 1338.0, + 294.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1335.0, + 1408.0, + 1335.0, + 1408.0, + 1369.0, + 294.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1819.0, + 1403.0, + 1819.0, + 1403.0, + 1855.0, + 294.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1847.0, + 601.0, + 1847.0, + 601.0, + 1888.0, + 294.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1207.0, + 1298.0, + 1207.0, + 1298.0, + 1246.0, + 246.0, + 1246.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 560, + 247, + 1142, + 247, + 1142, + 617, + 560, + 617 + ], + "score": 0.981, + "html": "
Bits
Integer multiplication384
Probabilistic truncation Nearest truncation1,536 4,462
Comparison1,369
Division (prob. truncation)29,866
Division (nearest truncation)57,798
Exponentiation (prob. truncation)77,684
Exponentiation (nearest truncation)
171,638
Invert square root (prob. truncation) Invert square root (nearest truncation)20,073 27,699
" + }, + { + "category_id": 1, + "poly": [ + 241, + 1179, + 1406, + 1179, + 1406, + 1306, + 241, + 1306 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 240, + 1026, + 1405, + 1026, + 1405, + 1151, + 240, + 1151 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 239, + 1583, + 1404, + 1583, + 1404, + 1706, + 239, + 1706 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 289, + 764, + 1406, + 764, + 1406, + 860, + 289, + 860 + ], + "score": 0.966 + }, + { + "category_id": 1, + "poly": [ + 253, + 671, + 1405, + 671, + 1405, + 736, + 253, + 736 + ], + "score": 0.954 + }, + { + "category_id": 8, + "poly": [ + 682, + 938, + 1014, + 938, + 1014, + 1013, + 682, + 1013 + ], + "score": 0.951 + }, + { + "category_id": 8, + "poly": [ + 619, + 1789, + 1081, + 1789, + 1081, + 1869, + 619, + 1869 + ], + "score": 0.946 + }, + { + "category_id": 8, + "poly": [ + 462, + 1410, + 1238, + 1410, + 1238, + 1494, + 462, + 1494 + ], + "score": 0.945 + }, + { + "category_id": 8, + "poly": [ + 747, + 1704, + 949, + 1704, + 949, + 1753, + 747, + 1753 + ], + "score": 0.944 + }, + { + "category_id": 1, + "poly": [ + 254, + 1335, + 1404, + 1335, + 1404, + 1398, + 254, + 1398 + ], + "score": 0.941 + }, + { + "category_id": 1, + "poly": [ + 247, + 1508, + 1409, + 1508, + 1409, + 1571, + 247, + 1571 + ], + "score": 0.917 + }, + { + "category_id": 1, + "poly": [ + 242, + 1761, + 501, + 1761, + 501, + 1792, + 242, + 1792 + ], + "score": 0.916 + }, + { + "category_id": 1, + "poly": [ + 240, + 888, + 1253, + 888, + 1253, + 925, + 240, + 925 + ], + "score": 0.915 + }, + { + "category_id": 1, + "poly": [ + 240, + 1873, + 1406, + 1873, + 1406, + 2009, + 240, + 2009 + ], + "score": 0.881 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1432, + 1400, + 1432, + 1400, + 1461, + 1366, + 1461 + ], + "score": 0.867 + }, + { + "category_id": 6, + "poly": [ + 348, + 197, + 1347, + 197, + 1347, + 232, + 348, + 232 + ], + "score": 0.846 + }, + { + "category_id": 2, + "poly": [ + 840, + 2061, + 859, + 2061, + 859, + 2086, + 840, + 2086 + ], + "score": 0.755 + }, + { + "category_id": 2, + "poly": [ + 839, + 2062, + 859, + 2062, + 859, + 2086, + 839, + 2086 + ], + "score": 0.106 + }, + { + "category_id": 14, + "poly": [ + 617, + 1787, + 1082, + 1787, + 1082, + 1870, + 617, + 1870 + ], + "score": 0.94, + "latex": "{ \\frac { e ^ { x _ { i } - m } } { \\sum _ { j } e ^ { x _ { j } - m } } } = { \\frac { e ^ { x _ { i } } e ^ { - m } } { ( \\sum _ { j } e ^ { x _ { j } } ) e ^ { - m } } } = { \\frac { e ^ { x _ { i } } } { \\sum _ { j } e ^ { x _ { j } } } } ." + }, + { + "category_id": 14, + "poly": [ + 680, + 936, + 1012, + 936, + 1012, + 1014, + 680, + 1014 + ], + "score": 0.94, + "latex": "{ \\mathsf { R e L U } } ( x ) : = { \\left\\{ \\begin{array} { l l } { x , } & { { \\mathrm { i f ~ } } x > 0 } \\\\ { 0 . } & { { \\mathrm { o t h e r w i s e } } } \\end{array} \\right. }" + }, + { + "category_id": 14, + "poly": [ + 462, + 1408, + 1236, + 1408, + 1236, + 1497, + 462, + 1497 + ], + "score": 0.93, + "latex": "\\nabla i : = \\frac { \\partial \\ell } { \\partial x _ { i } } = \\frac { \\partial } { \\partial x _ { i } } \\Big ( - \\sum _ { k } y _ { k } \\cdot x _ { k } + \\log \\sum _ { j } e ^ { x _ { j } } \\Big ) = - y _ { i } + \\frac { e ^ { x _ { i } } } { \\sum _ { j } e ^ { x _ { j } } } ," + }, + { + "category_id": 14, + "poly": [ + 747, + 1701, + 952, + 1701, + 952, + 1754, + 747, + 1754 + ], + "score": 0.92, + "latex": "m = \\operatorname* { m a x } _ { j } ( \\{ x _ { j } \\} ) ." + }, + { + "category_id": 13, + "poly": [ + 969, + 199, + 1054, + 199, + 1054, + 231, + 969, + 231 + ], + "score": 0.91, + "latex": "f = 3 2" + }, + { + "category_id": 13, + "poly": [ + 1094, + 1513, + 1123, + 1513, + 1123, + 1539, + 1094, + 1539 + ], + "score": 0.86, + "latex": "x _ { i }" + }, + { + "category_id": 13, + "poly": [ + 369, + 1513, + 396, + 1513, + 396, + 1540, + 369, + 1540 + ], + "score": 0.85, + "latex": "y _ { i }" + }, + { + "category_id": 13, + "poly": [ + 1290, + 195, + 1340, + 195, + 1340, + 227, + 1290, + 227 + ], + "score": 0.84, + "latex": "2 ^ { 1 2 8 }" + }, + { + "category_id": 13, + "poly": [ + 864, + 1933, + 1269, + 1933, + 1269, + 1981, + 864, + 1981 + ], + "score": 0.56, + "latex": "\\begin{array} { r } { \\mathrm { \\begin{array} { r } { \\mathrm { i d } ( x ) = \\frac { 1 } { 1 + \\exp ( - x ) } = \\frac { \\exp ( 0 ) } { \\exp ( 0 ) + \\exp ( x ) } } } \\end{array} } \\end{array}" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 188.0, + 968.0, + 188.0, + 968.0, + 238.0, + 343.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 188.0, + 1289.0, + 188.0, + 1289.0, + 238.0, + 1055.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1341.0, + 188.0, + 1351.0, + 188.0, + 1351.0, + 238.0, + 1341.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2060.0, + 862.0, + 2060.0, + 862.0, + 2090.0, + 840.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2059.0, + 861.0, + 2059.0, + 861.0, + 2090.0, + 840.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1187.0, + 278.0, + 1187.0, + 278.0, + 1214.0, + 240.0, + 1214.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1180.0, + 1409.0, + 1180.0, + 1409.0, + 1220.0, + 294.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1216.0, + 278.0, + 1216.0, + 278.0, + 1247.0, + 237.0, + 1247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1211.0, + 1403.0, + 1211.0, + 1403.0, + 1247.0, + 294.0, + 1247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1245.0, + 279.0, + 1245.0, + 279.0, + 1277.0, + 238.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1243.0, + 1405.0, + 1243.0, + 1405.0, + 1278.0, + 294.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1278.0, + 278.0, + 1278.0, + 278.0, + 1305.0, + 240.0, + 1305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1272.0, + 1268.0, + 1272.0, + 1268.0, + 1308.0, + 294.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1031.0, + 280.0, + 1031.0, + 280.0, + 1060.0, + 237.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1027.0, + 1408.0, + 1027.0, + 1408.0, + 1062.0, + 293.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1062.0, + 280.0, + 1062.0, + 280.0, + 1091.0, + 237.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1058.0, + 1405.0, + 1058.0, + 1405.0, + 1093.0, + 294.0, + 1093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1091.0, + 280.0, + 1091.0, + 280.0, + 1121.0, + 237.0, + 1121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1087.0, + 1405.0, + 1087.0, + 1405.0, + 1122.0, + 294.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1122.0, + 280.0, + 1122.0, + 280.0, + 1152.0, + 237.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1115.0, + 711.0, + 1115.0, + 711.0, + 1155.0, + 294.0, + 1155.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1590.0, + 277.0, + 1590.0, + 277.0, + 1617.0, + 239.0, + 1617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1582.0, + 1405.0, + 1582.0, + 1405.0, + 1621.0, + 294.0, + 1621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1619.0, + 277.0, + 1619.0, + 277.0, + 1646.0, + 239.0, + 1646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1617.0, + 1403.0, + 1617.0, + 1403.0, + 1647.0, + 296.0, + 1647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1650.0, + 277.0, + 1650.0, + 277.0, + 1677.0, + 240.0, + 1677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1639.0, + 1405.0, + 1639.0, + 1405.0, + 1684.0, + 293.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1681.0, + 276.0, + 1681.0, + 276.0, + 1707.0, + 239.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1674.0, + 678.0, + 1674.0, + 678.0, + 1711.0, + 293.0, + 1711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 762.0, + 1404.0, + 762.0, + 1404.0, + 803.0, + 294.0, + 803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 799.0, + 1404.0, + 799.0, + 1404.0, + 833.0, + 295.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 827.0, + 395.0, + 827.0, + 395.0, + 859.0, + 290.0, + 859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 671.0, + 1405.0, + 671.0, + 1405.0, + 707.0, + 250.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 702.0, + 1407.0, + 702.0, + 1407.0, + 738.0, + 250.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1333.0, + 1406.0, + 1333.0, + 1406.0, + 1373.0, + 247.0, + 1373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 242.0, + 1367.0, + 506.0, + 1367.0, + 506.0, + 1401.0, + 242.0, + 1401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1505.0, + 368.0, + 1505.0, + 368.0, + 1543.0, + 235.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 1505.0, + 1093.0, + 1505.0, + 1093.0, + 1543.0, + 397.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 1505.0, + 1406.0, + 1505.0, + 1406.0, + 1543.0, + 1124.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1533.0, + 414.0, + 1533.0, + 414.0, + 1577.0, + 234.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1757.0, + 504.0, + 1757.0, + 504.0, + 1797.0, + 235.0, + 1797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 888.0, + 1257.0, + 888.0, + 1257.0, + 928.0, + 234.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1877.0, + 280.0, + 1877.0, + 280.0, + 1904.0, + 239.0, + 1904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1873.0, + 1405.0, + 1873.0, + 1405.0, + 1904.0, + 294.0, + 1904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1907.0, + 278.0, + 1907.0, + 278.0, + 1934.0, + 239.0, + 1934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1904.0, + 1405.0, + 1904.0, + 1405.0, + 1935.0, + 296.0, + 1935.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1945.0, + 278.0, + 1945.0, + 278.0, + 1973.0, + 239.0, + 1973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1270.0, + 1929.0, + 1407.0, + 1929.0, + 1407.0, + 1984.0, + 1270.0, + 1984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1982.0, + 278.0, + 1982.0, + 278.0, + 2005.0, + 240.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1975.0, + 483.0, + 1975.0, + 483.0, + 2007.0, + 293.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 1929.5, + 1093.0, + 1929.5, + 1093.0, + 1985.5, + 287.0, + 1985.5 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 240, + 202, + 1404, + 202, + 1404, + 453, + 240, + 453 + ], + "score": 0.967 + }, + { + "category_id": 1, + "poly": [ + 256, + 836, + 1405, + 836, + 1405, + 1021, + 256, + 1021 + ], + "score": 0.96 + }, + { + "category_id": 1, + "poly": [ + 249, + 1914, + 1404, + 1914, + 1404, + 2008, + 249, + 2008 + ], + "score": 0.954 + }, + { + "category_id": 1, + "poly": [ + 276, + 499, + 1405, + 499, + 1405, + 592, + 276, + 592 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 296, + 1035, + 1406, + 1035, + 1406, + 1127, + 296, + 1127 + ], + "score": 0.945 + }, + { + "category_id": 1, + "poly": [ + 287, + 1808, + 1403, + 1808, + 1403, + 1901, + 287, + 1901 + ], + "score": 0.943 + }, + { + "category_id": 1, + "poly": [ + 276, + 639, + 1405, + 639, + 1405, + 703, + 276, + 703 + ], + "score": 0.941 + }, + { + "category_id": 1, + "poly": [ + 285, + 1384, + 1403, + 1384, + 1403, + 1598, + 285, + 1598 + ], + "score": 0.936 + }, + { + "category_id": 1, + "poly": [ + 293, + 1611, + 1404, + 1611, + 1404, + 1795, + 293, + 1795 + ], + "score": 0.93 + }, + { + "category_id": 0, + "poly": [ + 261, + 760, + 577, + 760, + 577, + 797, + 261, + 797 + ], + "score": 0.899 + }, + { + "category_id": 1, + "poly": [ + 277, + 1141, + 1405, + 1141, + 1405, + 1204, + 277, + 1204 + ], + "score": 0.894 + }, + { + "category_id": 1, + "poly": [ + 250, + 1215, + 1115, + 1215, + 1115, + 1248, + 250, + 1248 + ], + "score": 0.892 + }, + { + "category_id": 0, + "poly": [ + 266, + 1307, + 457, + 1307, + 457, + 1342, + 266, + 1342 + ], + "score": 0.838 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 858, + 2061, + 858, + 2084, + 841, + 2084 + ], + "score": 0.695 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2084, + 841, + 2084 + ], + "score": 0.17 + }, + { + "category_id": 13, + "poly": [ + 1052, + 1703, + 1136, + 1703, + 1136, + 1734, + 1052, + 1734 + ], + "score": 0.92, + "latex": "f = 1 6" + }, + { + "category_id": 13, + "poly": [ + 297, + 234, + 570, + 234, + 570, + 275, + 297, + 275 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { \\theta _ { j } \\theta _ { j } - \\frac { \\gamma } { B } \\sum _ { i = 1 } ^ { B } \\bigtriangledown _ { i j } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1306, + 1673, + 1398, + 1673, + 1398, + 1705, + 1306, + 1705 + ], + "score": 0.91, + "latex": "f = 3 2" + }, + { + "category_id": 13, + "poly": [ + 546, + 1733, + 631, + 1733, + 631, + 1764, + 546, + 1764 + ], + "score": 0.91, + "latex": "f = 3 2" + }, + { + "category_id": 13, + "poly": [ + 1156, + 1673, + 1250, + 1673, + 1250, + 1704, + 1156, + 1704 + ], + "score": 0.91, + "latex": "f = 1 6" + }, + { + "category_id": 13, + "poly": [ + 587, + 1765, + 672, + 1765, + 672, + 1796, + 587, + 1796 + ], + "score": 0.91, + "latex": "f = 3 2" + }, + { + "category_id": 13, + "poly": [ + 297, + 1673, + 391, + 1673, + 391, + 1704, + 297, + 1704 + ], + "score": 0.91, + "latex": "f = 6 4" + }, + { + "category_id": 13, + "poly": [ + 1305, + 240, + 1375, + 240, + 1375, + 271, + 1305, + 271 + ], + "score": 0.91, + "latex": "\\gamma > 0" + }, + { + "category_id": 13, + "poly": [ + 1186, + 1705, + 1270, + 1705, + 1270, + 1735, + 1186, + 1735 + ], + "score": 0.9, + "latex": "f = 3 2" + }, + { + "category_id": 13, + "poly": [ + 537, + 275, + 583, + 275, + 583, + 303, + 537, + 303 + ], + "score": 0.89, + "latex": "\\nabla i j" + }, + { + "category_id": 13, + "poly": [ + 649, + 239, + 677, + 239, + 677, + 273, + 649, + 273 + ], + "score": 0.87, + "latex": "\\theta _ { j }" + }, + { + "category_id": 13, + "poly": [ + 500, + 301, + 527, + 301, + 527, + 333, + 500, + 333 + ], + "score": 0.87, + "latex": "\\theta _ { j }" + }, + { + "category_id": 13, + "poly": [ + 1078, + 272, + 1095, + 272, + 1095, + 297, + 1078, + 297 + ], + "score": 0.69, + "latex": "i ^ { \\because }" + }, + { + "category_id": 13, + "poly": [ + 996, + 241, + 1013, + 241, + 1013, + 272, + 996, + 272 + ], + "score": 0.57, + "latex": "j" + }, + { + "category_id": 13, + "poly": [ + 1024, + 240, + 1049, + 240, + 1049, + 268, + 1024, + 268 + ], + "score": 0.51, + "latex": "B" + }, + { + "category_id": 13, + "poly": [ + 995, + 239, + 1050, + 239, + 1050, + 270, + 995, + 270 + ], + "score": 0.3, + "latex": "j , B" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 754.0, + 582.0, + 754.0, + 582.0, + 808.0, + 248.0, + 808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1305.0, + 463.0, + 1305.0, + 463.0, + 1350.0, + 251.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2059.0, + 860.0, + 2059.0, + 860.0, + 2091.0, + 839.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2059.0, + 860.0, + 2059.0, + 860.0, + 2092.0, + 840.0, + 2092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 210.0, + 277.0, + 210.0, + 277.0, + 234.0, + 240.0, + 234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 201.0, + 1409.0, + 201.0, + 1409.0, + 240.0, + 293.0, + 240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 245.0, + 277.0, + 245.0, + 277.0, + 271.0, + 240.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 222.0, + 296.0, + 222.0, + 296.0, + 290.0, + 291.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 571.0, + 222.0, + 648.0, + 222.0, + 648.0, + 290.0, + 571.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 222.0, + 994.0, + 222.0, + 994.0, + 290.0, + 678.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 222.0, + 1304.0, + 222.0, + 1304.0, + 290.0, + 1051.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1376.0, + 222.0, + 1416.0, + 222.0, + 1416.0, + 290.0, + 1376.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 274.0, + 280.0, + 274.0, + 280.0, + 303.0, + 237.0, + 303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 267.0, + 536.0, + 267.0, + 536.0, + 306.0, + 294.0, + 306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 267.0, + 1077.0, + 267.0, + 1077.0, + 306.0, + 584.0, + 306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 267.0, + 1405.0, + 267.0, + 1405.0, + 306.0, + 1096.0, + 306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 303.0, + 280.0, + 303.0, + 280.0, + 333.0, + 237.0, + 333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 296.0, + 499.0, + 296.0, + 499.0, + 337.0, + 291.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 528.0, + 296.0, + 1405.0, + 296.0, + 1405.0, + 337.0, + 528.0, + 337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 334.0, + 277.0, + 334.0, + 277.0, + 360.0, + 239.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 329.0, + 1405.0, + 329.0, + 1405.0, + 363.0, + 295.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 365.0, + 280.0, + 365.0, + 280.0, + 393.0, + 237.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 360.0, + 1405.0, + 360.0, + 1405.0, + 395.0, + 295.0, + 395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 395.0, + 280.0, + 395.0, + 280.0, + 425.0, + 237.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 389.0, + 1407.0, + 389.0, + 1407.0, + 426.0, + 294.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 425.0, + 280.0, + 425.0, + 280.0, + 454.0, + 237.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 421.0, + 1405.0, + 421.0, + 1405.0, + 455.0, + 294.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 845.0, + 278.0, + 845.0, + 278.0, + 870.0, + 250.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 838.0, + 1407.0, + 838.0, + 1407.0, + 874.0, + 293.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 876.0, + 276.0, + 876.0, + 276.0, + 898.0, + 252.0, + 898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 869.0, + 1406.0, + 869.0, + 1406.0, + 904.0, + 294.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 906.0, + 279.0, + 906.0, + 279.0, + 931.0, + 249.0, + 931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 897.0, + 1406.0, + 897.0, + 1406.0, + 934.0, + 292.0, + 934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 935.0, + 279.0, + 935.0, + 279.0, + 960.0, + 249.0, + 960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 923.0, + 1408.0, + 923.0, + 1408.0, + 967.0, + 292.0, + 967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 967.0, + 279.0, + 967.0, + 279.0, + 992.0, + 249.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 958.0, + 1407.0, + 958.0, + 1407.0, + 996.0, + 292.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 996.0, + 279.0, + 996.0, + 279.0, + 1021.0, + 249.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 987.0, + 450.0, + 987.0, + 450.0, + 1028.0, + 291.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 245.0, + 1921.0, + 276.0, + 1921.0, + 276.0, + 1945.0, + 245.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1912.0, + 1407.0, + 1912.0, + 1407.0, + 1951.0, + 291.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 245.0, + 1953.0, + 276.0, + 1953.0, + 276.0, + 1977.0, + 245.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1946.0, + 1406.0, + 1946.0, + 1406.0, + 1980.0, + 294.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 245.0, + 1984.0, + 277.0, + 1984.0, + 277.0, + 2007.0, + 245.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1977.0, + 1368.0, + 1977.0, + 1368.0, + 2011.0, + 293.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 498.0, + 1404.0, + 498.0, + 1404.0, + 532.0, + 295.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 531.0, + 1404.0, + 531.0, + 1404.0, + 565.0, + 294.0, + 565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 558.0, + 390.0, + 558.0, + 390.0, + 599.0, + 293.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1034.0, + 1404.0, + 1034.0, + 1404.0, + 1071.0, + 294.0, + 1071.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1067.0, + 1408.0, + 1067.0, + 1408.0, + 1100.0, + 295.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1095.0, + 1092.0, + 1095.0, + 1092.0, + 1129.0, + 295.0, + 1129.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1805.0, + 1405.0, + 1805.0, + 1405.0, + 1844.0, + 294.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1838.0, + 1405.0, + 1838.0, + 1405.0, + 1875.0, + 294.0, + 1875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1873.0, + 394.0, + 1873.0, + 394.0, + 1905.0, + 293.0, + 1905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 638.0, + 1405.0, + 638.0, + 1405.0, + 675.0, + 294.0, + 675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 672.0, + 1406.0, + 672.0, + 1406.0, + 704.0, + 296.0, + 704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1384.0, + 1405.0, + 1384.0, + 1405.0, + 1419.0, + 295.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1414.0, + 1406.0, + 1414.0, + 1406.0, + 1449.0, + 293.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1443.0, + 1407.0, + 1443.0, + 1407.0, + 1480.0, + 293.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1476.0, + 1406.0, + 1476.0, + 1406.0, + 1511.0, + 295.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1504.0, + 1406.0, + 1504.0, + 1406.0, + 1539.0, + 295.0, + 1539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1536.0, + 1406.0, + 1536.0, + 1406.0, + 1569.0, + 293.0, + 1569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1567.0, + 1279.0, + 1567.0, + 1279.0, + 1601.0, + 293.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1609.0, + 1407.0, + 1609.0, + 1407.0, + 1646.0, + 292.0, + 1646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1640.0, + 1409.0, + 1640.0, + 1409.0, + 1676.0, + 292.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1671.0, + 296.0, + 1671.0, + 296.0, + 1706.0, + 293.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1671.0, + 1155.0, + 1671.0, + 1155.0, + 1706.0, + 392.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 1671.0, + 1305.0, + 1671.0, + 1305.0, + 1706.0, + 1251.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 1671.0, + 1409.0, + 1671.0, + 1409.0, + 1706.0, + 1399.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1702.0, + 1051.0, + 1702.0, + 1051.0, + 1737.0, + 293.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1702.0, + 1185.0, + 1702.0, + 1185.0, + 1737.0, + 1137.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1271.0, + 1702.0, + 1406.0, + 1702.0, + 1406.0, + 1737.0, + 1271.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1732.0, + 545.0, + 1732.0, + 545.0, + 1768.0, + 293.0, + 1768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 632.0, + 1732.0, + 1405.0, + 1732.0, + 1405.0, + 1768.0, + 632.0, + 1768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1761.0, + 586.0, + 1761.0, + 586.0, + 1799.0, + 294.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 673.0, + 1761.0, + 1124.0, + 1761.0, + 1124.0, + 1799.0, + 673.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1142.0, + 1403.0, + 1142.0, + 1403.0, + 1174.0, + 296.0, + 1174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1171.0, + 1409.0, + 1171.0, + 1409.0, + 1206.0, + 294.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1214.0, + 1115.0, + 1214.0, + 1115.0, + 1252.0, + 240.0, + 1252.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1719, + 1405, + 1719, + 1405, + 1932, + 297, + 1932 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 285, + 1517, + 1403, + 1517, + 1403, + 1611, + 285, + 1611 + ], + "score": 0.971 + }, + { + "category_id": 3, + "poly": [ + 345, + 192, + 1357, + 192, + 1357, + 726, + 345, + 726 + ], + "score": 0.956 + }, + { + "category_id": 4, + "poly": [ + 295, + 1392, + 1401, + 1392, + 1401, + 1455, + 295, + 1455 + ], + "score": 0.934 + }, + { + "category_id": 4, + "poly": [ + 298, + 758, + 1403, + 758, + 1403, + 793, + 298, + 793 + ], + "score": 0.917 + }, + { + "category_id": 2, + "poly": [ + 330, + 1978, + 858, + 1978, + 858, + 2007, + 330, + 2007 + ], + "score": 0.917 + }, + { + "category_id": 0, + "poly": [ + 299, + 1658, + 598, + 1658, + 598, + 1690, + 299, + 1690 + ], + "score": 0.891 + }, + { + "category_id": 2, + "poly": [ + 840, + 2062, + 858, + 2062, + 858, + 2085, + 840, + 2085 + ], + "score": 0.804 + }, + { + "category_id": 3, + "poly": [ + 338, + 827, + 1316, + 827, + 1316, + 1359, + 338, + 1359 + ], + "score": 0.695 + }, + { + "category_id": 3, + "poly": [ + 338, + 829, + 1075, + 829, + 1075, + 1360, + 338, + 1360 + ], + "score": 0.324 + }, + { + "category_id": 13, + "poly": [ + 1249, + 1720, + 1350, + 1720, + 1350, + 1752, + 1249, + 1752 + ], + "score": 0.91, + "latex": "f \\ = \\ 3 2" + }, + { + "category_id": 13, + "poly": [ + 1120, + 1394, + 1204, + 1394, + 1204, + 1426, + 1120, + 1426 + ], + "score": 0.91, + "latex": "f = 3 2" + }, + { + "category_id": 13, + "poly": [ + 445, + 1780, + 520, + 1780, + 520, + 1809, + 445, + 1809 + ], + "score": 0.87, + "latex": "9 8 . 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 1167, + 311, + 1251, + 311, + 1251, + 342, + 1167, + 342 + ], + "score": 0.84, + "latex": "f = 1 6" + }, + { + "category_id": 13, + "poly": [ + 1167, + 345, + 1251, + 345, + 1251, + 377, + 1167, + 377 + ], + "score": 0.8, + "latex": "f = 3 2" + }, + { + "category_id": 13, + "poly": [ + 1167, + 276, + 1251, + 276, + 1251, + 308, + 1167, + 308 + ], + "score": 0.8, + "latex": "f = 6 4" + }, + { + "category_id": 13, + "poly": [ + 1167, + 207, + 1251, + 207, + 1251, + 239, + 1167, + 239 + ], + "score": 0.57, + "latex": "f = 1 6" + }, + { + "category_id": 13, + "poly": [ + 1167, + 242, + 1251, + 242, + 1251, + 274, + 1167, + 274 + ], + "score": 0.43, + "latex": "f = 3 2" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 199.0, + 458.0, + 199.0, + 458.0, + 248.0, + 399.0, + 248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1158.0, + 202.0, + 1166.0, + 202.0, + 1166.0, + 241.0, + 1158.0, + 241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 202.0, + 1327.0, + 202.0, + 1327.0, + 241.0, + 1252.0, + 241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 242.0, + 1067.0, + 242.0, + 1067.0, + 287.0, + 998.0, + 287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1159.0, + 233.0, + 1166.0, + 233.0, + 1166.0, + 278.0, + 1159.0, + 278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 233.0, + 1329.0, + 233.0, + 1329.0, + 278.0, + 1252.0, + 278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1159.0, + 270.0, + 1166.0, + 270.0, + 1166.0, + 312.0, + 1159.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 270.0, + 1328.0, + 270.0, + 1328.0, + 312.0, + 1252.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 317.0, + 455.0, + 317.0, + 455.0, + 350.0, + 381.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 997.0, + 326.0, + 1066.0, + 326.0, + 1066.0, + 363.0, + 997.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1157.0, + 304.0, + 1166.0, + 304.0, + 1166.0, + 381.0, + 1157.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 304.0, + 1353.0, + 304.0, + 1353.0, + 381.0, + 1252.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 394.0, + 382.0, + 394.0, + 382.0, + 454.0, + 345.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 997.0, + 403.0, + 1066.0, + 403.0, + 1066.0, + 445.0, + 997.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 389.0, + 1099.0, + 389.0, + 1099.0, + 459.0, + 1068.0, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 483.0, + 1066.0, + 483.0, + 1066.0, + 524.0, + 996.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 565.0, + 1051.0, + 565.0, + 1051.0, + 604.0, + 998.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 493.0, + 654.0, + 515.0, + 654.0, + 515.0, + 679.0, + 493.0, + 679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 633.0, + 648.0, + 676.0, + 648.0, + 676.0, + 686.0, + 633.0, + 686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 648.0, + 835.0, + 648.0, + 835.0, + 686.0, + 777.0, + 686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 645.0, + 988.0, + 645.0, + 988.0, + 687.0, + 926.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 673.0, + 687.0, + 784.0, + 687.0, + 784.0, + 734.0, + 673.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 380.0, + 421.5, + 461.0, + 421.5, + 461.0, + 462.5, + 380.0, + 462.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 379.0, + 529.0, + 461.0, + 529.0, + 461.0, + 570.0, + 379.0, + 570.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1390.0, + 1119.0, + 1390.0, + 1119.0, + 1430.0, + 294.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1205.0, + 1390.0, + 1406.0, + 1390.0, + 1406.0, + 1430.0, + 1205.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1422.0, + 422.0, + 1422.0, + 422.0, + 1455.0, + 293.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 754.0, + 1409.0, + 754.0, + 1409.0, + 797.0, + 293.0, + 797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 1970.0, + 860.0, + 1970.0, + 860.0, + 2014.0, + 330.0, + 2014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1656.0, + 601.0, + 1656.0, + 601.0, + 1696.0, + 295.0, + 1696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2090.0, + 838.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 837.0, + 1017.0, + 837.0, + 1017.0, + 864.0, + 994.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1152.0, + 835.0, + 1323.0, + 835.0, + 1323.0, + 874.0, + 1152.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 882.0, + 449.0, + 882.0, + 449.0, + 928.0, + 389.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1150.0, + 868.0, + 1327.0, + 868.0, + 1327.0, + 940.0, + 1150.0, + 940.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1151.0, + 936.0, + 1327.0, + 936.0, + 1327.0, + 977.0, + 1151.0, + 977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 968.0, + 1043.0, + 968.0, + 1043.0, + 1009.0, + 989.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1152.0, + 970.0, + 1324.0, + 970.0, + 1324.0, + 1010.0, + 1152.0, + 1010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 997.0, + 445.0, + 997.0, + 445.0, + 1031.0, + 370.0, + 1031.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1025.0, + 376.0, + 1025.0, + 376.0, + 1090.0, + 341.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1025.0, + 1072.0, + 1025.0, + 1072.0, + 1091.0, + 1045.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 1106.0, + 1041.0, + 1106.0, + 1041.0, + 1143.0, + 990.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 1242.0, + 1045.0, + 1242.0, + 1045.0, + 1282.0, + 986.0, + 1282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 488.0, + 1287.0, + 509.0, + 1287.0, + 509.0, + 1314.0, + 488.0, + 1314.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 1283.0, + 618.0, + 1283.0, + 618.0, + 1318.0, + 575.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 673.0, + 1283.0, + 717.0, + 1283.0, + 717.0, + 1318.0, + 673.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 771.0, + 1283.0, + 815.0, + 1283.0, + 815.0, + 1318.0, + 771.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 1283.0, + 912.0, + 1283.0, + 912.0, + 1318.0, + 869.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 960.0, + 1282.0, + 1018.0, + 1282.0, + 1018.0, + 1318.0, + 960.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 663.0, + 1324.0, + 772.0, + 1324.0, + 772.0, + 1366.0, + 663.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.25, + 1100.5, + 452.25, + 1100.5, + 452.25, + 1143.5, + 369.25, + 1143.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.25, + 1209.5, + 451.25, + 1209.5, + 451.25, + 1251.0, + 369.25, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 969.0, + 1231.5, + 976.0, + 1231.5, + 976.0, + 1243.0, + 969.0, + 1243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 841.0, + 1012.0, + 841.0, + 1012.0, + 862.0, + 996.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 883.0, + 449.0, + 883.0, + 449.0, + 929.0, + 391.0, + 929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 967.0, + 1043.0, + 967.0, + 1043.0, + 1008.0, + 989.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 995.0, + 446.0, + 995.0, + 446.0, + 1033.0, + 368.0, + 1033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 1028.0, + 373.0, + 1028.0, + 373.0, + 1088.0, + 336.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1024.0, + 1072.0, + 1024.0, + 1072.0, + 1090.0, + 1045.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 1105.0, + 447.0, + 1105.0, + 447.0, + 1139.0, + 369.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.0, + 1104.0, + 981.0, + 1104.0, + 981.0, + 1134.0, + 914.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 991.0, + 1105.0, + 1042.0, + 1105.0, + 1042.0, + 1144.0, + 991.0, + 1144.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 1239.0, + 1045.0, + 1239.0, + 1045.0, + 1284.0, + 983.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 487.0, + 1286.0, + 510.0, + 1286.0, + 510.0, + 1316.0, + 487.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 1283.0, + 618.0, + 1283.0, + 618.0, + 1318.0, + 576.0, + 1318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 1284.0, + 718.0, + 1284.0, + 718.0, + 1317.0, + 672.0, + 1317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 1284.0, + 815.0, + 1284.0, + 815.0, + 1317.0, + 768.0, + 1317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 868.0, + 1284.0, + 913.0, + 1284.0, + 913.0, + 1317.0, + 868.0, + 1317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 956.0, + 1281.0, + 1018.0, + 1281.0, + 1018.0, + 1320.0, + 956.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 664.0, + 1323.0, + 774.0, + 1323.0, + 774.0, + 1367.0, + 664.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 1208.0, + 452.0, + 1208.0, + 452.0, + 1251.0, + 369.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1719.0, + 1248.0, + 1719.0, + 1248.0, + 1753.0, + 295.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1351.0, + 1719.0, + 1403.0, + 1719.0, + 1403.0, + 1753.0, + 1351.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1750.0, + 1406.0, + 1750.0, + 1406.0, + 1784.0, + 294.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1780.0, + 444.0, + 1780.0, + 444.0, + 1814.0, + 295.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 521.0, + 1780.0, + 1403.0, + 1780.0, + 1403.0, + 1814.0, + 521.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1808.0, + 1408.0, + 1808.0, + 1408.0, + 1846.0, + 294.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1838.0, + 1406.0, + 1838.0, + 1406.0, + 1876.0, + 294.0, + 1876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1867.0, + 1408.0, + 1867.0, + 1408.0, + 1907.0, + 292.0, + 1907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1901.0, + 801.0, + 1901.0, + 801.0, + 1936.0, + 296.0, + 1936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1515.0, + 1406.0, + 1515.0, + 1406.0, + 1553.0, + 295.0, + 1553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1549.0, + 1405.0, + 1549.0, + 1405.0, + 1581.0, + 295.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1579.0, + 1278.0, + 1579.0, + 1278.0, + 1613.0, + 297.0, + 1613.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 313, + 989, + 1387, + 989, + 1387, + 1678, + 313, + 1678 + ], + "score": 0.986, + "html": "
NetworkEpoch time (s)Acc. (# epochs)Precision(f)
AMohassel and Zhang [2017]283/19333*93.4% (15)13
Mohassel and Rindal [2018]18094.0% (15)N/A
Agrawal et al. [2019]31392*95.0% (10)N/A
Wagh et al. [2019]247N/A13
Wagh et al. [2021]41N/A13
Ours3197.9% (15)16
Ours5097.7% (15)32
BWagh et al. [2019]4176N/A13
Wagh et al. [2021]101N/A13
Ours14493.6% (15)16
Ours24994.7% (15)32
CWagh et al. [2019]7188N/A13
Wagh et al. [2021]891N/A13
Tan et al. [2021]103694.0% (5)20
Ours34494.9% (5)16
Ours64393.8% (5)32
DMohassel and Rindal [2018]234N/AN/A
Ours4196.8% (15)16
Ours6896.8% (15)32
" + }, + { + "category_id": 1, + "poly": [ + 241, + 1853, + 1404, + 1853, + 1404, + 2008, + 241, + 2008 + ], + "score": 0.976 + }, + { + "category_id": 3, + "poly": [ + 370, + 192, + 1335, + 192, + 1335, + 726, + 370, + 726 + ], + "score": 0.965 + }, + { + "category_id": 6, + "poly": [ + 297, + 849, + 1405, + 849, + 1405, + 974, + 297, + 974 + ], + "score": 0.932 + }, + { + "category_id": 4, + "poly": [ + 362, + 758, + 1335, + 758, + 1335, + 793, + 362, + 793 + ], + "score": 0.924 + }, + { + "category_id": 0, + "poly": [ + 241, + 1762, + 523, + 1762, + 523, + 1800, + 241, + 1800 + ], + "score": 0.911 + }, + { + "category_id": 2, + "poly": [ + 840, + 2061, + 859, + 2061, + 859, + 2084, + 840, + 2084 + ], + "score": 0.802 + }, + { + "category_id": 13, + "poly": [ + 925, + 760, + 1010, + 760, + 1010, + 792, + 925, + 792 + ], + "score": 0.91, + "latex": "f = 3 2" + }, + { + "category_id": 13, + "poly": [ + 985, + 851, + 1020, + 851, + 1020, + 882, + 985, + 882 + ], + "score": 0.7, + "latex": "( ^ { * } )" + }, + { + "category_id": 15, + "poly": [ + 422.0, + 209.0, + 481.0, + 209.0, + 481.0, + 250.0, + 422.0, + 250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 200.0, + 1327.0, + 200.0, + 1327.0, + 240.0, + 1185.0, + 240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 233.0, + 1091.0, + 233.0, + 1091.0, + 272.0, + 1022.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 234.0, + 1327.0, + 234.0, + 1327.0, + 274.0, + 1185.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 637.0, + 265.0, + 754.0, + 265.0, + 754.0, + 309.0, + 637.0, + 309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 269.0, + 1327.0, + 269.0, + 1327.0, + 308.0, + 1185.0, + 308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 308.0, + 1091.0, + 308.0, + 1091.0, + 347.0, + 1021.0, + 347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 304.0, + 1327.0, + 304.0, + 1327.0, + 343.0, + 1185.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 391.0, + 410.0, + 391.0, + 410.0, + 456.0, + 374.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1020.0, + 377.0, + 1124.0, + 377.0, + 1124.0, + 460.0, + 1020.0, + 460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 456.0, + 1091.0, + 456.0, + 1091.0, + 495.0, + 1021.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 528.0, + 1077.0, + 528.0, + 1077.0, + 569.0, + 1021.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 603.0, + 1091.0, + 603.0, + 1091.0, + 643.0, + 1022.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 519.0, + 653.0, + 540.0, + 653.0, + 540.0, + 678.0, + 519.0, + 678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 649.0, + 701.0, + 649.0, + 701.0, + 685.0, + 659.0, + 685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 802.0, + 643.0, + 861.0, + 643.0, + 861.0, + 686.0, + 802.0, + 686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 952.0, + 646.0, + 1011.0, + 646.0, + 1011.0, + 686.0, + 952.0, + 686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 699.0, + 690.0, + 807.0, + 690.0, + 807.0, + 731.0, + 699.0, + 731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 758.0, + 287.5, + 819.0, + 287.5, + 819.0, + 311.5, + 758.0, + 311.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.25, + 317.5, + 484.25, + 317.5, + 484.25, + 360.5, + 405.25, + 360.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 425.5, + 488.0, + 425.5, + 488.0, + 468.5, + 403.0, + 468.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.25, + 533.0, + 487.25, + 533.0, + 487.25, + 575.5, + 404.25, + 575.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 849.0, + 984.0, + 849.0, + 984.0, + 886.0, + 292.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 849.0, + 1406.0, + 849.0, + 1406.0, + 886.0, + 1021.0, + 886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 882.0, + 1406.0, + 882.0, + 1406.0, + 915.0, + 295.0, + 915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 909.0, + 1406.0, + 909.0, + 1406.0, + 949.0, + 292.0, + 949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 943.0, + 1289.0, + 943.0, + 1289.0, + 976.0, + 296.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 757.0, + 924.0, + 757.0, + 924.0, + 796.0, + 359.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1011.0, + 757.0, + 1338.0, + 757.0, + 1338.0, + 796.0, + 1011.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1758.0, + 527.0, + 1758.0, + 527.0, + 1807.0, + 235.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2060.0, + 861.0, + 2060.0, + 861.0, + 2090.0, + 839.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1860.0, + 278.0, + 1860.0, + 278.0, + 1885.0, + 240.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1852.0, + 1405.0, + 1852.0, + 1405.0, + 1891.0, + 292.0, + 1891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1891.0, + 278.0, + 1891.0, + 278.0, + 1916.0, + 240.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1885.0, + 1404.0, + 1885.0, + 1404.0, + 1919.0, + 295.0, + 1919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1920.0, + 278.0, + 1920.0, + 278.0, + 1945.0, + 240.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1912.0, + 1404.0, + 1912.0, + 1404.0, + 1950.0, + 294.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1952.0, + 278.0, + 1952.0, + 278.0, + 1977.0, + 240.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1942.0, + 1405.0, + 1942.0, + 1405.0, + 1982.0, + 292.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1981.0, + 278.0, + 1981.0, + 278.0, + 2006.0, + 240.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1974.0, + 1386.0, + 1974.0, + 1386.0, + 2010.0, + 291.0, + 2010.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 836, + 2061, + 866, + 2061, + 866, + 2086, + 836, + 2086 + ], + "score": 0.804 + }, + { + "category_id": 1, + "poly": [ + 240, + 209, + 1412, + 209, + 1412, + 2023, + 240, + 2023 + ], + "score": 0.778 + }, + { + "category_id": 0, + "poly": [ + 258, + 200, + 456, + 200, + 456, + 235, + 258, + 235 + ], + "score": 0.701 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 866, + 2061, + 866, + 2086, + 836, + 2086 + ], + "score": 0.134 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2096.0, + 832.0, + 2096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 199.0, + 459.0, + 199.0, + 459.0, + 240.0, + 252.0, + 240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2096.0, + 832.0, + 2096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 209.0, + 283.0, + 209.0, + 283.0, + 239.0, + 238.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 203.0, + 460.0, + 203.0, + 460.0, + 241.0, + 293.0, + 241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 255.0, + 283.0, + 255.0, + 283.0, + 291.0, + 236.0, + 291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 253.0, + 1406.0, + 253.0, + 1406.0, + 291.0, + 291.0, + 291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 287.0, + 283.0, + 287.0, + 283.0, + 322.0, + 236.0, + 322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 281.0, + 1406.0, + 281.0, + 1406.0, + 322.0, + 317.0, + 322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 319.0, + 279.0, + 319.0, + 279.0, + 350.0, + 238.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 315.0, + 1295.0, + 315.0, + 1295.0, + 352.0, + 321.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 376.0, + 281.0, + 376.0, + 281.0, + 406.0, + 238.0, + 406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 368.0, + 1410.0, + 368.0, + 1410.0, + 412.0, + 289.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 406.0, + 279.0, + 406.0, + 279.0, + 438.0, + 238.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 400.0, + 860.0, + 400.0, + 860.0, + 438.0, + 319.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 462.0, + 281.0, + 462.0, + 281.0, + 492.0, + 238.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 450.0, + 1410.0, + 450.0, + 1410.0, + 502.0, + 285.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 492.0, + 279.0, + 492.0, + 279.0, + 524.0, + 238.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 484.0, + 1406.0, + 484.0, + 1406.0, + 530.0, + 319.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 522.0, + 279.0, + 522.0, + 279.0, + 554.0, + 238.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 516.0, + 1239.0, + 516.0, + 1239.0, + 560.0, + 321.0, + 560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 578.0, + 281.0, + 578.0, + 281.0, + 608.0, + 238.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 572.0, + 1406.0, + 572.0, + 1406.0, + 612.0, + 289.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 608.0, + 279.0, + 608.0, + 279.0, + 639.0, + 238.0, + 639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 602.0, + 1410.0, + 602.0, + 1410.0, + 645.0, + 319.0, + 645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 639.0, + 279.0, + 639.0, + 279.0, + 671.0, + 238.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 315.0, + 632.0, + 1410.0, + 632.0, + 1410.0, + 673.0, + 315.0, + 673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 669.0, + 279.0, + 669.0, + 279.0, + 701.0, + 238.0, + 701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 663.0, + 637.0, + 663.0, + 637.0, + 701.0, + 321.0, + 701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 723.0, + 283.0, + 723.0, + 283.0, + 759.0, + 236.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 711.0, + 1408.0, + 711.0, + 1408.0, + 765.0, + 286.0, + 765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 755.0, + 283.0, + 755.0, + 283.0, + 791.0, + 234.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 747.0, + 1410.0, + 747.0, + 1410.0, + 793.0, + 317.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 787.0, + 279.0, + 787.0, + 279.0, + 819.0, + 238.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 779.0, + 1329.0, + 779.0, + 1329.0, + 821.0, + 313.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 843.0, + 281.0, + 843.0, + 281.0, + 873.0, + 238.0, + 873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 835.0, + 1408.0, + 835.0, + 1408.0, + 879.0, + 289.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 873.0, + 279.0, + 873.0, + 279.0, + 905.0, + 238.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 869.0, + 1353.0, + 869.0, + 1353.0, + 907.0, + 321.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 929.0, + 281.0, + 929.0, + 281.0, + 958.0, + 238.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 923.0, + 1408.0, + 923.0, + 1408.0, + 963.0, + 287.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 958.0, + 279.0, + 958.0, + 279.0, + 990.0, + 238.0, + 990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 955.0, + 1408.0, + 955.0, + 1408.0, + 992.0, + 321.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 988.0, + 279.0, + 988.0, + 279.0, + 1020.0, + 238.0, + 1020.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 984.0, + 631.0, + 984.0, + 631.0, + 1022.0, + 319.0, + 1022.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1046.0, + 281.0, + 1046.0, + 281.0, + 1076.0, + 238.0, + 1076.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1040.0, + 1406.0, + 1040.0, + 1406.0, + 1078.0, + 293.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1076.0, + 279.0, + 1076.0, + 279.0, + 1108.0, + 238.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1072.0, + 1174.0, + 1072.0, + 1174.0, + 1110.0, + 321.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1130.0, + 283.0, + 1130.0, + 283.0, + 1166.0, + 236.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1128.0, + 1408.0, + 1128.0, + 1408.0, + 1166.0, + 293.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1162.0, + 279.0, + 1162.0, + 279.0, + 1194.0, + 238.0, + 1194.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1156.0, + 1412.0, + 1156.0, + 1412.0, + 1200.0, + 319.0, + 1200.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1194.0, + 279.0, + 1194.0, + 279.0, + 1226.0, + 238.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1190.0, + 854.0, + 1190.0, + 854.0, + 1228.0, + 323.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1250.0, + 281.0, + 1250.0, + 281.0, + 1279.0, + 238.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1244.0, + 1408.0, + 1244.0, + 1408.0, + 1284.0, + 289.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1279.0, + 279.0, + 1279.0, + 279.0, + 1311.0, + 238.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1271.0, + 1410.0, + 1271.0, + 1410.0, + 1317.0, + 317.0, + 1317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1301.0, + 397.0, + 1301.0, + 397.0, + 1343.0, + 319.0, + 1343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1309.0, + 279.0, + 1309.0, + 279.0, + 1341.0, + 238.0, + 1341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1363.0, + 283.0, + 1363.0, + 283.0, + 1399.0, + 236.0, + 1399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1359.0, + 1406.0, + 1359.0, + 1406.0, + 1397.0, + 291.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1397.0, + 279.0, + 1397.0, + 279.0, + 1427.0, + 238.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1387.0, + 1128.0, + 1387.0, + 1128.0, + 1431.0, + 317.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1453.0, + 281.0, + 1453.0, + 281.0, + 1483.0, + 238.0, + 1483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1445.0, + 1406.0, + 1445.0, + 1406.0, + 1489.0, + 291.0, + 1489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1483.0, + 279.0, + 1483.0, + 279.0, + 1515.0, + 238.0, + 1515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1477.0, + 1406.0, + 1477.0, + 1406.0, + 1521.0, + 319.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1513.0, + 279.0, + 1513.0, + 279.0, + 1545.0, + 238.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1507.0, + 1410.0, + 1507.0, + 1410.0, + 1551.0, + 319.0, + 1551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1543.0, + 279.0, + 1543.0, + 279.0, + 1575.0, + 238.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1539.0, + 719.0, + 1539.0, + 719.0, + 1577.0, + 321.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1598.0, + 281.0, + 1598.0, + 281.0, + 1628.0, + 238.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1595.0, + 1406.0, + 1595.0, + 1406.0, + 1632.0, + 291.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1630.0, + 279.0, + 1630.0, + 279.0, + 1662.0, + 238.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1626.0, + 1404.0, + 1626.0, + 1404.0, + 1664.0, + 321.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1660.0, + 279.0, + 1660.0, + 279.0, + 1692.0, + 238.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1654.0, + 1408.0, + 1654.0, + 1408.0, + 1698.0, + 319.0, + 1698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1690.0, + 279.0, + 1690.0, + 279.0, + 1722.0, + 238.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1684.0, + 655.0, + 1684.0, + 655.0, + 1722.0, + 321.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1744.0, + 283.0, + 1744.0, + 283.0, + 1780.0, + 236.0, + 1780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1738.0, + 1408.0, + 1738.0, + 1408.0, + 1782.0, + 289.0, + 1782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1776.0, + 281.0, + 1776.0, + 281.0, + 1812.0, + 234.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1770.0, + 1406.0, + 1770.0, + 1406.0, + 1814.0, + 319.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1808.0, + 279.0, + 1808.0, + 279.0, + 1838.0, + 238.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1800.0, + 1283.0, + 1800.0, + 1283.0, + 1842.0, + 319.0, + 1842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1864.0, + 281.0, + 1864.0, + 281.0, + 1894.0, + 238.0, + 1894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1856.0, + 1410.0, + 1856.0, + 1410.0, + 1900.0, + 289.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1894.0, + 279.0, + 1894.0, + 279.0, + 1925.0, + 238.0, + 1925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1884.0, + 1412.0, + 1884.0, + 1412.0, + 1931.0, + 317.0, + 1931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1925.0, + 279.0, + 1925.0, + 279.0, + 1957.0, + 238.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 315.0, + 1916.0, + 1108.0, + 1916.0, + 1108.0, + 1961.0, + 315.0, + 1961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1975.0, + 1293.0, + 1975.0, + 1293.0, + 2013.0, + 240.0, + 2013.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 835, + 2060, + 864, + 2060, + 864, + 2087, + 835, + 2087 + ], + "score": 0.831 + }, + { + "category_id": 1, + "poly": [ + 236, + 131, + 1413, + 131, + 1413, + 2022, + 236, + 2022 + ], + "score": 0.826 + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2058.0, + 869.0, + 2058.0, + 869.0, + 2097.0, + 831.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 208.0, + 283.0, + 208.0, + 283.0, + 239.0, + 239.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 201.0, + 1408.0, + 201.0, + 1408.0, + 241.0, + 285.0, + 241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 239.0, + 281.0, + 239.0, + 281.0, + 270.0, + 236.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 232.0, + 1408.0, + 232.0, + 1408.0, + 272.0, + 320.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 259.0, + 400.0, + 259.0, + 400.0, + 301.0, + 318.0, + 301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 268.0, + 281.0, + 268.0, + 281.0, + 299.0, + 236.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 230.0, + 315.0, + 1408.0, + 315.0, + 1408.0, + 359.0, + 230.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 355.0, + 281.0, + 355.0, + 281.0, + 386.0, + 239.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 346.0, + 1392.0, + 346.0, + 1392.0, + 386.0, + 316.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 409.0, + 283.0, + 409.0, + 283.0, + 440.0, + 236.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 407.0, + 1406.0, + 407.0, + 1406.0, + 440.0, + 293.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 440.0, + 281.0, + 440.0, + 281.0, + 471.0, + 236.0, + 471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 438.0, + 1408.0, + 438.0, + 1408.0, + 471.0, + 318.0, + 471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 471.0, + 281.0, + 471.0, + 281.0, + 502.0, + 236.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 463.0, + 635.0, + 463.0, + 635.0, + 502.0, + 320.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 527.0, + 283.0, + 527.0, + 283.0, + 556.0, + 239.0, + 556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 523.0, + 1211.0, + 523.0, + 1211.0, + 556.0, + 291.0, + 556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 573.0, + 1408.0, + 573.0, + 1408.0, + 614.0, + 236.0, + 614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 612.0, + 281.0, + 612.0, + 281.0, + 643.0, + 236.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 606.0, + 1404.0, + 606.0, + 1404.0, + 645.0, + 320.0, + 645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 641.0, + 278.0, + 641.0, + 278.0, + 674.0, + 236.0, + 674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 637.0, + 675.0, + 637.0, + 675.0, + 676.0, + 318.0, + 676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 697.0, + 283.0, + 697.0, + 283.0, + 728.0, + 239.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 691.0, + 1406.0, + 691.0, + 1406.0, + 730.0, + 293.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 728.0, + 281.0, + 728.0, + 281.0, + 761.0, + 236.0, + 761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 724.0, + 1047.0, + 724.0, + 1047.0, + 761.0, + 316.0, + 761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 782.0, + 283.0, + 782.0, + 283.0, + 813.0, + 236.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 780.0, + 1406.0, + 780.0, + 1406.0, + 813.0, + 295.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 813.0, + 281.0, + 813.0, + 281.0, + 844.0, + 236.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 809.0, + 1408.0, + 809.0, + 1408.0, + 848.0, + 320.0, + 848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 844.0, + 281.0, + 844.0, + 281.0, + 875.0, + 236.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 838.0, + 1224.0, + 838.0, + 1224.0, + 877.0, + 320.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 896.0, + 1404.0, + 896.0, + 1404.0, + 929.0, + 239.0, + 929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 929.0, + 281.0, + 929.0, + 281.0, + 960.0, + 236.0, + 960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 927.0, + 1404.0, + 927.0, + 1404.0, + 960.0, + 323.0, + 960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 960.0, + 281.0, + 960.0, + 281.0, + 991.0, + 236.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 954.0, + 1278.0, + 954.0, + 1278.0, + 994.0, + 320.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1016.0, + 283.0, + 1016.0, + 283.0, + 1047.0, + 236.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1010.0, + 1406.0, + 1010.0, + 1406.0, + 1050.0, + 293.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1043.0, + 281.0, + 1043.0, + 281.0, + 1077.0, + 236.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1041.0, + 1408.0, + 1041.0, + 1408.0, + 1081.0, + 320.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1074.0, + 281.0, + 1074.0, + 281.0, + 1108.0, + 236.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1070.0, + 1408.0, + 1070.0, + 1408.0, + 1110.0, + 320.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1099.0, + 1410.0, + 1099.0, + 1410.0, + 1139.0, + 320.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1137.0, + 281.0, + 1137.0, + 281.0, + 1170.0, + 236.0, + 1170.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1132.0, + 965.0, + 1132.0, + 965.0, + 1166.0, + 325.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1191.0, + 283.0, + 1191.0, + 283.0, + 1222.0, + 239.0, + 1222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1180.0, + 1408.0, + 1180.0, + 1408.0, + 1228.0, + 289.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1222.0, + 281.0, + 1222.0, + 281.0, + 1255.0, + 236.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1218.0, + 1410.0, + 1218.0, + 1410.0, + 1257.0, + 320.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1253.0, + 281.0, + 1253.0, + 281.0, + 1284.0, + 236.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1251.0, + 764.0, + 1251.0, + 764.0, + 1284.0, + 323.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1296.0, + 1411.0, + 1296.0, + 1411.0, + 1346.0, + 236.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1338.0, + 281.0, + 1338.0, + 281.0, + 1371.0, + 236.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1334.0, + 1408.0, + 1334.0, + 1408.0, + 1373.0, + 320.0, + 1373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1369.0, + 281.0, + 1369.0, + 281.0, + 1402.0, + 236.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1367.0, + 743.0, + 1367.0, + 743.0, + 1400.0, + 323.0, + 1400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1425.0, + 283.0, + 1425.0, + 283.0, + 1456.0, + 236.0, + 1456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1419.0, + 1406.0, + 1419.0, + 1406.0, + 1458.0, + 295.0, + 1458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1454.0, + 281.0, + 1454.0, + 281.0, + 1487.0, + 236.0, + 1487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 1446.0, + 1411.0, + 1446.0, + 1411.0, + 1493.0, + 316.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1485.0, + 281.0, + 1485.0, + 281.0, + 1516.0, + 236.0, + 1516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1477.0, + 696.0, + 1477.0, + 696.0, + 1518.0, + 321.0, + 1518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1535.0, + 1406.0, + 1535.0, + 1406.0, + 1574.0, + 236.0, + 1574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1570.0, + 278.0, + 1570.0, + 278.0, + 1603.0, + 236.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1568.0, + 1408.0, + 1568.0, + 1408.0, + 1601.0, + 323.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1601.0, + 281.0, + 1601.0, + 281.0, + 1632.0, + 236.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1597.0, + 415.0, + 1597.0, + 415.0, + 1632.0, + 323.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1657.0, + 283.0, + 1657.0, + 283.0, + 1688.0, + 236.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 1649.0, + 1408.0, + 1649.0, + 1408.0, + 1690.0, + 287.0, + 1690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1686.0, + 281.0, + 1686.0, + 281.0, + 1719.0, + 236.0, + 1719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1682.0, + 1408.0, + 1682.0, + 1408.0, + 1722.0, + 320.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1715.0, + 281.0, + 1715.0, + 281.0, + 1748.0, + 236.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 1707.0, + 1408.0, + 1707.0, + 1408.0, + 1755.0, + 316.0, + 1755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1740.0, + 400.0, + 1740.0, + 400.0, + 1782.0, + 320.0, + 1782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1748.0, + 281.0, + 1748.0, + 281.0, + 1780.0, + 236.0, + 1780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1802.0, + 283.0, + 1802.0, + 283.0, + 1833.0, + 236.0, + 1833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 1794.0, + 1406.0, + 1794.0, + 1406.0, + 1836.0, + 287.0, + 1836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1833.0, + 281.0, + 1833.0, + 281.0, + 1865.0, + 236.0, + 1865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1827.0, + 1406.0, + 1827.0, + 1406.0, + 1867.0, + 323.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1865.0, + 281.0, + 1865.0, + 281.0, + 1896.0, + 236.0, + 1896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 1856.0, + 1307.0, + 1856.0, + 1307.0, + 1898.0, + 316.0, + 1898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1919.0, + 281.0, + 1919.0, + 281.0, + 1950.0, + 236.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1912.0, + 1408.0, + 1912.0, + 1408.0, + 1952.0, + 291.0, + 1952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1950.0, + 278.0, + 1950.0, + 278.0, + 1981.0, + 236.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1945.0, + 1408.0, + 1945.0, + 1408.0, + 1985.0, + 320.0, + 1985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1981.0, + 281.0, + 1981.0, + 281.0, + 2012.0, + 236.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1975.0, + 795.0, + 1975.0, + 795.0, + 2014.0, + 320.0, + 2014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1103.5, + 277.0, + 1103.5, + 277.0, + 1137.0, + 236.0, + 1137.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 407, + 1390, + 1404, + 1390, + 1404, + 1657, + 407, + 1657 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 407, + 1048, + 1405, + 1048, + 1405, + 1339, + 407, + 1339 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 407, + 618, + 1404, + 618, + 1404, + 879, + 407, + 879 + ], + "score": 0.969 + }, + { + "category_id": 1, + "poly": [ + 406, + 1706, + 1404, + 1706, + 1404, + 1903, + 406, + 1903 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 406, + 929, + 1257, + 929, + 1257, + 999, + 406, + 999 + ], + "score": 0.936 + }, + { + "category_id": 1, + "poly": [ + 365, + 888, + 840, + 888, + 840, + 920, + 365, + 920 + ], + "score": 0.911 + }, + { + "category_id": 1, + "poly": [ + 363, + 1007, + 670, + 1007, + 670, + 1039, + 363, + 1039 + ], + "score": 0.904 + }, + { + "category_id": 1, + "poly": [ + 364, + 577, + 580, + 577, + 580, + 608, + 364, + 608 + ], + "score": 0.895 + }, + { + "category_id": 1, + "poly": [ + 266, + 202, + 1407, + 202, + 1407, + 462, + 266, + 462 + ], + "score": 0.873 + }, + { + "category_id": 1, + "poly": [ + 364, + 1666, + 1200, + 1666, + 1200, + 1698, + 364, + 1698 + ], + "score": 0.866 + }, + { + "category_id": 1, + "poly": [ + 369, + 1349, + 1401, + 1349, + 1401, + 1381, + 369, + 1381 + ], + "score": 0.863 + }, + { + "category_id": 0, + "poly": [ + 246, + 517, + 435, + 517, + 435, + 553, + 246, + 553 + ], + "score": 0.824 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2085, + 836, + 2085 + ], + "score": 0.82 + }, + { + "category_id": 2, + "poly": [ + 237, + 246, + 286, + 246, + 286, + 1944, + 237, + 1944 + ], + "score": 0.17 + }, + { + "category_id": 2, + "poly": [ + 239, + 204, + 279, + 204, + 279, + 461, + 239, + 461 + ], + "score": 0.113 + }, + { + "category_id": 15, + "poly": [ + 241.0, + 525.0, + 281.0, + 525.0, + 281.0, + 553.0, + 241.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 514.0, + 439.0, + 514.0, + 439.0, + 557.0, + 289.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 237.0, + 285.0, + 237.0, + 285.0, + 273.0, + 234.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 290.0, + 280.0, + 290.0, + 280.0, + 321.0, + 236.0, + 321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 320.0, + 282.0, + 320.0, + 282.0, + 363.0, + 234.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 372.0, + 280.0, + 372.0, + 280.0, + 404.0, + 236.0, + 404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 404.0, + 280.0, + 404.0, + 280.0, + 436.0, + 236.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 432.0, + 282.0, + 432.0, + 282.0, + 469.0, + 234.0, + 469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 528.0, + 280.0, + 528.0, + 280.0, + 559.0, + 236.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 580.0, + 282.0, + 580.0, + 282.0, + 617.0, + 234.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 623.0, + 280.0, + 623.0, + 280.0, + 655.0, + 236.0, + 655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 653.0, + 280.0, + 653.0, + 280.0, + 685.0, + 236.0, + 685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 687.0, + 282.0, + 687.0, + 282.0, + 724.0, + 234.0, + 724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 722.0, + 285.0, + 722.0, + 285.0, + 758.0, + 234.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 754.0, + 285.0, + 754.0, + 285.0, + 790.0, + 234.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 786.0, + 280.0, + 786.0, + 280.0, + 818.0, + 238.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 822.0, + 282.0, + 822.0, + 282.0, + 852.0, + 236.0, + 852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 850.0, + 282.0, + 850.0, + 282.0, + 887.0, + 234.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 891.0, + 282.0, + 891.0, + 282.0, + 932.0, + 234.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 934.0, + 278.0, + 934.0, + 278.0, + 966.0, + 236.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 970.0, + 285.0, + 970.0, + 285.0, + 1005.0, + 234.0, + 1005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1011.0, + 282.0, + 1011.0, + 282.0, + 1050.0, + 234.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1054.0, + 282.0, + 1054.0, + 282.0, + 1084.0, + 236.0, + 1084.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1084.0, + 282.0, + 1084.0, + 282.0, + 1114.0, + 236.0, + 1114.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1114.0, + 285.0, + 1114.0, + 285.0, + 1149.0, + 234.0, + 1149.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1149.0, + 280.0, + 1149.0, + 280.0, + 1181.0, + 236.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1181.0, + 282.0, + 1181.0, + 282.0, + 1211.0, + 236.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1217.0, + 282.0, + 1217.0, + 282.0, + 1247.0, + 236.0, + 1247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1247.0, + 280.0, + 1247.0, + 280.0, + 1279.0, + 236.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1282.0, + 278.0, + 1282.0, + 278.0, + 1312.0, + 236.0, + 1312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1310.0, + 282.0, + 1310.0, + 282.0, + 1352.0, + 234.0, + 1352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1352.0, + 282.0, + 1352.0, + 282.0, + 1389.0, + 234.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1393.0, + 282.0, + 1393.0, + 282.0, + 1430.0, + 234.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1428.0, + 282.0, + 1428.0, + 282.0, + 1466.0, + 234.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1464.0, + 285.0, + 1464.0, + 285.0, + 1500.0, + 234.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1494.0, + 282.0, + 1494.0, + 282.0, + 1531.0, + 234.0, + 1531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1531.0, + 285.0, + 1531.0, + 285.0, + 1567.0, + 234.0, + 1567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1563.0, + 282.0, + 1563.0, + 282.0, + 1599.0, + 238.0, + 1599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1599.0, + 280.0, + 1599.0, + 280.0, + 1629.0, + 236.0, + 1629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1629.0, + 280.0, + 1629.0, + 280.0, + 1664.0, + 236.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1668.0, + 282.0, + 1668.0, + 282.0, + 1707.0, + 234.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1711.0, + 280.0, + 1711.0, + 280.0, + 1743.0, + 236.0, + 1743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1741.0, + 285.0, + 1741.0, + 285.0, + 1779.0, + 234.0, + 1779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1777.0, + 280.0, + 1777.0, + 280.0, + 1809.0, + 236.0, + 1809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1807.0, + 285.0, + 1807.0, + 285.0, + 1842.0, + 234.0, + 1842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1844.0, + 280.0, + 1844.0, + 280.0, + 1876.0, + 236.0, + 1876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1874.0, + 280.0, + 1874.0, + 280.0, + 1906.0, + 236.0, + 1906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 208.0, + 279.0, + 208.0, + 279.0, + 238.0, + 237.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 237.0, + 281.0, + 237.0, + 281.0, + 273.0, + 236.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 290.0, + 279.0, + 290.0, + 279.0, + 322.0, + 237.0, + 322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 319.0, + 281.0, + 319.0, + 281.0, + 356.0, + 236.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 373.0, + 279.0, + 373.0, + 279.0, + 405.0, + 237.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 403.0, + 279.0, + 403.0, + 279.0, + 436.0, + 237.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 434.0, + 279.0, + 434.0, + 279.0, + 466.0, + 237.0, + 466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1388.0, + 1342.0, + 1388.0, + 1342.0, + 1424.0, + 404.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1425.0, + 1408.0, + 1425.0, + 1408.0, + 1460.0, + 404.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 1460.0, + 1404.0, + 1460.0, + 1404.0, + 1496.0, + 407.0, + 1496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 1493.0, + 987.0, + 1493.0, + 987.0, + 1525.0, + 451.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1526.0, + 1405.0, + 1526.0, + 1405.0, + 1565.0, + 406.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 1559.0, + 706.0, + 1559.0, + 706.0, + 1592.0, + 451.0, + 1592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 1593.0, + 1403.0, + 1593.0, + 1403.0, + 1629.0, + 407.0, + 1629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1623.0, + 903.0, + 1623.0, + 903.0, + 1658.0, + 450.0, + 1658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1047.0, + 1408.0, + 1047.0, + 1408.0, + 1081.0, + 405.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1079.0, + 1404.0, + 1079.0, + 1404.0, + 1110.0, + 450.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1109.0, + 1376.0, + 1109.0, + 1376.0, + 1140.0, + 450.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1143.0, + 1405.0, + 1143.0, + 1405.0, + 1180.0, + 406.0, + 1180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 1176.0, + 848.0, + 1176.0, + 848.0, + 1207.0, + 448.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 1207.0, + 1410.0, + 1207.0, + 1410.0, + 1248.0, + 408.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1240.0, + 977.0, + 1240.0, + 977.0, + 1275.0, + 450.0, + 1275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 1271.0, + 1405.0, + 1271.0, + 1405.0, + 1315.0, + 407.0, + 1315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 1308.0, + 1192.0, + 1308.0, + 1192.0, + 1339.0, + 452.0, + 1339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 617.0, + 1401.0, + 617.0, + 1401.0, + 651.0, + 406.0, + 651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 648.0, + 802.0, + 648.0, + 802.0, + 683.0, + 450.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 682.0, + 1212.0, + 682.0, + 1212.0, + 720.0, + 402.0, + 720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 717.0, + 1405.0, + 717.0, + 1405.0, + 755.0, + 404.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 447.0, + 748.0, + 1406.0, + 748.0, + 1406.0, + 786.0, + 447.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 780.0, + 603.0, + 780.0, + 603.0, + 816.0, + 450.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 813.0, + 1405.0, + 813.0, + 1405.0, + 855.0, + 402.0, + 855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 846.0, + 596.0, + 846.0, + 596.0, + 882.0, + 450.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1704.0, + 1408.0, + 1704.0, + 1408.0, + 1742.0, + 404.0, + 1742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 1738.0, + 660.0, + 1738.0, + 660.0, + 1772.0, + 451.0, + 1772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1772.0, + 1405.0, + 1772.0, + 1405.0, + 1808.0, + 405.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1803.0, + 951.0, + 1803.0, + 951.0, + 1839.0, + 449.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 409.0, + 1838.0, + 1405.0, + 1838.0, + 1405.0, + 1874.0, + 409.0, + 1874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1871.0, + 927.0, + 1871.0, + 927.0, + 1905.0, + 450.0, + 1905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 928.0, + 1259.0, + 928.0, + 1259.0, + 964.0, + 405.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 964.0, + 1180.0, + 964.0, + 1180.0, + 1001.0, + 403.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 885.0, + 846.0, + 885.0, + 846.0, + 925.0, + 359.0, + 925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 1001.0, + 675.0, + 1001.0, + 675.0, + 1046.0, + 358.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 572.0, + 586.0, + 572.0, + 586.0, + 612.0, + 361.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 268.0, + 201.0, + 1404.0, + 201.0, + 1404.0, + 243.0, + 268.0, + 243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 261.0, + 245.0, + 275.0, + 245.0, + 275.0, + 261.0, + 261.0, + 261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 230.0, + 466.0, + 230.0, + 466.0, + 270.0, + 321.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 264.0, + 299.0, + 275.0, + 299.0, + 275.0, + 312.0, + 264.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 285.0, + 1407.0, + 285.0, + 1407.0, + 322.0, + 287.0, + 322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 260.0, + 327.0, + 275.0, + 327.0, + 275.0, + 346.0, + 260.0, + 346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 316.0, + 1201.0, + 316.0, + 1201.0, + 353.0, + 322.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 366.0, + 1405.0, + 366.0, + 1405.0, + 407.0, + 287.0, + 407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 261.0, + 410.0, + 275.0, + 410.0, + 275.0, + 427.0, + 261.0, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 397.0, + 1413.0, + 397.0, + 1413.0, + 439.0, + 321.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 261.0, + 441.0, + 275.0, + 441.0, + 275.0, + 458.0, + 261.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 429.0, + 696.0, + 429.0, + 696.0, + 468.0, + 322.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 1661.0, + 1205.0, + 1661.0, + 1205.0, + 1705.0, + 357.0, + 1705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1344.0, + 1409.0, + 1344.0, + 1409.0, + 1387.0, + 361.0, + 1387.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/vlm/train/0-EYBhgw80y/0.png b/vlm/train/0-EYBhgw80y/0.png new file mode 100644 index 0000000000000000000000000000000000000000..1143c074911e1dd5a5aa9bc005f48a268f66bb97 --- /dev/null +++ b/vlm/train/0-EYBhgw80y/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3fdf3b60683fde518a569043ac6b8371f6c45bcd9742872418e61d688b6bc21d +size 501157 diff --git a/vlm/train/0-EYBhgw80y/1.png b/vlm/train/0-EYBhgw80y/1.png new file mode 100644 index 0000000000000000000000000000000000000000..b80916856f561f7c779b487c9c5f9457a2021712 --- /dev/null +++ b/vlm/train/0-EYBhgw80y/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:be2c24d6bb9e50130a786f14626232cb27b002db3f8bfa47ba1e78cf00c5de2a +size 607887 diff --git a/vlm/train/0-EYBhgw80y/10.png b/vlm/train/0-EYBhgw80y/10.png new file mode 100644 index 0000000000000000000000000000000000000000..47e3c3323cfa96be828ac389fefe08fedfd88a71 --- /dev/null +++ b/vlm/train/0-EYBhgw80y/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5baf7ea71c5762ad4dbc2f8d5e6b7ec1bf411740321eee56b9e3f5eb5e63b29a +size 464453 diff --git a/vlm/train/0-EYBhgw80y/11.png b/vlm/train/0-EYBhgw80y/11.png new file mode 100644 index 0000000000000000000000000000000000000000..e60fe7b4032be9938438be5939d9a4b7240bbfa2 --- /dev/null +++ b/vlm/train/0-EYBhgw80y/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:307669e69483acdfe6caf468b3581dfe4d114c384b8944f540aad3bbb7fc5311 +size 1593324 diff --git a/vlm/train/0-EYBhgw80y/12.png b/vlm/train/0-EYBhgw80y/12.png new file mode 100644 index 0000000000000000000000000000000000000000..4fa183ecfba1896ed89fb4e1f6537ff319eab830 --- /dev/null +++ b/vlm/train/0-EYBhgw80y/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7f3de41851bb4c76666159d5140ac83c6a38dbeb2979d7902dd64bcd0ea0646e +size 1757404 diff --git a/vlm/train/0-EYBhgw80y/13.png b/vlm/train/0-EYBhgw80y/13.png new file mode 100644 index 0000000000000000000000000000000000000000..d030cec6b5130b17b5c6c28e61f5689eb04546de --- /dev/null +++ b/vlm/train/0-EYBhgw80y/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d589b13d32bdb9d50776ee02502bb5c0d084e202c679a4528fae04e74d3a7460 +size 401742 diff --git a/vlm/train/0-EYBhgw80y/14.png b/vlm/train/0-EYBhgw80y/14.png new file mode 100644 index 0000000000000000000000000000000000000000..ebe9e356ab002d2862bc4355d8ed850967c8c0c1 --- /dev/null +++ b/vlm/train/0-EYBhgw80y/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:242ef74ea989cf9b878ba05f3f37ab635c501ee88c9faae2f480d057e9285c48 +size 178971 diff --git a/vlm/train/0-EYBhgw80y/2.png b/vlm/train/0-EYBhgw80y/2.png new file mode 100644 index 0000000000000000000000000000000000000000..68d97ad17699de2da0149872ed1f46b2e992dd97 --- /dev/null +++ b/vlm/train/0-EYBhgw80y/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:053f21bd5b0003cc5fb80582105a828008f6d14d0d32c320db9e89012ceebf72 +size 504927 diff --git a/vlm/train/0-EYBhgw80y/3.png b/vlm/train/0-EYBhgw80y/3.png new file mode 100644 index 0000000000000000000000000000000000000000..28f40814bfa01c99f397571da11b51b283137420 --- /dev/null +++ b/vlm/train/0-EYBhgw80y/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3dd9b7043faa7bd134749877a4894b7ef75d6aae0f6f2baabeb8fc4d52955a76 +size 457149 diff --git a/vlm/train/0-EYBhgw80y/4.png b/vlm/train/0-EYBhgw80y/4.png new file mode 100644 index 0000000000000000000000000000000000000000..70745f8708bbde6ba9c28f080a3b69f8e7a5502f --- /dev/null +++ b/vlm/train/0-EYBhgw80y/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f4dbe848f7284ce25047939d410c77f35ebb1b06d144f4b5c08e9e13c7a6e3d7 +size 448252 diff --git a/vlm/train/0-EYBhgw80y/5.png b/vlm/train/0-EYBhgw80y/5.png new file mode 100644 index 0000000000000000000000000000000000000000..0affd631e4a433c6fd310c74eb88e3b6a49b917c --- /dev/null +++ b/vlm/train/0-EYBhgw80y/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dcd797b9a6577d91735d57a2bd4fbc886675a4356c94930570bc77b75403be2b +size 533620 diff --git a/vlm/train/0-EYBhgw80y/6.png b/vlm/train/0-EYBhgw80y/6.png new file mode 100644 index 0000000000000000000000000000000000000000..b87ef17886d6c787953b85d7d67e3c6928492cd9 --- /dev/null +++ b/vlm/train/0-EYBhgw80y/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eef8fe85fc1c092caf43d9f06599161f3ee172e4f0e942b1b77f72d08c439f10 +size 540910 diff --git a/vlm/train/0-EYBhgw80y/7.png b/vlm/train/0-EYBhgw80y/7.png new file mode 100644 index 0000000000000000000000000000000000000000..b4c418616992af7033193a6c4cf48c57506866b0 --- /dev/null +++ b/vlm/train/0-EYBhgw80y/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ac738211b88da9352f08e35d19a7d7d7d7125ebbfa057f42c4ed689d504ae094 +size 489202 diff --git a/vlm/train/0-EYBhgw80y/8.png b/vlm/train/0-EYBhgw80y/8.png new file mode 100644 index 0000000000000000000000000000000000000000..82944fc185e8d1f710eed3e6b4f93ae449c11ca0 --- /dev/null +++ b/vlm/train/0-EYBhgw80y/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:764d0a74214e0aa1decd21f33ca9d6c1f71c783500b4f158723f5d8fde7d3051 +size 506508 diff --git a/vlm/train/0-EYBhgw80y/9.png b/vlm/train/0-EYBhgw80y/9.png new file mode 100644 index 0000000000000000000000000000000000000000..bc923b8402f859a1996b97a92baa6f1b6c7c9063 --- /dev/null +++ b/vlm/train/0-EYBhgw80y/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0d9f183a53c8dcc8bdbf427eda238a5fd835053114d4151177c4f060ab6e2865 +size 507475 diff --git a/vlm/train/3SV-ZePhnZM/0.png b/vlm/train/3SV-ZePhnZM/0.png new file mode 100644 index 0000000000000000000000000000000000000000..fbe36b79b5418341bfffbe2e419d5515677e7a50 --- /dev/null +++ b/vlm/train/3SV-ZePhnZM/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:43e5d5625fe86375d22768c6146a1b3a1ff2a13ed91cb22e0c19b13f0258a6bb +size 477007 diff --git a/vlm/train/3SV-ZePhnZM/1.png b/vlm/train/3SV-ZePhnZM/1.png new file mode 100644 index 0000000000000000000000000000000000000000..26b66eef1a7e55e3b2c38b2c65add8303e5dfc46 --- /dev/null +++ b/vlm/train/3SV-ZePhnZM/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f85d0b27a2e10e0d9855e01d15cbb06dad49e881de9591441d97b2cf6e1919a8 +size 613522 diff --git a/vlm/train/3SV-ZePhnZM/10.png b/vlm/train/3SV-ZePhnZM/10.png new file mode 100644 index 0000000000000000000000000000000000000000..4133014af984f781e116c3bbc31ff403b562862b --- /dev/null +++ b/vlm/train/3SV-ZePhnZM/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:269310edc3bda69c8038fcdea9fc9f91971690ce20e7207969f36091a53ec794 +size 581375 diff --git a/vlm/train/3SV-ZePhnZM/11.png b/vlm/train/3SV-ZePhnZM/11.png new file mode 100644 index 0000000000000000000000000000000000000000..053c0f442e15c66722e4cf7acf2e283c3e16f0f1 --- /dev/null +++ b/vlm/train/3SV-ZePhnZM/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ce3ba2a016dc4c51ff1bb083d4983509fd259d96c3c96cb87c736aceefa6ee09 +size 405250 diff --git a/vlm/train/3SV-ZePhnZM/12.png b/vlm/train/3SV-ZePhnZM/12.png new file mode 100644 index 0000000000000000000000000000000000000000..aaa0a76fdfa4aff0a28ca270c3cd5b0cc5b0c71d --- /dev/null +++ b/vlm/train/3SV-ZePhnZM/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c4d7be6bdf839f2dcd09f66635766c8d50713f7f766ec0af34ff8fa30099a2d4 +size 519104 diff --git a/vlm/train/3SV-ZePhnZM/13.png b/vlm/train/3SV-ZePhnZM/13.png new file mode 100644 index 0000000000000000000000000000000000000000..875e8cef6e4f3a0f7a27b296cc4a3ba80a6445b6 --- /dev/null +++ b/vlm/train/3SV-ZePhnZM/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cc442783c381400d2a9ba1d978532191efe011662f274dae40ca76a2de4f7664 +size 455846 diff --git a/vlm/train/3SV-ZePhnZM/14.png b/vlm/train/3SV-ZePhnZM/14.png new file mode 100644 index 0000000000000000000000000000000000000000..b70f964f1f95ce261a7bcf7e88580aa76d49b9cc --- /dev/null +++ b/vlm/train/3SV-ZePhnZM/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c5fd812695b49f3203f6e6f0dcc1fc418e6955e131219a6a6f4a6b782585d990 +size 287794 diff --git a/vlm/train/3SV-ZePhnZM/15.png b/vlm/train/3SV-ZePhnZM/15.png new file mode 100644 index 0000000000000000000000000000000000000000..451e313226d8e016eebd080015d3dd4bbd88086a --- /dev/null +++ b/vlm/train/3SV-ZePhnZM/15.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:210bfb8edbb994595ebf392b493cbdeaaa7366e60caa843c1bc879e7b80c365d +size 359405 diff --git a/vlm/train/3SV-ZePhnZM/2.png b/vlm/train/3SV-ZePhnZM/2.png new file mode 100644 index 0000000000000000000000000000000000000000..8c9ffcbcf9ee6bb9d0add8770159ff4c2a664a96 --- /dev/null +++ b/vlm/train/3SV-ZePhnZM/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:be4f646ae6020f450d00e178b611dbc7ef6221792c83185fe272218e13d293cd +size 541406 diff --git a/vlm/train/3SV-ZePhnZM/3.png b/vlm/train/3SV-ZePhnZM/3.png new file mode 100644 index 0000000000000000000000000000000000000000..b40b2ea0906f6389a2169b97ef2ba8735443aaa4 --- /dev/null +++ b/vlm/train/3SV-ZePhnZM/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3004fbbfb680c435f29e3f97305e7ba85159ab1de2367990468dfd7185f34363 +size 544312 diff --git a/vlm/train/3SV-ZePhnZM/4.png b/vlm/train/3SV-ZePhnZM/4.png new file mode 100644 index 0000000000000000000000000000000000000000..2abe566a1208311f01e6e404ec21e48e88947d7d --- /dev/null +++ b/vlm/train/3SV-ZePhnZM/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3f9fa942c9b219a0c1a1aa5ba659efd1e49976775d761988f4bc113bb36f7593 +size 543410 diff --git a/vlm/train/3SV-ZePhnZM/5.png b/vlm/train/3SV-ZePhnZM/5.png new file mode 100644 index 0000000000000000000000000000000000000000..eb625e51ba7d7d6e995cb8ba48f17fc49ad8b514 --- /dev/null +++ b/vlm/train/3SV-ZePhnZM/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8214b0d3290bebfb3c6020a81a84d933bf10a258a97fef8fd247118c5e9c3174 +size 507619 diff --git a/vlm/train/3SV-ZePhnZM/6.png b/vlm/train/3SV-ZePhnZM/6.png new file mode 100644 index 0000000000000000000000000000000000000000..929d7317a79e87b71bea3ff586bf0632a0891c68 --- /dev/null +++ b/vlm/train/3SV-ZePhnZM/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2e32d8d2f658e58dfe592ae68dec0317e56dba424e1a242433ed307f52a0736f +size 553234 diff --git a/vlm/train/3SV-ZePhnZM/7.png b/vlm/train/3SV-ZePhnZM/7.png new file mode 100644 index 0000000000000000000000000000000000000000..f7539bbe029212b0565f1c174d71b620744a6b81 --- /dev/null +++ b/vlm/train/3SV-ZePhnZM/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e3c2bd87c5700f32807b3160e9708c526f96405ac5b59926d31312d64ff579ee +size 524337 diff --git a/vlm/train/3SV-ZePhnZM/8.png b/vlm/train/3SV-ZePhnZM/8.png new file mode 100644 index 0000000000000000000000000000000000000000..1442e84d4734062d797b1cfc543eed815f4e2b7e --- /dev/null +++ b/vlm/train/3SV-ZePhnZM/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ce20efe360b2cb1e64d82e2ba6805dc63f6081c7c50eaaa05f65647bce766a33 +size 500731 diff --git a/vlm/train/3SV-ZePhnZM/9.png b/vlm/train/3SV-ZePhnZM/9.png new file mode 100644 index 0000000000000000000000000000000000000000..ef902e5e38c2737148e23a7367d034db99331fe7 --- /dev/null +++ b/vlm/train/3SV-ZePhnZM/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4ebd8cb0093a340af22971471d76bd4eadac8b5a791e606580b8e7657b4bf5fb +size 560181 diff --git a/vlm/train/4c0J6lwQ4_/0.png b/vlm/train/4c0J6lwQ4_/0.png new file mode 100644 index 0000000000000000000000000000000000000000..6c65f96398bb4a4bec1043cee6cfaa3ed5001f93 --- /dev/null +++ b/vlm/train/4c0J6lwQ4_/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:82e9d985270f55e90115df882a8108ecec1900d18e73e544d05fa48a34d42b0b +size 498716 diff --git a/vlm/train/4c0J6lwQ4_/1.png b/vlm/train/4c0J6lwQ4_/1.png new file mode 100644 index 0000000000000000000000000000000000000000..58a9e3c05c67e1c1b6447f1a034fdb9c27054a68 --- /dev/null +++ b/vlm/train/4c0J6lwQ4_/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:43fe73f910b9032f4c54b36b7f832c1224ca15ead941be38ce0d068d04972658 +size 602149 diff --git a/vlm/train/4c0J6lwQ4_/10.png b/vlm/train/4c0J6lwQ4_/10.png new file mode 100644 index 0000000000000000000000000000000000000000..c8ea83d3fbe45b9e6ab607338ad73597688c17b0 --- /dev/null +++ b/vlm/train/4c0J6lwQ4_/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:488d8f308f28b4005caa9f3bfbf475ed7c77157811fa55482af9e395fc9a4d82 +size 585724 diff --git a/vlm/train/4c0J6lwQ4_/11.png b/vlm/train/4c0J6lwQ4_/11.png new file mode 100644 index 0000000000000000000000000000000000000000..943deecda5f261fc0dffbdff667942648337027c --- /dev/null +++ b/vlm/train/4c0J6lwQ4_/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:74b34de21c758d4265f7d83beb00d13409ae898aa3ecdfc83f3dc9ed655487ae +size 437345 diff --git a/vlm/train/4c0J6lwQ4_/14.png b/vlm/train/4c0J6lwQ4_/14.png new file mode 100644 index 0000000000000000000000000000000000000000..a1ddea43cc8e482eee6bfdb9caba7c9fcb7e0ed8 --- /dev/null +++ b/vlm/train/4c0J6lwQ4_/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a3b7642914e9a5e6db34b2a3b33e0995cf4be8bcade7231993ff3d64b34ebcc7 +size 125725 diff --git a/vlm/train/4c0J6lwQ4_/2.png b/vlm/train/4c0J6lwQ4_/2.png new file mode 100644 index 0000000000000000000000000000000000000000..16bd2371f9a04e2552753f412bc363b321006023 --- /dev/null +++ b/vlm/train/4c0J6lwQ4_/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:89345fafe350742e1c3e422a408ab9ded5828e27ce6c820e8bc03a881789a281 +size 555504 diff --git a/vlm/train/4c0J6lwQ4_/3.png b/vlm/train/4c0J6lwQ4_/3.png new file mode 100644 index 0000000000000000000000000000000000000000..164a17e35a727284b825893bde3865efbcabc616 --- /dev/null +++ b/vlm/train/4c0J6lwQ4_/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0c40b458325d07883921eb990f7d419c3b9aa831619bae6c0d8ba4a065372059 +size 479098 diff --git a/vlm/train/4c0J6lwQ4_/6.png b/vlm/train/4c0J6lwQ4_/6.png new file mode 100644 index 0000000000000000000000000000000000000000..22b2f13da155bfc22824fea7e362f0d00bf842f4 --- /dev/null +++ b/vlm/train/4c0J6lwQ4_/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1ae050e8c3a3e91b6b9a371e09092c01542e1a5d84054560e7bcf4afeb2f509e +size 607762 diff --git a/vlm/train/4c0J6lwQ4_/9.png b/vlm/train/4c0J6lwQ4_/9.png new file mode 100644 index 0000000000000000000000000000000000000000..a9ef0cb76a653ab2870c9b291b441236bc500608 --- /dev/null +++ b/vlm/train/4c0J6lwQ4_/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8229bb44b02cb1c4054656c12cc6cc65c6fdf50606689166fa73b090f4529eaf +size 550136 diff --git a/vlm/train/519VBzfEaKW/0.png b/vlm/train/519VBzfEaKW/0.png new file mode 100644 index 0000000000000000000000000000000000000000..e319db9c6d373aa4b97068e5bcef9911da8d1e26 --- /dev/null +++ b/vlm/train/519VBzfEaKW/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:38fa5c88c71b64797a71c147884178dabafa27b2e7cf1df2ab46804dd17a70e5 +size 431040 diff --git a/vlm/train/519VBzfEaKW/1.png b/vlm/train/519VBzfEaKW/1.png new file mode 100644 index 0000000000000000000000000000000000000000..17cffffedf0d783438eb7c3c28d1a8ebd3d49d03 --- /dev/null +++ b/vlm/train/519VBzfEaKW/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ca12f85f1883ddc7555ff408140188c0a2e6fa052eee7c10eceb9a8bebbc5817 +size 556845 diff --git a/vlm/train/519VBzfEaKW/10.png b/vlm/train/519VBzfEaKW/10.png new file mode 100644 index 0000000000000000000000000000000000000000..95a22e6da2344612ef37fa45617247c7f2c0f1a6 --- /dev/null +++ b/vlm/train/519VBzfEaKW/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1d9b6c9bfc8b3435b3d2699c0dd8319b0204382f74f186d2ed58aac8c8b664a1 +size 498745 diff --git a/vlm/train/519VBzfEaKW/11.png b/vlm/train/519VBzfEaKW/11.png new file mode 100644 index 0000000000000000000000000000000000000000..a42acff80330b0fb1ea212e92168dadf9dc34d0d --- /dev/null +++ b/vlm/train/519VBzfEaKW/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bd5726ca2a3d31efa310987f60ab78e7719945cc2a45da5f278eb1b5c4a98507 +size 134778 diff --git a/vlm/train/519VBzfEaKW/2.png b/vlm/train/519VBzfEaKW/2.png new file mode 100644 index 0000000000000000000000000000000000000000..a55c302e354f5f7dd966e9a0edacae7e3e26c4c4 --- /dev/null +++ b/vlm/train/519VBzfEaKW/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:67008403cc9d135215f8c5c22818b275df244a58582eabc6edc877135c871a33 +size 585938 diff --git a/vlm/train/519VBzfEaKW/3.png b/vlm/train/519VBzfEaKW/3.png new file mode 100644 index 0000000000000000000000000000000000000000..4d847b90ffa161c7bac09eeffa343ad13ee2fac5 --- /dev/null +++ b/vlm/train/519VBzfEaKW/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:60d1ace6d3a313a2389c031649eaba516cccf8d3361d601e9edad4f2c7917eb3 +size 494887 diff --git a/vlm/train/519VBzfEaKW/4.png b/vlm/train/519VBzfEaKW/4.png new file mode 100644 index 0000000000000000000000000000000000000000..1bd1a5ac94a48f3ac61547e78e4c578967f4c5cf --- /dev/null +++ b/vlm/train/519VBzfEaKW/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:022057d665fa9b007941c7b0904bfbbae2a5fc059fc5f2fc91bff0764f680e8d +size 521514 diff --git a/vlm/train/519VBzfEaKW/5.png b/vlm/train/519VBzfEaKW/5.png new file mode 100644 index 0000000000000000000000000000000000000000..9760ecc13abb2f2e0efd7a2de90e60a1b7846828 --- /dev/null +++ b/vlm/train/519VBzfEaKW/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0b02758e4032e6df1e1bb385de9117972531ff5565e2f47278ebabfe02852e65 +size 531889 diff --git a/vlm/train/519VBzfEaKW/6.png b/vlm/train/519VBzfEaKW/6.png new file mode 100644 index 0000000000000000000000000000000000000000..5bdb5349febe422d5e79be86c60e1eda2998a29b --- /dev/null +++ b/vlm/train/519VBzfEaKW/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d661ac0654d0255a5b4cb6d7a9632c24149b39f34818aedf5679b168473b41e3 +size 502529 diff --git a/vlm/train/519VBzfEaKW/7.png b/vlm/train/519VBzfEaKW/7.png new file mode 100644 index 0000000000000000000000000000000000000000..e14a081c81f5dc92639e07e8eaf09fff7d0b80d3 --- /dev/null +++ b/vlm/train/519VBzfEaKW/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f5bb6f5a1e45ec5bcc9ad1d9c4c25a8dee5b00e8d7d7bce2f3c72d0d4be812e4 +size 418435 diff --git a/vlm/train/519VBzfEaKW/8.png b/vlm/train/519VBzfEaKW/8.png new file mode 100644 index 0000000000000000000000000000000000000000..ec6426d2f7c7d8cb0e9c36b5aaccfa62c30719a5 --- /dev/null +++ b/vlm/train/519VBzfEaKW/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ead42542ffedfc485fb027f1dfb8c42ff7f862f2ebca0dd1873779203540dd0a +size 534279 diff --git a/vlm/train/519VBzfEaKW/9.png b/vlm/train/519VBzfEaKW/9.png new file mode 100644 index 0000000000000000000000000000000000000000..47abe8e2c25d9ea7886e276b3c9789ee7c91855a --- /dev/null +++ b/vlm/train/519VBzfEaKW/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:65d8967235e13e3db7b11c7155658c3cfb63d96ae4d2c1bd53b5ec3b7a5084da +size 439020 diff --git a/vlm/train/6xHJ37MVxxp/0.png b/vlm/train/6xHJ37MVxxp/0.png new file mode 100644 index 0000000000000000000000000000000000000000..c9e5c719ba4d0c235577d2f3f7cbc9d3e1e52c1a --- /dev/null +++ b/vlm/train/6xHJ37MVxxp/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2b72806b974fd0eeb248d677fdb874f9e5d2732610f3b909513bcc4e8a5648d3 +size 497698 diff --git a/vlm/train/6xHJ37MVxxp/1.png b/vlm/train/6xHJ37MVxxp/1.png new file mode 100644 index 0000000000000000000000000000000000000000..c7599290b2ee7d96400a0f30f7768e28b78ca470 --- /dev/null +++ b/vlm/train/6xHJ37MVxxp/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0b3f9114343632376cb5f0179050af34d3989ce4cfb4aed91c64454075d5c568 +size 598027 diff --git a/vlm/train/6xHJ37MVxxp/10.png b/vlm/train/6xHJ37MVxxp/10.png new file mode 100644 index 0000000000000000000000000000000000000000..c0a0022e99b7ea3f8b5580c1e716f066cfe1294c --- /dev/null +++ b/vlm/train/6xHJ37MVxxp/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3e765330823b0939fdb4bc60821adb008c8bba60e1ccc34e500c3be4eb65c747 +size 501738 diff --git a/vlm/train/6xHJ37MVxxp/11.png b/vlm/train/6xHJ37MVxxp/11.png new file mode 100644 index 0000000000000000000000000000000000000000..e16a37931de3335162fc612c2b528dbab6c7721d --- /dev/null +++ b/vlm/train/6xHJ37MVxxp/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dff51e0cc9c565414f965e932508595682b40132664104331efa557232a3f952 +size 297284 diff --git a/vlm/train/6xHJ37MVxxp/12.png b/vlm/train/6xHJ37MVxxp/12.png new file mode 100644 index 0000000000000000000000000000000000000000..4e2c2901e2aee5f6ea92737d0c651c76a526cf20 --- /dev/null +++ b/vlm/train/6xHJ37MVxxp/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:45e6c274d9d9edc64e51ebb262a11428a8f710b2d4d1f42148f686c6680a4a31 +size 230894 diff --git a/vlm/train/6xHJ37MVxxp/14.png b/vlm/train/6xHJ37MVxxp/14.png new file mode 100644 index 0000000000000000000000000000000000000000..df291f0e4e412b4a97d475641ce9bf0d874badcc --- /dev/null +++ b/vlm/train/6xHJ37MVxxp/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:33b095b1ba63e855b50c7fd1005f202ba2bcaf4f139d57875105d862083d54e4 +size 150343 diff --git a/vlm/train/6xHJ37MVxxp/2.png b/vlm/train/6xHJ37MVxxp/2.png new file mode 100644 index 0000000000000000000000000000000000000000..c9715f0e148a5e2bbd29c218bfe08e5c226bb9ac --- /dev/null +++ b/vlm/train/6xHJ37MVxxp/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c3081f106fa5c5af7dcc6e044a7f1f23596e76f1fa4bc48a056965336318852f +size 502139 diff --git a/vlm/train/6xHJ37MVxxp/3.png b/vlm/train/6xHJ37MVxxp/3.png new file mode 100644 index 0000000000000000000000000000000000000000..1b27465804785c0c52b70d42e4d2a31ce4ee3dd4 --- /dev/null +++ b/vlm/train/6xHJ37MVxxp/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f2f4c0e893ab4b12015b45d6db4633845018d01032e1ab1e1356c2de7602106c +size 560265 diff --git a/vlm/train/6xHJ37MVxxp/4.png b/vlm/train/6xHJ37MVxxp/4.png new file mode 100644 index 0000000000000000000000000000000000000000..35a87a5eb65d88cea6a8332c3fbf08f86d4c56be --- /dev/null +++ b/vlm/train/6xHJ37MVxxp/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:42a5ed1f41e8b3eea7f798fcbcb7734e9126fa31b73fc170a0d0a9ffd1c14342 +size 591665 diff --git a/vlm/train/6xHJ37MVxxp/5.png b/vlm/train/6xHJ37MVxxp/5.png new file mode 100644 index 0000000000000000000000000000000000000000..402193be8ae6f9f5c3756280b5745d54d555a510 --- /dev/null +++ b/vlm/train/6xHJ37MVxxp/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b776a6d0e1d76250ae3babbd7dd5b68af8c397619c6f0d7ece3c183f362bdeb3 +size 547513 diff --git a/vlm/train/6xHJ37MVxxp/6.png b/vlm/train/6xHJ37MVxxp/6.png new file mode 100644 index 0000000000000000000000000000000000000000..c6b4836c2c301c0db5fddb4a5086dda1173fa539 --- /dev/null +++ b/vlm/train/6xHJ37MVxxp/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5ac9024933368536aefd38f4c5303bc48eb5c7ac4bac35724f94aa4553cc0135 +size 540890 diff --git a/vlm/train/6xHJ37MVxxp/7.png b/vlm/train/6xHJ37MVxxp/7.png new file mode 100644 index 0000000000000000000000000000000000000000..7aea53016434358e4d241ea625d5d79e99081fc8 --- /dev/null +++ b/vlm/train/6xHJ37MVxxp/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9e9b8ed5d353bb07eb88cd2cf925ca5d5ea444e515cfa03873b765055cfdd54f +size 564298 diff --git a/vlm/train/6xHJ37MVxxp/8.png b/vlm/train/6xHJ37MVxxp/8.png new file mode 100644 index 0000000000000000000000000000000000000000..ada9578a02ce59dd241f73c6b6f5dd1ec28bbed1 --- /dev/null +++ b/vlm/train/6xHJ37MVxxp/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e9ab8e0bad812ab655c68448d60c12135c8f8caafa2d3b1e3449541eab93f3bf +size 572744 diff --git a/vlm/train/6xHJ37MVxxp/9.png b/vlm/train/6xHJ37MVxxp/9.png new file mode 100644 index 0000000000000000000000000000000000000000..a14794dba0bd47b38333cd1568852a88e83bb2a3 --- /dev/null +++ b/vlm/train/6xHJ37MVxxp/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:814732df41e742b39b52e47550d1ba55627f25d6cc334af5c558caeafc11e943 +size 479717 diff --git a/vlm/train/AAes_3W-2z/0.png b/vlm/train/AAes_3W-2z/0.png new file mode 100644 index 0000000000000000000000000000000000000000..119bb360146f05f677b44cfba7d1dbcba78b4b86 --- /dev/null +++ b/vlm/train/AAes_3W-2z/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c5a76f4d10aed3930f95780945e1d018ebed94e8a7477c342307e2834b110d8a +size 518434 diff --git a/vlm/train/AAes_3W-2z/1.png b/vlm/train/AAes_3W-2z/1.png new file mode 100644 index 0000000000000000000000000000000000000000..60ecb709a142ba9b061d8b19a803b6e4413de435 --- /dev/null +++ b/vlm/train/AAes_3W-2z/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:07dc8d1d3c351634a47a6b0c94c59be8b52f67ead33dbd941b8a7015792eada6 +size 573664 diff --git a/vlm/train/AAes_3W-2z/10.png b/vlm/train/AAes_3W-2z/10.png new file mode 100644 index 0000000000000000000000000000000000000000..9b410ed7314f560f50c3a6ba82320cd25d1144d7 --- /dev/null +++ b/vlm/train/AAes_3W-2z/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3e217c815c11c890c9b5459825f4707f9014c149140b1feeed30ee00cd20f8a9 +size 544306 diff --git a/vlm/train/AAes_3W-2z/11.png b/vlm/train/AAes_3W-2z/11.png new file mode 100644 index 0000000000000000000000000000000000000000..137d7db0792a72d5ce3d3edef795903fea0f641a --- /dev/null +++ b/vlm/train/AAes_3W-2z/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bad15a0926396d2ad717e76a2fd81528a32ff29c2c6ff4a10c7bd8b9ca60aa97 +size 553902 diff --git a/vlm/train/AAes_3W-2z/12.png b/vlm/train/AAes_3W-2z/12.png new file mode 100644 index 0000000000000000000000000000000000000000..cdbc998d3c87fdeb90bcbc9225fd7a8b4252bc81 --- /dev/null +++ b/vlm/train/AAes_3W-2z/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:56976f09233a9778910d843bf6a27b9606acdec76dc7b2b9960a607c21eb9d67 +size 339340 diff --git a/vlm/train/AAes_3W-2z/13.png b/vlm/train/AAes_3W-2z/13.png new file mode 100644 index 0000000000000000000000000000000000000000..2878661353ed3b1eff277e9f9f3653a997624795 --- /dev/null +++ b/vlm/train/AAes_3W-2z/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d9dfe9945e3fd7b814f88c17aaf09657756c590d0c6a76c8ea590f7af8662835 +size 448017 diff --git a/vlm/train/AAes_3W-2z/14.png b/vlm/train/AAes_3W-2z/14.png new file mode 100644 index 0000000000000000000000000000000000000000..30a796c18186e273ed0bfb92803343f708a7af67 --- /dev/null +++ b/vlm/train/AAes_3W-2z/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:10c01f970691fa9d2ab491d677437e21262c493521ba23efaadf501664995a9c +size 595992 diff --git a/vlm/train/AAes_3W-2z/15.png b/vlm/train/AAes_3W-2z/15.png new file mode 100644 index 0000000000000000000000000000000000000000..3f679a0db02eb5983992ca4f67e18337c852e2ce --- /dev/null +++ b/vlm/train/AAes_3W-2z/15.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:245e79c655c5e66c8809f6510b532862541619d3eb90a2ba8696c65093316738 +size 462131 diff --git a/vlm/train/AAes_3W-2z/16.png b/vlm/train/AAes_3W-2z/16.png new file mode 100644 index 0000000000000000000000000000000000000000..c8719b6877da4bae7a15b83b31f665dd26e6337c --- /dev/null +++ b/vlm/train/AAes_3W-2z/16.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8a6b13dd95ea6f5c64cb900d8dde8d1aa24b4eb706d130a62d6b5c5584522642 +size 510863 diff --git a/vlm/train/AAes_3W-2z/17.png b/vlm/train/AAes_3W-2z/17.png new file mode 100644 index 0000000000000000000000000000000000000000..2e5ad4f3e9b80025dfa68bec196b864d161a69b3 --- /dev/null +++ b/vlm/train/AAes_3W-2z/17.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:84a3f721937f6a404acd883000545b23563950dfbac23ab2302b06ac9732c159 +size 548616 diff --git a/vlm/train/AAes_3W-2z/18.png b/vlm/train/AAes_3W-2z/18.png new file mode 100644 index 0000000000000000000000000000000000000000..109464f9f8923b6880fd14502f0c2969d9edbb5b --- /dev/null +++ b/vlm/train/AAes_3W-2z/18.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ae8611305dd23fa5c5fff0220b9c3d32c041defd2326652b4d16748e78c74b59 +size 142309 diff --git a/vlm/train/AAes_3W-2z/19.png b/vlm/train/AAes_3W-2z/19.png new file mode 100644 index 0000000000000000000000000000000000000000..9c34814b75edbc5ef122e4456cea2e550f7283f2 --- /dev/null +++ b/vlm/train/AAes_3W-2z/19.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:00590152c08bb0342e35698e26313ecdbec5dd7cbd8efe07016f3678a557a8cf +size 291619 diff --git a/vlm/train/AAes_3W-2z/2.png b/vlm/train/AAes_3W-2z/2.png new file mode 100644 index 0000000000000000000000000000000000000000..f10ac264b279a0e7b310975133055b8e7bd39e1c --- /dev/null +++ b/vlm/train/AAes_3W-2z/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4f08d8975b4864198b7fdc0137167eb644805fde60c0f6cb0ee60fe549bdb840 +size 589972 diff --git a/vlm/train/AAes_3W-2z/3.png b/vlm/train/AAes_3W-2z/3.png new file mode 100644 index 0000000000000000000000000000000000000000..97e06146e4ed13af1bc83d0db2d829ae53c8df52 --- /dev/null +++ b/vlm/train/AAes_3W-2z/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:045804dc7ef2db9e361d3c2a47f1a7f8129ca8c9115ad4fdf850866d923be8dc +size 587315 diff --git a/vlm/train/AAes_3W-2z/4.png b/vlm/train/AAes_3W-2z/4.png new file mode 100644 index 0000000000000000000000000000000000000000..fc262019eba83f5db1ea250800c148b80f8d76e3 --- /dev/null +++ b/vlm/train/AAes_3W-2z/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5c279fe8c15ee10532a3bce537b9968887508e4ee6949ca5a3db3a6d8dfe4913 +size 541308 diff --git a/vlm/train/AAes_3W-2z/5.png b/vlm/train/AAes_3W-2z/5.png new file mode 100644 index 0000000000000000000000000000000000000000..38434c04d2dcecf8ac5113864aab04c901a687f5 --- /dev/null +++ b/vlm/train/AAes_3W-2z/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9b816fd890abf98deefee29667e9221b0c7a334272322a43601b852ae9abc310 +size 584921 diff --git a/vlm/train/AAes_3W-2z/6.png b/vlm/train/AAes_3W-2z/6.png new file mode 100644 index 0000000000000000000000000000000000000000..f6c605be1be60305073ef16603bd9034b1010b89 --- /dev/null +++ b/vlm/train/AAes_3W-2z/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e701f45829a93912edb7480196d12872e27f071bffaa989c309a5463600b40b7 +size 554288 diff --git a/vlm/train/AAes_3W-2z/7.png b/vlm/train/AAes_3W-2z/7.png new file mode 100644 index 0000000000000000000000000000000000000000..4d22d14fe765fb1465009737d812e0bfe9c3d261 --- /dev/null +++ b/vlm/train/AAes_3W-2z/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5f748218528cd6f64c5e52cd81cd072e08562c8c08ed124c446d830a3dbfdda2 +size 566650 diff --git a/vlm/train/AAes_3W-2z/8.png b/vlm/train/AAes_3W-2z/8.png new file mode 100644 index 0000000000000000000000000000000000000000..7ac281e611eba1b2169aaac9ffc0ea68710e25de --- /dev/null +++ b/vlm/train/AAes_3W-2z/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:00e8b68a05ee81532acf66e72ceaba4a1f178f48fd2e72e7a96c310bd79c9419 +size 511168 diff --git a/vlm/train/AAes_3W-2z/9.png b/vlm/train/AAes_3W-2z/9.png new file mode 100644 index 0000000000000000000000000000000000000000..9907acc84d3191dab0d137461366a3541b581c2a --- /dev/null +++ b/vlm/train/AAes_3W-2z/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8a82eed09a542c7ff2d86d8162cdd2501cebf86e35334434bf9a76c0dd05870c +size 573587 diff --git a/vlm/train/AYAgKFl78z/0.png b/vlm/train/AYAgKFl78z/0.png new file mode 100644 index 0000000000000000000000000000000000000000..ece341be5b199ccafcc45c78d35a957be0748509 --- /dev/null +++ b/vlm/train/AYAgKFl78z/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7d48c065ea9eefd0f5703784b3d2c5dd603328fd8d07be7eaefcabf9ccd41570 +size 408937 diff --git a/vlm/train/AYAgKFl78z/1.png b/vlm/train/AYAgKFl78z/1.png new file mode 100644 index 0000000000000000000000000000000000000000..2a419b0d5fc17cf6cb591e17816083b1a4c74c48 --- /dev/null +++ b/vlm/train/AYAgKFl78z/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2aedcffc55e9b489366c81e9954f7c48df8801c4d248d5d170734859168e33af +size 545193 diff --git a/vlm/train/AYAgKFl78z/10.png b/vlm/train/AYAgKFl78z/10.png new file mode 100644 index 0000000000000000000000000000000000000000..3ba4ca770df9b737889b9fe0c56de4d107d4d9f1 --- /dev/null +++ b/vlm/train/AYAgKFl78z/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:523f3e39ded44507c700a71bf7999e4ad29975d061e6ef64ea9a5d3fa7880515 +size 495235 diff --git a/vlm/train/AYAgKFl78z/11.png b/vlm/train/AYAgKFl78z/11.png new file mode 100644 index 0000000000000000000000000000000000000000..c908d3205a0cb18b777510bd6039ca6401829782 --- /dev/null +++ b/vlm/train/AYAgKFl78z/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ed6b1825671247c0160caefd7033e53efc2de51c35eef738a89191a6bf0243e0 +size 546716 diff --git a/vlm/train/AYAgKFl78z/12.png b/vlm/train/AYAgKFl78z/12.png new file mode 100644 index 0000000000000000000000000000000000000000..a51e964e441ce2557bab44e0f0bbf6f891ab67ea --- /dev/null +++ b/vlm/train/AYAgKFl78z/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:470d84f624e86b362fc7e315c9775098dfbe454efe93fc361e2e30b54525f76d +size 203183 diff --git a/vlm/train/AYAgKFl78z/2.png b/vlm/train/AYAgKFl78z/2.png new file mode 100644 index 0000000000000000000000000000000000000000..ec3b689c16e5d3790ced68feb742d9e1a24dce0a --- /dev/null +++ b/vlm/train/AYAgKFl78z/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:77eb8024cd206337b9bf5f46909f282f02be51e6f67cc67c20af5f32200ecc22 +size 485195 diff --git a/vlm/train/AYAgKFl78z/3.png b/vlm/train/AYAgKFl78z/3.png new file mode 100644 index 0000000000000000000000000000000000000000..f23f2a6e54d94ffaa186d45e25afbc7bee36ba00 --- /dev/null +++ b/vlm/train/AYAgKFl78z/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fbdfa5d992ca5739812db646b2f7ffa98887615c392155dfcd74b182170c84eb +size 493256 diff --git a/vlm/train/AYAgKFl78z/4.png b/vlm/train/AYAgKFl78z/4.png new file mode 100644 index 0000000000000000000000000000000000000000..5eb5977174f57322e4989bb50cb416bbc6e23e6a --- /dev/null +++ b/vlm/train/AYAgKFl78z/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7a6e1c0c487d5197c21f15e3f574833184e3142936393b67f01e0bea11b29d17 +size 429280 diff --git a/vlm/train/AYAgKFl78z/5.png b/vlm/train/AYAgKFl78z/5.png new file mode 100644 index 0000000000000000000000000000000000000000..57d6592a6e66ad7d234909848ab2c4c97b7973df --- /dev/null +++ b/vlm/train/AYAgKFl78z/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:33468791524d412ac5b7203c94cf607d37a6fa8582b721aba133a7c7c9ab068b +size 416112 diff --git a/vlm/train/AYAgKFl78z/6.png b/vlm/train/AYAgKFl78z/6.png new file mode 100644 index 0000000000000000000000000000000000000000..d48cd0a05cc24d0cd1b7ad4181da7e38d0e0b76f --- /dev/null +++ b/vlm/train/AYAgKFl78z/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7774e39ec378daaff0f3fb07aaedd88260b82cbd1c50fe30b199555bb65645f5 +size 405745 diff --git a/vlm/train/AYAgKFl78z/7.png b/vlm/train/AYAgKFl78z/7.png new file mode 100644 index 0000000000000000000000000000000000000000..97594de8992f450b58cab6e39862343db7ff584e --- /dev/null +++ b/vlm/train/AYAgKFl78z/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:81c1b24f986e7ca0429e827fd5ada55d3f0928f09632a5bd63d43b4b2a939162 +size 439302 diff --git a/vlm/train/AYAgKFl78z/8.png b/vlm/train/AYAgKFl78z/8.png new file mode 100644 index 0000000000000000000000000000000000000000..482234ec746415ad3729dd18a30ecb57a4ae2cf9 --- /dev/null +++ b/vlm/train/AYAgKFl78z/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:28a0ddf19f5b808f5976f7bd232ec69a1f3ba1a4271563cbd7bb4939b1821156 +size 382284 diff --git a/vlm/train/AYAgKFl78z/9.png b/vlm/train/AYAgKFl78z/9.png new file mode 100644 index 0000000000000000000000000000000000000000..b9b106f22a4f02f112ec067b79d6eba27356dfa6 --- /dev/null +++ b/vlm/train/AYAgKFl78z/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c670f3fbc189d985ab4d563386c68fca8c39b757c9f7fceebeac07c0f9277d31 +size 406310 diff --git a/vlm/train/B14rPj0qY7/0.png b/vlm/train/B14rPj0qY7/0.png new file mode 100644 index 0000000000000000000000000000000000000000..52b24ffc873ff0edea45d4942e585ce89c1d0a23 --- /dev/null +++ b/vlm/train/B14rPj0qY7/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7109900f9d60b0e33c8e87e73530e3b6b1a8bc310ae08f0bf357c9923fab4dfb +size 486000 diff --git a/vlm/train/B14rPj0qY7/1.png b/vlm/train/B14rPj0qY7/1.png new file mode 100644 index 0000000000000000000000000000000000000000..3c5358d9e0417276ddf7ee9bed766e7872c173fa --- /dev/null +++ b/vlm/train/B14rPj0qY7/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8511415f477781aaebe6439a29fe828ddba1a1237473a978fcab73411dce4ba5 +size 523315 diff --git a/vlm/train/B14rPj0qY7/2.png b/vlm/train/B14rPj0qY7/2.png new file mode 100644 index 0000000000000000000000000000000000000000..682815fc6cc5a76b8a00c4a32098f54585ed9db4 --- /dev/null +++ b/vlm/train/B14rPj0qY7/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4f61fa9ba9dc14f27d7509f250aca289c44405f575d8bdc14be827d7a52f8998 +size 543128 diff --git a/vlm/train/B14rPj0qY7/3.png b/vlm/train/B14rPj0qY7/3.png new file mode 100644 index 0000000000000000000000000000000000000000..741d70e99757c88b2d5b8a2d2221decad6b2d58b --- /dev/null +++ b/vlm/train/B14rPj0qY7/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:49301830913d1b08b49e0063d19a8968c4584b1a3e1fb209ff3effa21bc5266f +size 390076 diff --git a/vlm/train/B14rPj0qY7/4.png b/vlm/train/B14rPj0qY7/4.png new file mode 100644 index 0000000000000000000000000000000000000000..3d5c848748933ea3477204cc2588a012504a6ba1 --- /dev/null +++ b/vlm/train/B14rPj0qY7/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:234bdb4066fb0a1c73f090d4bc898b28a489931b1fcad23133163bab156c5b1a +size 436550 diff --git a/vlm/train/B14rPj0qY7/5.png b/vlm/train/B14rPj0qY7/5.png new file mode 100644 index 0000000000000000000000000000000000000000..2e79d59687a7829b514c61843b87623526aaee15 --- /dev/null +++ b/vlm/train/B14rPj0qY7/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9dda99fd306bca46280fc7c86c83586d530b6748646a96789f4a9340aaff1e73 +size 463319 diff --git a/vlm/train/B14rPj0qY7/6.png b/vlm/train/B14rPj0qY7/6.png new file mode 100644 index 0000000000000000000000000000000000000000..5af9e7f11274b6deaa74a02a6371b5b083f5b249 --- /dev/null +++ b/vlm/train/B14rPj0qY7/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5db030f4715b995b02e5470484644a9c3cf3b219f2bfe9322b3558ec6a5f1c43 +size 436863 diff --git a/vlm/train/B14rPj0qY7/7.png b/vlm/train/B14rPj0qY7/7.png new file mode 100644 index 0000000000000000000000000000000000000000..8e6aed1d87f660769cac6fcd25e18746466d5c34 --- /dev/null +++ b/vlm/train/B14rPj0qY7/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bbcf39e9255fd4953e08bffe39d20c95647dba80ad67bba4e5effdf6c87e03ec +size 435670 diff --git a/vlm/train/B14rPj0qY7/8.png b/vlm/train/B14rPj0qY7/8.png new file mode 100644 index 0000000000000000000000000000000000000000..20cc4e2280e848d47bac628d7fd1684ab9e1afa1 --- /dev/null +++ b/vlm/train/B14rPj0qY7/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b3b56d514c0e3e875339c81b0760951b386fe83e3fb5e110b216a91867ec51ea +size 387642 diff --git a/vlm/train/B14rPj0qY7/9.png b/vlm/train/B14rPj0qY7/9.png new file mode 100644 index 0000000000000000000000000000000000000000..a9c4c52d8d4007f7086cbc4cb1e81afe2c5b3841 --- /dev/null +++ b/vlm/train/B14rPj0qY7/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:95088691bf0b40513d1446a98bd7a61b71ffffde0d558f7c5b519c958ab4217d +size 538660 diff --git a/vlm/train/BJMvBjC5YQ/0.png b/vlm/train/BJMvBjC5YQ/0.png new file mode 100644 index 0000000000000000000000000000000000000000..2f05a96b81618f35bb156ba2d94a129b04609634 --- /dev/null +++ b/vlm/train/BJMvBjC5YQ/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9c7a8e66623566ee110e0cf969630483e9dc67722528a3a617251c4eedd6b99b +size 523127 diff --git a/vlm/train/BJMvBjC5YQ/1.png b/vlm/train/BJMvBjC5YQ/1.png new file mode 100644 index 0000000000000000000000000000000000000000..f260f3f4e66cbd0f7d9ef478e35764a6b713f7cd --- /dev/null +++ b/vlm/train/BJMvBjC5YQ/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9283f8ec54b398c6df36d57a69890a5a70535755c47c8541a5907aaccfebfc81 +size 533921 diff --git a/vlm/train/BJMvBjC5YQ/10.png b/vlm/train/BJMvBjC5YQ/10.png new file mode 100644 index 0000000000000000000000000000000000000000..04e04d261569480968513784406b324c4228063b --- /dev/null +++ b/vlm/train/BJMvBjC5YQ/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:858875b1849ac168c41b685abf9a3ab744e0053cbc4bc24f0e1ae3e23cc3b5c8 +size 447940 diff --git a/vlm/train/BJMvBjC5YQ/11.png b/vlm/train/BJMvBjC5YQ/11.png new file mode 100644 index 0000000000000000000000000000000000000000..4fa9148e77b3177281a58af53a80909bf3a9f742 --- /dev/null +++ b/vlm/train/BJMvBjC5YQ/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d2ed55c6a3af39ed164d4094bec1bb0ec35651c99f3869f0df41ce4794cbada5 +size 487469 diff --git a/vlm/train/BJMvBjC5YQ/12.png b/vlm/train/BJMvBjC5YQ/12.png new file mode 100644 index 0000000000000000000000000000000000000000..57a0c50ef14596716934882c2e381cef541b9307 --- /dev/null +++ b/vlm/train/BJMvBjC5YQ/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6687d57696ad7d7e2d4420569e9c14db8290af3a8ccf7a5d014e227ab145e9ee +size 438126 diff --git a/vlm/train/BJMvBjC5YQ/13.png b/vlm/train/BJMvBjC5YQ/13.png new file mode 100644 index 0000000000000000000000000000000000000000..638e335cbe4feddd115fa21ba0f949673247e3a3 --- /dev/null +++ b/vlm/train/BJMvBjC5YQ/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9884b806694f46c74761ffebc4c1c8dd671cb22df5fcf2ff8b2f2e1de8ad8025 +size 464946 diff --git a/vlm/train/BJMvBjC5YQ/14.png b/vlm/train/BJMvBjC5YQ/14.png new file mode 100644 index 0000000000000000000000000000000000000000..04d8be168eea50c063ac654b02119978d2e92e14 --- /dev/null +++ b/vlm/train/BJMvBjC5YQ/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4661f7bae71b9184b202b9a0f56804d1af426254252e08d42fce6f3893d6f1f2 +size 479838 diff --git a/vlm/train/BJMvBjC5YQ/15.png b/vlm/train/BJMvBjC5YQ/15.png new file mode 100644 index 0000000000000000000000000000000000000000..dba96dab59bf1998f29c12b26035af17c6592db8 --- /dev/null +++ b/vlm/train/BJMvBjC5YQ/15.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d834a2b45ed77ebced7cd757dd2eeb20df3cb90fe0ab134691f2b65e27492abf +size 403441 diff --git a/vlm/train/BJMvBjC5YQ/16.png b/vlm/train/BJMvBjC5YQ/16.png new file mode 100644 index 0000000000000000000000000000000000000000..2463ae713e3feb561a3564c3d8c5b530b44d334b --- /dev/null +++ b/vlm/train/BJMvBjC5YQ/16.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:34880d3271198e2c6a13b3757b7d7ccc47ce03c7cdf371cab960b14fffca7e58 +size 492370 diff --git a/vlm/train/BJMvBjC5YQ/17.png b/vlm/train/BJMvBjC5YQ/17.png new file mode 100644 index 0000000000000000000000000000000000000000..eea3ea3c6ca68695f7754cb2e3d0601c43f8b037 --- /dev/null +++ b/vlm/train/BJMvBjC5YQ/17.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6b534a324a486c9bb51ab80de6323cd50b8fa7b9ebeae4feed1a8e07d1badde6 +size 302509 diff --git a/vlm/train/BJMvBjC5YQ/2.png b/vlm/train/BJMvBjC5YQ/2.png new file mode 100644 index 0000000000000000000000000000000000000000..bfdb62a1fa72acb23ffee63a25c5ed965c5bbf00 --- /dev/null +++ b/vlm/train/BJMvBjC5YQ/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b0c9bcec64cf95576f84f3748c5e1946614fb38bf7bd70b5a6b50b6498986707 +size 499868 diff --git a/vlm/train/BJMvBjC5YQ/3.png b/vlm/train/BJMvBjC5YQ/3.png new file mode 100644 index 0000000000000000000000000000000000000000..89654c73434ac10429fe6ff8e91635ee1de40123 --- /dev/null +++ b/vlm/train/BJMvBjC5YQ/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b896257dca36591088a18d9d5d058aee540e2f4cde8296f77a388a2c3f8c4923 +size 483004 diff --git a/vlm/train/BJMvBjC5YQ/4.png b/vlm/train/BJMvBjC5YQ/4.png new file mode 100644 index 0000000000000000000000000000000000000000..b6d50014535327f6261de47a79494e1accb6bd66 --- /dev/null +++ b/vlm/train/BJMvBjC5YQ/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2c32cd343d2b856b705c1ce300589005acb7dad86453aece231fed03a51d8573 +size 582210 diff --git a/vlm/train/BJMvBjC5YQ/5.png b/vlm/train/BJMvBjC5YQ/5.png new file mode 100644 index 0000000000000000000000000000000000000000..dcec3d5bd7a7d43396fe0369303d5db20f7551db --- /dev/null +++ b/vlm/train/BJMvBjC5YQ/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:afa963ddef7b9d427bf16230c159c70e2d9fac5e413093594d7b811a47054a87 +size 472554 diff --git a/vlm/train/BJMvBjC5YQ/6.png b/vlm/train/BJMvBjC5YQ/6.png new file mode 100644 index 0000000000000000000000000000000000000000..a1a87e9f331961d2e7279e4e431b8c9c30ce1178 --- /dev/null +++ b/vlm/train/BJMvBjC5YQ/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a1fc39600d7a137f94911e40ae1554a62932271b9bfc6b88670b11862eaa128d +size 431658 diff --git a/vlm/train/BJMvBjC5YQ/7.png b/vlm/train/BJMvBjC5YQ/7.png new file mode 100644 index 0000000000000000000000000000000000000000..4bb27b6c13cef80b50da5a604f50d75b034bff4c --- /dev/null +++ b/vlm/train/BJMvBjC5YQ/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a90cb956f04f7aa8c5ee2f5ec220eaa1e8814db16977f7b9a0df047d98257879 +size 619197 diff --git a/vlm/train/BJMvBjC5YQ/8.png b/vlm/train/BJMvBjC5YQ/8.png new file mode 100644 index 0000000000000000000000000000000000000000..54d46b082d7948a96d1ef2ef6749f5660f0b321d --- /dev/null +++ b/vlm/train/BJMvBjC5YQ/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bb33ada9b8d21d9f7f3fc7e406f72ad17f947cec604dfa19a30a82b6d75f5256 +size 569748 diff --git a/vlm/train/BJMvBjC5YQ/9.png b/vlm/train/BJMvBjC5YQ/9.png new file mode 100644 index 0000000000000000000000000000000000000000..10e7dd23265daccbc174f74b9d01b1754f2e4dc4 --- /dev/null +++ b/vlm/train/BJMvBjC5YQ/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3029b58b2ca3e49f4bbf3064089fc9734e33786f3c6f960de75423edc6ed04bd +size 380381 diff --git a/vlm/train/BJe1334YDH/0.png b/vlm/train/BJe1334YDH/0.png new file mode 100644 index 0000000000000000000000000000000000000000..8d8c0d43903ff8301dc6b832132db560ed41358a --- /dev/null +++ b/vlm/train/BJe1334YDH/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cf5d0e488757914cc474ba78f90ed0cbc89b6db124ffc53805b287a0e3b12654 +size 496565 diff --git a/vlm/train/BJe1334YDH/1.png b/vlm/train/BJe1334YDH/1.png new file mode 100644 index 0000000000000000000000000000000000000000..a190c41c4cda5d90f8bf44036dc9b575ab96d53f --- /dev/null +++ b/vlm/train/BJe1334YDH/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9e6e017be9239bad3a60f81e4a259f8da5bbab8fea97b380909afa1a1772b8c2 +size 564228 diff --git a/vlm/train/BJe1334YDH/10.png b/vlm/train/BJe1334YDH/10.png new file mode 100644 index 0000000000000000000000000000000000000000..939d61761bfc7de07c736a61d7f27faada5e89f9 --- /dev/null +++ b/vlm/train/BJe1334YDH/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9252baec7f0bcf277d5a55522b556c86cde2f07169e81f637e7bfe3842250baa +size 357211 diff --git a/vlm/train/BJe1334YDH/11.png b/vlm/train/BJe1334YDH/11.png new file mode 100644 index 0000000000000000000000000000000000000000..8ec01c7277ed1802e83c5283fad6ba27253f51d2 --- /dev/null +++ b/vlm/train/BJe1334YDH/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f0751a0db6729eaa4203c6fea86c34e7d2c4102a1743fef3382d02daba3deb7b +size 380744 diff --git a/vlm/train/BJe1334YDH/12.png b/vlm/train/BJe1334YDH/12.png new file mode 100644 index 0000000000000000000000000000000000000000..3ca0020cbc70aad03d4e909be3627a2be8571826 --- /dev/null +++ b/vlm/train/BJe1334YDH/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d9f98a64fbf84a61fdd6043bf79dc07002fbc63505208910fccabc67d3991057 +size 199696 diff --git a/vlm/train/BJe1334YDH/13.png b/vlm/train/BJe1334YDH/13.png new file mode 100644 index 0000000000000000000000000000000000000000..bbf2475edad496b70bb2b3ae944305e089b6732c --- /dev/null +++ b/vlm/train/BJe1334YDH/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:93655ad7fb24d27a9156c2f9b0827d518bcdf902e2eede191ce047ba9b9768a1 +size 293320 diff --git a/vlm/train/BJe1334YDH/14.png b/vlm/train/BJe1334YDH/14.png new file mode 100644 index 0000000000000000000000000000000000000000..d80d871f8557bd9293db162a53445469ad13a93b --- /dev/null +++ b/vlm/train/BJe1334YDH/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ca96c15937f6c670e1952bdf7287bc01d9551fcf96934ebde9e949f6bf692b4c +size 60741 diff --git a/vlm/train/BJe1334YDH/2.png b/vlm/train/BJe1334YDH/2.png new file mode 100644 index 0000000000000000000000000000000000000000..f48a8f489568cb7df287c027940e8b72d41b52a8 --- /dev/null +++ b/vlm/train/BJe1334YDH/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:33ab4e5bb6738005ae0526f372bebf478041ecc41dfaa648661ae52dff96e8bd +size 524508 diff --git a/vlm/train/BJe1334YDH/3.png b/vlm/train/BJe1334YDH/3.png new file mode 100644 index 0000000000000000000000000000000000000000..19e05f37a3670ea87268f33c3a0fab6fcd77ab1a --- /dev/null +++ b/vlm/train/BJe1334YDH/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a650d648fb6a6e7b0177f4b0927e3bd75805cc1bd1d682ee3db00d54c6340f08 +size 550908 diff --git a/vlm/train/BJe1334YDH/4.png b/vlm/train/BJe1334YDH/4.png new file mode 100644 index 0000000000000000000000000000000000000000..ca8fd2e534ad509d42312feed07f25cca3f2eb4c --- /dev/null +++ b/vlm/train/BJe1334YDH/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:06f13a1218ba07a8bed7c06546ff1f23562ab4ff49d298c07c84c56278197722 +size 497481 diff --git a/vlm/train/BJe1334YDH/5.png b/vlm/train/BJe1334YDH/5.png new file mode 100644 index 0000000000000000000000000000000000000000..b21ca13f1add484ae1af747b405f5e9cc7bf4072 --- /dev/null +++ b/vlm/train/BJe1334YDH/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c92397915bb961ce2dfc6fd70638c39f3ef48647efd036cce9b6d83b41778004 +size 506497 diff --git a/vlm/train/BJe1334YDH/6.png b/vlm/train/BJe1334YDH/6.png new file mode 100644 index 0000000000000000000000000000000000000000..ac0d05816b58a052fc93594f84170c36dc3c2f29 --- /dev/null +++ b/vlm/train/BJe1334YDH/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3124410b207436cae6163083d6b7ad3217b8fe618c7082c0d1ff6c5dceee34f7 +size 399382 diff --git a/vlm/train/BJe1334YDH/7.png b/vlm/train/BJe1334YDH/7.png new file mode 100644 index 0000000000000000000000000000000000000000..dec71b8c5c384b87bde141d2bf825d65219c7c7c --- /dev/null +++ b/vlm/train/BJe1334YDH/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b91aaaab685f9c7dc010542c0a3d7d586ae30f3317c1cb622520f36972e44b52 +size 472984 diff --git a/vlm/train/BJe1334YDH/8.png b/vlm/train/BJe1334YDH/8.png new file mode 100644 index 0000000000000000000000000000000000000000..13d45d61a42148afe815843f87f75c17dd7a1f28 --- /dev/null +++ b/vlm/train/BJe1334YDH/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6c6bfba07093bc26b45f0e8b2edc6022bf233a15ccdb74e98ff12ccadae70a9a +size 477774 diff --git a/vlm/train/BJe1334YDH/9.png b/vlm/train/BJe1334YDH/9.png new file mode 100644 index 0000000000000000000000000000000000000000..e353854d37a37400584f89e673ccb3116b45b332 --- /dev/null +++ b/vlm/train/BJe1334YDH/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4c4262daa06b776b26f3fdeb4d3079e18ea92173019802d4831e5b4bf250e41b +size 529649 diff --git a/vlm/train/BJgPCveAW/0.png b/vlm/train/BJgPCveAW/0.png new file mode 100644 index 0000000000000000000000000000000000000000..f94cd9e49731332cef0b4a960dcb4f3bbe4e88f4 --- /dev/null +++ b/vlm/train/BJgPCveAW/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2f13ba65dea2b4b4ed03ab72d47c6ec815d02a2a494b0444761f05b98a55b3ee +size 512410 diff --git a/vlm/train/BJgPCveAW/1.png b/vlm/train/BJgPCveAW/1.png new file mode 100644 index 0000000000000000000000000000000000000000..c796615f8e060968f7c7489f780384aac76dbc74 --- /dev/null +++ b/vlm/train/BJgPCveAW/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3aadfc2b6dc8a79c4e17ac6354793f789719314b39ffa3b682830426a5bd5b07 +size 579778 diff --git a/vlm/train/BJgPCveAW/2.png b/vlm/train/BJgPCveAW/2.png new file mode 100644 index 0000000000000000000000000000000000000000..7757705fb7a0467d581da5884bf955d68f128fbc --- /dev/null +++ b/vlm/train/BJgPCveAW/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3e1edeb006a1b575c810e58d59eac986ab0ad49ef83518ad97678e86c9036575 +size 649330 diff --git a/vlm/train/BJgPCveAW/3.png b/vlm/train/BJgPCveAW/3.png new file mode 100644 index 0000000000000000000000000000000000000000..06e367cc4dc56abca1f5e46be49be24f77383df1 --- /dev/null +++ b/vlm/train/BJgPCveAW/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ac911c6c2c4e3d7b010a3eaec269e6ea5e37dcb6cea194543c0225429bc15391 +size 610623 diff --git a/vlm/train/BJgPCveAW/4.png b/vlm/train/BJgPCveAW/4.png new file mode 100644 index 0000000000000000000000000000000000000000..c97a8159941f1f1bedca6f546044325c4fd56f8b --- /dev/null +++ b/vlm/train/BJgPCveAW/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2d36c3f78e48b09a1ea176328b0adacc594ceb58f364e9f78e918c1e4abbc4cf +size 541860 diff --git a/vlm/train/BJgPCveAW/5.png b/vlm/train/BJgPCveAW/5.png new file mode 100644 index 0000000000000000000000000000000000000000..9a03112ffd5daead929169e33052cb2c89aca254 --- /dev/null +++ b/vlm/train/BJgPCveAW/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bcb6f80fa7dabd0c37eba068ccf37ac96211573acba77cf4b871538cfb6a3652 +size 628038 diff --git a/vlm/train/BJgPCveAW/6.png b/vlm/train/BJgPCveAW/6.png new file mode 100644 index 0000000000000000000000000000000000000000..82e08c319ca280c8bda76f32eb4c85bc6d64c6f4 --- /dev/null +++ b/vlm/train/BJgPCveAW/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a35d0ba6e71fff67decf0c06d84f7cee50393d635296faf63c1e01e6ec570b93 +size 447521 diff --git a/vlm/train/BJgPCveAW/7.png b/vlm/train/BJgPCveAW/7.png new file mode 100644 index 0000000000000000000000000000000000000000..1f79a0c86d04b8f98638b1b3cc1bbd97b5035c99 --- /dev/null +++ b/vlm/train/BJgPCveAW/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5d933426309cde3f4d1c10b883403a36ea51726530c7205491afb81a6336524c +size 585750 diff --git a/vlm/train/BJgPCveAW/8.png b/vlm/train/BJgPCveAW/8.png new file mode 100644 index 0000000000000000000000000000000000000000..6cdb38d510eb427bf48beec3e1137409899a9999 --- /dev/null +++ b/vlm/train/BJgPCveAW/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0ca172b268869f1b7bcec65f6312a0e76ed25d5903a8228d8fe20aaee5c1739c +size 515607 diff --git a/vlm/train/BJgPCveAW/9.png b/vlm/train/BJgPCveAW/9.png new file mode 100644 index 0000000000000000000000000000000000000000..3125709dcf2397ec2bfb43846f22b8986155f70f --- /dev/null +++ b/vlm/train/BJgPCveAW/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4dc3d9e28dbad32231310404af9714127ab1cbddff7fb34cb0aab864acbeafab +size 337934 diff --git a/vlm/train/Byk-VI9eg/0.png b/vlm/train/Byk-VI9eg/0.png new file mode 100644 index 0000000000000000000000000000000000000000..9d88e09039d1eb3c5073cac6fa15de9829a63ed2 --- /dev/null +++ b/vlm/train/Byk-VI9eg/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f23f3916972bdb8e7131e6e0f5be52fbffa2e1359f194298268a221d5d1e7c89 +size 561337 diff --git a/vlm/train/Byk-VI9eg/1.png b/vlm/train/Byk-VI9eg/1.png new file mode 100644 index 0000000000000000000000000000000000000000..c263ce769a5cec38b1f018779beecc95d6a78ea6 --- /dev/null +++ b/vlm/train/Byk-VI9eg/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3f9c2ee2dd49aa788d7ca53d93e30d9b9638e8e735df19119e079a6f78458881 +size 562618 diff --git a/vlm/train/Byk-VI9eg/10.png b/vlm/train/Byk-VI9eg/10.png new file mode 100644 index 0000000000000000000000000000000000000000..287f0cf181e99d8cd656e892bbf46820d5ed5f99 --- /dev/null +++ b/vlm/train/Byk-VI9eg/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e3125b26e720737db6969058916a5bb6c43807cc0bd1465c6af09280fe49fe28 +size 533695 diff --git a/vlm/train/Byk-VI9eg/11.png b/vlm/train/Byk-VI9eg/11.png new file mode 100644 index 0000000000000000000000000000000000000000..abe028c2f641c12ecfda676656e20188bae59978 --- /dev/null +++ b/vlm/train/Byk-VI9eg/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:379bac7d701431fa82f4acca4bf7a88bd0e581377ae85c229f6ee2c6fc7e336e +size 823908 diff --git a/vlm/train/Byk-VI9eg/12.png b/vlm/train/Byk-VI9eg/12.png new file mode 100644 index 0000000000000000000000000000000000000000..a273369936927f5ee795aea333e3df4809171ef8 --- /dev/null +++ b/vlm/train/Byk-VI9eg/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e1cf348f3b21c3a3f9c5dbd484bd42abc2de0f014133f4799e07b17821023d8f +size 1066655 diff --git a/vlm/train/Byk-VI9eg/13.png b/vlm/train/Byk-VI9eg/13.png new file mode 100644 index 0000000000000000000000000000000000000000..8d2fb5e19897e70481c70ba3ff54be0f7741dc92 --- /dev/null +++ b/vlm/train/Byk-VI9eg/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:74e0cb52108c301c69e16732100b410e48b1fb43eadefa5f7f43b0adff14df87 +size 509642 diff --git a/vlm/train/Byk-VI9eg/2.png b/vlm/train/Byk-VI9eg/2.png new file mode 100644 index 0000000000000000000000000000000000000000..a8e52afe5af9978eb478c6bb32e33b5c46e3aa1d --- /dev/null +++ b/vlm/train/Byk-VI9eg/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eef43a238cd8859836fc31120f5c862ffa123aa5e88a11d8becb35142a2180f8 +size 451233 diff --git a/vlm/train/Byk-VI9eg/3.png b/vlm/train/Byk-VI9eg/3.png new file mode 100644 index 0000000000000000000000000000000000000000..f05b2913a888f196cbfd4dac496970ff524ebfa2 --- /dev/null +++ b/vlm/train/Byk-VI9eg/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:419a9f9af3928bf08449221815b5b10bdb00a474d6bf46132b7acb0d1ca24336 +size 500210 diff --git a/vlm/train/Byk-VI9eg/4.png b/vlm/train/Byk-VI9eg/4.png new file mode 100644 index 0000000000000000000000000000000000000000..2ef5ebc7b2221502700f2bb042033607af87792d --- /dev/null +++ b/vlm/train/Byk-VI9eg/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7b679daa42a4a159cdd0dc2379560eccb7b90f39d8f77667d10e36d4485c3f06 +size 485999 diff --git a/vlm/train/Byk-VI9eg/5.png b/vlm/train/Byk-VI9eg/5.png new file mode 100644 index 0000000000000000000000000000000000000000..3fb766d50f2ccedf34637bf9a53f8a913f1e6005 --- /dev/null +++ b/vlm/train/Byk-VI9eg/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ddb69846c5ea332d4c9350c2a0b531f36017ef8adc1bdc80a4db3a649313751c +size 527511 diff --git a/vlm/train/Byk-VI9eg/6.png b/vlm/train/Byk-VI9eg/6.png new file mode 100644 index 0000000000000000000000000000000000000000..d4b20684357a6e500a4f560b89ca6b015840b1d1 --- /dev/null +++ b/vlm/train/Byk-VI9eg/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:af1e2695658b47978aa752bec50be10ee1614188075644fca91aa39a8aa2a4b1 +size 549462 diff --git a/vlm/train/Byk-VI9eg/7.png b/vlm/train/Byk-VI9eg/7.png new file mode 100644 index 0000000000000000000000000000000000000000..8bad16f5c68a632bd954e0af4683ee61493fb60e --- /dev/null +++ b/vlm/train/Byk-VI9eg/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3d42af1029ac5e87c12dc7567e127ec781fc35c9019dd55fc2bd28708626ffe5 +size 1069560 diff --git a/vlm/train/Byk-VI9eg/8.png b/vlm/train/Byk-VI9eg/8.png new file mode 100644 index 0000000000000000000000000000000000000000..adb8df5f31be63de56db91f8903c82caa26acf73 --- /dev/null +++ b/vlm/train/Byk-VI9eg/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0c9b01c4bcbdcca1938e5471e62c50415852b1e45686d14e8c473b99b1a151e9 +size 514819 diff --git a/vlm/train/Byk-VI9eg/9.png b/vlm/train/Byk-VI9eg/9.png new file mode 100644 index 0000000000000000000000000000000000000000..b9743f349ef94546825f98c7a02a0d11f872d7c3 --- /dev/null +++ b/vlm/train/Byk-VI9eg/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:837860e165d1f885cb13741921de857ead91a629a3414194e3b1723137399d87 +size 337522 diff --git a/vlm/train/CmI7NqBR4Ua/0.png b/vlm/train/CmI7NqBR4Ua/0.png new file mode 100644 index 0000000000000000000000000000000000000000..f1cb450f8b2c4cdb92e7f9940c853e709928b96e --- /dev/null +++ b/vlm/train/CmI7NqBR4Ua/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:16b6a6d6cad5fad7beb91dfee3589f1393ad58c1d23fbe0c1be35c0b342fc93d +size 419921 diff --git a/vlm/train/CmI7NqBR4Ua/1.png b/vlm/train/CmI7NqBR4Ua/1.png new file mode 100644 index 0000000000000000000000000000000000000000..4638f558e5d749ac4b4d50285738f486c6932bdc --- /dev/null +++ b/vlm/train/CmI7NqBR4Ua/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3f6101d8132626670dc174564fcd9616aec145a777fdd96506fe97f4751598c6 +size 480202 diff --git a/vlm/train/CmI7NqBR4Ua/10.png b/vlm/train/CmI7NqBR4Ua/10.png new file mode 100644 index 0000000000000000000000000000000000000000..f2a29c30dfac887359559ae8c975bac44763961a --- /dev/null +++ b/vlm/train/CmI7NqBR4Ua/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9cfaef473353009d47fe2c78627e7373bbffa081fa3aa18c63097cfe0909feff +size 556126 diff --git a/vlm/train/CmI7NqBR4Ua/11.png b/vlm/train/CmI7NqBR4Ua/11.png new file mode 100644 index 0000000000000000000000000000000000000000..60bbf9e56ed8120ffecde401edd790a92fde1f95 --- /dev/null +++ b/vlm/train/CmI7NqBR4Ua/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:364be03911db5fab7245482e5c517319c2dd180505c4eb7e065eac8fff694327 +size 588988 diff --git a/vlm/train/CmI7NqBR4Ua/12.png b/vlm/train/CmI7NqBR4Ua/12.png new file mode 100644 index 0000000000000000000000000000000000000000..445998367703745216cada36783b6a078fe78ed7 --- /dev/null +++ b/vlm/train/CmI7NqBR4Ua/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec10fbcb5aa0590ae6be28a9c487fdab2edd4194acc34e1c973758e39f459876 +size 557425 diff --git a/vlm/train/CmI7NqBR4Ua/13.png b/vlm/train/CmI7NqBR4Ua/13.png new file mode 100644 index 0000000000000000000000000000000000000000..7a6e883179679603786399c2f99c7300974b3de8 --- /dev/null +++ b/vlm/train/CmI7NqBR4Ua/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:45f5e01879af1592d59b4a858001f591c41ee63cccf021aeb8f8ed5b6b4b7703 +size 196790 diff --git a/vlm/train/CmI7NqBR4Ua/2.png b/vlm/train/CmI7NqBR4Ua/2.png new file mode 100644 index 0000000000000000000000000000000000000000..e186636630a31d558dd3e0e91e13670337dacb29 --- /dev/null +++ b/vlm/train/CmI7NqBR4Ua/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d5e057fd20575762c159aa4dcc76ea37f53902b602973b01fabf54fa176db04d +size 509204 diff --git a/vlm/train/CmI7NqBR4Ua/3.png b/vlm/train/CmI7NqBR4Ua/3.png new file mode 100644 index 0000000000000000000000000000000000000000..9551f668f6adff882d900d3fc84510181d06ecd2 --- /dev/null +++ b/vlm/train/CmI7NqBR4Ua/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:17eac0c1a54d6be523b1890103d05f90fedbbe63c9bc74bd87557b86010377b8 +size 436863 diff --git a/vlm/train/CmI7NqBR4Ua/4.png b/vlm/train/CmI7NqBR4Ua/4.png new file mode 100644 index 0000000000000000000000000000000000000000..59527ec71fdb36866dd6ac5018fb417e03e917f4 --- /dev/null +++ b/vlm/train/CmI7NqBR4Ua/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2b99113c6f4e9164141d63cab9ba9e4cb95eabfa62f0aeeac6d6edc8905c2e47 +size 487316 diff --git a/vlm/train/CmI7NqBR4Ua/5.png b/vlm/train/CmI7NqBR4Ua/5.png new file mode 100644 index 0000000000000000000000000000000000000000..fc0cbb45826a18ff0f22ca20a6af7ac17e007c23 --- /dev/null +++ b/vlm/train/CmI7NqBR4Ua/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:33384a6bd24b853a2f24a68d62a5e437c167ce9f54f8d5849d4d71b46bce7389 +size 526889 diff --git a/vlm/train/CmI7NqBR4Ua/6.png b/vlm/train/CmI7NqBR4Ua/6.png new file mode 100644 index 0000000000000000000000000000000000000000..9e6a85b45d350044ad49c412d54054f8d4055c53 --- /dev/null +++ b/vlm/train/CmI7NqBR4Ua/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6bcedd75f92ade322b48913532f4435cf2910e1d275a93e7df1e30e791b4c2c2 +size 399287 diff --git a/vlm/train/CmI7NqBR4Ua/7.png b/vlm/train/CmI7NqBR4Ua/7.png new file mode 100644 index 0000000000000000000000000000000000000000..70eaab8293e6a8d81523178bdf184a5e708db3c4 --- /dev/null +++ b/vlm/train/CmI7NqBR4Ua/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:700ead4b136052993f5330796123821346df5729102fd2913e924febe7e35cb0 +size 417795 diff --git a/vlm/train/CmI7NqBR4Ua/8.png b/vlm/train/CmI7NqBR4Ua/8.png new file mode 100644 index 0000000000000000000000000000000000000000..229610963e27d01b16d7cb3a855593ad7c8f9155 --- /dev/null +++ b/vlm/train/CmI7NqBR4Ua/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d4002e91e13804dc3f6272bc1044bbc79b963cb4112ef685cfa1658123936e25 +size 381281 diff --git a/vlm/train/CmI7NqBR4Ua/9.png b/vlm/train/CmI7NqBR4Ua/9.png new file mode 100644 index 0000000000000000000000000000000000000000..39588b03201838a79c85b35706905f0176a3e801 --- /dev/null +++ b/vlm/train/CmI7NqBR4Ua/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f96c0311dad315641335db666e5bdd5f7259e2f409e68dd2a1581e089a8e42b5 +size 439404 diff --git a/vlm/train/DHSNrGhAY7W/0.png b/vlm/train/DHSNrGhAY7W/0.png new file mode 100644 index 0000000000000000000000000000000000000000..f9f60bbe71505d362c2b1ea48363a15329349799 --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9057b8c17ff710418a9119ed4012c2dbe515af66c5c4c81e15e0f2a6aaaf86a6 +size 528399 diff --git a/vlm/train/DHSNrGhAY7W/1.png b/vlm/train/DHSNrGhAY7W/1.png new file mode 100644 index 0000000000000000000000000000000000000000..6e44ae434e80ff0174e893eaafa1eedd9d35a5aa --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:950fb416edc6fbc2cbb6293b8dcebd03b8d1980affd5639dae6e8b2f273c6234 +size 559568 diff --git a/vlm/train/DHSNrGhAY7W/10.png b/vlm/train/DHSNrGhAY7W/10.png new file mode 100644 index 0000000000000000000000000000000000000000..a2949c6c5d908bab4da8509b081cd3f55c5213af --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:192a80fa099a70ebf716a171658fa8af58ae791570a9cab82024c6c65eae708d +size 393541 diff --git a/vlm/train/DHSNrGhAY7W/11.png b/vlm/train/DHSNrGhAY7W/11.png new file mode 100644 index 0000000000000000000000000000000000000000..3f696a195234ad26c761d94c7592e9f385d84635 --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:44cd6b64282874d25244f2c0387f7bb34d86a253d70387339884ebf136c43704 +size 449430 diff --git a/vlm/train/DHSNrGhAY7W/12.png b/vlm/train/DHSNrGhAY7W/12.png new file mode 100644 index 0000000000000000000000000000000000000000..0a7a389ebd65fc4c2ccd007e08b529a5b707cb80 --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9423fc4a397b416131a77d340308c75134d2f768731005ea3e36ebab5993eb6c +size 301963 diff --git a/vlm/train/DHSNrGhAY7W/13.png b/vlm/train/DHSNrGhAY7W/13.png new file mode 100644 index 0000000000000000000000000000000000000000..0fd9a73610e63129434e58c4ec5ade3d9c838e9d --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f3f6c0e845b597723731ac69f1d99f5e695b001a1a4566febacb134c44f2101c +size 331558 diff --git a/vlm/train/DHSNrGhAY7W/14.png b/vlm/train/DHSNrGhAY7W/14.png new file mode 100644 index 0000000000000000000000000000000000000000..4e7c39c05dfd781c13c46e052b0d77a2d2fc3cdd --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:51e900c086e8011c005c483eeed6013ceb5ecd7066fb148a1bc4d4898c6c81bc +size 400803 diff --git a/vlm/train/DHSNrGhAY7W/15.png b/vlm/train/DHSNrGhAY7W/15.png new file mode 100644 index 0000000000000000000000000000000000000000..77bff167c6e092faddc1fc423a687eae31ca706d --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/15.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:40d1ddc8010b4e12e032b322f5b88bb48704888cc821dc6f170c39b08a203811 +size 415174 diff --git a/vlm/train/DHSNrGhAY7W/16.png b/vlm/train/DHSNrGhAY7W/16.png new file mode 100644 index 0000000000000000000000000000000000000000..105319a1a7dacc0f40484de011a31f25f713bd2f --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/16.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:28078dc5fdcb7e6ee762243e89c6eaa72a0dc7fb39dcb5a3baacf7ede4663f7b +size 368125 diff --git a/vlm/train/DHSNrGhAY7W/17.png b/vlm/train/DHSNrGhAY7W/17.png new file mode 100644 index 0000000000000000000000000000000000000000..c5420a9f883d135ae81332fdb2e4ff8eacf2c46c --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/17.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:49ddad488cd722050a0e69367bd275f2dfd1245e1a3c4160eedd3f5a2457a813 +size 349593 diff --git a/vlm/train/DHSNrGhAY7W/18.png b/vlm/train/DHSNrGhAY7W/18.png new file mode 100644 index 0000000000000000000000000000000000000000..200f06fe14bce729bcf0ac7ae3e53cba32cea893 --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/18.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:63d9390b9aa0114f9141b55c5606c2eb0dd383af18bdd89f0e0af53107f5ebed +size 361276 diff --git a/vlm/train/DHSNrGhAY7W/19.png b/vlm/train/DHSNrGhAY7W/19.png new file mode 100644 index 0000000000000000000000000000000000000000..5397585cf996cceae47a480d382887ac82898eef --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/19.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:da1bfc5f3c1df70ba4f35d58208537063d5e9f426d8b774c9b98e215131d624c +size 489827 diff --git a/vlm/train/DHSNrGhAY7W/2.png b/vlm/train/DHSNrGhAY7W/2.png new file mode 100644 index 0000000000000000000000000000000000000000..0ed745f584cd43bcb320ab2218b31952d8997896 --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1c8e4043ddd196f47c6988580504dddf84ece27499afd7c256e137394426f037 +size 529491 diff --git a/vlm/train/DHSNrGhAY7W/20.png b/vlm/train/DHSNrGhAY7W/20.png new file mode 100644 index 0000000000000000000000000000000000000000..35975f13abad005dd53dab41787ae01f4b4be16c --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/20.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:048b5dd59df9de77709435e3920f55afbf6488df4e9d018b5da85b231b3d94b8 +size 288013 diff --git a/vlm/train/DHSNrGhAY7W/21.png b/vlm/train/DHSNrGhAY7W/21.png new file mode 100644 index 0000000000000000000000000000000000000000..d7c52c17bdaf5996827c939e11ace82d994695dd --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/21.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3869542a9baecbae18095967801911883a2638789d9ebd969e31f95bf1804bbd +size 309004 diff --git a/vlm/train/DHSNrGhAY7W/22.png b/vlm/train/DHSNrGhAY7W/22.png new file mode 100644 index 0000000000000000000000000000000000000000..02cc7ec3aa9dcc89af22673a1c70c075c77e9f23 --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/22.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:554bbf617befbfab9b7a7c29c14401ceab26ef0800146c0349801af3ad1603e8 +size 498199 diff --git a/vlm/train/DHSNrGhAY7W/23.png b/vlm/train/DHSNrGhAY7W/23.png new file mode 100644 index 0000000000000000000000000000000000000000..6f84516291919c25eb237200adcc441d220ebb82 --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/23.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:56c8b06c56ba6e2234ee1739513484253af2ebfbec2d733de91cc291126aa21a +size 238520 diff --git a/vlm/train/DHSNrGhAY7W/3.png b/vlm/train/DHSNrGhAY7W/3.png new file mode 100644 index 0000000000000000000000000000000000000000..abacfea830cb012ec131cf92705664a21ec9e773 --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:73a4f5819094dfe10ba189cd504a641163be0b029fd5987ab562b846a12ddb0b +size 573694 diff --git a/vlm/train/DHSNrGhAY7W/4.png b/vlm/train/DHSNrGhAY7W/4.png new file mode 100644 index 0000000000000000000000000000000000000000..74c7bb93cccc8efd10262312054bc3570494e7d5 --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7374dde2b603aeea81f4eb3c81ae5179fe53f18cf7fd140b9ac49b738133980d +size 545550 diff --git a/vlm/train/DHSNrGhAY7W/5.png b/vlm/train/DHSNrGhAY7W/5.png new file mode 100644 index 0000000000000000000000000000000000000000..b8a1d2df5ac6871d8f7157641b92766c8798558f --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b250c47c08f089ff0bebd1da8fd9f6040c13d1aa293a24e1ec5010f7c02164c0 +size 591285 diff --git a/vlm/train/DHSNrGhAY7W/6.png b/vlm/train/DHSNrGhAY7W/6.png new file mode 100644 index 0000000000000000000000000000000000000000..8ae8579963f39b26f5a74c707e02482b36aabc5e --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:212689913f0e18decb0d61e2f76ad5a2f07881f25f7bb3d9727b5d16ff82a93d +size 621188 diff --git a/vlm/train/DHSNrGhAY7W/7.png b/vlm/train/DHSNrGhAY7W/7.png new file mode 100644 index 0000000000000000000000000000000000000000..be91d5ae53d1c345d8758faafa140a1279616223 --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e617aaace17d92dd1c7e9a14cac8b59a4724d8fa65f5916bba91ca35d2dd7b33 +size 620026 diff --git a/vlm/train/DHSNrGhAY7W/8.png b/vlm/train/DHSNrGhAY7W/8.png new file mode 100644 index 0000000000000000000000000000000000000000..e8879cfdc8b1aab5304d5a66039de057f3148e97 --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eaecdd9ad2a53e268f5c38944aa1ae8c40976487d4c2d364a6b29f952504b1fc +size 557116 diff --git a/vlm/train/DHSNrGhAY7W/9.png b/vlm/train/DHSNrGhAY7W/9.png new file mode 100644 index 0000000000000000000000000000000000000000..2b5727aa33ca1b7a461466893fdd74208a017f63 --- /dev/null +++ b/vlm/train/DHSNrGhAY7W/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c2b9f33b0bef38cfe02ba25a2e5a4366dc3094175cda38bb60e7100e7affbb9c +size 450448 diff --git a/vlm/train/H135uzZ0-/0.png b/vlm/train/H135uzZ0-/0.png new file mode 100644 index 0000000000000000000000000000000000000000..207bcc6295ac3b1ca8f9db12da3c7614adacd12a --- /dev/null +++ b/vlm/train/H135uzZ0-/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6b22ffcf79604987d1d7bc530272e3d97c6d70eaeeabc2d7fc2b1b26aea7da61 +size 419558 diff --git a/vlm/train/H135uzZ0-/1.png b/vlm/train/H135uzZ0-/1.png new file mode 100644 index 0000000000000000000000000000000000000000..c8a6027e53c530abcc39f2c8ca841825367d65d8 --- /dev/null +++ b/vlm/train/H135uzZ0-/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a425cec2439de4b106e7db13a531273909f345833aeb56cf8cc3dbee58e3faed +size 581722 diff --git a/vlm/train/H135uzZ0-/10.png b/vlm/train/H135uzZ0-/10.png new file mode 100644 index 0000000000000000000000000000000000000000..6d8b541394357b2339a25fed4d1fcd94b13f0c5d --- /dev/null +++ b/vlm/train/H135uzZ0-/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1a4fdbfe7e16fa76fe63ad5bb847e149e3f98f29eddaa2f074feef596e8ecc63 +size 146846 diff --git a/vlm/train/H135uzZ0-/2.png b/vlm/train/H135uzZ0-/2.png new file mode 100644 index 0000000000000000000000000000000000000000..fbbc6d921a8c0edc6014f5d244ce8c783eb85369 --- /dev/null +++ b/vlm/train/H135uzZ0-/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fc983f89e6d787a9039db6004290271742e38114f494b910b5ce165bb080ccd8 +size 597081 diff --git a/vlm/train/H135uzZ0-/3.png b/vlm/train/H135uzZ0-/3.png new file mode 100644 index 0000000000000000000000000000000000000000..3af762f4ed5a962b1bf802589cc7bc7cf86bdf7e --- /dev/null +++ b/vlm/train/H135uzZ0-/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0de8984d3ab9503a5edd1e3d28edae75db505b4cb719a4efc59e71cf690bae39 +size 447852 diff --git a/vlm/train/H135uzZ0-/4.png b/vlm/train/H135uzZ0-/4.png new file mode 100644 index 0000000000000000000000000000000000000000..b99979b564ff0b123179a90dc8f91e30abbfc05d --- /dev/null +++ b/vlm/train/H135uzZ0-/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ed1f3acbb864d91457d78695740dc54e8f4f764832edd7511ebf5c361a29899b +size 515399 diff --git a/vlm/train/H135uzZ0-/5.png b/vlm/train/H135uzZ0-/5.png new file mode 100644 index 0000000000000000000000000000000000000000..7fc00f5e7a39a64b52c7f22b20be6a92e14c36be --- /dev/null +++ b/vlm/train/H135uzZ0-/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cea19d30817048f1d3a0dcb8d78cf1776c146c71f93066070b3b3be7ddd20a1f +size 502262 diff --git a/vlm/train/H135uzZ0-/6.png b/vlm/train/H135uzZ0-/6.png new file mode 100644 index 0000000000000000000000000000000000000000..78ef17128452453550efe2ca53ec7af2f29b4c15 --- /dev/null +++ b/vlm/train/H135uzZ0-/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:04fd3b79e2f1c4207d76cfb11ea3f5a28a3da119714a9080b22d126626a68c12 +size 433790 diff --git a/vlm/train/H135uzZ0-/7.png b/vlm/train/H135uzZ0-/7.png new file mode 100644 index 0000000000000000000000000000000000000000..3db39d851647d68829b23075b706a6dd6dc98b37 --- /dev/null +++ b/vlm/train/H135uzZ0-/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a0f44c10bb917b2e9d365599e992d22f96b89cbea468064fd390a13300c1d76f +size 497733 diff --git a/vlm/train/H135uzZ0-/8.png b/vlm/train/H135uzZ0-/8.png new file mode 100644 index 0000000000000000000000000000000000000000..fe7b892f2967227fdd05eb9d851fdbcb2aaa4f1e --- /dev/null +++ b/vlm/train/H135uzZ0-/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:59e66352545a426f3dfaf5a4031756ea95cc113598b8f4aec8a259806670f211 +size 534305 diff --git a/vlm/train/H135uzZ0-/9.png b/vlm/train/H135uzZ0-/9.png new file mode 100644 index 0000000000000000000000000000000000000000..3aeabaed59f5fef8f6aa1a5e25fcd31074cd5229 --- /dev/null +++ b/vlm/train/H135uzZ0-/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:61163f9f50f9015bd8ebdf6e5dc4ad1a61ff3beddce197f4dfd7dbc90978e149 +size 536787 diff --git a/vlm/train/H1gDNyrKDS/0.png b/vlm/train/H1gDNyrKDS/0.png new file mode 100644 index 0000000000000000000000000000000000000000..2951b5a96de6ab1fb30e16b7b5b4a39c68e78ca5 --- /dev/null +++ b/vlm/train/H1gDNyrKDS/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a426e50910dc3a3e675dc1644b49a809cf96fe1991fe2103a59f52d9eb0637b6 +size 508789 diff --git a/vlm/train/H1gDNyrKDS/1.png b/vlm/train/H1gDNyrKDS/1.png new file mode 100644 index 0000000000000000000000000000000000000000..c2bdf784174f95f120e0aebfabe70e62ece9107a --- /dev/null +++ b/vlm/train/H1gDNyrKDS/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3883ecf4395676f4914d737c4e134a1f6e327e4705a1c67ddba01abf851e8748 +size 500175 diff --git a/vlm/train/H1gDNyrKDS/10.png b/vlm/train/H1gDNyrKDS/10.png new file mode 100644 index 0000000000000000000000000000000000000000..a3b4d509548f7db23a716e49dd431761849f3eca --- /dev/null +++ b/vlm/train/H1gDNyrKDS/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3ca1a53d763a6a8aeca62f7d3a1eeef257db2174602e758978676bbcfbc50883 +size 562408 diff --git a/vlm/train/H1gDNyrKDS/11.png b/vlm/train/H1gDNyrKDS/11.png new file mode 100644 index 0000000000000000000000000000000000000000..a87a5c5570edceadd1055804a0bd47b030dbd169 --- /dev/null +++ b/vlm/train/H1gDNyrKDS/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3562690395b7cd0affa1d12db54ef6abd236eef6fb2e0b5d726be7a79a50cb2d +size 485140 diff --git a/vlm/train/H1gDNyrKDS/12.png b/vlm/train/H1gDNyrKDS/12.png new file mode 100644 index 0000000000000000000000000000000000000000..197423544872adb7cbad9596a70ec80532b22c66 --- /dev/null +++ b/vlm/train/H1gDNyrKDS/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2af3433c81f55a3e965acb63133ff11b0f6afe5960b851915b83fde652f02caf +size 442539 diff --git a/vlm/train/H1gDNyrKDS/13.png b/vlm/train/H1gDNyrKDS/13.png new file mode 100644 index 0000000000000000000000000000000000000000..c5d37564c50667c5c11afc97377b0c2b38f05cc3 --- /dev/null +++ b/vlm/train/H1gDNyrKDS/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:53cbb41f389e140f9019152c1fd65243a8b0d54ca6a61f7f6b4eabbf12a4fb31 +size 486307 diff --git a/vlm/train/H1gDNyrKDS/14.png b/vlm/train/H1gDNyrKDS/14.png new file mode 100644 index 0000000000000000000000000000000000000000..ccef873028e51a9b71ad2b83caad304e7cd56698 --- /dev/null +++ b/vlm/train/H1gDNyrKDS/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d4e4a574d60be803ce198abb6858b49673e3a5c8c52ff5ba798049c54396334e +size 394918 diff --git a/vlm/train/H1gDNyrKDS/15.png b/vlm/train/H1gDNyrKDS/15.png new file mode 100644 index 0000000000000000000000000000000000000000..ddcd4023b87af3636923c35178958ee4c1639550 --- /dev/null +++ b/vlm/train/H1gDNyrKDS/15.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d544395540bdb8aded4b9fdf097b2bdc9c01056de70f7fb06ccb2c54513b5ea5 +size 550685 diff --git a/vlm/train/H1gDNyrKDS/16.png b/vlm/train/H1gDNyrKDS/16.png new file mode 100644 index 0000000000000000000000000000000000000000..e9eef849085cc1d9894c625ee4578ebe51fa552a --- /dev/null +++ b/vlm/train/H1gDNyrKDS/16.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:675d455c3b47a16846a8896f9f0c7bb284457df9deb29181a2f7c3c026f9708c +size 314663 diff --git a/vlm/train/H1gDNyrKDS/17.png b/vlm/train/H1gDNyrKDS/17.png new file mode 100644 index 0000000000000000000000000000000000000000..dfae0ca8215caa8c7dff36e59f2a2e8873e61e77 --- /dev/null +++ b/vlm/train/H1gDNyrKDS/17.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9c85895b641e223d043e0321712097555599be99a21afc0add3b90dd15ddbb0e +size 225765 diff --git a/vlm/train/H1gDNyrKDS/18.png b/vlm/train/H1gDNyrKDS/18.png new file mode 100644 index 0000000000000000000000000000000000000000..2b083439617ce886c107f5ebf760920004278faf --- /dev/null +++ b/vlm/train/H1gDNyrKDS/18.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aa0a96bdd935141b7c525136535ba19a9cab544807fee542a3cccc61e1b07d87 +size 559089 diff --git a/vlm/train/H1gDNyrKDS/19.png b/vlm/train/H1gDNyrKDS/19.png new file mode 100644 index 0000000000000000000000000000000000000000..040f0a715b6b64896b27e0f7e9b6bce4c7d67d22 --- /dev/null +++ b/vlm/train/H1gDNyrKDS/19.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f6ec5a7e377eee99cf01c727f706b6b51270836c1d4f47195792ae0419d393a8 +size 526855 diff --git a/vlm/train/H1gDNyrKDS/2.png b/vlm/train/H1gDNyrKDS/2.png new file mode 100644 index 0000000000000000000000000000000000000000..6bcb938661e7c68912da76d6d3354de9550ece91 --- /dev/null +++ b/vlm/train/H1gDNyrKDS/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e20867b04f5391a16d18e739d3d5b5367ea3c2df590599508e5bd4783c7c7d76 +size 565306 diff --git a/vlm/train/H1gDNyrKDS/20.png b/vlm/train/H1gDNyrKDS/20.png new file mode 100644 index 0000000000000000000000000000000000000000..3fbf3e5db90d3dd34e32b88ce52f23d3a76ff20a --- /dev/null +++ b/vlm/train/H1gDNyrKDS/20.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f9a425d2b9cdf4b17787e57893c04b6ea49758a8afbdc76e72e24954252d7360 +size 549949 diff --git a/vlm/train/H1gDNyrKDS/21.png b/vlm/train/H1gDNyrKDS/21.png new file mode 100644 index 0000000000000000000000000000000000000000..50476046bf998b316fec606559c788b65c718cd7 --- /dev/null +++ b/vlm/train/H1gDNyrKDS/21.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d02d49c1d1da92a72075cacc012f3aa6bae28224b3230730c95328f85939330f +size 339222 diff --git a/vlm/train/H1gDNyrKDS/22.png b/vlm/train/H1gDNyrKDS/22.png new file mode 100644 index 0000000000000000000000000000000000000000..1ddd16523ab44551a301fae602d559897fb1e7f8 --- /dev/null +++ b/vlm/train/H1gDNyrKDS/22.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fdbadd6d332a5d35ddfa82fb2925daee919d9fd1735382bff7dc3acb13248b60 +size 515119 diff --git a/vlm/train/H1gDNyrKDS/23.png b/vlm/train/H1gDNyrKDS/23.png new file mode 100644 index 0000000000000000000000000000000000000000..0067c2df7ac20e7cf7a6552ddeac49136dcab8b9 --- /dev/null +++ b/vlm/train/H1gDNyrKDS/23.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:60a29deeb60f46dbda0c6c7c1620a2efc661ec2b8280998b615f1a2f9f5cd27c +size 397010 diff --git a/vlm/train/H1gDNyrKDS/24.png b/vlm/train/H1gDNyrKDS/24.png new file mode 100644 index 0000000000000000000000000000000000000000..6d52c832f2fc5f116e521a8ad9d8fcb8452843b7 --- /dev/null +++ b/vlm/train/H1gDNyrKDS/24.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ba774f4d34ad6360b3b6239e82fe70f95a34bf0a309be8d182483b168678c295 +size 419542 diff --git a/vlm/train/H1gDNyrKDS/25.png b/vlm/train/H1gDNyrKDS/25.png new file mode 100644 index 0000000000000000000000000000000000000000..5e5bbedcd198d5a0b1fba166be2df92df4d48ae8 --- /dev/null +++ b/vlm/train/H1gDNyrKDS/25.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:774a25616d5286e518020805f536d05cca9d5c15741078f6dee1646e38551a5b +size 428835 diff --git a/vlm/train/H1gDNyrKDS/26.png b/vlm/train/H1gDNyrKDS/26.png new file mode 100644 index 0000000000000000000000000000000000000000..eac13de89ef3f34f06836db9b214b4489c74bd8c --- /dev/null +++ b/vlm/train/H1gDNyrKDS/26.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0f86ba01c460ae0ddad66df1dcec11bdc68cae94195431e51556c779641ef5ee +size 280182 diff --git a/vlm/train/H1gDNyrKDS/27.png b/vlm/train/H1gDNyrKDS/27.png new file mode 100644 index 0000000000000000000000000000000000000000..54fd7f2e495919e7c2d96c8e39236f8db5093161 --- /dev/null +++ b/vlm/train/H1gDNyrKDS/27.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c8d6b7121938c31410a2f472367a11c0b59d8cdab700c6d3e3762cb7398706ed +size 298260 diff --git a/vlm/train/H1gDNyrKDS/3.png b/vlm/train/H1gDNyrKDS/3.png new file mode 100644 index 0000000000000000000000000000000000000000..0e498c873799f4df46e0333cbe19b37c0f3b62be --- /dev/null +++ b/vlm/train/H1gDNyrKDS/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2d1baeafd06b8acb23d8d5497a006bada6441cee035464bf3f0c8671de9f4ede +size 619229 diff --git a/vlm/train/H1gDNyrKDS/4.png b/vlm/train/H1gDNyrKDS/4.png new file mode 100644 index 0000000000000000000000000000000000000000..7788316a141a553941d68b7bdc6ea6f1f1a7aee0 --- /dev/null +++ b/vlm/train/H1gDNyrKDS/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0cc7790e2289c78bfca96e20010d239f76d1e8b726d4f120919dfd243ce17260 +size 580413 diff --git a/vlm/train/H1gDNyrKDS/5.png b/vlm/train/H1gDNyrKDS/5.png new file mode 100644 index 0000000000000000000000000000000000000000..326a9e24b827536b0a042872084cc8cce44bd96c --- /dev/null +++ b/vlm/train/H1gDNyrKDS/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:06918a2a132efbadeace2a2e7fc92d1b65bf2b95e43493ed5b25c0280bcb9346 +size 515715 diff --git a/vlm/train/H1gDNyrKDS/6.png b/vlm/train/H1gDNyrKDS/6.png new file mode 100644 index 0000000000000000000000000000000000000000..e2dc97ea11283f5e0ac161ec511e02e9f2a0dda6 --- /dev/null +++ b/vlm/train/H1gDNyrKDS/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5eaed4f17e8e3da79148b1ecd62b8106721edaec2fae359faf10ce97254df643 +size 564187 diff --git a/vlm/train/H1gDNyrKDS/7.png b/vlm/train/H1gDNyrKDS/7.png new file mode 100644 index 0000000000000000000000000000000000000000..4c4db81726e196a31c1f80a20dcb87dbd7b730b5 --- /dev/null +++ b/vlm/train/H1gDNyrKDS/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1deea2619df1912fb0b9b34bcbf698d437493e9969530aaa5139b179db270ce0 +size 566880 diff --git a/vlm/train/H1gDNyrKDS/8.png b/vlm/train/H1gDNyrKDS/8.png new file mode 100644 index 0000000000000000000000000000000000000000..fe4465c22c429c586e032e18dc3b0c2fce7205e9 --- /dev/null +++ b/vlm/train/H1gDNyrKDS/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:540bb6ea1a26ecaa098ace7f3f731dc2e92cf537c4cf8f2305ace3034108b74a +size 624201 diff --git a/vlm/train/H1gDNyrKDS/9.png b/vlm/train/H1gDNyrKDS/9.png new file mode 100644 index 0000000000000000000000000000000000000000..0766ce21dc24ca93adaad1ad2bb0902813fe787f --- /dev/null +++ b/vlm/train/H1gDNyrKDS/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3d678118a53af99242fd1f119f3c3dd844cbdb3e002175d268f3edaf3759f100 +size 544664 diff --git a/vlm/train/H1ldzA4tPr/0.png b/vlm/train/H1ldzA4tPr/0.png new file mode 100644 index 0000000000000000000000000000000000000000..fd2502874504b6a1011300eeac3bf231a2d73602 --- /dev/null +++ b/vlm/train/H1ldzA4tPr/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4f9d9dbed833390395d1ab067ea9fb10a2080d554d31e45ee4d4eec37aebb565 +size 499934 diff --git a/vlm/train/H1ldzA4tPr/1.png b/vlm/train/H1ldzA4tPr/1.png new file mode 100644 index 0000000000000000000000000000000000000000..d5d8a6c2f834dcfa9ebd977e813d9bdc690d906a --- /dev/null +++ b/vlm/train/H1ldzA4tPr/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b02871562d789aa42bef5c6b17dacdc8f3db00499b835aaf2a10e092f4e096db +size 621691 diff --git a/vlm/train/H1ldzA4tPr/10.png b/vlm/train/H1ldzA4tPr/10.png new file mode 100644 index 0000000000000000000000000000000000000000..c4f41fa0c3bd6662dc0bd5264ced08697dca3cb3 --- /dev/null +++ b/vlm/train/H1ldzA4tPr/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b5ffcb1cccc83d5364ddbde5675d67290bb2515b7807500612dc9b8776724814 +size 518255 diff --git a/vlm/train/H1ldzA4tPr/11.png b/vlm/train/H1ldzA4tPr/11.png new file mode 100644 index 0000000000000000000000000000000000000000..7a1d215fd55b77a13727ddcf8f499c4d9cf2936f --- /dev/null +++ b/vlm/train/H1ldzA4tPr/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fac34b561708452c8fc2ed47a7baa0ec8bc8bafbf5a5106ceef88f70861c9f01 +size 92216 diff --git a/vlm/train/H1ldzA4tPr/12.png b/vlm/train/H1ldzA4tPr/12.png new file mode 100644 index 0000000000000000000000000000000000000000..5487325fe357926828c1134dd44889a9cfa944b2 --- /dev/null +++ b/vlm/train/H1ldzA4tPr/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:93b7609366eb669176347579ddcc2e04596a127f65d7625852374960fb47865d +size 551551 diff --git a/vlm/train/H1ldzA4tPr/13.png b/vlm/train/H1ldzA4tPr/13.png new file mode 100644 index 0000000000000000000000000000000000000000..2bc6a41f8bb7abe148888e4ca706b10f423b2d95 --- /dev/null +++ b/vlm/train/H1ldzA4tPr/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5a28e2a8dfba6f89565cf51c9ee7de0be7e25a4c1170c31b8c8d36483528cf13 +size 312962 diff --git a/vlm/train/H1ldzA4tPr/2.png b/vlm/train/H1ldzA4tPr/2.png new file mode 100644 index 0000000000000000000000000000000000000000..f45fe5e0e6491eb4bddbee9ad3178c7e16410196 --- /dev/null +++ b/vlm/train/H1ldzA4tPr/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9c8008c94cf8db81a7554c5c197f3931b554582878546ffb2ac153e403d0cf45 +size 548709 diff --git a/vlm/train/H1ldzA4tPr/3.png b/vlm/train/H1ldzA4tPr/3.png new file mode 100644 index 0000000000000000000000000000000000000000..71237419561f10ab30d0bf9b0b7efdcb09570a4d --- /dev/null +++ b/vlm/train/H1ldzA4tPr/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d3088e79acaf40ab5a34e98c8c4512ea8c46426bd6b1ede1340af5d3a6ea3cde +size 482269 diff --git a/vlm/train/H1ldzA4tPr/4.png b/vlm/train/H1ldzA4tPr/4.png new file mode 100644 index 0000000000000000000000000000000000000000..3797cf2a5c45e18c265645e17bf3cedec0d3ba4d --- /dev/null +++ b/vlm/train/H1ldzA4tPr/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e24117f56ff356205bd6531f748362582298a93787bdc7293a074bbf81ea2926 +size 558896 diff --git a/vlm/train/H1ldzA4tPr/5.png b/vlm/train/H1ldzA4tPr/5.png new file mode 100644 index 0000000000000000000000000000000000000000..b4e2332d6e0f75258cf41c0c0480938948741a94 --- /dev/null +++ b/vlm/train/H1ldzA4tPr/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:85d2fd151bcfac4fff9a588c5547f16094f444e02965863490ba047fdf131459 +size 487466 diff --git a/vlm/train/H1ldzA4tPr/6.png b/vlm/train/H1ldzA4tPr/6.png new file mode 100644 index 0000000000000000000000000000000000000000..def5c93c54a3194e899df881f3a7b1ac2248bc0d --- /dev/null +++ b/vlm/train/H1ldzA4tPr/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b109602c0836adb0f1ce38a47711a6c818028b822693a2ed14ec6231250791ff +size 700568 diff --git a/vlm/train/H1ldzA4tPr/7.png b/vlm/train/H1ldzA4tPr/7.png new file mode 100644 index 0000000000000000000000000000000000000000..0c5769df47f0e65ccf94fca9c20f99f4f835748a --- /dev/null +++ b/vlm/train/H1ldzA4tPr/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0900a46fe2a629e39e3bcb10036a412d8402713a701890870f8c7ea54fe783b7 +size 564983 diff --git a/vlm/train/H1ldzA4tPr/8.png b/vlm/train/H1ldzA4tPr/8.png new file mode 100644 index 0000000000000000000000000000000000000000..877781b05d6fa6b4b6cea82d8e731201e4a5f8eb --- /dev/null +++ b/vlm/train/H1ldzA4tPr/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9932c2608215c53b5e8ed01908742d890d2c6430129c83925ab349adb7776f3e +size 585910 diff --git a/vlm/train/H1ldzA4tPr/9.png b/vlm/train/H1ldzA4tPr/9.png new file mode 100644 index 0000000000000000000000000000000000000000..aed8f0d31d9679874dfeb3930a3ee31df3d93e98 --- /dev/null +++ b/vlm/train/H1ldzA4tPr/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:05abaf9745ee196128ec373c281219e060e3df8ca62fbc56264014df216d3c3b +size 488876 diff --git a/vlm/train/H1xQSjCqFQ/0.png b/vlm/train/H1xQSjCqFQ/0.png new file mode 100644 index 0000000000000000000000000000000000000000..84aed91daa29e985f5513506a7ef29883b8b2298 --- /dev/null +++ b/vlm/train/H1xQSjCqFQ/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ee45c81d83acd1f16c05faaa829244b399d8caa0e6ffd638c868511f6ef28bf3 +size 494317 diff --git a/vlm/train/H1xQSjCqFQ/1.png b/vlm/train/H1xQSjCqFQ/1.png new file mode 100644 index 0000000000000000000000000000000000000000..8e2bb0b54d9ce06d7e4a90ef345f4916780def68 --- /dev/null +++ b/vlm/train/H1xQSjCqFQ/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f1820876e5e6865f01ee9554e190f3a5958fc0f697486bb9f2bb5cc1d1eb40a6 +size 539248 diff --git a/vlm/train/H1xQSjCqFQ/10.png b/vlm/train/H1xQSjCqFQ/10.png new file mode 100644 index 0000000000000000000000000000000000000000..229f0243380271833aeba25216474c4417115e18 --- /dev/null +++ b/vlm/train/H1xQSjCqFQ/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f682ae1c824218e869080df1cd90fa2722c02ff5eabcd9de3d7c52aa156d685b +size 444765 diff --git a/vlm/train/H1xQSjCqFQ/11.png b/vlm/train/H1xQSjCqFQ/11.png new file mode 100644 index 0000000000000000000000000000000000000000..1d181cf3bfb70994fc66df6f696a773f133027bc --- /dev/null +++ b/vlm/train/H1xQSjCqFQ/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d09e0d3de7dcc29b4f46d1f9a4b678a7b7b7dbd6d5ada484e2a36abc1aead90a +size 578226 diff --git a/vlm/train/H1xQSjCqFQ/12.png b/vlm/train/H1xQSjCqFQ/12.png new file mode 100644 index 0000000000000000000000000000000000000000..022f8f7a30610d76984bfe17bdec205b402c8a0b --- /dev/null +++ b/vlm/train/H1xQSjCqFQ/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b64454fd8ecb3bc01a24d43209c79a860e6224fe91c5d73fed4a2e0fe2c0e493 +size 850509 diff --git a/vlm/train/H1xQSjCqFQ/13.png b/vlm/train/H1xQSjCqFQ/13.png new file mode 100644 index 0000000000000000000000000000000000000000..63d11b5569d87572a5d3e3414e43feb711129163 --- /dev/null +++ b/vlm/train/H1xQSjCqFQ/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4248382881e1d86f3285e70735933695ded0e395d27dc54edafa9e90b83a0ec0 +size 1684658 diff --git a/vlm/train/H1xQSjCqFQ/14.png b/vlm/train/H1xQSjCqFQ/14.png new file mode 100644 index 0000000000000000000000000000000000000000..d89314319daa308af68ed1633c7386ef32a2d4b1 --- /dev/null +++ b/vlm/train/H1xQSjCqFQ/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c03f4d6abd699eef077ac306c387ff1bddf39a86630ceec4cd7d2bfc4d070661 +size 525297 diff --git a/vlm/train/H1xQSjCqFQ/15.png b/vlm/train/H1xQSjCqFQ/15.png new file mode 100644 index 0000000000000000000000000000000000000000..bf75535fff6f0bc44750958c299c53e069bd6000 --- /dev/null +++ b/vlm/train/H1xQSjCqFQ/15.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:10ad0b1510eb1604c787efe79184720757e14a960e715fd7404917d23c2d9f48 +size 462048 diff --git a/vlm/train/H1xQSjCqFQ/16.png b/vlm/train/H1xQSjCqFQ/16.png new file mode 100644 index 0000000000000000000000000000000000000000..919303bdba4791bd4fc19fc9a901e9b6eef1b026 --- /dev/null +++ b/vlm/train/H1xQSjCqFQ/16.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:92b2f572eb84617acb037887c702e46151f5eddfcc14a9950ee3694035206cc8 +size 425052 diff --git a/vlm/train/H1xQSjCqFQ/2.png b/vlm/train/H1xQSjCqFQ/2.png new file mode 100644 index 0000000000000000000000000000000000000000..2db59fc6d9abeaf6294986537ae4ea2e3cb63b61 --- /dev/null +++ b/vlm/train/H1xQSjCqFQ/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9784fe46f9f91c368b50c074879ef326c7a8da9ce722d306ac3c70320c6ae8cd +size 508966 diff --git a/vlm/train/H1xQSjCqFQ/3.png b/vlm/train/H1xQSjCqFQ/3.png new file mode 100644 index 0000000000000000000000000000000000000000..f113458bd62ebdb9493a22e0863b2b099ea023a4 --- /dev/null +++ b/vlm/train/H1xQSjCqFQ/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0ec0b9c26e40736d98fa99d8a8e5f7f2d8404c63410343de18b1d3f15a2d07d8 +size 483331 diff --git a/vlm/train/H1xQSjCqFQ/4.png b/vlm/train/H1xQSjCqFQ/4.png new file mode 100644 index 0000000000000000000000000000000000000000..947cda44029dd020f734fb214e3391baf73331f1 --- /dev/null +++ b/vlm/train/H1xQSjCqFQ/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7fd97da0902ca3d8dfc49df78c9db6aac96c5bc056df9340b58bad44daa3e0bc +size 702609 diff --git a/vlm/train/H1xQSjCqFQ/5.png b/vlm/train/H1xQSjCqFQ/5.png new file mode 100644 index 0000000000000000000000000000000000000000..470b8907b26f71667d8da769e6cfb5e2df146048 --- /dev/null +++ b/vlm/train/H1xQSjCqFQ/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:50f3c744646647a0a7a3397996a20532722bec9674864c60ee42054d93acb749 +size 577023 diff --git a/vlm/train/H1xQSjCqFQ/6.png b/vlm/train/H1xQSjCqFQ/6.png new file mode 100644 index 0000000000000000000000000000000000000000..0a894a2838074a8c6c38cba1acb3cd336fc21b27 --- /dev/null +++ b/vlm/train/H1xQSjCqFQ/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cd63a019c46c5e9c8be7100c3d132a45d0f5f5590241f5b76e5901d7907f1b63 +size 972323 diff --git a/vlm/train/H1xQSjCqFQ/7.png b/vlm/train/H1xQSjCqFQ/7.png new file mode 100644 index 0000000000000000000000000000000000000000..6b447253487e7871aa8532af0d3372f56694101f --- /dev/null +++ b/vlm/train/H1xQSjCqFQ/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b9dbc7de8ed30759d42c36fd0555ece8d6eed5ed347c158890aa8a694c89b7e4 +size 645418 diff --git a/vlm/train/H1xQSjCqFQ/8.png b/vlm/train/H1xQSjCqFQ/8.png new file mode 100644 index 0000000000000000000000000000000000000000..506c1befd14d9b0ce466df1c082c4ce5a7977af7 --- /dev/null +++ b/vlm/train/H1xQSjCqFQ/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a37902715459ca5dc07c1793451ccb16d6d6a10d06793029e9ad870f28dd2e69 +size 487425 diff --git a/vlm/train/H1xQSjCqFQ/9.png b/vlm/train/H1xQSjCqFQ/9.png new file mode 100644 index 0000000000000000000000000000000000000000..005a7dfc6a6c6d1c8689488f1877862c0c535575 --- /dev/null +++ b/vlm/train/H1xQSjCqFQ/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c40cb861c9173e558324cb2d4b93fb862316d14c78603cde2704c88b20623a0b +size 497902 diff --git a/vlm/train/Hk6WhagRW/0.png b/vlm/train/Hk6WhagRW/0.png new file mode 100644 index 0000000000000000000000000000000000000000..5150648964a4c3fd9732095d5ceb97249d3cc2e4 --- /dev/null +++ b/vlm/train/Hk6WhagRW/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d512e72b69eba2ed3f960817bcb5c4b85efd0977e24055ffc4141e5c715aded7 +size 461471 diff --git a/vlm/train/Hk6WhagRW/1.png b/vlm/train/Hk6WhagRW/1.png new file mode 100644 index 0000000000000000000000000000000000000000..81c61166478338726398267f548bc312c396c26d --- /dev/null +++ b/vlm/train/Hk6WhagRW/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9a7f52f153024c8f8fc8918713160ee55234ade8e9238f5e71c7cc74e0350f51 +size 555631 diff --git a/vlm/train/Hk6WhagRW/10.png b/vlm/train/Hk6WhagRW/10.png new file mode 100644 index 0000000000000000000000000000000000000000..9e35ee34e2fe0f4afe91111f3cd119f9ab08eb25 --- /dev/null +++ b/vlm/train/Hk6WhagRW/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c9cefabc0d28d3661e479dc2c7c0d7c2bacceb821c9fac09904a2e768ef03ff8 +size 505256 diff --git a/vlm/train/Hk6WhagRW/11.png b/vlm/train/Hk6WhagRW/11.png new file mode 100644 index 0000000000000000000000000000000000000000..9f9d0e98ef7546b15bdfabf120539135687cc3af --- /dev/null +++ b/vlm/train/Hk6WhagRW/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:858789ca92f8fe142435417beaa4e41295ab9b2e07737a403ca3bd87870025fb +size 538173 diff --git a/vlm/train/Hk6WhagRW/12.png b/vlm/train/Hk6WhagRW/12.png new file mode 100644 index 0000000000000000000000000000000000000000..69047ef2c5840262944342c47a2bbe7607bf85da --- /dev/null +++ b/vlm/train/Hk6WhagRW/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7f2d0ab1aa65884561d6b4f1d8144f82c276258821f2dfe63df6968943a8b09d +size 356097 diff --git a/vlm/train/Hk6WhagRW/13.png b/vlm/train/Hk6WhagRW/13.png new file mode 100644 index 0000000000000000000000000000000000000000..7afe4e4fd46f3da0e2c8f416592e174267525de8 --- /dev/null +++ b/vlm/train/Hk6WhagRW/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:051a1a08ed9d8d0ef4977600f106189222ec6a635becff1346bef9d87e08d164 +size 245500 diff --git a/vlm/train/Hk6WhagRW/14.png b/vlm/train/Hk6WhagRW/14.png new file mode 100644 index 0000000000000000000000000000000000000000..51be8258a2e7620cfe0929e8d3e296da3cdb460f --- /dev/null +++ b/vlm/train/Hk6WhagRW/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b8abae4f6ff359452c969e6086fcc04440048489dc15877ac2ea20112661b3ee +size 250640 diff --git a/vlm/train/Hk6WhagRW/2.png b/vlm/train/Hk6WhagRW/2.png new file mode 100644 index 0000000000000000000000000000000000000000..0998c253f927a22013b6f06ef81cd5c28de0916b --- /dev/null +++ b/vlm/train/Hk6WhagRW/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7c31ea7a3962879096e4b06c4221ce55bffabbe6f09ff090815c967faecd9e1a +size 540222 diff --git a/vlm/train/Hk6WhagRW/3.png b/vlm/train/Hk6WhagRW/3.png new file mode 100644 index 0000000000000000000000000000000000000000..8def7dff965f83b06793464cc644725b97c85365 --- /dev/null +++ b/vlm/train/Hk6WhagRW/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9eae9859dcaa52f6a8078a01035286b5981cc790bac2a0652f3e1677ca745110 +size 522417 diff --git a/vlm/train/Hk6WhagRW/4.png b/vlm/train/Hk6WhagRW/4.png new file mode 100644 index 0000000000000000000000000000000000000000..bc75628b156702945976ab75f501e0fcd3050a6c --- /dev/null +++ b/vlm/train/Hk6WhagRW/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:31c95591e6a0c42e1cd44bb985da1061e0ec92890883f2a2717e2016977defb8 +size 486240 diff --git a/vlm/train/Hk6WhagRW/5.png b/vlm/train/Hk6WhagRW/5.png new file mode 100644 index 0000000000000000000000000000000000000000..7b6659f489a19fd65dcf695e41dd8f8e434a911f --- /dev/null +++ b/vlm/train/Hk6WhagRW/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:17355a46adc1a5805f534ed7a9adce3bd7b4aa6c4dc92cc584aaf615b7dcdeaa +size 434521 diff --git a/vlm/train/Hk6WhagRW/6.png b/vlm/train/Hk6WhagRW/6.png new file mode 100644 index 0000000000000000000000000000000000000000..2641fc15819605dc7106bb101a0e9c2950dd87bc --- /dev/null +++ b/vlm/train/Hk6WhagRW/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2aa95997c3ab6e1c1031c85c9b699fd761f895ccc91f190ed22bcd43f24e4184 +size 436885 diff --git a/vlm/train/Hk6WhagRW/7.png b/vlm/train/Hk6WhagRW/7.png new file mode 100644 index 0000000000000000000000000000000000000000..3dd5473222369442fa5442b9c85b4aa38bd17b34 --- /dev/null +++ b/vlm/train/Hk6WhagRW/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6b92ca4d1d20778fc9b9d4dfb11111c071606a82cccd9c859aee168bba2a21a2 +size 358834 diff --git a/vlm/train/Hk6WhagRW/8.png b/vlm/train/Hk6WhagRW/8.png new file mode 100644 index 0000000000000000000000000000000000000000..2649f4f633d6a5cd98b478c24041c397eb650e1b --- /dev/null +++ b/vlm/train/Hk6WhagRW/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:183f5252e475253311bc2cb7fc24f3e4b2200e7c04acba54233e3593655b7f87 +size 503502 diff --git a/vlm/train/Hk6WhagRW/9.png b/vlm/train/Hk6WhagRW/9.png new file mode 100644 index 0000000000000000000000000000000000000000..b10330af234c16a9becd3c7f32a6bbc4c6a34305 --- /dev/null +++ b/vlm/train/Hk6WhagRW/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2a6def88a7f53f151adb8b24c1b5810d9d632655b7a2da90c2925e9087e14f97 +size 453442 diff --git a/vlm/train/Hkla1eHFvS/0.png b/vlm/train/Hkla1eHFvS/0.png new file mode 100644 index 0000000000000000000000000000000000000000..e009d8705cc96b06bb0d234203e61d4844d010c6 --- /dev/null +++ b/vlm/train/Hkla1eHFvS/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec2bc01d8edde5644c5fe2bbcad52c4b14195eba1812b593c719b6e2359ac2f6 +size 487397 diff --git a/vlm/train/Hkla1eHFvS/1.png b/vlm/train/Hkla1eHFvS/1.png new file mode 100644 index 0000000000000000000000000000000000000000..66a8db0d58249457f4c6dba7654ffdb8efa8222a --- /dev/null +++ b/vlm/train/Hkla1eHFvS/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c641989d353d127925226f8fa7410ad2bc9fdd49ad8ee84048f77898eb7bfd9f +size 623303 diff --git a/vlm/train/Hkla1eHFvS/10.png b/vlm/train/Hkla1eHFvS/10.png new file mode 100644 index 0000000000000000000000000000000000000000..235776afa682880bc71dee2e54ea244c9e16f47b --- /dev/null +++ b/vlm/train/Hkla1eHFvS/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6ad40ea10094ae33d47fe222695dcdd8c2a34186f81cc57704a55e9aafb14f86 +size 517931 diff --git a/vlm/train/Hkla1eHFvS/11.png b/vlm/train/Hkla1eHFvS/11.png new file mode 100644 index 0000000000000000000000000000000000000000..9fb23df520fde07153157f806d8f486e713186bb --- /dev/null +++ b/vlm/train/Hkla1eHFvS/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ee56555f11767d5856265503c82d611027759db70ef0fbce54872c99780fa90f +size 460396 diff --git a/vlm/train/Hkla1eHFvS/12.png b/vlm/train/Hkla1eHFvS/12.png new file mode 100644 index 0000000000000000000000000000000000000000..f5e0f10e93ffb9c768f60bdf196b5d45d733c1a5 --- /dev/null +++ b/vlm/train/Hkla1eHFvS/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:810a0a3060a58573e55a04feebe01767bc578334332634afb37bdb1772796be7 +size 532070 diff --git a/vlm/train/Hkla1eHFvS/13.png b/vlm/train/Hkla1eHFvS/13.png new file mode 100644 index 0000000000000000000000000000000000000000..32d270557af56bf796a87992d71fdea4000e756b --- /dev/null +++ b/vlm/train/Hkla1eHFvS/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:231d4f67fdb91aee240eb299adddb987a85877eaa61fb0779c25807ef2ded98a +size 480196 diff --git a/vlm/train/Hkla1eHFvS/14.png b/vlm/train/Hkla1eHFvS/14.png new file mode 100644 index 0000000000000000000000000000000000000000..eec0b75b7b36524b3be454e6136d78aa4766f720 --- /dev/null +++ b/vlm/train/Hkla1eHFvS/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f9fe038c9e09672185694fddfd985b7609b302f143110569ccdbd7e46cfea96e +size 423353 diff --git a/vlm/train/Hkla1eHFvS/15.png b/vlm/train/Hkla1eHFvS/15.png new file mode 100644 index 0000000000000000000000000000000000000000..19be210c5da3bc9f8defca08dd4ac545262ea502 --- /dev/null +++ b/vlm/train/Hkla1eHFvS/15.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ea04d27709c55b90c0630a70c60a9a20f550ba2e939c1b3764062450fa006218 +size 505415 diff --git a/vlm/train/Hkla1eHFvS/16.png b/vlm/train/Hkla1eHFvS/16.png new file mode 100644 index 0000000000000000000000000000000000000000..99ba36edff3d3a5aa869fa54cb8f743f39726b1f --- /dev/null +++ b/vlm/train/Hkla1eHFvS/16.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3d80470faf76b89403644354066bdfaf237b08b803077e5efce7ce32c7112210 +size 450083 diff --git a/vlm/train/Hkla1eHFvS/17.png b/vlm/train/Hkla1eHFvS/17.png new file mode 100644 index 0000000000000000000000000000000000000000..8b92ef457c2794872c4e16a41756a88719875b08 --- /dev/null +++ b/vlm/train/Hkla1eHFvS/17.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b7c82ec24247082437cc8d9c30c1f5fe09dbb371d679957c9ca3659ccbec02e3 +size 589817 diff --git a/vlm/train/Hkla1eHFvS/18.png b/vlm/train/Hkla1eHFvS/18.png new file mode 100644 index 0000000000000000000000000000000000000000..8de52eb75068ad02e2e12744e7327aeb7e366ede --- /dev/null +++ b/vlm/train/Hkla1eHFvS/18.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:19f7d75ab019ab78b6e935d6380c2a31d67f9bee2f825e4e0a2ba95205cba38f +size 364419 diff --git a/vlm/train/Hkla1eHFvS/19.png b/vlm/train/Hkla1eHFvS/19.png new file mode 100644 index 0000000000000000000000000000000000000000..a5bf38c9564b05eb4705a1ca16b61cad2a2d5a24 --- /dev/null +++ b/vlm/train/Hkla1eHFvS/19.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:098442125126357c98cec695496b5193d97e5a46ddf415762e5fc07d929be648 +size 299005 diff --git a/vlm/train/Hkla1eHFvS/2.png b/vlm/train/Hkla1eHFvS/2.png new file mode 100644 index 0000000000000000000000000000000000000000..e685ef2515cee7770b086dfe2213822d6489b1ce --- /dev/null +++ b/vlm/train/Hkla1eHFvS/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4ccf11e0f903a71d3c910393d886837659093dd669b5fdb5bb53da3224046f0f +size 592365 diff --git a/vlm/train/Hkla1eHFvS/20.png b/vlm/train/Hkla1eHFvS/20.png new file mode 100644 index 0000000000000000000000000000000000000000..7d10200acb73da2c12482bd67a4631b05dd42841 --- /dev/null +++ b/vlm/train/Hkla1eHFvS/20.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3778bf649f0af1b961038322d9fc1341d41f82d69870ed48ad29b71669e20748 +size 488695 diff --git a/vlm/train/Hkla1eHFvS/3.png b/vlm/train/Hkla1eHFvS/3.png new file mode 100644 index 0000000000000000000000000000000000000000..70628b6047ed39b84000a6d11096935be346f7ff --- /dev/null +++ b/vlm/train/Hkla1eHFvS/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:abc9050f517a659d95982978e4373a1a40e8da8f6cc74f9f7cfa45cede71275b +size 571514 diff --git a/vlm/train/Hkla1eHFvS/4.png b/vlm/train/Hkla1eHFvS/4.png new file mode 100644 index 0000000000000000000000000000000000000000..6621b525af108ab843972b4f484eac9d58e36ab3 --- /dev/null +++ b/vlm/train/Hkla1eHFvS/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6e2c9400dec3bb0bda7d3f0fde7f308161d7a7e1705f49a2a60896b91fc3d3e6 +size 537203 diff --git a/vlm/train/Hkla1eHFvS/5.png b/vlm/train/Hkla1eHFvS/5.png new file mode 100644 index 0000000000000000000000000000000000000000..6ea0dc9fa87573c1fce41e71e3d88723546a3755 --- /dev/null +++ b/vlm/train/Hkla1eHFvS/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a896662fc521195700cc592c123ccb3f0e8710bfa98c1763c0cce3a6393ce592 +size 615462 diff --git a/vlm/train/Hkla1eHFvS/6.png b/vlm/train/Hkla1eHFvS/6.png new file mode 100644 index 0000000000000000000000000000000000000000..25d60cddb8255a07cb69568a58bf45461d3f6459 --- /dev/null +++ b/vlm/train/Hkla1eHFvS/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1ff8fe595033ace368e6a9db3c24ec176a2215d6a5ba01394d3a4af585ab6ea3 +size 614383 diff --git a/vlm/train/Hkla1eHFvS/7.png b/vlm/train/Hkla1eHFvS/7.png new file mode 100644 index 0000000000000000000000000000000000000000..d6cc3816590634f20131b066d9ca9aa5b76c00e6 --- /dev/null +++ b/vlm/train/Hkla1eHFvS/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ea4d74f266974542d01253cec892ad69e4c9ccbb3e4f342f88aa111c32dc1525 +size 645809 diff --git a/vlm/train/Hkla1eHFvS/8.png b/vlm/train/Hkla1eHFvS/8.png new file mode 100644 index 0000000000000000000000000000000000000000..a22d6d898827a83723554b8c83c50d98d68da100 --- /dev/null +++ b/vlm/train/Hkla1eHFvS/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:11e1b472c41506ded4ae16a5673a351917c2dae6cb415cb2fb97049e7dff6199 +size 539293 diff --git a/vlm/train/Hkla1eHFvS/9.png b/vlm/train/Hkla1eHFvS/9.png new file mode 100644 index 0000000000000000000000000000000000000000..2713fed500fd495ce955d38559fe7787977f0d65 --- /dev/null +++ b/vlm/train/Hkla1eHFvS/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5e7847880142f802a342255c0cf8dba8096d1c665cc6c6e0638d03c8db6459d7 +size 532605 diff --git a/vlm/train/HkxStoC5F7/0.png b/vlm/train/HkxStoC5F7/0.png new file mode 100644 index 0000000000000000000000000000000000000000..5de30b385c758c2e67d7f5cd689791524871da59 --- /dev/null +++ b/vlm/train/HkxStoC5F7/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e0734e0971c9f381689ff1070d2f8fde181014d1d959d24467b8ff60a1731eb1 +size 464259 diff --git a/vlm/train/HkxStoC5F7/1.png b/vlm/train/HkxStoC5F7/1.png new file mode 100644 index 0000000000000000000000000000000000000000..a26021d9f8a7fe5dd78e45dcd7ee4f99b1f49fe1 --- /dev/null +++ b/vlm/train/HkxStoC5F7/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:40b4736a92dd9eb1f44e113e5930e282a9fbda88e37c76af9792c727be1f3a09 +size 447397 diff --git a/vlm/train/HkxStoC5F7/10.png b/vlm/train/HkxStoC5F7/10.png new file mode 100644 index 0000000000000000000000000000000000000000..0667bb3064921897d6e7c5c72ad80a4ec4b1cfe9 --- /dev/null +++ b/vlm/train/HkxStoC5F7/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:53e96b38b01163f8f08c13b78ff449791291e57ebae92775ef8d2de7ff34acba +size 507339 diff --git a/vlm/train/HkxStoC5F7/11.png b/vlm/train/HkxStoC5F7/11.png new file mode 100644 index 0000000000000000000000000000000000000000..02352836d5b443f1891ff8f74ee77e9d4a9574ac --- /dev/null +++ b/vlm/train/HkxStoC5F7/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b816b4690fb228cdf8873da3e84a0665a5019df896b655d1efbc28adecee6041 +size 516281 diff --git a/vlm/train/HkxStoC5F7/12.png b/vlm/train/HkxStoC5F7/12.png new file mode 100644 index 0000000000000000000000000000000000000000..cc1e17236d27ff66600b5ddab60dd7fc5ba516ff --- /dev/null +++ b/vlm/train/HkxStoC5F7/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:822c6dac0563c6c3781cd65c75c2b8893afcb953200433620d3f322eb129e99e +size 550029 diff --git a/vlm/train/HkxStoC5F7/13.png b/vlm/train/HkxStoC5F7/13.png new file mode 100644 index 0000000000000000000000000000000000000000..562e866bb6582047b694709b3adfd0ae692e3303 --- /dev/null +++ b/vlm/train/HkxStoC5F7/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bc05e0485a472a368c4946528cf37611ee2fb1d26eddaedf801641deb753df13 +size 473863 diff --git a/vlm/train/HkxStoC5F7/14.png b/vlm/train/HkxStoC5F7/14.png new file mode 100644 index 0000000000000000000000000000000000000000..640d298705bc669ba2c163479a679a79b6867a63 --- /dev/null +++ b/vlm/train/HkxStoC5F7/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e0c292398f287166e89b34400810681f0b3c3ddb8c3ae233e920b8bad1e01070 +size 523538 diff --git a/vlm/train/HkxStoC5F7/15.png b/vlm/train/HkxStoC5F7/15.png new file mode 100644 index 0000000000000000000000000000000000000000..3d32b9b6331a922de29ae642274a971596de4951 --- /dev/null +++ b/vlm/train/HkxStoC5F7/15.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:64611ddf0414f8208fa1056741a3195b593cadaf597b9b7e1caa1a4a0fab0efb +size 488475 diff --git a/vlm/train/HkxStoC5F7/16.png b/vlm/train/HkxStoC5F7/16.png new file mode 100644 index 0000000000000000000000000000000000000000..308817cfa9e6bc3c53e1d3dba7968ad94e2636da --- /dev/null +++ b/vlm/train/HkxStoC5F7/16.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9411b1fae6356cfb57c4c354be3dca0cb84edb09650f89a371f66d4417632840 +size 271236 diff --git a/vlm/train/HkxStoC5F7/17.png b/vlm/train/HkxStoC5F7/17.png new file mode 100644 index 0000000000000000000000000000000000000000..9d1e0e6a26a8b995904d8bba8a443581ddf74470 --- /dev/null +++ b/vlm/train/HkxStoC5F7/17.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:49316fe5a426df658b660ef60fe98eda878c2c4f0f2c6c7baa7923b87a74a625 +size 528076 diff --git a/vlm/train/HkxStoC5F7/18.png b/vlm/train/HkxStoC5F7/18.png new file mode 100644 index 0000000000000000000000000000000000000000..a8d19f70ecc63481b5c91ce114f564e5fd9a0c9d --- /dev/null +++ b/vlm/train/HkxStoC5F7/18.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fbc360eaac99c8a4e67f0614795fcd488090694546a91d144bec572d6a00237c +size 501588 diff --git a/vlm/train/HkxStoC5F7/19.png b/vlm/train/HkxStoC5F7/19.png new file mode 100644 index 0000000000000000000000000000000000000000..4c8c9a0bbbc53c7f8f3b21d33fec3c6a26aa7947 --- /dev/null +++ b/vlm/train/HkxStoC5F7/19.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f5b763fc890031023a944047bf324c4bfc9a2b78b4f456bb237f0d71a380f72f +size 363747 diff --git a/vlm/train/HkxStoC5F7/2.png b/vlm/train/HkxStoC5F7/2.png new file mode 100644 index 0000000000000000000000000000000000000000..188001a2013d463f97b349c7dcfacc555bfe35de --- /dev/null +++ b/vlm/train/HkxStoC5F7/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4e34883f6ae9b583ed936847f9e263917a17f7cf83ee249a87cd2cc8c86f4808 +size 583622 diff --git a/vlm/train/HkxStoC5F7/20.png b/vlm/train/HkxStoC5F7/20.png new file mode 100644 index 0000000000000000000000000000000000000000..0e840ee66b0d5177567512bd9bf27d75cf74e56e --- /dev/null +++ b/vlm/train/HkxStoC5F7/20.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ef8d38132b6208a5af9dd1ec3fa1a3e33e4397b3b9a282e49455eb508b4852bf +size 277303 diff --git a/vlm/train/HkxStoC5F7/3.png b/vlm/train/HkxStoC5F7/3.png new file mode 100644 index 0000000000000000000000000000000000000000..95792a22cf8123ccdea6e46dee9a9e4f3b7b06e1 --- /dev/null +++ b/vlm/train/HkxStoC5F7/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6ba382f4ce075a423c282e8b16be1c4062105e7c89fd0cfc5b11029454253fdf +size 564111 diff --git a/vlm/train/HkxStoC5F7/4.png b/vlm/train/HkxStoC5F7/4.png new file mode 100644 index 0000000000000000000000000000000000000000..6522c400fc3592b2a5062e27466fb374442da9fe --- /dev/null +++ b/vlm/train/HkxStoC5F7/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a73861827fc4a23dcb4a91bf97cfbf56f328e489a99aec785c1ef54e82e9721d +size 479456 diff --git a/vlm/train/HkxStoC5F7/5.png b/vlm/train/HkxStoC5F7/5.png new file mode 100644 index 0000000000000000000000000000000000000000..3b39c249428cbc39d31913a13e63bc090f93aa6f --- /dev/null +++ b/vlm/train/HkxStoC5F7/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e1d82face4382cc5061f1bd1ea5f29d02f999e8c49584c6cb2a18eab289070e9 +size 546476 diff --git a/vlm/train/HkxStoC5F7/6.png b/vlm/train/HkxStoC5F7/6.png new file mode 100644 index 0000000000000000000000000000000000000000..ba3ea30e9731b28e545f26e19318700b73772f9e --- /dev/null +++ b/vlm/train/HkxStoC5F7/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c5d2635aebb91a9f3c3d0493aa8a491ceaa36b67b4ed996fc9ceae31cab714b9 +size 523975 diff --git a/vlm/train/HkxStoC5F7/7.png b/vlm/train/HkxStoC5F7/7.png new file mode 100644 index 0000000000000000000000000000000000000000..73349620dbd35cfdbc7c94f7c5d07df24b4a3123 --- /dev/null +++ b/vlm/train/HkxStoC5F7/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0e96a6412fb771bb9499678c108e717fe6fc5ee1ac79c437b6025ae0e547ece1 +size 539443 diff --git a/vlm/train/HkxStoC5F7/8.png b/vlm/train/HkxStoC5F7/8.png new file mode 100644 index 0000000000000000000000000000000000000000..a366d6bac846ab2e6c4557bf7d05e6729e8a1726 --- /dev/null +++ b/vlm/train/HkxStoC5F7/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:432d19d8407c015d35fad7c9a6ce946a3e3b63cbd29cfbec187108ea4b61d690 +size 548567 diff --git a/vlm/train/HkxStoC5F7/9.png b/vlm/train/HkxStoC5F7/9.png new file mode 100644 index 0000000000000000000000000000000000000000..5e5800e0fb5cafe1d003e8d9852783a3f5c6468f --- /dev/null +++ b/vlm/train/HkxStoC5F7/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:de81e17f5d3fd6bb79f9b0b08e835e1980bc860085ad5b6f24e441ed85e35676 +size 397719 diff --git a/vlm/train/Hy1d-ebAb/1.png b/vlm/train/Hy1d-ebAb/1.png new file mode 100644 index 0000000000000000000000000000000000000000..e7f30871b6c3339e365fa23029da87fd5dd647d1 --- /dev/null +++ b/vlm/train/Hy1d-ebAb/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4d4f5c35ad9258138354dee0c6fb8a13aa9ad5587e4e3d803a2d92e058249dad +size 601909 diff --git a/vlm/train/Hy1d-ebAb/12.png b/vlm/train/Hy1d-ebAb/12.png new file mode 100644 index 0000000000000000000000000000000000000000..82bb6a0d589f4badc134b5b451528df0ffe7ec61 --- /dev/null +++ b/vlm/train/Hy1d-ebAb/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c339ea338c755d4cb5d2463c0abc599ad2475fecb3b6cb6b126f33457237192c +size 399670 diff --git a/vlm/train/Hy1d-ebAb/13.png b/vlm/train/Hy1d-ebAb/13.png new file mode 100644 index 0000000000000000000000000000000000000000..5c85fecfba01c3b2e65c89c6c444ec2c147d5f23 --- /dev/null +++ b/vlm/train/Hy1d-ebAb/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f54df384e02528e7bee2143ed4439f97de4f743bcb7d9544cebce81b97246844 +size 386369 diff --git a/vlm/train/Hy1d-ebAb/16.png b/vlm/train/Hy1d-ebAb/16.png new file mode 100644 index 0000000000000000000000000000000000000000..5eafcccd875792b3f47b22a010560eb1cbe59820 --- /dev/null +++ b/vlm/train/Hy1d-ebAb/16.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6b4b342709d0f14a0bba10c022549f6d784572199cfd7817102ceaf7b3bf5656 +size 577795 diff --git a/vlm/train/Hy1d-ebAb/18.png b/vlm/train/Hy1d-ebAb/18.png new file mode 100644 index 0000000000000000000000000000000000000000..084a8f0298f5e8dc5530dd86e12a7a7438915604 --- /dev/null +++ b/vlm/train/Hy1d-ebAb/18.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c86ab25db6816aea1cb16171dca69a5291891944f0beb0eb56df20259b3175b9 +size 218969 diff --git a/vlm/train/Hy1d-ebAb/19.png b/vlm/train/Hy1d-ebAb/19.png new file mode 100644 index 0000000000000000000000000000000000000000..de35ca2d50a7c13ac103e5a971fb6ab1ea72f58e --- /dev/null +++ b/vlm/train/Hy1d-ebAb/19.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:add6d3a17208d45a138b2f077d0ee6214a89ef466ac38351433216154afa45d9 +size 309425 diff --git a/vlm/train/Hy1d-ebAb/20.png b/vlm/train/Hy1d-ebAb/20.png new file mode 100644 index 0000000000000000000000000000000000000000..598468058dbbbb35c88ad2e69b154e5e1b0ace72 --- /dev/null +++ b/vlm/train/Hy1d-ebAb/20.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:78e6385749e3072ecb722a57dd638c5453a03d7a78444233245be3560dc9215c +size 188096 diff --git a/vlm/train/Hy1d-ebAb/4.png b/vlm/train/Hy1d-ebAb/4.png new file mode 100644 index 0000000000000000000000000000000000000000..9fc00b03aa1f4d63758171cd580d3e112a4473b9 --- /dev/null +++ b/vlm/train/Hy1d-ebAb/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9337db3a9a8200af486aa84cecac6af99d2069ab16424e967ee60f27797d5ccb +size 567682 diff --git a/vlm/train/Hy1d-ebAb/5.png b/vlm/train/Hy1d-ebAb/5.png new file mode 100644 index 0000000000000000000000000000000000000000..de0761420adaacea5507058758ee44bcd6987a35 --- /dev/null +++ b/vlm/train/Hy1d-ebAb/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b1317ed0cd56b3c4f7faf4f94c9a31e28314319c6fa913ccda80e087421d58d5 +size 544093 diff --git a/vlm/train/Hy1d-ebAb/7.png b/vlm/train/Hy1d-ebAb/7.png new file mode 100644 index 0000000000000000000000000000000000000000..9190744112283f346baadec755a1b6a6acef831d --- /dev/null +++ b/vlm/train/Hy1d-ebAb/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:de0506e02b7808b33162a282490b8926fa185b8b9299adbf87cc9b39f29f1e54 +size 541016 diff --git a/vlm/train/Hy1d-ebAb/8.png b/vlm/train/Hy1d-ebAb/8.png new file mode 100644 index 0000000000000000000000000000000000000000..96d80ef25a88f52bd6277e1a9aab05e9eb0e439a --- /dev/null +++ b/vlm/train/Hy1d-ebAb/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d8ec2419a668bc2cf1316340536f1457677f0d53a827a82c9c24ba357084b2b3 +size 443303 diff --git a/vlm/train/HyEi7bWR-/0.png b/vlm/train/HyEi7bWR-/0.png new file mode 100644 index 0000000000000000000000000000000000000000..399acd0d476364fe2ec51945bf74b95c9d9d522e --- /dev/null +++ b/vlm/train/HyEi7bWR-/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3b3d27c238b246e9293d42d53f567ed518410ee0c3e2b0eedc48a1e0462b09a3 +size 530496 diff --git a/vlm/train/HyEi7bWR-/1.png b/vlm/train/HyEi7bWR-/1.png new file mode 100644 index 0000000000000000000000000000000000000000..8a27b27529d7085327b003b710267c5ad793e750 --- /dev/null +++ b/vlm/train/HyEi7bWR-/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:63beb2bfce4eab529d17788cd7e0d3e1eabe295a82652c870234dd7f2dabac8a +size 459581 diff --git a/vlm/train/HyEi7bWR-/10.png b/vlm/train/HyEi7bWR-/10.png new file mode 100644 index 0000000000000000000000000000000000000000..8eb7b6041a808ed5c6db70eee72d82c9517a44b5 --- /dev/null +++ b/vlm/train/HyEi7bWR-/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:480094e2c2c775905a117da37e325440f17bfb8a89a68385d9287356237bfe08 +size 286443 diff --git a/vlm/train/HyEi7bWR-/11.png b/vlm/train/HyEi7bWR-/11.png new file mode 100644 index 0000000000000000000000000000000000000000..b4c2adedd6015dcdb454211c57a91b20e84b3e9f --- /dev/null +++ b/vlm/train/HyEi7bWR-/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:039028e1c196c2313f69e85aa94b7ee30c37cbad10e51a022f288cd6d8559a9a +size 389590 diff --git a/vlm/train/HyEi7bWR-/12.png b/vlm/train/HyEi7bWR-/12.png new file mode 100644 index 0000000000000000000000000000000000000000..04c1ed851068d0d04d0d5a2c9bbba5114c7368f5 --- /dev/null +++ b/vlm/train/HyEi7bWR-/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c750ca33cb95592d0db5506a1592fd2ef69835ebba38084bd85cd76858bb3d6d +size 383297 diff --git a/vlm/train/HyEi7bWR-/13.png b/vlm/train/HyEi7bWR-/13.png new file mode 100644 index 0000000000000000000000000000000000000000..13929ef80c8fd13883d94ed164094d120ccbfd72 --- /dev/null +++ b/vlm/train/HyEi7bWR-/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b5e1c7d0da875954a8865888ef376421eb6161ef1d486838efad586839a54416 +size 92597 diff --git a/vlm/train/HyEi7bWR-/2.png b/vlm/train/HyEi7bWR-/2.png new file mode 100644 index 0000000000000000000000000000000000000000..1d341961a9e4a69d6a0938c6bc020af2b33df38e --- /dev/null +++ b/vlm/train/HyEi7bWR-/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a8b219b4f38570f3777f822f2f36c4829a63183ea3ff6c404347c549ed63c689 +size 464009 diff --git a/vlm/train/HyEi7bWR-/3.png b/vlm/train/HyEi7bWR-/3.png new file mode 100644 index 0000000000000000000000000000000000000000..9419e11929f85fccaaf98e3c3d980093f03843dd --- /dev/null +++ b/vlm/train/HyEi7bWR-/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ca383366c24ba9b8baa92bceff8f20c3d020d2d662b30aed3410911cc8d7e933 +size 436367 diff --git a/vlm/train/HyEi7bWR-/4.png b/vlm/train/HyEi7bWR-/4.png new file mode 100644 index 0000000000000000000000000000000000000000..9f820517e40476261b8c86815878a3b4530272f2 --- /dev/null +++ b/vlm/train/HyEi7bWR-/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3f6176a768e45a50a6be7cbbfa2a8e744a0be911ff7e03015e8b14d430c2c953 +size 457508 diff --git a/vlm/train/HyEi7bWR-/5.png b/vlm/train/HyEi7bWR-/5.png new file mode 100644 index 0000000000000000000000000000000000000000..73b2ac3a7be92e1ef2a3483f17b1fb544f59b01c --- /dev/null +++ b/vlm/train/HyEi7bWR-/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:404063833982d48ec55680b81d6b136ad4977631779b42e211052226c71b1ec2 +size 517448 diff --git a/vlm/train/HyEi7bWR-/6.png b/vlm/train/HyEi7bWR-/6.png new file mode 100644 index 0000000000000000000000000000000000000000..fd243958177b764bcf11c6a936bd83dbcf720eb7 --- /dev/null +++ b/vlm/train/HyEi7bWR-/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6764a8ffbd88580224156c979c437f622fc2a9ce7aff228a8611c49c375b44bf +size 543510 diff --git a/vlm/train/HyEi7bWR-/7.png b/vlm/train/HyEi7bWR-/7.png new file mode 100644 index 0000000000000000000000000000000000000000..68d7faf371ef9f81f37a381ebd9c358d2b94d72b --- /dev/null +++ b/vlm/train/HyEi7bWR-/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9e08b148a96b40f967477e89b85dd1285482888426175e8c4f2ad5a92690bde9 +size 507117 diff --git a/vlm/train/HyEi7bWR-/8.png b/vlm/train/HyEi7bWR-/8.png new file mode 100644 index 0000000000000000000000000000000000000000..36c2c566ba778aedc0c898fb9099a4eedcff183a --- /dev/null +++ b/vlm/train/HyEi7bWR-/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eca8df65e9c634cd8e2146d660a648a91c2aedc85072cafdbc9c764072c6d59d +size 528743 diff --git a/vlm/train/HyEi7bWR-/9.png b/vlm/train/HyEi7bWR-/9.png new file mode 100644 index 0000000000000000000000000000000000000000..db8a8af7dab84604d6f3fdf7baf264adbcc925cf --- /dev/null +++ b/vlm/train/HyEi7bWR-/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f6b34c9d7214cc984c0eedd293822835b77bf4b8291a5da3be753923bd48f861 +size 155025 diff --git a/vlm/train/Hye9lnCct7/16.png b/vlm/train/Hye9lnCct7/16.png new file mode 100644 index 0000000000000000000000000000000000000000..65ff7d3da3280f10ba6b7b4a3b15ce799c8eccb6 --- /dev/null +++ b/vlm/train/Hye9lnCct7/16.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:94be63cacb94090a47883b18b4ec8bae5a732268a6357a8d63ba04fdf31d955d +size 456672 diff --git a/vlm/train/Hyewf3AqYX/0.png b/vlm/train/Hyewf3AqYX/0.png new file mode 100644 index 0000000000000000000000000000000000000000..c4fbaf4b7e2343bc88574c2d76168cf9181a6495 --- /dev/null +++ b/vlm/train/Hyewf3AqYX/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:813dd1cbe894ac680c44d9eda8a3e0d059b8cf3bea55ff84232ea14a6e164b06 +size 526591 diff --git a/vlm/train/Hyewf3AqYX/1.png b/vlm/train/Hyewf3AqYX/1.png new file mode 100644 index 0000000000000000000000000000000000000000..2098cdb6608531514d9d0d58f241fd3dbc3884f7 --- /dev/null +++ b/vlm/train/Hyewf3AqYX/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c23d4d6977b597e3047ea6b459d3cf68e4fa36252b313a37a6461a5dcb7e52e1 +size 599258 diff --git a/vlm/train/Hyewf3AqYX/10.png b/vlm/train/Hyewf3AqYX/10.png new file mode 100644 index 0000000000000000000000000000000000000000..c0ca5a824f5bb0d705df980a6c9c8e2d7baced1b --- /dev/null +++ b/vlm/train/Hyewf3AqYX/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:646b6bbbeeb311e20143f8386de8384d5eb10cb48f19941ef62ae348153d45dc +size 556994 diff --git a/vlm/train/Hyewf3AqYX/11.png b/vlm/train/Hyewf3AqYX/11.png new file mode 100644 index 0000000000000000000000000000000000000000..1c3ea11d1cb6b3118b019390623ec47c6cbaaba9 --- /dev/null +++ b/vlm/train/Hyewf3AqYX/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1077b2ba2aba4b5e2ae18531855b69fd7e3ee0d0c8565b2ec4653a6ee0a37977 +size 283803 diff --git a/vlm/train/Hyewf3AqYX/13.png b/vlm/train/Hyewf3AqYX/13.png new file mode 100644 index 0000000000000000000000000000000000000000..39e8fa4188a92dc4239ed86dba3ca217349f7591 --- /dev/null +++ b/vlm/train/Hyewf3AqYX/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2a5ae9b26f5a9369db10710e2d65e895197226a7ed70a86fe4bc73b92d876470 +size 303928 diff --git a/vlm/train/Hyewf3AqYX/14.png b/vlm/train/Hyewf3AqYX/14.png new file mode 100644 index 0000000000000000000000000000000000000000..13177a91732532f706283a6c4ca070d9aef8d507 --- /dev/null +++ b/vlm/train/Hyewf3AqYX/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:45fb1dff635ea1050d3308c7d0964988ea83f7981dc7b032ad9771544af5e997 +size 278252 diff --git a/vlm/train/Hyewf3AqYX/15.png b/vlm/train/Hyewf3AqYX/15.png new file mode 100644 index 0000000000000000000000000000000000000000..ffba37dc166d7ee7adcefc3b3299118173d9e635 --- /dev/null +++ b/vlm/train/Hyewf3AqYX/15.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:15645cd86df5f94e8e8a99aea4e7189b8b130ab41c7210e976639f9ff14bde85 +size 334921 diff --git a/vlm/train/Hyewf3AqYX/16.png b/vlm/train/Hyewf3AqYX/16.png new file mode 100644 index 0000000000000000000000000000000000000000..523624378d3f7ef49ec022652dba4fb5c2fa2279 --- /dev/null +++ b/vlm/train/Hyewf3AqYX/16.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6a0da5b031a1fb61ffeee72bceb8b0fb1b1a6aa80727307668c544d8f6b14a6b +size 459474 diff --git a/vlm/train/Hyewf3AqYX/17.png b/vlm/train/Hyewf3AqYX/17.png new file mode 100644 index 0000000000000000000000000000000000000000..43a35be2ad268a96d68e5c743fe6120ec58a16f9 --- /dev/null +++ b/vlm/train/Hyewf3AqYX/17.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a40095938e5b408b52d63d3f5dde8bdc6de6be76affd5c7334eb18f98f6ab1f9 +size 348655 diff --git a/vlm/train/Hyewf3AqYX/18.png b/vlm/train/Hyewf3AqYX/18.png new file mode 100644 index 0000000000000000000000000000000000000000..e8712eea76ae2fe2d981f2dcd6380009607ab7a0 --- /dev/null +++ b/vlm/train/Hyewf3AqYX/18.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dad0c3544d21326927ab943b506d5931b38dd24067d8159ce6fdb67b26b29538 +size 2674950 diff --git a/vlm/train/Hyewf3AqYX/19.png b/vlm/train/Hyewf3AqYX/19.png new file mode 100644 index 0000000000000000000000000000000000000000..406bc3c077abc8195150e8c51834f05153dffd31 --- /dev/null +++ b/vlm/train/Hyewf3AqYX/19.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:95963b689d1f68efe01bfa0ab84586bf5c8ff489da0e7b92c906bbe6d88a0924 +size 2326893 diff --git a/vlm/train/Hyewf3AqYX/2.png b/vlm/train/Hyewf3AqYX/2.png new file mode 100644 index 0000000000000000000000000000000000000000..1a6bc32ffa124970040c9b6f46e2529a6d0ebda9 --- /dev/null +++ b/vlm/train/Hyewf3AqYX/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9fa89638ff47c16ade2098b8c5317d5e25bbe3eed04e0013dbcd69831cf6b9d1 +size 533050 diff --git a/vlm/train/Hyewf3AqYX/3.png b/vlm/train/Hyewf3AqYX/3.png new file mode 100644 index 0000000000000000000000000000000000000000..c7a638cd17d1d2182daaa923ea0405dc9f9cb69c --- /dev/null +++ b/vlm/train/Hyewf3AqYX/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3aea18c1b068d4d236b2e90edb13994bdcb8c64d62db2f1e149c9ca737ec6111 +size 483740 diff --git a/vlm/train/Hyewf3AqYX/4.png b/vlm/train/Hyewf3AqYX/4.png new file mode 100644 index 0000000000000000000000000000000000000000..838ba18a453e73a9e791e3a6a5a9d9006abecc16 --- /dev/null +++ b/vlm/train/Hyewf3AqYX/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e5e254771984dd222492ccbd3ef4847bfd46719ac4322ee0ea6c19e4cead8b7b +size 410700 diff --git a/vlm/train/Hyewf3AqYX/6.png b/vlm/train/Hyewf3AqYX/6.png new file mode 100644 index 0000000000000000000000000000000000000000..7e9a0a4b1cb1744875aeb49c7f046845af299c30 --- /dev/null +++ b/vlm/train/Hyewf3AqYX/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:08a3e6b4852b594f368afd59d00d0d6bdae202ee6ac04cb2155326ec42937e92 +size 582712 diff --git a/vlm/train/Hyewf3AqYX/9.png b/vlm/train/Hyewf3AqYX/9.png new file mode 100644 index 0000000000000000000000000000000000000000..435d95771c557edf2e3869911a33d2ba6d3f1ced --- /dev/null +++ b/vlm/train/Hyewf3AqYX/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d3b23b7d0f7820a93a43d88b8df81ff33bd3a80e0bbaa307b43e60e7d054d38d +size 537925 diff --git a/vlm/train/IBVBtz_sRSm/0.png b/vlm/train/IBVBtz_sRSm/0.png new file mode 100644 index 0000000000000000000000000000000000000000..7327c20244e58b0037944e9d6b159f95dce45645 --- /dev/null +++ b/vlm/train/IBVBtz_sRSm/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5e3a9e112b7a20a725e01d41fe0f2b1225930a01f1a142b3af3550b4db1678b3 +size 423679 diff --git a/vlm/train/IBVBtz_sRSm/1.png b/vlm/train/IBVBtz_sRSm/1.png new file mode 100644 index 0000000000000000000000000000000000000000..623ed48340b0d46a511d856914f3a1afe54709c6 --- /dev/null +++ b/vlm/train/IBVBtz_sRSm/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d45d131c6db09acd44600d55e990031f55fb266f87e60ecc47552b3101cb3ad2 +size 563034 diff --git a/vlm/train/IBVBtz_sRSm/10.png b/vlm/train/IBVBtz_sRSm/10.png new file mode 100644 index 0000000000000000000000000000000000000000..bc323582f8b7af693f51c2eb8c9f7f56dcbc3126 --- /dev/null +++ b/vlm/train/IBVBtz_sRSm/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d10d7eff167d8c863f00009a79b9dc817184a40b7eed1a37287c489203835559 +size 530525 diff --git a/vlm/train/IBVBtz_sRSm/11.png b/vlm/train/IBVBtz_sRSm/11.png new file mode 100644 index 0000000000000000000000000000000000000000..843826ffb739a5a247f8f4292f69b7ac69040f2a --- /dev/null +++ b/vlm/train/IBVBtz_sRSm/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8cde904a0b3dc42146dd48d28cc1faa7c886e2018eb64be8b422c570c8eb9b8a +size 596668 diff --git a/vlm/train/IBVBtz_sRSm/12.png b/vlm/train/IBVBtz_sRSm/12.png new file mode 100644 index 0000000000000000000000000000000000000000..a411cad7f204ee3e958bb4f007b5922e656ac618 --- /dev/null +++ b/vlm/train/IBVBtz_sRSm/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:be469065e123a2381b36ec9bab211f512d02d01c988d5469d10721efe15849dc +size 580471 diff --git a/vlm/train/IBVBtz_sRSm/13.png b/vlm/train/IBVBtz_sRSm/13.png new file mode 100644 index 0000000000000000000000000000000000000000..2a69ccadc4fec6002bef66897d5173e1740b0195 --- /dev/null +++ b/vlm/train/IBVBtz_sRSm/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eb95da334f933f57b93eec4542f883bdc752a10f28ce50391a6e90cca1655c68 +size 332171 diff --git a/vlm/train/IBVBtz_sRSm/2.png b/vlm/train/IBVBtz_sRSm/2.png new file mode 100644 index 0000000000000000000000000000000000000000..77854418afff6bdd14e35ef652f86aebab3411bf --- /dev/null +++ b/vlm/train/IBVBtz_sRSm/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:48ac0724628356e22475bf3c399f9c4cea56d91aab576c00605cb3b2ee9da8e0 +size 525435 diff --git a/vlm/train/IBVBtz_sRSm/3.png b/vlm/train/IBVBtz_sRSm/3.png new file mode 100644 index 0000000000000000000000000000000000000000..e21a49005ac671cf4964fbcb766ce55b30ef6017 --- /dev/null +++ b/vlm/train/IBVBtz_sRSm/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:33c890f004d5890a8bca44f5af2f0a4394a41a656157aa9c62e26e8a998683ef +size 599320 diff --git a/vlm/train/IBVBtz_sRSm/4.png b/vlm/train/IBVBtz_sRSm/4.png new file mode 100644 index 0000000000000000000000000000000000000000..b227c9446a8cdfca6940b31ed79fc501b1c12daf --- /dev/null +++ b/vlm/train/IBVBtz_sRSm/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d033c208e3d794d9f6a85785f6adb4c40f39da31b8175a841409f64a84c3e13f +size 600524 diff --git a/vlm/train/IBVBtz_sRSm/5.png b/vlm/train/IBVBtz_sRSm/5.png new file mode 100644 index 0000000000000000000000000000000000000000..a8a1fe7965386bfe7e45eaff7afacdb68a2afec2 --- /dev/null +++ b/vlm/train/IBVBtz_sRSm/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec030fa1802af9f3ec283ac9a3ee5c24155624a04d331c88836df18b664066ea +size 477147 diff --git a/vlm/train/IBVBtz_sRSm/6.png b/vlm/train/IBVBtz_sRSm/6.png new file mode 100644 index 0000000000000000000000000000000000000000..a57b3ea2557ab95da6f49f8d13e9cbed23c887a3 --- /dev/null +++ b/vlm/train/IBVBtz_sRSm/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:db3004ac479b758beec509756d2261ef9f1ade492341b5d8b7b5535d5fce1d64 +size 625544 diff --git a/vlm/train/IBVBtz_sRSm/7.png b/vlm/train/IBVBtz_sRSm/7.png new file mode 100644 index 0000000000000000000000000000000000000000..debc3ce90ef8f00ff882b75d6ff86d950af49b31 --- /dev/null +++ b/vlm/train/IBVBtz_sRSm/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d71e6d3ee7b957cec90b692ed866304ef49d548c1b536ef45543367669c35b4b +size 549035 diff --git a/vlm/train/IBVBtz_sRSm/8.png b/vlm/train/IBVBtz_sRSm/8.png new file mode 100644 index 0000000000000000000000000000000000000000..2ddebd849aa6093d59398c72aa2feb8bdf97f0c8 --- /dev/null +++ b/vlm/train/IBVBtz_sRSm/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:410a7061f2a4855a70efd906fa5dfd1261792feb7b67f5f0970c62e13e38321b +size 598490 diff --git a/vlm/train/IBVBtz_sRSm/9.png b/vlm/train/IBVBtz_sRSm/9.png new file mode 100644 index 0000000000000000000000000000000000000000..3c95ed717da4254653261325871c56ac7c05f63e --- /dev/null +++ b/vlm/train/IBVBtz_sRSm/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c377d97633df6487191cfa1ca209e1ab525d5879d10cd21e7767ada1c69522b8 +size 552251 diff --git a/vlm/train/KJSC_AsN14/0.png b/vlm/train/KJSC_AsN14/0.png new file mode 100644 index 0000000000000000000000000000000000000000..34ad9e608c45b94b452ff2a1ecaf4b6dedd6546d --- /dev/null +++ b/vlm/train/KJSC_AsN14/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9ca49bbec451c21b02b660b39a1b742945c4c8bafafe805b826a622e3a74d12c +size 485078 diff --git a/vlm/train/KJSC_AsN14/1.png b/vlm/train/KJSC_AsN14/1.png new file mode 100644 index 0000000000000000000000000000000000000000..6584430b1ff60c22a090f5e7edf29bb2db1c51b9 --- /dev/null +++ b/vlm/train/KJSC_AsN14/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ce17042f4ae75d93c5626c662405b24ba0b562faa618ac2d499ca8bba77e91ce +size 618336 diff --git a/vlm/train/KJSC_AsN14/10.png b/vlm/train/KJSC_AsN14/10.png new file mode 100644 index 0000000000000000000000000000000000000000..f87beec9f9e596229e48a68c1d8c3b5e8ee6eaed --- /dev/null +++ b/vlm/train/KJSC_AsN14/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:677043032909e5ab5db1d7798511b20a4fcbcc8b6cb9aff40234639c5652a0d8 +size 524421 diff --git a/vlm/train/KJSC_AsN14/2.png b/vlm/train/KJSC_AsN14/2.png new file mode 100644 index 0000000000000000000000000000000000000000..d8285fdd7b497bf0af59f9a8985d96c14d361e5f --- /dev/null +++ b/vlm/train/KJSC_AsN14/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e62863022fc86b9af4cd8d0b243f8c8123e78b7e67c105d0714fcf6c2b2e27dc +size 464618 diff --git a/vlm/train/KJSC_AsN14/3.png b/vlm/train/KJSC_AsN14/3.png new file mode 100644 index 0000000000000000000000000000000000000000..2336bb10d356f40f78b21d16ebaf1b914a0a2aac --- /dev/null +++ b/vlm/train/KJSC_AsN14/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7e0d799be73c2df070cdeb7aeeb5e10c8336379a8c54d57d6efa73a3fc3e53bc +size 885115 diff --git a/vlm/train/KJSC_AsN14/4.png b/vlm/train/KJSC_AsN14/4.png new file mode 100644 index 0000000000000000000000000000000000000000..9bc04881bff728bf05cc97d27d1bc3c4a29c3b01 --- /dev/null +++ b/vlm/train/KJSC_AsN14/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:efe477cee3d90817c8a57771321b91c54dc5afb5c9a9c3fd822d8feb3a700368 +size 517033 diff --git a/vlm/train/KJSC_AsN14/5.png b/vlm/train/KJSC_AsN14/5.png new file mode 100644 index 0000000000000000000000000000000000000000..640fb5d46ee04274d8f4c6fa4b11cd228dd2492b --- /dev/null +++ b/vlm/train/KJSC_AsN14/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:982f834fbbce18fa3a43b1e6582de82632ae31201a7d1a1edcd5225e8bea0606 +size 504732 diff --git a/vlm/train/KJSC_AsN14/6.png b/vlm/train/KJSC_AsN14/6.png new file mode 100644 index 0000000000000000000000000000000000000000..454c9a0ec502996bd248be6d9a01b8b1b1c8e054 --- /dev/null +++ b/vlm/train/KJSC_AsN14/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e8d1681e2e4c67eccb5ea31095ca5a1ecdae1f292ce7a8ea1b03b1eaff405f3d +size 557052 diff --git a/vlm/train/KJSC_AsN14/7.png b/vlm/train/KJSC_AsN14/7.png new file mode 100644 index 0000000000000000000000000000000000000000..ccc7a623f94635844ce3ab24d7be30accd00ccc1 --- /dev/null +++ b/vlm/train/KJSC_AsN14/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f580271a24ce630c36ff452189e6d43ac50a44f715bd79f84dfed879f777c909 +size 489849 diff --git a/vlm/train/KJSC_AsN14/8.png b/vlm/train/KJSC_AsN14/8.png new file mode 100644 index 0000000000000000000000000000000000000000..37b143627de550ab909f380b7e23cfaa02f618fa --- /dev/null +++ b/vlm/train/KJSC_AsN14/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b3c64ef45872bbd66cee5b05655869c87a03ed3d8be92727f23af6a15aeec898 +size 536941 diff --git a/vlm/train/KJSC_AsN14/9.png b/vlm/train/KJSC_AsN14/9.png new file mode 100644 index 0000000000000000000000000000000000000000..bd4762f6dcdd9fae7ba22f6baf115d57f40f6150 --- /dev/null +++ b/vlm/train/KJSC_AsN14/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0c97f38bac4bddc428a656218184e2b60f88107b673f6a6093191efc15cc57cb +size 577723 diff --git a/vlm/train/MQQeeDiO5vv/0.png b/vlm/train/MQQeeDiO5vv/0.png new file mode 100644 index 0000000000000000000000000000000000000000..59e0df9c7a7a3b52df535c2e1e07828c2a7901c8 --- /dev/null +++ b/vlm/train/MQQeeDiO5vv/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2cd7a2833496ef4b1f52c389c614a032edd8269ba34389ac9c3377e78ac48b7a +size 439015 diff --git a/vlm/train/MQQeeDiO5vv/1.png b/vlm/train/MQQeeDiO5vv/1.png new file mode 100644 index 0000000000000000000000000000000000000000..625647f2990ed7d1f105ed3b47aea217d78ac78e --- /dev/null +++ b/vlm/train/MQQeeDiO5vv/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a7cc99825791be970be23488b0949293aa9df39aa679395ca67588c3485d88be +size 582052 diff --git a/vlm/train/MQQeeDiO5vv/10.png b/vlm/train/MQQeeDiO5vv/10.png new file mode 100644 index 0000000000000000000000000000000000000000..4d7b51fe611600dc97cf5e65c23a4484ab9ef4ba --- /dev/null +++ b/vlm/train/MQQeeDiO5vv/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:66bdf2b621cbe8d6aae2fa7ef0eb39491d6645e0e8f2674ed3146300bb9b23fe +size 598374 diff --git a/vlm/train/MQQeeDiO5vv/11.png b/vlm/train/MQQeeDiO5vv/11.png new file mode 100644 index 0000000000000000000000000000000000000000..9042447ea2ee21be983886652bbcfe952b2fe00c --- /dev/null +++ b/vlm/train/MQQeeDiO5vv/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a2b4e1cb33369288d60dc48e62aa3c999abb8d6e6a6e30bab5f395b7c277d785 +size 549778 diff --git a/vlm/train/MQQeeDiO5vv/12.png b/vlm/train/MQQeeDiO5vv/12.png new file mode 100644 index 0000000000000000000000000000000000000000..7b31b5d4ef11d03148e5d420a4bd616ea84c5750 --- /dev/null +++ b/vlm/train/MQQeeDiO5vv/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:77d885d68cde9dfbd36ca1cafe295597fb81343e32edba9a367089750ac8b558 +size 454743 diff --git a/vlm/train/MQQeeDiO5vv/2.png b/vlm/train/MQQeeDiO5vv/2.png new file mode 100644 index 0000000000000000000000000000000000000000..4ae1f772b72a8bceaae8538fd6fa879dd8245b24 --- /dev/null +++ b/vlm/train/MQQeeDiO5vv/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3e257d184357cfc7cf25bf1ae08bc7c9e7af4d8db6f41423a8c81f99210ac930 +size 444865 diff --git a/vlm/train/MQQeeDiO5vv/3.png b/vlm/train/MQQeeDiO5vv/3.png new file mode 100644 index 0000000000000000000000000000000000000000..71be1b4bc4397816fc444a4b354e746c54c3d8ff --- /dev/null +++ b/vlm/train/MQQeeDiO5vv/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:95d946b84537a6811427adbe7bd64c456a8028553595806626b48c6dc661e989 +size 508563 diff --git a/vlm/train/MQQeeDiO5vv/4.png b/vlm/train/MQQeeDiO5vv/4.png new file mode 100644 index 0000000000000000000000000000000000000000..d57e47e59e62a76c917bdf8e858031239a9fbf78 --- /dev/null +++ b/vlm/train/MQQeeDiO5vv/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b57a7f0535dc5e9758d9252d18361337da8a1fe8ff9bc8e26018901831ee2547 +size 559939 diff --git a/vlm/train/MQQeeDiO5vv/5.png b/vlm/train/MQQeeDiO5vv/5.png new file mode 100644 index 0000000000000000000000000000000000000000..02c9ef3feb71c42cbd5e705f6d7803d5156cdf7c --- /dev/null +++ b/vlm/train/MQQeeDiO5vv/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e1c284e975ea5c0be24b9bd2001f95526aa485a577e3f1d1dca1ac873abf506c +size 562712 diff --git a/vlm/train/MQQeeDiO5vv/6.png b/vlm/train/MQQeeDiO5vv/6.png new file mode 100644 index 0000000000000000000000000000000000000000..443431e6302acc9644856542c4e5b98510aa6f23 --- /dev/null +++ b/vlm/train/MQQeeDiO5vv/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9bb90bc12eb0b679b799d2d04b9cd9e544707d9a2c41a3ce801eaf77c503b01c +size 476750 diff --git a/vlm/train/MQQeeDiO5vv/7.png b/vlm/train/MQQeeDiO5vv/7.png new file mode 100644 index 0000000000000000000000000000000000000000..8ca3dce02c686264218a925425c1d209354a56cb --- /dev/null +++ b/vlm/train/MQQeeDiO5vv/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:373e3910c98f15fb9c23bb9c95279309650f884acbb76153f099b220207062ed +size 516160 diff --git a/vlm/train/MQQeeDiO5vv/8.png b/vlm/train/MQQeeDiO5vv/8.png new file mode 100644 index 0000000000000000000000000000000000000000..431da1d14da4b04d819809cb86cf6a26170ec904 --- /dev/null +++ b/vlm/train/MQQeeDiO5vv/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a05f2305cb51a5daf62f664f7071945cc984fc67c3844f3c6000666249299b3a +size 501909 diff --git a/vlm/train/MQQeeDiO5vv/9.png b/vlm/train/MQQeeDiO5vv/9.png new file mode 100644 index 0000000000000000000000000000000000000000..cabd32e34ee3f0ed375e473793120040b85a5d18 --- /dev/null +++ b/vlm/train/MQQeeDiO5vv/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:10a682f2247757052517c92dc1b149013dd5cb82f9f46460fc68a2fb28282356 +size 558602 diff --git a/vlm/train/PhtFY9plHk/0.png b/vlm/train/PhtFY9plHk/0.png new file mode 100644 index 0000000000000000000000000000000000000000..df587d10c2ae73791565a9276f8a5540b109c840 --- /dev/null +++ b/vlm/train/PhtFY9plHk/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4c5bbdd796ca9c8a9602c4e9e4a6a39c8b9e1ccb6639cc7653a688176282800b +size 460122 diff --git a/vlm/train/PhtFY9plHk/1.png b/vlm/train/PhtFY9plHk/1.png new file mode 100644 index 0000000000000000000000000000000000000000..566920e6cc8fc5ebd3c2ace69b6b3859696d9bfb --- /dev/null +++ b/vlm/train/PhtFY9plHk/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:75b617d2ccecd3ffeaa25a9c51e8a76fbc05a878c6b309c5bebcdf751ed81cea +size 605753 diff --git a/vlm/train/PhtFY9plHk/10.png b/vlm/train/PhtFY9plHk/10.png new file mode 100644 index 0000000000000000000000000000000000000000..b90ba8eb680ecb76c9598355374588869871ec7b --- /dev/null +++ b/vlm/train/PhtFY9plHk/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:918db6fe407f954d3a9c73cb09ee1046c6a3d034b4344290223ac0b1ed8d0dea +size 600573 diff --git a/vlm/train/PhtFY9plHk/11.png b/vlm/train/PhtFY9plHk/11.png new file mode 100644 index 0000000000000000000000000000000000000000..c3c4f35e80125b8aa780c9330851145b30488486 --- /dev/null +++ b/vlm/train/PhtFY9plHk/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bd74426e817b391e051cdf773047f93e7b74f2d1ebcdb6cc1b7dcfca96d3bef7 +size 421514 diff --git a/vlm/train/PhtFY9plHk/2.png b/vlm/train/PhtFY9plHk/2.png new file mode 100644 index 0000000000000000000000000000000000000000..e790cdd64f251120de97c97cd3b3cad3dfcd73cc --- /dev/null +++ b/vlm/train/PhtFY9plHk/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:73a7b2e5a6d77c9f9d6e0076a625c9597c7c892788fd7eb981b5e22a7617da94 +size 502130 diff --git a/vlm/train/PhtFY9plHk/3.png b/vlm/train/PhtFY9plHk/3.png new file mode 100644 index 0000000000000000000000000000000000000000..4b6bcb5bceb59542549d7047cd61090e3613c271 --- /dev/null +++ b/vlm/train/PhtFY9plHk/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d37651cbe1c5717435c7561abbed375baae8a21caad72061d62ef850bc7c2d71 +size 445160 diff --git a/vlm/train/PhtFY9plHk/4.png b/vlm/train/PhtFY9plHk/4.png new file mode 100644 index 0000000000000000000000000000000000000000..b964d4b413835d79dc5b7764f673da56d70615e2 --- /dev/null +++ b/vlm/train/PhtFY9plHk/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:785fc5b29604e9f3a39285a41f88b57eafaaa61a9c69006bf95952a0e6736914 +size 509302 diff --git a/vlm/train/PhtFY9plHk/5.png b/vlm/train/PhtFY9plHk/5.png new file mode 100644 index 0000000000000000000000000000000000000000..29a3cd495a1ada8ca8ab8f29d04767d74f49e7d5 --- /dev/null +++ b/vlm/train/PhtFY9plHk/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:51d074b187827cf673e28b0f5c5532a483863cf595a68944b07bac1643c209fc +size 456614 diff --git a/vlm/train/PhtFY9plHk/6.png b/vlm/train/PhtFY9plHk/6.png new file mode 100644 index 0000000000000000000000000000000000000000..76cbb7e174aa7f496388dabb708f0723bebfd0e0 --- /dev/null +++ b/vlm/train/PhtFY9plHk/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2e1e57a39363ba5506f6a8bd4ece79261d4b22be662905817f15c97576386a6c +size 626084 diff --git a/vlm/train/PhtFY9plHk/7.png b/vlm/train/PhtFY9plHk/7.png new file mode 100644 index 0000000000000000000000000000000000000000..12f2c5f8f7e665b8a745136ebf5b0006319d4ffe --- /dev/null +++ b/vlm/train/PhtFY9plHk/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e18c88c8386491397d7d7da09deec9f47cedd34ca9163bd88d00eba6334265d4 +size 352673 diff --git a/vlm/train/PhtFY9plHk/8.png b/vlm/train/PhtFY9plHk/8.png new file mode 100644 index 0000000000000000000000000000000000000000..68ed5310a3f0ee040f6bceaca8d93140580e8f8b --- /dev/null +++ b/vlm/train/PhtFY9plHk/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f6f2131c3bf3b749f4e90093aa3aa8b2ea75ac4e7a3d8d445e3ff20c8bfc82c6 +size 378632 diff --git a/vlm/train/PhtFY9plHk/9.png b/vlm/train/PhtFY9plHk/9.png new file mode 100644 index 0000000000000000000000000000000000000000..537fec8dc944fa2605afc278ecd7b64411888c15 --- /dev/null +++ b/vlm/train/PhtFY9plHk/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d4ca7d6cff4f001ab1168450aafff149686c289ca7c8694bfd6328a1ec696464 +size 580367 diff --git a/vlm/train/Re_VXFOyyO/10.png b/vlm/train/Re_VXFOyyO/10.png new file mode 100644 index 0000000000000000000000000000000000000000..8578098d4688b3500e07c34e09a00d0cf67cf72d --- /dev/null +++ b/vlm/train/Re_VXFOyyO/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:253644fcbdd10fde5b969638e85691ca3f012f3ce89d9a9df144f66e0c311166 +size 539893 diff --git a/vlm/train/Re_VXFOyyO/3.png b/vlm/train/Re_VXFOyyO/3.png new file mode 100644 index 0000000000000000000000000000000000000000..2ebb69c84ab6de7f3e8c785d035a4a4b3d0163be --- /dev/null +++ b/vlm/train/Re_VXFOyyO/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c29d39b3d5c581cd41a4559778d734e8255360c6cf1260bd73a9c289998af13a +size 531864 diff --git a/vlm/train/Re_VXFOyyO/6.png b/vlm/train/Re_VXFOyyO/6.png new file mode 100644 index 0000000000000000000000000000000000000000..f2ec1d1fe96c5427f80fc2612c4c47261ac044c6 --- /dev/null +++ b/vlm/train/Re_VXFOyyO/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bb3b01ceb0b4aa656bece7c85441d178f776827e47138194bac562c9171cb7ea +size 568233 diff --git a/vlm/train/Re_VXFOyyO/9.png b/vlm/train/Re_VXFOyyO/9.png new file mode 100644 index 0000000000000000000000000000000000000000..c1a063b602e5b5a96403740f4f81523711630b7d --- /dev/null +++ b/vlm/train/Re_VXFOyyO/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9a8c6eaf2ebc7c659e3597a07f790296746389faca1b45ce8f503357552cb7e2 +size 480397 diff --git a/vlm/train/S1X7nhsxl/0.png b/vlm/train/S1X7nhsxl/0.png new file mode 100644 index 0000000000000000000000000000000000000000..637cccd0aa822595ab0ea1e97ff8c8e71e2314c9 --- /dev/null +++ b/vlm/train/S1X7nhsxl/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9267d6d50c69b46456121b90d0a419d8ad4ef253769639b4d5145d8a5ae7eab5 +size 480581 diff --git a/vlm/train/S1X7nhsxl/1.png b/vlm/train/S1X7nhsxl/1.png new file mode 100644 index 0000000000000000000000000000000000000000..d824173151046d26ea4b0b9840eca6c331194841 --- /dev/null +++ b/vlm/train/S1X7nhsxl/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bf2fc16843316c35cd3e33053ababe24e7d888e326edd1e007334eeeb8c18882 +size 522662 diff --git a/vlm/train/S1X7nhsxl/10.png b/vlm/train/S1X7nhsxl/10.png new file mode 100644 index 0000000000000000000000000000000000000000..afa392989b7bf1f63791d6b6bc87b1eeb6d94f72 --- /dev/null +++ b/vlm/train/S1X7nhsxl/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b11170ba4669959ef7e88323b8e886b3ef03009bbb5c9511eba38458380f0a38 +size 447110 diff --git a/vlm/train/S1X7nhsxl/2.png b/vlm/train/S1X7nhsxl/2.png new file mode 100644 index 0000000000000000000000000000000000000000..dd83d3d5f1ec1f531a6adbb560593cdfda6c4630 --- /dev/null +++ b/vlm/train/S1X7nhsxl/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:76f952dcddc1558fe0f08dc51fc57b3562320268e8483be38f8b1f1eaf6e0a8f +size 616465 diff --git a/vlm/train/S1X7nhsxl/3.png b/vlm/train/S1X7nhsxl/3.png new file mode 100644 index 0000000000000000000000000000000000000000..d233013ac4c00a306bf4091683e42f7066b4e71d --- /dev/null +++ b/vlm/train/S1X7nhsxl/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:451b55e1b5d902b853d6e004a00fafd844d82db8aa839bde3614e8ce1dc5f07a +size 427004 diff --git a/vlm/train/S1X7nhsxl/4.png b/vlm/train/S1X7nhsxl/4.png new file mode 100644 index 0000000000000000000000000000000000000000..6a597659bc2d3f75df5bf7616baf00f84c58ea7c --- /dev/null +++ b/vlm/train/S1X7nhsxl/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fc9d14c5f8e88efe78d4e38182d68f6fb8b90aa2f1724deefe81b10082712105 +size 549985 diff --git a/vlm/train/S1X7nhsxl/5.png b/vlm/train/S1X7nhsxl/5.png new file mode 100644 index 0000000000000000000000000000000000000000..674ad3928793d6bf6642d5ab1602014f2ab133a0 --- /dev/null +++ b/vlm/train/S1X7nhsxl/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5d1c2b519f0cd5219f1374a68047e96a8be420a204945632b77c4ad220d2368a +size 519399 diff --git a/vlm/train/S1X7nhsxl/6.png b/vlm/train/S1X7nhsxl/6.png new file mode 100644 index 0000000000000000000000000000000000000000..dc780935b11844babcec85fcdf958541f1a86cda --- /dev/null +++ b/vlm/train/S1X7nhsxl/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2cf5207860de1079c89451c2dd44685e60f5f7fce060061a8179b6aaa67dd8a2 +size 665896 diff --git a/vlm/train/S1X7nhsxl/7.png b/vlm/train/S1X7nhsxl/7.png new file mode 100644 index 0000000000000000000000000000000000000000..65b92ed3bec0493160dd2b777f82f7fa83f923c0 --- /dev/null +++ b/vlm/train/S1X7nhsxl/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:96d3a28f9e9ddf41c011941460837fca7a2a7893e15e601a2749e71ced384a6e +size 1563670 diff --git a/vlm/train/S1X7nhsxl/8.png b/vlm/train/S1X7nhsxl/8.png new file mode 100644 index 0000000000000000000000000000000000000000..4f25c1931a0a9df4eed583db4aa8f998fef58f5c --- /dev/null +++ b/vlm/train/S1X7nhsxl/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c184f2f14ae164b9099ea77364bd26bd6ac6756656949bdcb3d6d3b6864a8dbe +size 521367 diff --git a/vlm/train/S1X7nhsxl/9.png b/vlm/train/S1X7nhsxl/9.png new file mode 100644 index 0000000000000000000000000000000000000000..997c22c585e8e939ffe43090467d41aa24bf0719 --- /dev/null +++ b/vlm/train/S1X7nhsxl/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e9c71a2923663d0b778fbf336bfe8bcd4c522873c4cd5da0e8f297597b4a9f67 +size 541883 diff --git a/vlm/train/S1xh5sYgx/0.png b/vlm/train/S1xh5sYgx/0.png new file mode 100644 index 0000000000000000000000000000000000000000..af2d023588c33643628a82d7f69f835571c3b844 --- /dev/null +++ b/vlm/train/S1xh5sYgx/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:743a3752184db78d46d3eb4186f24aa0eac44fe51cc09b167cb64fe8728dcfad +size 506594 diff --git a/vlm/train/S1xh5sYgx/1.png b/vlm/train/S1xh5sYgx/1.png new file mode 100644 index 0000000000000000000000000000000000000000..9ac511cb0caee700a9564b641faffda543c3bca7 --- /dev/null +++ b/vlm/train/S1xh5sYgx/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1a78292d1bd4de6d2b44de6b2152a74122faa6bb944945c67d92e7032872bcf9 +size 609161 diff --git a/vlm/train/S1xh5sYgx/10.png b/vlm/train/S1xh5sYgx/10.png new file mode 100644 index 0000000000000000000000000000000000000000..91fff0bc147ac3f1da758d071bca6b60a4336558 --- /dev/null +++ b/vlm/train/S1xh5sYgx/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3b99121ba0f3ea801ba7da5922f41ee7ffad716ea01cf7a93d44a24ee70a99af +size 497619 diff --git a/vlm/train/S1xh5sYgx/11.png b/vlm/train/S1xh5sYgx/11.png new file mode 100644 index 0000000000000000000000000000000000000000..26b14e390ce8dfe9e90e74fbcbc402d62c2e8bd3 --- /dev/null +++ b/vlm/train/S1xh5sYgx/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:78d0ae3d5a8c846aaca27f2ed773712c5a47da8a8ad5e678896368e6c5095a62 +size 497164 diff --git a/vlm/train/S1xh5sYgx/12.png b/vlm/train/S1xh5sYgx/12.png new file mode 100644 index 0000000000000000000000000000000000000000..1c78b4442e76214db6ee853246e4563b4a46a74c --- /dev/null +++ b/vlm/train/S1xh5sYgx/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2dbff9dca123783839614268c5b6e77bbf8fc025c58ddd2cb91ff0bbbaccbf98 +size 145544 diff --git a/vlm/train/S1xh5sYgx/2.png b/vlm/train/S1xh5sYgx/2.png new file mode 100644 index 0000000000000000000000000000000000000000..81320cd09d85909a1a0bd557859725cc2e6cd409 --- /dev/null +++ b/vlm/train/S1xh5sYgx/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6f53ec80a8ce1a899685467de6c20f1d5446636a3a1dfa0d99bcb57ed0f647a2 +size 590948 diff --git a/vlm/train/S1xh5sYgx/3.png b/vlm/train/S1xh5sYgx/3.png new file mode 100644 index 0000000000000000000000000000000000000000..522a2f25d80689e7188c29fd2040fd64b8cb3088 --- /dev/null +++ b/vlm/train/S1xh5sYgx/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3f33ee12d19d9e705b78a8c6fd5f4b82e3043310400e054cdd4a8372e3423f20 +size 493789 diff --git a/vlm/train/S1xh5sYgx/4.png b/vlm/train/S1xh5sYgx/4.png new file mode 100644 index 0000000000000000000000000000000000000000..855959d0cde778edd0250d251444fd4754ef3042 --- /dev/null +++ b/vlm/train/S1xh5sYgx/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7f06f2ba9a24d1e2a4a0694b8029b99ce4f707d9dd51db31719149d2a1425557 +size 450728 diff --git a/vlm/train/S1xh5sYgx/5.png b/vlm/train/S1xh5sYgx/5.png new file mode 100644 index 0000000000000000000000000000000000000000..d60abeb14caac02529dbbfad386249cca15d9dbc --- /dev/null +++ b/vlm/train/S1xh5sYgx/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:63541fdbfec4791ce14c758e694c2704951519424675f9c136710c6d97047741 +size 508500 diff --git a/vlm/train/S1xh5sYgx/6.png b/vlm/train/S1xh5sYgx/6.png new file mode 100644 index 0000000000000000000000000000000000000000..7a2bf3e1b6ec023e28fa9288d743590b29b8cf6e --- /dev/null +++ b/vlm/train/S1xh5sYgx/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:920033cbe7c7f5b1060d01f37df993f2e2892c2fdd52fc3efbda0cb25d4e4435 +size 562094 diff --git a/vlm/train/S1xh5sYgx/7.png b/vlm/train/S1xh5sYgx/7.png new file mode 100644 index 0000000000000000000000000000000000000000..642848a8905feb6204cca7bfeb78d8b5c6a04866 --- /dev/null +++ b/vlm/train/S1xh5sYgx/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:737cd9ab9c2fea09398cd8060c12ba442610c54943d9a121f849098f9e8e5f08 +size 587662 diff --git a/vlm/train/S1xh5sYgx/8.png b/vlm/train/S1xh5sYgx/8.png new file mode 100644 index 0000000000000000000000000000000000000000..87d1c89d461312d0adcfced259bff2eaa23c88ec --- /dev/null +++ b/vlm/train/S1xh5sYgx/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:63abefd825e43eaf02d2ad79d40625eb52e8a2d302fdf445ef607b3367d93e27 +size 565132 diff --git a/vlm/train/S1xh5sYgx/9.png b/vlm/train/S1xh5sYgx/9.png new file mode 100644 index 0000000000000000000000000000000000000000..409ae119b2a7594a7a9ab6da598436e43541c947 --- /dev/null +++ b/vlm/train/S1xh5sYgx/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:71b243d9397d4065ab6cd2b35bcdf756a5ff3b0280a5f1fc417da766eed18b3e +size 545762 diff --git a/vlm/train/SJgf6Z-0W/0.png b/vlm/train/SJgf6Z-0W/0.png new file mode 100644 index 0000000000000000000000000000000000000000..9de116784e024fddb1762757937bd482e30c457f --- /dev/null +++ b/vlm/train/SJgf6Z-0W/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1aa427824f28fed945b36956235a83b9947220c77c25cde558836e7534a8a7df +size 468238 diff --git a/vlm/train/SJgf6Z-0W/1.png b/vlm/train/SJgf6Z-0W/1.png new file mode 100644 index 0000000000000000000000000000000000000000..44534d1ef69b70dc67e36638bb1aa67a36db1401 --- /dev/null +++ b/vlm/train/SJgf6Z-0W/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:42a9297fb59b202b9ce3095c0e1f91278b53b8f4958b25c4abc282ab45490973 +size 566272 diff --git a/vlm/train/SJgf6Z-0W/10.png b/vlm/train/SJgf6Z-0W/10.png new file mode 100644 index 0000000000000000000000000000000000000000..11c400eff711558eee97f4a8c07657d5ca93f0dc --- /dev/null +++ b/vlm/train/SJgf6Z-0W/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2a0c0679177b9efca8b8b128ad4f15b71113859e711d79035aa9614f907faee7 +size 86150 diff --git a/vlm/train/SJgf6Z-0W/2.png b/vlm/train/SJgf6Z-0W/2.png new file mode 100644 index 0000000000000000000000000000000000000000..0cb132badd850a7cbd7b9205f0a91bd4c9e32e17 --- /dev/null +++ b/vlm/train/SJgf6Z-0W/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c258e556017c5ff770b224bc9f6efc143e70ba0dceac5af1d8c9d4a1d27f6df8 +size 436845 diff --git a/vlm/train/SJgf6Z-0W/3.png b/vlm/train/SJgf6Z-0W/3.png new file mode 100644 index 0000000000000000000000000000000000000000..fd297dec9b84c3a5f7d0ac71cb829852db89f6f9 --- /dev/null +++ b/vlm/train/SJgf6Z-0W/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b1448b9727cea004a6061c53dec11aa9c6d2dff8a211c54c233266c203b51d66 +size 455760 diff --git a/vlm/train/SJgf6Z-0W/4.png b/vlm/train/SJgf6Z-0W/4.png new file mode 100644 index 0000000000000000000000000000000000000000..dbe495afece96961f905f6ace3cc8a72bbca3fca --- /dev/null +++ b/vlm/train/SJgf6Z-0W/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b228b9c06b570e20dddad36797be32da07dbf3dc9b719f5e286c70c693e2ff56 +size 472704 diff --git a/vlm/train/SJgf6Z-0W/5.png b/vlm/train/SJgf6Z-0W/5.png new file mode 100644 index 0000000000000000000000000000000000000000..8c63776c18902e79049b341582e0085acd6663ee --- /dev/null +++ b/vlm/train/SJgf6Z-0W/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:701dd423536e5334cd49c0a17cb5254e063798474405592eb9a6fc26ee24db22 +size 443708 diff --git a/vlm/train/SJgf6Z-0W/6.png b/vlm/train/SJgf6Z-0W/6.png new file mode 100644 index 0000000000000000000000000000000000000000..675b7086e57d23c4b547a421448074129abc061f --- /dev/null +++ b/vlm/train/SJgf6Z-0W/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d3080ff3ec95dea64561621d07678c01bedec3f6ee5600063771b2ae06c45e8c +size 365531 diff --git a/vlm/train/SJgf6Z-0W/7.png b/vlm/train/SJgf6Z-0W/7.png new file mode 100644 index 0000000000000000000000000000000000000000..cb0ac4487cf8c27a310299d9273a0cdb67f54004 --- /dev/null +++ b/vlm/train/SJgf6Z-0W/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9f5718d6e5e685b6f3b6e962962d9b9422bdba26e0ba8f828786d6955e934b9b +size 449048 diff --git a/vlm/train/SJgf6Z-0W/8.png b/vlm/train/SJgf6Z-0W/8.png new file mode 100644 index 0000000000000000000000000000000000000000..1faecbbab9ad86a0da35c881fb6f66f93e9565ce --- /dev/null +++ b/vlm/train/SJgf6Z-0W/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8f93cfbe7f48aefc42d3fb3a28e3e331eabf60c54e053d31147a17463bdc1778 +size 576414 diff --git a/vlm/train/SJgf6Z-0W/9.png b/vlm/train/SJgf6Z-0W/9.png new file mode 100644 index 0000000000000000000000000000000000000000..4a36637ba89e2f54e4d8b7523e8dd92ff336fd6a --- /dev/null +++ b/vlm/train/SJgf6Z-0W/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6be1f82dff44735fe703ff43ebf0110743fbf4308be57451658d75a3e7a5f6ed +size 121451 diff --git a/vlm/train/SkFEGHx0Z/0.png b/vlm/train/SkFEGHx0Z/0.png new file mode 100644 index 0000000000000000000000000000000000000000..1ae74e9ced637544ff192a201f606f7b00ec9fa1 --- /dev/null +++ b/vlm/train/SkFEGHx0Z/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:df6eaf2b7b9fd2740a759a646467b5d453ad07840d7dcfa27308668f316a0257 +size 492494 diff --git a/vlm/train/SkFEGHx0Z/1.png b/vlm/train/SkFEGHx0Z/1.png new file mode 100644 index 0000000000000000000000000000000000000000..35b070b1abd67c7427c1e04ec2fc506106fc37e6 --- /dev/null +++ b/vlm/train/SkFEGHx0Z/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:191b560b68f58aa63436bfec52edf4b47fbea69a2020424c3953bb200437e178 +size 551680 diff --git a/vlm/train/SkFEGHx0Z/10.png b/vlm/train/SkFEGHx0Z/10.png new file mode 100644 index 0000000000000000000000000000000000000000..fe03a31f74d525348ba957289816f6a9eb44de6a --- /dev/null +++ b/vlm/train/SkFEGHx0Z/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:76012113d56a49b0cc70f178c1465b251c5536869e7bb55ac9adf756d5705b84 +size 551022 diff --git a/vlm/train/SkFEGHx0Z/11.png b/vlm/train/SkFEGHx0Z/11.png new file mode 100644 index 0000000000000000000000000000000000000000..7452d4fa27364cdf4304580cb38a9716599a0e9e --- /dev/null +++ b/vlm/train/SkFEGHx0Z/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7492ccc72bcfff2722653d570e0d338033649860a727ab5ba3ae64c2885d903f +size 409647 diff --git a/vlm/train/SkFEGHx0Z/2.png b/vlm/train/SkFEGHx0Z/2.png new file mode 100644 index 0000000000000000000000000000000000000000..9defa3fb2598b68c160955caeebbd77c3f4dad91 --- /dev/null +++ b/vlm/train/SkFEGHx0Z/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3e706ce1a42665145fd7e63536bf271b77ddd490683980176c3d9dfe11f2cda7 +size 589525 diff --git a/vlm/train/SkFEGHx0Z/3.png b/vlm/train/SkFEGHx0Z/3.png new file mode 100644 index 0000000000000000000000000000000000000000..7fc6c9a32f31aaa71f33d64c71d6083001af9555 --- /dev/null +++ b/vlm/train/SkFEGHx0Z/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec6bf2a2f4fd62718efc511476a962c145b3423a750a9aae1d144b988b230a94 +size 560698 diff --git a/vlm/train/SkFEGHx0Z/4.png b/vlm/train/SkFEGHx0Z/4.png new file mode 100644 index 0000000000000000000000000000000000000000..0f8111d37d4f65465be78f1ca2ec82a076c9c0e0 --- /dev/null +++ b/vlm/train/SkFEGHx0Z/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c30966f82a3caa8378a726ab18648a5de498c7c733deb06d64f9b4e8da9ee252 +size 586859 diff --git a/vlm/train/SkFEGHx0Z/5.png b/vlm/train/SkFEGHx0Z/5.png new file mode 100644 index 0000000000000000000000000000000000000000..b6fd3bf5c6805e86556a90ddfdb6ef2629aa4a86 --- /dev/null +++ b/vlm/train/SkFEGHx0Z/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7908937164ed92b417605f32ea2476f68d26f7cb32ce808f72b95a716fc83487 +size 496026 diff --git a/vlm/train/SkFEGHx0Z/6.png b/vlm/train/SkFEGHx0Z/6.png new file mode 100644 index 0000000000000000000000000000000000000000..9c32ea49ba7943d67b7cb8cc2e05b7969905a927 --- /dev/null +++ b/vlm/train/SkFEGHx0Z/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0070271b34d53bfd0022d1c3a29f9ec5fc34f58d33354731d61a563edd351904 +size 1263728 diff --git a/vlm/train/SkFEGHx0Z/7.png b/vlm/train/SkFEGHx0Z/7.png new file mode 100644 index 0000000000000000000000000000000000000000..85ecbb62b0f0e1e5df4a3a457d7c85e255e7c06e --- /dev/null +++ b/vlm/train/SkFEGHx0Z/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d20476bdccfb3a86d48bdd903b492a4a813f4021314d7588c3ce98a515452b63 +size 444052 diff --git a/vlm/train/SkFEGHx0Z/8.png b/vlm/train/SkFEGHx0Z/8.png new file mode 100644 index 0000000000000000000000000000000000000000..0eeecce1ec1b86de4ee6f795150edc44c4731b9b --- /dev/null +++ b/vlm/train/SkFEGHx0Z/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:af0909ddbad61ed3ca95f44893335a4ec66647bd64ff3f852fdf3b3a1a3aae0f +size 498889 diff --git a/vlm/train/SkFEGHx0Z/9.png b/vlm/train/SkFEGHx0Z/9.png new file mode 100644 index 0000000000000000000000000000000000000000..a4f58c4c28358bcd7ec987ec1242840da4b9dc4b --- /dev/null +++ b/vlm/train/SkFEGHx0Z/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:10734f1952c0ac06172a57d8b884ee656c6b058c0a38379007405e316d3447ea +size 425900 diff --git a/vlm/train/SknC0bW0-/0.png b/vlm/train/SknC0bW0-/0.png new file mode 100644 index 0000000000000000000000000000000000000000..fecad5698785559ca15f34c3a86d128314919c2a --- /dev/null +++ b/vlm/train/SknC0bW0-/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fdd90f54bee8cc4c14db79ddb91b81334561f4d104aa3180c419a8e03c3e6559 +size 483227 diff --git a/vlm/train/SknC0bW0-/1.png b/vlm/train/SknC0bW0-/1.png new file mode 100644 index 0000000000000000000000000000000000000000..76834e4a4a7a4aee28fba8e368fabe0accbc19e9 --- /dev/null +++ b/vlm/train/SknC0bW0-/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:892b0d518954334de3b14bfa8b08d3a67c746b92d7c0e9092ae456c8b5234b33 +size 556328 diff --git a/vlm/train/SknC0bW0-/2.png b/vlm/train/SknC0bW0-/2.png new file mode 100644 index 0000000000000000000000000000000000000000..f792cf2ca0959218973e350ecd9f923f970f090c --- /dev/null +++ b/vlm/train/SknC0bW0-/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:82dd1bcb227bbcd3fc855f294c6ffa2bb8e89d806b7c1b4b29a70e3f9a2899d7 +size 578207 diff --git a/vlm/train/SknC0bW0-/3.png b/vlm/train/SknC0bW0-/3.png new file mode 100644 index 0000000000000000000000000000000000000000..87c02a8cdc51483017aaf20d1ec796766eb05a4c --- /dev/null +++ b/vlm/train/SknC0bW0-/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:825364de1cab19c90d89e9acb175b0be73e9fc0048f95db67c86639f86c81c7f +size 531275 diff --git a/vlm/train/SknC0bW0-/4.png b/vlm/train/SknC0bW0-/4.png new file mode 100644 index 0000000000000000000000000000000000000000..30ad6a9a48399afe2d8f62bc1b85d9b49aa0134e --- /dev/null +++ b/vlm/train/SknC0bW0-/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f2800c38d626b78801e06360fd6ccc1b24b74527767c3b372bb4674431f49d2e +size 479445 diff --git a/vlm/train/SknC0bW0-/5.png b/vlm/train/SknC0bW0-/5.png new file mode 100644 index 0000000000000000000000000000000000000000..a1ca266ddd0cba8feeaac4c3aec6c550fcc3fa1d --- /dev/null +++ b/vlm/train/SknC0bW0-/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f8e4f0fc65af78d67aec1607e88624509aa4cf480eb90e120aad66e0b9fae9d5 +size 482018 diff --git a/vlm/train/SknC0bW0-/6.png b/vlm/train/SknC0bW0-/6.png new file mode 100644 index 0000000000000000000000000000000000000000..4a5416f38b18c63de9d93acbb591f3e5ad2893a6 --- /dev/null +++ b/vlm/train/SknC0bW0-/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4a693905b359e1073c5286f0fbbd88c8d6269b7cd554ea39b6b957be3f0c6b3c +size 568380 diff --git a/vlm/train/SknC0bW0-/7.png b/vlm/train/SknC0bW0-/7.png new file mode 100644 index 0000000000000000000000000000000000000000..f103adae489f296c63ad6d64b47d300bf422c554 --- /dev/null +++ b/vlm/train/SknC0bW0-/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:96444201bf4d3ef71a58b8f7ccbed1949b7395e749e92200998b087af28ca88f +size 621333 diff --git a/vlm/train/SknC0bW0-/8.png b/vlm/train/SknC0bW0-/8.png new file mode 100644 index 0000000000000000000000000000000000000000..5d9ad68f051706f2bda808e04d6659f4f176c452 --- /dev/null +++ b/vlm/train/SknC0bW0-/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cbeca383b0ec7ff47d7f82dd7382176fcb134f653d96675c9063d74e62cbf870 +size 563462 diff --git a/vlm/train/SknC0bW0-/9.png b/vlm/train/SknC0bW0-/9.png new file mode 100644 index 0000000000000000000000000000000000000000..424c7423f60994cb922c467346b31da5e92adde2 --- /dev/null +++ b/vlm/train/SknC0bW0-/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0028722e8d7946b86b0ceab5100e04974181e43ca89694a4754af95a5b23797f +size 116575 diff --git a/vlm/train/Svfh1_hYEtF/0.png b/vlm/train/Svfh1_hYEtF/0.png new file mode 100644 index 0000000000000000000000000000000000000000..9f88d2d08f3f6a9334b534ede69bbe0e73934f45 --- /dev/null +++ b/vlm/train/Svfh1_hYEtF/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5dc0d160dd44f4f85e28fbe138b5b5bb043426f1789284530619cfa630c1a35c +size 535973 diff --git a/vlm/train/Svfh1_hYEtF/1.png b/vlm/train/Svfh1_hYEtF/1.png new file mode 100644 index 0000000000000000000000000000000000000000..7e80ae5acddcaecdf144b8aa888cfb8d66315c28 --- /dev/null +++ b/vlm/train/Svfh1_hYEtF/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:462e3085444a81fdc53477c6a2e14661611d8892eb180d9dd4040d88db2c1a05 +size 798443 diff --git a/vlm/train/Svfh1_hYEtF/10.png b/vlm/train/Svfh1_hYEtF/10.png new file mode 100644 index 0000000000000000000000000000000000000000..3d04ec90d1f26f0916e0503ed116690f37a3a3c2 --- /dev/null +++ b/vlm/train/Svfh1_hYEtF/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:45896dbb49f632a5fc75a0c8296769ecccce39c00b48fadb06855f0202428cb6 +size 147761 diff --git a/vlm/train/Svfh1_hYEtF/11.png b/vlm/train/Svfh1_hYEtF/11.png new file mode 100644 index 0000000000000000000000000000000000000000..854274d32485443fa088b0ea2d4b103f6c2710bd --- /dev/null +++ b/vlm/train/Svfh1_hYEtF/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d39e4f7cf34c840575bba487ca694c64abcda1a5bcbd7cf7a36a95b5ec4e2df2 +size 534702 diff --git a/vlm/train/Svfh1_hYEtF/12.png b/vlm/train/Svfh1_hYEtF/12.png new file mode 100644 index 0000000000000000000000000000000000000000..6f1c98c81ca8e350bade6d1f25942a449eaa8af4 --- /dev/null +++ b/vlm/train/Svfh1_hYEtF/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:31b454979af2cfdf1ac25bf8b46b8ebc6a63c406905ab162e77e0c854b778ddf +size 507377 diff --git a/vlm/train/Svfh1_hYEtF/13.png b/vlm/train/Svfh1_hYEtF/13.png new file mode 100644 index 0000000000000000000000000000000000000000..8272bb9f7337f510ea0b5adfd7ec7fcf08eeb480 --- /dev/null +++ b/vlm/train/Svfh1_hYEtF/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c4befd3d74899fa82b2d79da088cde523e968fcbd4ea60e2ab5c6c0d7bef5f42 +size 713218 diff --git a/vlm/train/Svfh1_hYEtF/14.png b/vlm/train/Svfh1_hYEtF/14.png new file mode 100644 index 0000000000000000000000000000000000000000..44f8f98c987ec51a59a11a1fa32911796a1a02ac --- /dev/null +++ b/vlm/train/Svfh1_hYEtF/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:86775020c30253c0699bbcaf009a6195688bcecf86b310431aae3bed49f6f51a +size 399390 diff --git a/vlm/train/Svfh1_hYEtF/15.png b/vlm/train/Svfh1_hYEtF/15.png new file mode 100644 index 0000000000000000000000000000000000000000..3e043fd65edeaad97af1209697c85abebfad7902 --- /dev/null +++ b/vlm/train/Svfh1_hYEtF/15.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bb1650eda8fcf7b65c142215e3c3ad87083c98516f9db9b976dad0be9f9271e6 +size 259705 diff --git a/vlm/train/Svfh1_hYEtF/2.png b/vlm/train/Svfh1_hYEtF/2.png new file mode 100644 index 0000000000000000000000000000000000000000..cda1333b089ae63f40326b87a4422f7ac2d25dad --- /dev/null +++ b/vlm/train/Svfh1_hYEtF/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b1ebefe4d3cc49b185bf44e5f04e307a4bfaee5df1e7b6623f1dd4dbea73f8d5 +size 648901 diff --git a/vlm/train/Svfh1_hYEtF/3.png b/vlm/train/Svfh1_hYEtF/3.png new file mode 100644 index 0000000000000000000000000000000000000000..4d39c10ccb71ba7bfff10851fbd8141b30c0f90e --- /dev/null +++ b/vlm/train/Svfh1_hYEtF/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fb229d661a9c37fe454e1580b1747e50b75a96d693dfb4c52712fdfd227fcce1 +size 604708 diff --git a/vlm/train/Svfh1_hYEtF/4.png b/vlm/train/Svfh1_hYEtF/4.png new file mode 100644 index 0000000000000000000000000000000000000000..db0c830bed487602129e3e49bbc635e12a55ea65 --- /dev/null +++ b/vlm/train/Svfh1_hYEtF/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7c04c2d92392540c185e14b553cbca2342fe5bf289ffbf5533e2b0f577ec3ca4 +size 623452 diff --git a/vlm/train/Svfh1_hYEtF/5.png b/vlm/train/Svfh1_hYEtF/5.png new file mode 100644 index 0000000000000000000000000000000000000000..68ef794c7823e715a8fa03575d25815bf534a2fb --- /dev/null +++ b/vlm/train/Svfh1_hYEtF/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8fd170bf13d1fcb59eb343974170e62495f49dfbec866f6f8fc8e8eb206d4770 +size 644501 diff --git a/vlm/train/Svfh1_hYEtF/6.png b/vlm/train/Svfh1_hYEtF/6.png new file mode 100644 index 0000000000000000000000000000000000000000..ada845a3da5e1dc056071d159ac6422ab622e5c7 --- /dev/null +++ b/vlm/train/Svfh1_hYEtF/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:06216a15ea63d8aba1505402f7714baeb5f82dddd73a8b83ad86e0c3cb269d55 +size 707918 diff --git a/vlm/train/Svfh1_hYEtF/7.png b/vlm/train/Svfh1_hYEtF/7.png new file mode 100644 index 0000000000000000000000000000000000000000..54a37a4dafaf3eb65670e382818b3d5e443e213f --- /dev/null +++ b/vlm/train/Svfh1_hYEtF/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:74b162609ec0a826362c15385a59b6bf15452bd48dccfe2062ee78152b8c8c0b +size 693234 diff --git a/vlm/train/Svfh1_hYEtF/8.png b/vlm/train/Svfh1_hYEtF/8.png new file mode 100644 index 0000000000000000000000000000000000000000..2062e2b71a07e25f2bba6194f6382ed10b972fc9 --- /dev/null +++ b/vlm/train/Svfh1_hYEtF/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:33c29f8cc1f07bbc3022bf65014e8d16b5cd56fac934374ae1f6cf16e9635c0d +size 531694 diff --git a/vlm/train/Svfh1_hYEtF/9.png b/vlm/train/Svfh1_hYEtF/9.png new file mode 100644 index 0000000000000000000000000000000000000000..9882e90f4ea945891aa76a5601881bbacacf87e3 --- /dev/null +++ b/vlm/train/Svfh1_hYEtF/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7fe230c0b7705087328d58683a9927989da1dfd6d4e12abfe8c33794eab5bf88 +size 593842 diff --git a/vlm/train/Sy4lojC9tm/0.png b/vlm/train/Sy4lojC9tm/0.png new file mode 100644 index 0000000000000000000000000000000000000000..e35e74e3d0b2b8bedf19ee6ff38c7cc05e2697c8 --- /dev/null +++ b/vlm/train/Sy4lojC9tm/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3e48de047368610f00f8cc857e56f56d24fa8c9e05cb87db764bc8962c0c974e +size 524635 diff --git a/vlm/train/Sy4lojC9tm/1.png b/vlm/train/Sy4lojC9tm/1.png new file mode 100644 index 0000000000000000000000000000000000000000..b567d1393c8ae6b74b1f36d8053cebd89227ae51 --- /dev/null +++ b/vlm/train/Sy4lojC9tm/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:98e5dddfa7c970c73ba59c18c1dddb424bed4b510c66c841d803f12b3dd896ee +size 941042 diff --git a/vlm/train/Sy4lojC9tm/10.png b/vlm/train/Sy4lojC9tm/10.png new file mode 100644 index 0000000000000000000000000000000000000000..70adee1281f2f0023b89797e81ab2fe68074ce15 --- /dev/null +++ b/vlm/train/Sy4lojC9tm/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f4a44bb66f86edb238dcbdfbd1c31bb232fb8d7c68e09a069aa90082b9a33b2d +size 482019 diff --git a/vlm/train/Sy4lojC9tm/11.png b/vlm/train/Sy4lojC9tm/11.png new file mode 100644 index 0000000000000000000000000000000000000000..edd97adfb1b0656e2ecc2cb5769d184b340ec87c --- /dev/null +++ b/vlm/train/Sy4lojC9tm/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aa7734e06c4bab16470e2843e6235d3f99739b8cc50745acd5267ea85f54d076 +size 129012 diff --git a/vlm/train/Sy4lojC9tm/12.png b/vlm/train/Sy4lojC9tm/12.png new file mode 100644 index 0000000000000000000000000000000000000000..ecc0814cad57e702fc81dcd84bbf1f4425e2f8f8 --- /dev/null +++ b/vlm/train/Sy4lojC9tm/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8df7362488b28bb8af48e03cdbeda110a68d184b6f6c32c812db8c9029de3c8a +size 167649 diff --git a/vlm/train/Sy4lojC9tm/2.png b/vlm/train/Sy4lojC9tm/2.png new file mode 100644 index 0000000000000000000000000000000000000000..7801b292af8416316314ac17daf6a931254fb184 --- /dev/null +++ b/vlm/train/Sy4lojC9tm/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:33cdb9766f70f07fbb025fd2e1621b62ef37f7ea95415800f96bfe1dffd805fc +size 580563 diff --git a/vlm/train/Sy4lojC9tm/3.png b/vlm/train/Sy4lojC9tm/3.png new file mode 100644 index 0000000000000000000000000000000000000000..9160935632bf39a737314830ed35303373415721 --- /dev/null +++ b/vlm/train/Sy4lojC9tm/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bbd2e3f6bba10de92155f0cf9f07d37a424a2ce65c12684c7ff7e42606818937 +size 474102 diff --git a/vlm/train/Sy4lojC9tm/4.png b/vlm/train/Sy4lojC9tm/4.png new file mode 100644 index 0000000000000000000000000000000000000000..b5eae6ded1694cd305ea381e7f9426c90c21b884 --- /dev/null +++ b/vlm/train/Sy4lojC9tm/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8557d52b3df734efed6854d27beaf613eb1a924ce44fd3a3b1604c571985e595 +size 520999 diff --git a/vlm/train/Sy4lojC9tm/5.png b/vlm/train/Sy4lojC9tm/5.png new file mode 100644 index 0000000000000000000000000000000000000000..5bdb91bbf99fa344f750660b835e5dd89e50b289 --- /dev/null +++ b/vlm/train/Sy4lojC9tm/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:90813aa3b014a8e412f1e11f0435e61d9e70dcde046f0f8be3ae8254b44de3c3 +size 574088 diff --git a/vlm/train/Sy4lojC9tm/6.png b/vlm/train/Sy4lojC9tm/6.png new file mode 100644 index 0000000000000000000000000000000000000000..fbb6126c35f3f3e163a2452dd98ffb6e2b25f545 --- /dev/null +++ b/vlm/train/Sy4lojC9tm/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1df18a0315b0da1a01047e5bfdb2a7ee19c7d24559a1ce635ba1a209b31d830b +size 1024930 diff --git a/vlm/train/Sy4lojC9tm/7.png b/vlm/train/Sy4lojC9tm/7.png new file mode 100644 index 0000000000000000000000000000000000000000..347987468e9513c0052002f1dd75af7bfa954359 --- /dev/null +++ b/vlm/train/Sy4lojC9tm/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ddd4f421d6d9970adb751e94e3714b7183f6477d23f90f7aa94c589b0bdbc30c +size 580783 diff --git a/vlm/train/Sy4lojC9tm/8.png b/vlm/train/Sy4lojC9tm/8.png new file mode 100644 index 0000000000000000000000000000000000000000..f57d0b8adfaa954ed4e6e79ba0f33dfe5e7518d8 --- /dev/null +++ b/vlm/train/Sy4lojC9tm/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:56db10066f1956d7e9df80dd15fb3124d90873aee1c3f38cdebccfcc68588dae +size 601618 diff --git a/vlm/train/Sy4lojC9tm/9.png b/vlm/train/Sy4lojC9tm/9.png new file mode 100644 index 0000000000000000000000000000000000000000..c3536a5832ea293b59b9928370611c6c16bfd14d --- /dev/null +++ b/vlm/train/Sy4lojC9tm/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:825fcb9c0bd23794651bc97fc949687f8da6bb1e8bed000b39c3881cef110d62 +size 480841 diff --git a/vlm/train/SyGjjsC5tQ/0.png b/vlm/train/SyGjjsC5tQ/0.png new file mode 100644 index 0000000000000000000000000000000000000000..60de7c839cb62ab903fb3bfdf9c54e695dd53e75 --- /dev/null +++ b/vlm/train/SyGjjsC5tQ/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aae384fe599b1a4b384fdeea93fcc7862c74886932b092e760bc3a543dfd48ef +size 529218 diff --git a/vlm/train/SyGjjsC5tQ/1.png b/vlm/train/SyGjjsC5tQ/1.png new file mode 100644 index 0000000000000000000000000000000000000000..f86a6c9ad711f80ab088a7914dd5f9ade4fd4b4f --- /dev/null +++ b/vlm/train/SyGjjsC5tQ/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:41ab6aa4a49662bbca7f6b68e01b754ea8d6d86af0557dfa7fa47029801da11f +size 588622 diff --git a/vlm/train/SyGjjsC5tQ/10.png b/vlm/train/SyGjjsC5tQ/10.png new file mode 100644 index 0000000000000000000000000000000000000000..1e3ef3053b00f3cd6494fefe4fb8763338950806 --- /dev/null +++ b/vlm/train/SyGjjsC5tQ/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cb8931144672370eafa6c31eee10d5527b18bd08fceaff66e17decf6100b96c2 +size 482728 diff --git a/vlm/train/SyGjjsC5tQ/11.png b/vlm/train/SyGjjsC5tQ/11.png new file mode 100644 index 0000000000000000000000000000000000000000..bc516c03d9033834b716f6d4c20297aef5dab2bc --- /dev/null +++ b/vlm/train/SyGjjsC5tQ/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:324ba0d55341b89fefd0f5fbdb2834b322467b89f695efc43c286f7c5d8aa5e4 +size 263820 diff --git a/vlm/train/SyGjjsC5tQ/12.png b/vlm/train/SyGjjsC5tQ/12.png new file mode 100644 index 0000000000000000000000000000000000000000..566cd98004736139616a2681ee0de2a9039f2fbf --- /dev/null +++ b/vlm/train/SyGjjsC5tQ/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:22a32b4b63c5aaaf42c6490a5c248e2ea066df3cbb2dcb3c3b70608b933eaf55 +size 387724 diff --git a/vlm/train/SyGjjsC5tQ/13.png b/vlm/train/SyGjjsC5tQ/13.png new file mode 100644 index 0000000000000000000000000000000000000000..b60f12bb675a3654ec8c737698485bf8cc54b256 --- /dev/null +++ b/vlm/train/SyGjjsC5tQ/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:841b41b00900191b0da88629528293d7c622e954e24da2533c4959d26e670e96 +size 373540 diff --git a/vlm/train/SyGjjsC5tQ/14.png b/vlm/train/SyGjjsC5tQ/14.png new file mode 100644 index 0000000000000000000000000000000000000000..c489008a89d5dd0343afd5e9f9ccb5770fbe30f2 --- /dev/null +++ b/vlm/train/SyGjjsC5tQ/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1e717baf1190eb0d1bf88ff2817e9882c556bc2aad37ce124cc09e7aded610f7 +size 414866 diff --git a/vlm/train/SyGjjsC5tQ/15.png b/vlm/train/SyGjjsC5tQ/15.png new file mode 100644 index 0000000000000000000000000000000000000000..5bd9ab50664b0f9206081563e21eb801e3c49a81 --- /dev/null +++ b/vlm/train/SyGjjsC5tQ/15.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:928e5f50d6511d5c7709ae6d7b62f4689f9082e670ecbca521f528c58738c62b +size 345432 diff --git a/vlm/train/SyGjjsC5tQ/16.png b/vlm/train/SyGjjsC5tQ/16.png new file mode 100644 index 0000000000000000000000000000000000000000..f7b055eddfef43610f79c96ece09920c3ce97173 --- /dev/null +++ b/vlm/train/SyGjjsC5tQ/16.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:70d0c5e98a9483149ba962dc53d2efec4e2b9f4116d2ef5ae08d79072aef31f2 +size 306089 diff --git a/vlm/train/SyGjjsC5tQ/17.png b/vlm/train/SyGjjsC5tQ/17.png new file mode 100644 index 0000000000000000000000000000000000000000..8b923523662839ced6fb66273b1a5140ea9e477d --- /dev/null +++ b/vlm/train/SyGjjsC5tQ/17.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:453bbb1e60a5c5a1da5e9786fe9d405eb7d14cd6c44d6708e87482f344636e33 +size 466347 diff --git a/vlm/train/SyGjjsC5tQ/18.png b/vlm/train/SyGjjsC5tQ/18.png new file mode 100644 index 0000000000000000000000000000000000000000..02824d17581890579a8a574df6004f721650b0f2 --- /dev/null +++ b/vlm/train/SyGjjsC5tQ/18.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:58cc2639994aa2a30745a330ccd851521ab12c3178b772aa342f9418b9eda0a6 +size 285369 diff --git a/vlm/train/SyGjjsC5tQ/19.png b/vlm/train/SyGjjsC5tQ/19.png new file mode 100644 index 0000000000000000000000000000000000000000..5c95f1ff3333c5f0edf4b3b2084d65abd154d68d --- /dev/null +++ b/vlm/train/SyGjjsC5tQ/19.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:df93442e11ede020313e5d6aa4de40d309e609d9b1b6095ab89c73c838f4f500 +size 511166 diff --git a/vlm/train/SyGjjsC5tQ/2.png b/vlm/train/SyGjjsC5tQ/2.png new file mode 100644 index 0000000000000000000000000000000000000000..0e7459cd1cd79f4fcd4c84ed0079534c44a54d9a --- /dev/null +++ b/vlm/train/SyGjjsC5tQ/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3e3dc0ce0cf4d400e5a96512c2ab839edcdbf05d1ef764e5b478ed40f6ee8484 +size 538355 diff --git a/vlm/train/SyGjjsC5tQ/3.png b/vlm/train/SyGjjsC5tQ/3.png new file mode 100644 index 0000000000000000000000000000000000000000..60a0e1eeca809614cef336ed90ed9add0b566ad5 --- /dev/null +++ b/vlm/train/SyGjjsC5tQ/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:39ac6b2dfdaa15d898fa9b63dc3536233e8a681f623075c69249c77cb2b7f39d +size 551093 diff --git a/vlm/train/SyGjjsC5tQ/4.png b/vlm/train/SyGjjsC5tQ/4.png new file mode 100644 index 0000000000000000000000000000000000000000..cf0ab0a35dfb7c94b13e78a3c0c730cc2e2f7a61 --- /dev/null +++ b/vlm/train/SyGjjsC5tQ/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:55fc7ebadda0da21d964f77bbff0b9f81da5a2f3723d1b76ed28654b0e9c98c2 +size 499961 diff --git a/vlm/train/SyGjjsC5tQ/5.png b/vlm/train/SyGjjsC5tQ/5.png new file mode 100644 index 0000000000000000000000000000000000000000..1629c6c8c8b993d07ed1eb0cde90b4ccc57177d0 --- /dev/null +++ b/vlm/train/SyGjjsC5tQ/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9f8754eea1a4e9ce3bf35929c53337921a8e944fba908c6ffae92f41c5fb9342 +size 555949 diff --git a/vlm/train/SyGjjsC5tQ/6.png b/vlm/train/SyGjjsC5tQ/6.png new file mode 100644 index 0000000000000000000000000000000000000000..e70942c806a6ec50d8617b16b4f051fd2af66d24 --- /dev/null +++ b/vlm/train/SyGjjsC5tQ/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d0eb8bd88304b605433ec468034e8949c943bb8a769eb786b60fbf0374757107 +size 489936 diff --git a/vlm/train/SyGjjsC5tQ/7.png b/vlm/train/SyGjjsC5tQ/7.png new file mode 100644 index 0000000000000000000000000000000000000000..f2186a5e39a81259dc5a5b1b201170f09b4c586c --- /dev/null +++ b/vlm/train/SyGjjsC5tQ/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5f5a04d81b39b594abb0888ce1c3157f66513e6fa60171351c2efc7bd404ab72 +size 540175 diff --git a/vlm/train/SyGjjsC5tQ/8.png b/vlm/train/SyGjjsC5tQ/8.png new file mode 100644 index 0000000000000000000000000000000000000000..c93fa1d2b0fe075ba37fe70cbe3511b9de0802c5 --- /dev/null +++ b/vlm/train/SyGjjsC5tQ/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d1292ff4df993351b2730eb008ca93868ec3e09de92740b027d295ffcc7c11f4 +size 496763 diff --git a/vlm/train/SyGjjsC5tQ/9.png b/vlm/train/SyGjjsC5tQ/9.png new file mode 100644 index 0000000000000000000000000000000000000000..e9b5b4f9b660d584b301ad03336360eb0bb8b2eb --- /dev/null +++ b/vlm/train/SyGjjsC5tQ/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:43df9162856ea62e88c1f0c9e5cc45f0b74d34e6afd96a94bf69381c2d7d52e6 +size 379153 diff --git a/vlm/train/SyMDXnCcF7/0.png b/vlm/train/SyMDXnCcF7/0.png new file mode 100644 index 0000000000000000000000000000000000000000..9d376124cd0dda7cfb1c9db773e5996edffc3ebf --- /dev/null +++ b/vlm/train/SyMDXnCcF7/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ab9777831f6dc3a452ff4b2bba25d4b03705be8c990c432ab57c678ddc3fac05 +size 517596 diff --git a/vlm/train/SyMDXnCcF7/1.png b/vlm/train/SyMDXnCcF7/1.png new file mode 100644 index 0000000000000000000000000000000000000000..c1d6a1e185928ec9e52e4bcda9e05bf9c153ab66 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0d752d77076dd282b0580176d08de9579e9efa72a730fd26c311253077e231c7 +size 612718 diff --git a/vlm/train/SyMDXnCcF7/10.png b/vlm/train/SyMDXnCcF7/10.png new file mode 100644 index 0000000000000000000000000000000000000000..c8cf9e794d3b7aeef0ae2514670b7e0e8ffa4b80 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:be965adedb82f6685128ccb28181ea17f53a06d011054bbc901581e6af9a09ef +size 542492 diff --git a/vlm/train/SyMDXnCcF7/11.png b/vlm/train/SyMDXnCcF7/11.png new file mode 100644 index 0000000000000000000000000000000000000000..3cc71425a488d92112667ab9412304f37ce5359a --- /dev/null +++ b/vlm/train/SyMDXnCcF7/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dca10e3fdc56b11592991953c55059009eb718309c7b0535a085fcc289ea050c +size 549276 diff --git a/vlm/train/SyMDXnCcF7/12.png b/vlm/train/SyMDXnCcF7/12.png new file mode 100644 index 0000000000000000000000000000000000000000..fe23bb495acdb978ee5a85be866414a59ef3907d --- /dev/null +++ b/vlm/train/SyMDXnCcF7/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:30c6237f0840f92c7e165e8e1a7d3d8ed1d70700e5b21f91ba215fa4eff8db44 +size 556847 diff --git a/vlm/train/SyMDXnCcF7/13.png b/vlm/train/SyMDXnCcF7/13.png new file mode 100644 index 0000000000000000000000000000000000000000..211e9cfce31f43babbc0532c7fb527ad893c6297 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:adaf54858d63a9e803edc563f2c027d5b7eff997925aaf85ef4b7cd9e63c1a5b +size 545690 diff --git a/vlm/train/SyMDXnCcF7/14.png b/vlm/train/SyMDXnCcF7/14.png new file mode 100644 index 0000000000000000000000000000000000000000..24cfc5e1b000185716988cd62ce1cadad55661f2 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6de06bd1378af7365be52b3da95850aa1f58dd5cd5bda36bf923a7bb9df65e28 +size 518009 diff --git a/vlm/train/SyMDXnCcF7/15.png b/vlm/train/SyMDXnCcF7/15.png new file mode 100644 index 0000000000000000000000000000000000000000..b52e0812a44f4a1626bf969c3486bd20b5484edc --- /dev/null +++ b/vlm/train/SyMDXnCcF7/15.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fa91fd379d1e2c1e3b8d63166fd9259b148ef850c7488dbc82bfe2e4596f2c99 +size 407491 diff --git a/vlm/train/SyMDXnCcF7/16.png b/vlm/train/SyMDXnCcF7/16.png new file mode 100644 index 0000000000000000000000000000000000000000..1efe95981792a741c15b3dac92dfd5b425a669ae --- /dev/null +++ b/vlm/train/SyMDXnCcF7/16.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d2e431a0373b5b5e18634b15a97441b47c19a749be7bc485d9850fd086a488ce +size 440569 diff --git a/vlm/train/SyMDXnCcF7/17.png b/vlm/train/SyMDXnCcF7/17.png new file mode 100644 index 0000000000000000000000000000000000000000..bfb759d1f293e6afcdbffe5172251342dc8ff137 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/17.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:90278d9f016baa6c6c3a0f2cf5794af8dc289c9db6698b72aeb51c706319660a +size 418694 diff --git a/vlm/train/SyMDXnCcF7/18.png b/vlm/train/SyMDXnCcF7/18.png new file mode 100644 index 0000000000000000000000000000000000000000..7435d66edd018a63489bc4502cfe6dba946a8491 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/18.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e1dc7c4691b83ed912a375dac50b978e4d28c48b6d326b85765c59795d6baeb5 +size 471530 diff --git a/vlm/train/SyMDXnCcF7/19.png b/vlm/train/SyMDXnCcF7/19.png new file mode 100644 index 0000000000000000000000000000000000000000..25b67b0b85c17e85226cebbca83d6fc1cfab87c7 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/19.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:acadf5c4eeabed5c3ce4bdb7df931261f81f31ae611c7efa89e4c3b474feed9d +size 517068 diff --git a/vlm/train/SyMDXnCcF7/2.png b/vlm/train/SyMDXnCcF7/2.png new file mode 100644 index 0000000000000000000000000000000000000000..425b80c3276f8f2dc6c274326ae69d208dcae15a --- /dev/null +++ b/vlm/train/SyMDXnCcF7/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:59f256fe9023e4c75ac3a7cb91c2f4ce8dc5151665097ae156a2e8488893616b +size 572080 diff --git a/vlm/train/SyMDXnCcF7/20.png b/vlm/train/SyMDXnCcF7/20.png new file mode 100644 index 0000000000000000000000000000000000000000..effb7a744e4481ce09460b17b566e0fa8b1f049c --- /dev/null +++ b/vlm/train/SyMDXnCcF7/20.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:98ec7c86dbe8669e003cf181e71f7356a405ab57ba7aa738ab8493e47cba1461 +size 389095 diff --git a/vlm/train/SyMDXnCcF7/21.png b/vlm/train/SyMDXnCcF7/21.png new file mode 100644 index 0000000000000000000000000000000000000000..300b050bfab90bdc46d5cfde2185c1c5cff6e02c --- /dev/null +++ b/vlm/train/SyMDXnCcF7/21.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cab740071dc550f5fb33abd22c0f1f87ad0ff98d239f6f521170cce553a9093a +size 387028 diff --git a/vlm/train/SyMDXnCcF7/22.png b/vlm/train/SyMDXnCcF7/22.png new file mode 100644 index 0000000000000000000000000000000000000000..31be29220e45df43a2840311286b639103a14cb7 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/22.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:48103a5287778f51427a2306d2bcd788288967187a4db0bf272e0966bd97496b +size 408449 diff --git a/vlm/train/SyMDXnCcF7/23.png b/vlm/train/SyMDXnCcF7/23.png new file mode 100644 index 0000000000000000000000000000000000000000..b65acecfc06b9cad4ea00feab4dbe5cf3179bfb9 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/23.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:516877437231ece0f981ed5c3b3acdf938b0f49b317ea50e559aeebe2cba7dac +size 337588 diff --git a/vlm/train/SyMDXnCcF7/24.png b/vlm/train/SyMDXnCcF7/24.png new file mode 100644 index 0000000000000000000000000000000000000000..d1b8f67a1c815c2455e04025b6b549514c32f3be --- /dev/null +++ b/vlm/train/SyMDXnCcF7/24.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d2a19e865970c584e72391916f9240e2db1da9f6dc6bef14f1643eca05828447 +size 362374 diff --git a/vlm/train/SyMDXnCcF7/25.png b/vlm/train/SyMDXnCcF7/25.png new file mode 100644 index 0000000000000000000000000000000000000000..6d1304e9f07c4eb91cfc7f56fd554f10afda06a6 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/25.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:59851bd6e5016089790d5812a8aa6244db374bfe92b25840b78c8507d0b553cd +size 380479 diff --git a/vlm/train/SyMDXnCcF7/26.png b/vlm/train/SyMDXnCcF7/26.png new file mode 100644 index 0000000000000000000000000000000000000000..30ac6adaf9781f173bede1cf261d61bd264ef238 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/26.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d645e15d73752a0223f6f762cf671becd5a3dd7878ece8a93f09904ea4ee2f24 +size 279124 diff --git a/vlm/train/SyMDXnCcF7/27.png b/vlm/train/SyMDXnCcF7/27.png new file mode 100644 index 0000000000000000000000000000000000000000..118aec0e46450ad659811512083f219006ddd718 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/27.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2f7f05899da6e08e07e3b896b413ff8facdc32351f7fe66662005742e691b627 +size 405868 diff --git a/vlm/train/SyMDXnCcF7/28.png b/vlm/train/SyMDXnCcF7/28.png new file mode 100644 index 0000000000000000000000000000000000000000..05efe27ef4d5a3475fb24b058e4d5bc6cc374cbc --- /dev/null +++ b/vlm/train/SyMDXnCcF7/28.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f9fd1d42b5c4912c4a8a6b73a30951b3783114acc57fad7148a87c5644f2940c +size 350961 diff --git a/vlm/train/SyMDXnCcF7/29.png b/vlm/train/SyMDXnCcF7/29.png new file mode 100644 index 0000000000000000000000000000000000000000..94e433323e7511ec5674f9c4226ee96072d40384 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/29.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d0d40f37dd8eed56b6273074e2e4384ea12f23646ac7b769889f1244c12779d6 +size 306291 diff --git a/vlm/train/SyMDXnCcF7/3.png b/vlm/train/SyMDXnCcF7/3.png new file mode 100644 index 0000000000000000000000000000000000000000..43d659b7ffb6466f423a75579852c43b9f7dd39c --- /dev/null +++ b/vlm/train/SyMDXnCcF7/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bde36fa2f704ed996e3e55a3b2bb9b2f82545b045d5a1341d83eefead1de2d73 +size 555719 diff --git a/vlm/train/SyMDXnCcF7/30.png b/vlm/train/SyMDXnCcF7/30.png new file mode 100644 index 0000000000000000000000000000000000000000..386273a5223000f396e963ee260609cc7cdb1997 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/30.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2a42eac2e905d940770a9c0ca06c2ebd6a608306c1d41ba1e15b2d0c8af495f1 +size 332009 diff --git a/vlm/train/SyMDXnCcF7/31.png b/vlm/train/SyMDXnCcF7/31.png new file mode 100644 index 0000000000000000000000000000000000000000..32f77e1a431a562bd77b6ed25c89a4396905b098 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/31.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6ed27a251e50e0a220f0b33d7a9e1111d56886bd91dd7c8881bd6e3d8b992efb +size 301633 diff --git a/vlm/train/SyMDXnCcF7/32.png b/vlm/train/SyMDXnCcF7/32.png new file mode 100644 index 0000000000000000000000000000000000000000..7d0250d1de576377d70582a23006ede45e720092 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/32.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a87bc9965de1064747db7b0d15206610223cba999a8f92bb0a79d403b09ce214 +size 231258 diff --git a/vlm/train/SyMDXnCcF7/33.png b/vlm/train/SyMDXnCcF7/33.png new file mode 100644 index 0000000000000000000000000000000000000000..dfbb4fdeae12ca76733c4bfaf9cb698f81897fdd --- /dev/null +++ b/vlm/train/SyMDXnCcF7/33.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:597f143f39b026bacf7afed0e75ab3dfe66c07f374e30c832d4301b9368426ba +size 238334 diff --git a/vlm/train/SyMDXnCcF7/34.png b/vlm/train/SyMDXnCcF7/34.png new file mode 100644 index 0000000000000000000000000000000000000000..07eb41ae795ba718aa640d24a05bb0e5f3c74f50 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/34.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:28493cb4e77cf2d5e28548a522738d1c129eabb160e34f9e77914f4e5004f06d +size 271222 diff --git a/vlm/train/SyMDXnCcF7/35.png b/vlm/train/SyMDXnCcF7/35.png new file mode 100644 index 0000000000000000000000000000000000000000..85851fdade23cc12c1adfa3a811704bb16cb270a --- /dev/null +++ b/vlm/train/SyMDXnCcF7/35.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c0ee16572cf442d389c56fe67f0540a6d7e116d324b08e9ea2ac5fcf5377d140 +size 358634 diff --git a/vlm/train/SyMDXnCcF7/36.png b/vlm/train/SyMDXnCcF7/36.png new file mode 100644 index 0000000000000000000000000000000000000000..e69747c4305c7f229332f4b2deecb57c60057b2b --- /dev/null +++ b/vlm/train/SyMDXnCcF7/36.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dd69aa72e5b30f9ed66933ad061af9e051564e3456a72149387aaace8d91493c +size 393112 diff --git a/vlm/train/SyMDXnCcF7/37.png b/vlm/train/SyMDXnCcF7/37.png new file mode 100644 index 0000000000000000000000000000000000000000..f5c0e422e727c1de7add8fc3f9f3c5f960d53ac1 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/37.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:86406cbe17946593d0cec2eb529adc73a92d36ed1138a9372956c38efacd033e +size 293659 diff --git a/vlm/train/SyMDXnCcF7/38.png b/vlm/train/SyMDXnCcF7/38.png new file mode 100644 index 0000000000000000000000000000000000000000..c0917ad89820edaa91c1e6fa76c6e24391a3e86a --- /dev/null +++ b/vlm/train/SyMDXnCcF7/38.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8f90582f7f57c8db6055e44b3f2a457448199d157d39acc2b58f7328c4eddd71 +size 339002 diff --git a/vlm/train/SyMDXnCcF7/39.png b/vlm/train/SyMDXnCcF7/39.png new file mode 100644 index 0000000000000000000000000000000000000000..d00222640082ba4603296c11ddfab0db6cc15142 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/39.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d87d630046a2111b4d1a990a40698d01e9db18a6d5811524d99c3e46f144cb59 +size 460914 diff --git a/vlm/train/SyMDXnCcF7/4.png b/vlm/train/SyMDXnCcF7/4.png new file mode 100644 index 0000000000000000000000000000000000000000..bcb5644bb9225df5a1044d16febab28771fde547 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:63400df0fe8eaa1c9605fa73ef0100dae29e0e1ad869bfc9227810252f516e0f +size 522168 diff --git a/vlm/train/SyMDXnCcF7/40.png b/vlm/train/SyMDXnCcF7/40.png new file mode 100644 index 0000000000000000000000000000000000000000..2df97ee03a39821657dcf1ff44ac63c5cf99c52a --- /dev/null +++ b/vlm/train/SyMDXnCcF7/40.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:272bfccf3cd2cdfba45647af85a5de4cb6e3100e5725d64e11a69b9538cb219a +size 275921 diff --git a/vlm/train/SyMDXnCcF7/41.png b/vlm/train/SyMDXnCcF7/41.png new file mode 100644 index 0000000000000000000000000000000000000000..8b5d6f5ea9de64629c687f3f4a0db919512ef174 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/41.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f89764535bb765ba51753ec081adf172e2cbab0ebf5d99560fb58b52d3bc0bd0 +size 329227 diff --git a/vlm/train/SyMDXnCcF7/42.png b/vlm/train/SyMDXnCcF7/42.png new file mode 100644 index 0000000000000000000000000000000000000000..e0cafa0b6c0a16c1c45b02f71dbed1c36cde0031 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/42.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4461eec110bb510c22ceea6dfbe0c2f616822cfa0866904e5cf6af9b2af4526e +size 258123 diff --git a/vlm/train/SyMDXnCcF7/43.png b/vlm/train/SyMDXnCcF7/43.png new file mode 100644 index 0000000000000000000000000000000000000000..4e6817425a9836821f1f9442f58174ee22966ea8 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/43.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3815b4aa9d454e5687fc479f943d6a168142e7f5f8d706c0b8ab941c4f6051e3 +size 322519 diff --git a/vlm/train/SyMDXnCcF7/44.png b/vlm/train/SyMDXnCcF7/44.png new file mode 100644 index 0000000000000000000000000000000000000000..409aa047dd2f152d8b14875420739c379c69d8e0 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/44.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e525466ef57725eaae6dbd992a7a199e6c0c07d83bc6bbe5f1b0738c759f6455 +size 303515 diff --git a/vlm/train/SyMDXnCcF7/45.png b/vlm/train/SyMDXnCcF7/45.png new file mode 100644 index 0000000000000000000000000000000000000000..a34cef4d387f11b6ff9264cac09c32568f803a95 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/45.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a6c6a705e0412076be114ab2d2a4e15373d503a30c9fdd087f95a2cdcb8e8101 +size 378780 diff --git a/vlm/train/SyMDXnCcF7/46.png b/vlm/train/SyMDXnCcF7/46.png new file mode 100644 index 0000000000000000000000000000000000000000..61a775624e3c3112e1e27cb7b05b491f930a679a --- /dev/null +++ b/vlm/train/SyMDXnCcF7/46.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:26a05d32cb7ce40cc57968a0ae5c44f8c0b4df1a9ab4ee1c94f4546b93ba8d42 +size 374482 diff --git a/vlm/train/SyMDXnCcF7/47.png b/vlm/train/SyMDXnCcF7/47.png new file mode 100644 index 0000000000000000000000000000000000000000..423fbf3892b64853421faa6ede531d049b4254fc --- /dev/null +++ b/vlm/train/SyMDXnCcF7/47.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aa34edea3a989475a8d5c5b76d17a0278540eb3910e168b793d29431f86b68e2 +size 339278 diff --git a/vlm/train/SyMDXnCcF7/48.png b/vlm/train/SyMDXnCcF7/48.png new file mode 100644 index 0000000000000000000000000000000000000000..6ce29d490d8f918acb5720edfc47b841ab685107 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/48.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cdd2224e0423d86e2727ee052826fb8fb5be95d7ea35f29acdb5a2edaa1b3794 +size 292612 diff --git a/vlm/train/SyMDXnCcF7/49.png b/vlm/train/SyMDXnCcF7/49.png new file mode 100644 index 0000000000000000000000000000000000000000..6ceb9e110e633d9e90399e4b9642bd6303bb324b --- /dev/null +++ b/vlm/train/SyMDXnCcF7/49.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3642d8ed5e1e3ec35af5bbe9a5fd975186c408c7c9485b6283e8acbdc9772982 +size 497855 diff --git a/vlm/train/SyMDXnCcF7/5.png b/vlm/train/SyMDXnCcF7/5.png new file mode 100644 index 0000000000000000000000000000000000000000..40519fa318110d736877f1bbde6343df5f51e0f6 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:394a1c894b95ee63fb9bb5544782649f583b1b5744750da211c7b169a36438c8 +size 523913 diff --git a/vlm/train/SyMDXnCcF7/50.png b/vlm/train/SyMDXnCcF7/50.png new file mode 100644 index 0000000000000000000000000000000000000000..60462e06b764e1353b299bb5c7eee036c2f20a19 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/50.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:251d487ec7132e5896c84a821ed0cc6db0fba107932f46b495c63ce03a683c2c +size 392486 diff --git a/vlm/train/SyMDXnCcF7/51.png b/vlm/train/SyMDXnCcF7/51.png new file mode 100644 index 0000000000000000000000000000000000000000..aa7263611665ebd5efffa01401b72fd9c6eb579c --- /dev/null +++ b/vlm/train/SyMDXnCcF7/51.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:02bd03d30e94407e5eb8ba93c5de91989f4e03c5f80cc8259242c9002884ef9b +size 362852 diff --git a/vlm/train/SyMDXnCcF7/52.png b/vlm/train/SyMDXnCcF7/52.png new file mode 100644 index 0000000000000000000000000000000000000000..c60873e51c2aff3591db744c59c2ed4793578f0a --- /dev/null +++ b/vlm/train/SyMDXnCcF7/52.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0ba8beb7168e8712df470483974d485747314e13f10799ad12a5e84fb88d21be +size 355518 diff --git a/vlm/train/SyMDXnCcF7/53.png b/vlm/train/SyMDXnCcF7/53.png new file mode 100644 index 0000000000000000000000000000000000000000..f3dbd7cca7ff45c8c7cb19d5911ddc9c6f830a2b --- /dev/null +++ b/vlm/train/SyMDXnCcF7/53.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a05aec8ecada6691339e718c7737d69a506b12d9764a18bba1b17fce8a346399 +size 359993 diff --git a/vlm/train/SyMDXnCcF7/54.png b/vlm/train/SyMDXnCcF7/54.png new file mode 100644 index 0000000000000000000000000000000000000000..acce1b280f396ab1ef4a1fe8b1c39a450189e3d0 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/54.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:339f37c4625b0e18e240f30c7cd50d10e5ad057bc30e7ce61d1f807cfb8fbeca +size 237797 diff --git a/vlm/train/SyMDXnCcF7/55.png b/vlm/train/SyMDXnCcF7/55.png new file mode 100644 index 0000000000000000000000000000000000000000..3f96f7afd2a4a3e272334e2401340166c4af11d9 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/55.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aa684d9014842d32dcc1f7bc8e236f0204264266bad2ebad40af15da66441a96 +size 361498 diff --git a/vlm/train/SyMDXnCcF7/56.png b/vlm/train/SyMDXnCcF7/56.png new file mode 100644 index 0000000000000000000000000000000000000000..773a6e7c21256c78d737846e2d3fbdb526909634 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/56.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e6488d2393d0d61911612fa98586d408256db12c59b253fef5f54a95c43fa1de +size 369484 diff --git a/vlm/train/SyMDXnCcF7/57.png b/vlm/train/SyMDXnCcF7/57.png new file mode 100644 index 0000000000000000000000000000000000000000..8305ab08b230651328959b18a0a6a6474aad4078 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/57.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6fdda42d6820109f6355aabdb28984f57c3250132d25160c5186635a3fcea000 +size 261893 diff --git a/vlm/train/SyMDXnCcF7/58.png b/vlm/train/SyMDXnCcF7/58.png new file mode 100644 index 0000000000000000000000000000000000000000..9e94f039be0d2762ad07befb2ce7406e3ca38bd4 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/58.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d0a2a1b01bcbdb79a1e4e2f24ced5b82a0adbd88becb73d1b56f33dcd91df908 +size 411624 diff --git a/vlm/train/SyMDXnCcF7/59.png b/vlm/train/SyMDXnCcF7/59.png new file mode 100644 index 0000000000000000000000000000000000000000..e8ba23e50801e4e001b259ccc3781c7e2d0387a8 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/59.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e918681ee715b2940c9fafd5cab93453b84c09b45ece0715aa10cb94cc36d436 +size 310100 diff --git a/vlm/train/SyMDXnCcF7/6.png b/vlm/train/SyMDXnCcF7/6.png new file mode 100644 index 0000000000000000000000000000000000000000..d81e11df0a21c143dc1f5bc0e43ad37b52b160b6 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e3334d038f5d1920379992455291ea342b20a8637970b5ad8cbeae318cd47a16 +size 487448 diff --git a/vlm/train/SyMDXnCcF7/60.png b/vlm/train/SyMDXnCcF7/60.png new file mode 100644 index 0000000000000000000000000000000000000000..96e877bd30ec0b7fd79ca025aad8da17ee1de83a --- /dev/null +++ b/vlm/train/SyMDXnCcF7/60.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bf4646a313714b7359d8b32d763e49c56ed8afad2295742d2641635f0dfceca9 +size 278368 diff --git a/vlm/train/SyMDXnCcF7/61.png b/vlm/train/SyMDXnCcF7/61.png new file mode 100644 index 0000000000000000000000000000000000000000..cf91a52c15accf7b8b7c55e81b579589565ca265 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/61.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:21a1d73566ab9a9853bc9b8619fa79a54465a79ee43a39f55b14196526b2c4cb +size 247944 diff --git a/vlm/train/SyMDXnCcF7/62.png b/vlm/train/SyMDXnCcF7/62.png new file mode 100644 index 0000000000000000000000000000000000000000..2aac0ccdb3f34d55ea965837c1a1e81f96c9feff --- /dev/null +++ b/vlm/train/SyMDXnCcF7/62.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3143c8e0e8666766fc04de4b6680d74dd5e8aa83a434c57bf4bdd92cd5552a2a +size 325353 diff --git a/vlm/train/SyMDXnCcF7/63.png b/vlm/train/SyMDXnCcF7/63.png new file mode 100644 index 0000000000000000000000000000000000000000..98d682f39f5caa10d41511cb2e1a102822f0c512 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/63.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7c6ac71cb794335729a0d1842d51239a66db8fbfcc6e9c1af5be1278e5bc0942 +size 327371 diff --git a/vlm/train/SyMDXnCcF7/64.png b/vlm/train/SyMDXnCcF7/64.png new file mode 100644 index 0000000000000000000000000000000000000000..78f04830d74bab6ea403fb73d889044506d8de04 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/64.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b74a66bbc30974276fe7f67e67201dc75459f3c5487f3ff979b7bfcf603d2981 +size 315897 diff --git a/vlm/train/SyMDXnCcF7/65.png b/vlm/train/SyMDXnCcF7/65.png new file mode 100644 index 0000000000000000000000000000000000000000..08b1e9df1192413919a605a535f93234e5354679 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/65.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eea231bb90bf2ca3b55de0d88e05729f176983f469d847c4c90dde770ced1ac0 +size 286642 diff --git a/vlm/train/SyMDXnCcF7/66.png b/vlm/train/SyMDXnCcF7/66.png new file mode 100644 index 0000000000000000000000000000000000000000..fbc556b0a3e75c1d6dc902d11d464dfe635afb30 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/66.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5800d726570d9bb7f6e1448edf2b2f7630367ae5f069c0e4d974ee28a5cdeffe +size 264515 diff --git a/vlm/train/SyMDXnCcF7/67.png b/vlm/train/SyMDXnCcF7/67.png new file mode 100644 index 0000000000000000000000000000000000000000..e7ebad452a0dc724f38a88669b6efbe554f6fa6f --- /dev/null +++ b/vlm/train/SyMDXnCcF7/67.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b2ec2bd297738b2bfd4aaa49d0d377e49d3ce4f2551c3c8ff473d4b466b21216 +size 387917 diff --git a/vlm/train/SyMDXnCcF7/68.png b/vlm/train/SyMDXnCcF7/68.png new file mode 100644 index 0000000000000000000000000000000000000000..09464d71be7632425fffb06742d0f91a29a15f84 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/68.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9690576e73bba337f142c211f301444c5e527d921e040f58072dbcb66fc45612 +size 298173 diff --git a/vlm/train/SyMDXnCcF7/69.png b/vlm/train/SyMDXnCcF7/69.png new file mode 100644 index 0000000000000000000000000000000000000000..385141301132ef2f649151d372177bb8cd5ee848 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/69.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:30e1e519afc25e4f7bee72b89ae84fd91aadf129d68bc68863869a3d94bebee1 +size 257202 diff --git a/vlm/train/SyMDXnCcF7/7.png b/vlm/train/SyMDXnCcF7/7.png new file mode 100644 index 0000000000000000000000000000000000000000..1dfa3a29857779aff613c62f17a725b364828a6a --- /dev/null +++ b/vlm/train/SyMDXnCcF7/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:759e883bfdc1f3ce211add8a249aba9d7e9a16208c2752902c9e56c433abdc2f +size 520509 diff --git a/vlm/train/SyMDXnCcF7/70.png b/vlm/train/SyMDXnCcF7/70.png new file mode 100644 index 0000000000000000000000000000000000000000..33cd1a0e70f643c7bcc9f8f5ee200b6cd1f3ebbd --- /dev/null +++ b/vlm/train/SyMDXnCcF7/70.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ea9777be9c7cb690b01dcabc4ff753debd6d91221ad55e6bfd69daa00626e689 +size 223858 diff --git a/vlm/train/SyMDXnCcF7/71.png b/vlm/train/SyMDXnCcF7/71.png new file mode 100644 index 0000000000000000000000000000000000000000..90f4030634fc572a995562548b69b57e93b1decb --- /dev/null +++ b/vlm/train/SyMDXnCcF7/71.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9252fc2fb2369329f079d25ddfaebec4dc014b85710574a19266dd18cc749c83 +size 305978 diff --git a/vlm/train/SyMDXnCcF7/72.png b/vlm/train/SyMDXnCcF7/72.png new file mode 100644 index 0000000000000000000000000000000000000000..b40bb75f06e23b81303526af6fce5878c641a342 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/72.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e9545d88f53956e4aeac11e2a7cc70fc49bf4b201f7d56ae21843903791fc6d7 +size 373285 diff --git a/vlm/train/SyMDXnCcF7/73.png b/vlm/train/SyMDXnCcF7/73.png new file mode 100644 index 0000000000000000000000000000000000000000..3c2ef62e56ac7b9e6194d9a3842de47e4e10154d --- /dev/null +++ b/vlm/train/SyMDXnCcF7/73.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:84d03791fa60998d5914912ff0d3f2d9d2899d94866d709158025e3b19c53296 +size 353653 diff --git a/vlm/train/SyMDXnCcF7/74.png b/vlm/train/SyMDXnCcF7/74.png new file mode 100644 index 0000000000000000000000000000000000000000..6968b5169a33974a01ecb8279ab6948cacdd847d --- /dev/null +++ b/vlm/train/SyMDXnCcF7/74.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9994e844c7188082cae783729c0fa67f86109e99140f89a51430aa287b012721 +size 230073 diff --git a/vlm/train/SyMDXnCcF7/75.png b/vlm/train/SyMDXnCcF7/75.png new file mode 100644 index 0000000000000000000000000000000000000000..90e0e886fdbceae7561ebf65e6a64ce6621b1ebe --- /dev/null +++ b/vlm/train/SyMDXnCcF7/75.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1b19d36d588bcff78a1f254ccd779bbc744c649d041991d3e223efbb80358e77 +size 281027 diff --git a/vlm/train/SyMDXnCcF7/76.png b/vlm/train/SyMDXnCcF7/76.png new file mode 100644 index 0000000000000000000000000000000000000000..b88c15ba666aca5d1feea50fcbb189e40a97e0d3 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/76.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:945ad409bcd897084640ca5a042d6afaf70ed90771969562e3ce290dd39d09de +size 271418 diff --git a/vlm/train/SyMDXnCcF7/77.png b/vlm/train/SyMDXnCcF7/77.png new file mode 100644 index 0000000000000000000000000000000000000000..ab1892eaacb8e830b152baa184cc6d5e73eb2dfe --- /dev/null +++ b/vlm/train/SyMDXnCcF7/77.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:80a87ddd4b905ce894ef28dd0b1b0d8666044bbdd099f4765ea2882c64983301 +size 373689 diff --git a/vlm/train/SyMDXnCcF7/78.png b/vlm/train/SyMDXnCcF7/78.png new file mode 100644 index 0000000000000000000000000000000000000000..e81e708400b5d9b8716956bfa42dbbea62076421 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/78.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bcb3ce1723eae9d71447ad8b4620a3adc24e61342243d84cfb2a2dcb92406389 +size 358416 diff --git a/vlm/train/SyMDXnCcF7/79.png b/vlm/train/SyMDXnCcF7/79.png new file mode 100644 index 0000000000000000000000000000000000000000..0161d1f785ec13ce987cf92b5f31458da49b39b4 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/79.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:46de7d823ae7763c512c10bebb30c58b5f4d7e4f2248c7c901ff5ec9fa061855 +size 325400 diff --git a/vlm/train/SyMDXnCcF7/8.png b/vlm/train/SyMDXnCcF7/8.png new file mode 100644 index 0000000000000000000000000000000000000000..2f306ee08606a33aadf189f951c3c502ba95cb81 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cd1e04927f385cf8cb41dd5d440650dd057c8dd31d857b922000e4b751624ef6 +size 495193 diff --git a/vlm/train/SyMDXnCcF7/80.png b/vlm/train/SyMDXnCcF7/80.png new file mode 100644 index 0000000000000000000000000000000000000000..58681b6ef0a3332b1bca77b188b3d29f6419d8ab --- /dev/null +++ b/vlm/train/SyMDXnCcF7/80.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f7b2eda010829975703d0ff253132916ba5481c3a7f5e2ed61d19a803ab4715c +size 359008 diff --git a/vlm/train/SyMDXnCcF7/81.png b/vlm/train/SyMDXnCcF7/81.png new file mode 100644 index 0000000000000000000000000000000000000000..ebe2390893a290cd4aaf4729915e2e70b137b157 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/81.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:abc0ccaf649a55078aee913dc85842dcf4827812c70233c41484b85ccbc75bb0 +size 342242 diff --git a/vlm/train/SyMDXnCcF7/82.png b/vlm/train/SyMDXnCcF7/82.png new file mode 100644 index 0000000000000000000000000000000000000000..72859e5e103496050a9920a2838c057d6650c0f0 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/82.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c18dd20e6babbe27c77d59e25a0ff34003822cf53476895993e89b76ee8fc689 +size 395770 diff --git a/vlm/train/SyMDXnCcF7/83.png b/vlm/train/SyMDXnCcF7/83.png new file mode 100644 index 0000000000000000000000000000000000000000..829d6a423c3683410e28c0cb3742a552f8a6caa4 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/83.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b27aea38c3b5edaef9392709b4e3a262fbb33f1e84c08c55697c2f53908069d0 +size 286484 diff --git a/vlm/train/SyMDXnCcF7/84.png b/vlm/train/SyMDXnCcF7/84.png new file mode 100644 index 0000000000000000000000000000000000000000..7229b838deb4a9c5ec9c0f393bb8bfe34b21521e --- /dev/null +++ b/vlm/train/SyMDXnCcF7/84.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fbf44e2f2242bde744d9a2cfd0f7c486ad9ece7644023de8a7f3daa7c9582a2d +size 427616 diff --git a/vlm/train/SyMDXnCcF7/85.png b/vlm/train/SyMDXnCcF7/85.png new file mode 100644 index 0000000000000000000000000000000000000000..0528b1d5ab0887eb17be267334d5d4ef71b60022 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/85.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2e90c5a33492b280f1c247a4d0f1c506377a56c71092b58f5ffc4479edb924cc +size 326261 diff --git a/vlm/train/SyMDXnCcF7/86.png b/vlm/train/SyMDXnCcF7/86.png new file mode 100644 index 0000000000000000000000000000000000000000..0b4d8733cbf7c4635ed13c2ac003cebdf688d00c --- /dev/null +++ b/vlm/train/SyMDXnCcF7/86.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f2b1817b6b85f9a60ec918cb26fb57b883ed1eaac22793ba9069860cb53b2a81 +size 308707 diff --git a/vlm/train/SyMDXnCcF7/87.png b/vlm/train/SyMDXnCcF7/87.png new file mode 100644 index 0000000000000000000000000000000000000000..0f2d3187b01d1e4f1a71b1dc88017340b622f272 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/87.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b07fd9cee618d603edfaf092f83ea7802359fd5c510cbc1f72230ad306ec22d3 +size 403075 diff --git a/vlm/train/SyMDXnCcF7/88.png b/vlm/train/SyMDXnCcF7/88.png new file mode 100644 index 0000000000000000000000000000000000000000..1b8dc58a15435e12d5908f6722f1fb933e59a48b --- /dev/null +++ b/vlm/train/SyMDXnCcF7/88.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:35ddea1a8047afc379bd9330d4e28103405dc7629674478544a3eb1e7577c2ca +size 332380 diff --git a/vlm/train/SyMDXnCcF7/89.png b/vlm/train/SyMDXnCcF7/89.png new file mode 100644 index 0000000000000000000000000000000000000000..b53fcc36fd5f68f9b602d37db6943de99ceb5bcf --- /dev/null +++ b/vlm/train/SyMDXnCcF7/89.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d1cf5d5ac5a287dcfa9df3e322a72cb0c242578d22c79801d578c43e8d044157 +size 288566 diff --git a/vlm/train/SyMDXnCcF7/9.png b/vlm/train/SyMDXnCcF7/9.png new file mode 100644 index 0000000000000000000000000000000000000000..3393f036701c38fa4267d99a5b99b0ee270815f5 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ee1a0ab4afad73d867f95efa97e7ba280dfa050170eeed9cc8b6e92b7a07b85b +size 486816 diff --git a/vlm/train/SyMDXnCcF7/90.png b/vlm/train/SyMDXnCcF7/90.png new file mode 100644 index 0000000000000000000000000000000000000000..fa7088ef3617aedb81c10eb07ddb3d9f52cffddb --- /dev/null +++ b/vlm/train/SyMDXnCcF7/90.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5cdae911275a36bf97c89670f600832df61d282923646238f33480f11f7b0668 +size 443328 diff --git a/vlm/train/SyMDXnCcF7/91.png b/vlm/train/SyMDXnCcF7/91.png new file mode 100644 index 0000000000000000000000000000000000000000..6ea6da9ed05fabe7d59a547a60d69ca0393a8692 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/91.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5afdf133eaad261aab9101755cf1f2703021e9caefc34198d7babf20b801a149 +size 425798 diff --git a/vlm/train/SyMDXnCcF7/92.png b/vlm/train/SyMDXnCcF7/92.png new file mode 100644 index 0000000000000000000000000000000000000000..0db77f76912072852acfc5658ed0799ff08d6e77 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/92.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ba1e4d42ce719b0fc9ae44740045584ce6e158f639b72f1c735cfa5c0fbe2b9c +size 380686 diff --git a/vlm/train/SyMDXnCcF7/93.png b/vlm/train/SyMDXnCcF7/93.png new file mode 100644 index 0000000000000000000000000000000000000000..323e73b8a04e00732cfa3068867a66b76bf85cd4 --- /dev/null +++ b/vlm/train/SyMDXnCcF7/93.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9b22960d0013fe1597d280cbbd39432cfabb8b5d486cc033feb6928283e1211d +size 357721 diff --git a/vlm/train/SyMDXnCcF7/94.png b/vlm/train/SyMDXnCcF7/94.png new file mode 100644 index 0000000000000000000000000000000000000000..e39d1fc2623baa9132b06f47663d807588aa72de --- /dev/null +++ b/vlm/train/SyMDXnCcF7/94.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3696ea71b843a2362265372ce9a9da5945b9f168ed518a37bbeb28b35f453cd9 +size 115398 diff --git a/vlm/train/SylO2yStDr/0.png b/vlm/train/SylO2yStDr/0.png new file mode 100644 index 0000000000000000000000000000000000000000..9fd6bff799a22cf5d0d8dfecc512129f82bc903d --- /dev/null +++ b/vlm/train/SylO2yStDr/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3a32cf475572bbb157438879a82b7f1b9de45e325f995b507b1e9f8706308ccf +size 503240 diff --git a/vlm/train/SylO2yStDr/10.png b/vlm/train/SylO2yStDr/10.png new file mode 100644 index 0000000000000000000000000000000000000000..91189fe36ce0fca9841690f4f29041b690403059 --- /dev/null +++ b/vlm/train/SylO2yStDr/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ff43088cbbc39e3c25d8b898c5a33f1436a0d3529ddb6b31180d5d17d1f4982d +size 514302 diff --git a/vlm/train/SylO2yStDr/11.png b/vlm/train/SylO2yStDr/11.png new file mode 100644 index 0000000000000000000000000000000000000000..47d1a0a0a3c69a6e7a096ea118d9ff6eef3e7fac --- /dev/null +++ b/vlm/train/SylO2yStDr/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2f2351a7e6977b6a152f7014c9e8c128823f36b787101c1bbeb56a55336fa266 +size 418016 diff --git a/vlm/train/SylO2yStDr/14.png b/vlm/train/SylO2yStDr/14.png new file mode 100644 index 0000000000000000000000000000000000000000..df32929caf8c89bb0bd57a3157c2624ee4189bd3 --- /dev/null +++ b/vlm/train/SylO2yStDr/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:80796c3913d802d64b8fb6d839572bc6f2bb40c33334e2fb39567adf35b833a2 +size 419052 diff --git a/vlm/train/SylO2yStDr/15.png b/vlm/train/SylO2yStDr/15.png new file mode 100644 index 0000000000000000000000000000000000000000..0f28f669564a72ed74006bf29fc751bd135441e2 --- /dev/null +++ b/vlm/train/SylO2yStDr/15.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0c88be70b66256d001efad38b4b56bdad8883c811a02974f19c45b36862d40ab +size 371376 diff --git a/vlm/train/SylO2yStDr/2.png b/vlm/train/SylO2yStDr/2.png new file mode 100644 index 0000000000000000000000000000000000000000..7b3e4b49d68bd82a72f21407c34a914d0a2cae09 --- /dev/null +++ b/vlm/train/SylO2yStDr/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c893064735d1efd9c8d991381de185dfc896c271aaabb9e967b7629829944f32 +size 489384 diff --git a/vlm/train/SylO2yStDr/3.png b/vlm/train/SylO2yStDr/3.png new file mode 100644 index 0000000000000000000000000000000000000000..e37f4cc0a6d008495cbb6da15c6a455772061700 --- /dev/null +++ b/vlm/train/SylO2yStDr/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:465a4dfc8b198b7c3db4a435be6feb6e6ab5dca9d0b39a97ebdde635c78642f6 +size 519076 diff --git a/vlm/train/SylO2yStDr/6.png b/vlm/train/SylO2yStDr/6.png new file mode 100644 index 0000000000000000000000000000000000000000..0d94bedbaefb63a15513f678d2b1b035d0869ddb --- /dev/null +++ b/vlm/train/SylO2yStDr/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:34d0aebd57fb98e8a336116f5825932eb5d30615c3953a929dfd08e7c4deac14 +size 498838 diff --git a/vlm/train/SylO2yStDr/9.png b/vlm/train/SylO2yStDr/9.png new file mode 100644 index 0000000000000000000000000000000000000000..8f2d844c7f7fafb991656868e1a5299aa8d499cc --- /dev/null +++ b/vlm/train/SylO2yStDr/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aa7cfb3f62a4e861901dac5b4480e037183759a7810392d8078e6e869ba1e3e5 +size 516862 diff --git a/vlm/train/Syx7A3NFvH/0.png b/vlm/train/Syx7A3NFvH/0.png new file mode 100644 index 0000000000000000000000000000000000000000..1328d5f380383a56746090a3b1ba54945fca2946 --- /dev/null +++ b/vlm/train/Syx7A3NFvH/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5076c8c580c23530337d521af0e9d2be984d1d4141188e8c2a0386db4f8b8820 +size 503928 diff --git a/vlm/train/Syx7A3NFvH/1.png b/vlm/train/Syx7A3NFvH/1.png new file mode 100644 index 0000000000000000000000000000000000000000..38aa41a8f09b64e2b76675fd51e1e7fe199cb153 --- /dev/null +++ b/vlm/train/Syx7A3NFvH/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f5c124b298a6211cbeb69fb74f1bf518218884d247885da7bd0ad448333bcfd2 +size 547570 diff --git a/vlm/train/Syx7A3NFvH/10.png b/vlm/train/Syx7A3NFvH/10.png new file mode 100644 index 0000000000000000000000000000000000000000..ad804ab346b3977e77a7f2a3f3c0c41f5a9d5389 --- /dev/null +++ b/vlm/train/Syx7A3NFvH/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:99bafae0421f9d120b93bd08aa9555d4bf2ce558988fb13745ed202c29ea220f +size 478729 diff --git a/vlm/train/Syx7A3NFvH/11.png b/vlm/train/Syx7A3NFvH/11.png new file mode 100644 index 0000000000000000000000000000000000000000..3e867892f11569967805f79c7cfa113a1abbcde9 --- /dev/null +++ b/vlm/train/Syx7A3NFvH/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:646c9857cbc0b842d47415de3aa2655eefe589b70d1b40fa01bf79b87865ff1b +size 415046 diff --git a/vlm/train/Syx7A3NFvH/12.png b/vlm/train/Syx7A3NFvH/12.png new file mode 100644 index 0000000000000000000000000000000000000000..464b933330e98097f8f2eecb7c016d8eb6e17dad --- /dev/null +++ b/vlm/train/Syx7A3NFvH/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7a04a8ff5c093536344d415164f606591f02d224972b2bc7f8c29746d120e5b3 +size 353875 diff --git a/vlm/train/Syx7A3NFvH/13.png b/vlm/train/Syx7A3NFvH/13.png new file mode 100644 index 0000000000000000000000000000000000000000..70c552563db0620f5395ec7cf4f80e115f3853ce --- /dev/null +++ b/vlm/train/Syx7A3NFvH/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:70694d4d4face3f158a371ad4cf91f4229bce4d65692845bdaa7cf2a461fb3fc +size 403960 diff --git a/vlm/train/Syx7A3NFvH/14.png b/vlm/train/Syx7A3NFvH/14.png new file mode 100644 index 0000000000000000000000000000000000000000..6ffcf2acce735706b3a3e8c9933565f68aeea0c0 --- /dev/null +++ b/vlm/train/Syx7A3NFvH/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1712ed117b928786e7265bae3d91cf085f4875b4a0ce5fe4c278a434e9dab57f +size 391317 diff --git a/vlm/train/Syx7A3NFvH/15.png b/vlm/train/Syx7A3NFvH/15.png new file mode 100644 index 0000000000000000000000000000000000000000..1bf9cc0ebe87fbab11bf5c986a6a12efc0602393 --- /dev/null +++ b/vlm/train/Syx7A3NFvH/15.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fb7e5d5750f14d522635458b6cbcd8c58f7a11a033e7898e86d2a87513ca4f5b +size 190914 diff --git a/vlm/train/Syx7A3NFvH/16.png b/vlm/train/Syx7A3NFvH/16.png new file mode 100644 index 0000000000000000000000000000000000000000..db978ba4547a315aeada72e2308e8f71b429bf1f --- /dev/null +++ b/vlm/train/Syx7A3NFvH/16.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6bca06f85c8f7531961dc1aa04c16ae6244824189a13ac247314a6960251e58c +size 169617 diff --git a/vlm/train/Syx7A3NFvH/2.png b/vlm/train/Syx7A3NFvH/2.png new file mode 100644 index 0000000000000000000000000000000000000000..2c612c22d13e2eb2bd947bc7897de58d380dd60b --- /dev/null +++ b/vlm/train/Syx7A3NFvH/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d19d0fa7b6fa8dda3b34361c918bb918a918e22e27f0aae5abd58ee8110f66c5 +size 568566 diff --git a/vlm/train/Syx7A3NFvH/3.png b/vlm/train/Syx7A3NFvH/3.png new file mode 100644 index 0000000000000000000000000000000000000000..b13444b7bb972b8b9fa7bfdd4a9ba1a50771759c --- /dev/null +++ b/vlm/train/Syx7A3NFvH/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0a16ea10c4595e52ca651b59607599d95a8c9a0b924ccadefc4e86c9d32ec7d2 +size 558829 diff --git a/vlm/train/Syx7A3NFvH/4.png b/vlm/train/Syx7A3NFvH/4.png new file mode 100644 index 0000000000000000000000000000000000000000..dbec5b3ed16fca4902388f4205bef6a63c838fd9 --- /dev/null +++ b/vlm/train/Syx7A3NFvH/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:de1c2680f7fa12f31aa9aad5085e78c1db326a1cc7a4ccd04e1d8112b5f16c10 +size 554566 diff --git a/vlm/train/Syx7A3NFvH/5.png b/vlm/train/Syx7A3NFvH/5.png new file mode 100644 index 0000000000000000000000000000000000000000..259fcc34308d5b3db5aba8cc2a92a80ca279af1d --- /dev/null +++ b/vlm/train/Syx7A3NFvH/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e3ebcd46b749af75413a9a852c16168e088804b78975e26a2ab98dc5e7971368 +size 629159 diff --git a/vlm/train/Syx7A3NFvH/6.png b/vlm/train/Syx7A3NFvH/6.png new file mode 100644 index 0000000000000000000000000000000000000000..28000b4e290caba8e19831913fdbc3541edd89a2 --- /dev/null +++ b/vlm/train/Syx7A3NFvH/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b87a14830ec763fe488602f44950bee5a4287effb01856080fca737884b047c9 +size 577700 diff --git a/vlm/train/Syx7A3NFvH/7.png b/vlm/train/Syx7A3NFvH/7.png new file mode 100644 index 0000000000000000000000000000000000000000..f18e44105e350aefafb637bfba9527bbbf9fd06c --- /dev/null +++ b/vlm/train/Syx7A3NFvH/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d272687e41e06f7bf07e4094e14012ca02e514f7590e62cefd76b0cef05eb8a6 +size 755444 diff --git a/vlm/train/Syx7A3NFvH/8.png b/vlm/train/Syx7A3NFvH/8.png new file mode 100644 index 0000000000000000000000000000000000000000..173b45d93f03c1c09f5f6c83893fc5dfc0dddf1e --- /dev/null +++ b/vlm/train/Syx7A3NFvH/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d8b3f7b302ac408d8fbfe2ba951e56bf4f15b1b9145292c4a9860d8e50d87ae4 +size 580667 diff --git a/vlm/train/Syx7A3NFvH/9.png b/vlm/train/Syx7A3NFvH/9.png new file mode 100644 index 0000000000000000000000000000000000000000..853d8fa42fe4819ea9897863c7dfce82cc300052 --- /dev/null +++ b/vlm/train/Syx7A3NFvH/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7dd29895b50c49c0224d513e0597a1fcd918b7532492cebaf5b6ddfd5568159f +size 518453 diff --git a/vlm/train/XWYJ25-yTRS/0.png b/vlm/train/XWYJ25-yTRS/0.png new file mode 100644 index 0000000000000000000000000000000000000000..31e4c9d5eef817d89176234ada16f7050a65e929 --- /dev/null +++ b/vlm/train/XWYJ25-yTRS/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:66443bd8276fde0dded4edcbdb6323b4a7876d806c006b75bb8b19a99b6cc788 +size 421232 diff --git a/vlm/train/XWYJ25-yTRS/1.png b/vlm/train/XWYJ25-yTRS/1.png new file mode 100644 index 0000000000000000000000000000000000000000..947f792c82521f7a015958a69a43ead11ab68693 --- /dev/null +++ b/vlm/train/XWYJ25-yTRS/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:21f2e3b7fdfd394fe97587300f90433013cc9cf182f0c151af05d63e401054ae +size 575558 diff --git a/vlm/train/XWYJ25-yTRS/10.png b/vlm/train/XWYJ25-yTRS/10.png new file mode 100644 index 0000000000000000000000000000000000000000..a50d3ce5d5b439c1b1fe72258d8c6df45404353a --- /dev/null +++ b/vlm/train/XWYJ25-yTRS/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:027582590455f81a6c68687e75184a82fef74709db181e187085de93b3ccf62a +size 541517 diff --git a/vlm/train/XWYJ25-yTRS/11.png b/vlm/train/XWYJ25-yTRS/11.png new file mode 100644 index 0000000000000000000000000000000000000000..f0110340284dededcce2acf7b559106d39b06d65 --- /dev/null +++ b/vlm/train/XWYJ25-yTRS/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5823fea8c846244c7d5d5b233f01b14c1b3dbe02157d3280bb5efb505b9e89f5 +size 520684 diff --git a/vlm/train/XWYJ25-yTRS/12.png b/vlm/train/XWYJ25-yTRS/12.png new file mode 100644 index 0000000000000000000000000000000000000000..507f200582531d2db717ed0a3222734c32268df0 --- /dev/null +++ b/vlm/train/XWYJ25-yTRS/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9fe2326b0ddd45e44ba430aec6447ee96281a3160b3a655cb325a2832171eb02 +size 522718 diff --git a/vlm/train/XWYJ25-yTRS/2.png b/vlm/train/XWYJ25-yTRS/2.png new file mode 100644 index 0000000000000000000000000000000000000000..53617e6ede6e8de3d741dba0fb43352b7f8338b5 --- /dev/null +++ b/vlm/train/XWYJ25-yTRS/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:89a2296e88266e96816d584a8bcd1097a2f049ad1f4a87cb99616daf7b3d8276 +size 589764 diff --git a/vlm/train/XWYJ25-yTRS/3.png b/vlm/train/XWYJ25-yTRS/3.png new file mode 100644 index 0000000000000000000000000000000000000000..763bc5e17cc90605bd0494c2f8c58972b7cb563a --- /dev/null +++ b/vlm/train/XWYJ25-yTRS/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:54420cecb0a1a36ebd7ee48c3863f204751e2d4b693781dd29fa302b4b4a5395 +size 558226 diff --git a/vlm/train/XWYJ25-yTRS/4.png b/vlm/train/XWYJ25-yTRS/4.png new file mode 100644 index 0000000000000000000000000000000000000000..593afcedeaa337392f9baf3b151ecfa755ea9820 --- /dev/null +++ b/vlm/train/XWYJ25-yTRS/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8bc80729b6251db3366be6184558e222861ee3f44db7225e291cf63068e3ac51 +size 571581 diff --git a/vlm/train/XWYJ25-yTRS/5.png b/vlm/train/XWYJ25-yTRS/5.png new file mode 100644 index 0000000000000000000000000000000000000000..f7d1350319897dbe81abe87ac1f31eb847e3f9fc --- /dev/null +++ b/vlm/train/XWYJ25-yTRS/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:42a1d228c1de73a29ae183df04b58437042f7714631cd244e66efd1ba0c7d654 +size 599434 diff --git a/vlm/train/XWYJ25-yTRS/6.png b/vlm/train/XWYJ25-yTRS/6.png new file mode 100644 index 0000000000000000000000000000000000000000..e3112d50bd2ad7524abbc09f574589c8efcbb6a7 --- /dev/null +++ b/vlm/train/XWYJ25-yTRS/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4969906e6e7184948aef2a36ba0ce8de3f7bd19bbe83e3a09b6a4a48429f56e0 +size 593933 diff --git a/vlm/train/XWYJ25-yTRS/7.png b/vlm/train/XWYJ25-yTRS/7.png new file mode 100644 index 0000000000000000000000000000000000000000..8cfd46c62a7b70cea842f95a765b26f0c6f18af7 --- /dev/null +++ b/vlm/train/XWYJ25-yTRS/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c0545164cb64d6999c34ac88927c760a46f07697583d3cde0f1302a6233d2377 +size 619689 diff --git a/vlm/train/XWYJ25-yTRS/8.png b/vlm/train/XWYJ25-yTRS/8.png new file mode 100644 index 0000000000000000000000000000000000000000..ffdbcfc4d584e478debb2756d863ac2ec2211117 --- /dev/null +++ b/vlm/train/XWYJ25-yTRS/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a2cb031d3b2033de240b726ad10fc04d668f557f218eabfbe3d1b8fb44eda092 +size 592982 diff --git a/vlm/train/XWYJ25-yTRS/9.png b/vlm/train/XWYJ25-yTRS/9.png new file mode 100644 index 0000000000000000000000000000000000000000..fe312eaba70b022e7c6650ed488c45e8aa8f478e --- /dev/null +++ b/vlm/train/XWYJ25-yTRS/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:726d5c0b71d24f70157a08b6f3d9bb19cf33bbb0542212a23da3009f80dbf1a2 +size 543797 diff --git a/vlm/train/aLE2sEtMNXv/0.png b/vlm/train/aLE2sEtMNXv/0.png new file mode 100644 index 0000000000000000000000000000000000000000..41d3a417d6adece383fdb701689d5394583a148e --- /dev/null +++ b/vlm/train/aLE2sEtMNXv/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a3401e2780bd329d16025554e207fdbb5aaa7d8d2cdb1de23cca8f2c5abcbde6 +size 465646 diff --git a/vlm/train/aLE2sEtMNXv/1.png b/vlm/train/aLE2sEtMNXv/1.png new file mode 100644 index 0000000000000000000000000000000000000000..4617e1289b921c4ed0279573183d90f828c499fd --- /dev/null +++ b/vlm/train/aLE2sEtMNXv/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1ad4cdc5a5a949a4302ee80feccba648e4fc4beb5f7e838f6438833610ecbc03 +size 509743 diff --git a/vlm/train/aLE2sEtMNXv/10.png b/vlm/train/aLE2sEtMNXv/10.png new file mode 100644 index 0000000000000000000000000000000000000000..ba240d4e74a4ca80c0436fc9dfb8941d87c18dbe --- /dev/null +++ b/vlm/train/aLE2sEtMNXv/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:399bbff0efff0f9e7daa1da4f781bbed6186d3cf5e31873af5bc9d87dddca08e +size 546118 diff --git a/vlm/train/aLE2sEtMNXv/11.png b/vlm/train/aLE2sEtMNXv/11.png new file mode 100644 index 0000000000000000000000000000000000000000..4538540a28b4738bc1c3ce670091e27362b46c6a --- /dev/null +++ b/vlm/train/aLE2sEtMNXv/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:998f6371d7ad012a37fc355a30b949a9b3c56f51c3b353d45843adb9d963d0c0 +size 585257 diff --git a/vlm/train/aLE2sEtMNXv/12.png b/vlm/train/aLE2sEtMNXv/12.png new file mode 100644 index 0000000000000000000000000000000000000000..c26cb5645b2390daea544f1213da35ca65c4a914 --- /dev/null +++ b/vlm/train/aLE2sEtMNXv/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1472d770f4769481b9a3a5724c6ecf9cbc568fb1606973adc29cdf6b850afeb2 +size 545203 diff --git a/vlm/train/aLE2sEtMNXv/13.png b/vlm/train/aLE2sEtMNXv/13.png new file mode 100644 index 0000000000000000000000000000000000000000..7c4760880a94f0561f8cb581f763de2ec22e7fbe --- /dev/null +++ b/vlm/train/aLE2sEtMNXv/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e2d823106f304919bb475518b8b5ff629cdd39411573c9f6de10c7526061a440 +size 326201 diff --git a/vlm/train/aLE2sEtMNXv/2.png b/vlm/train/aLE2sEtMNXv/2.png new file mode 100644 index 0000000000000000000000000000000000000000..d351a314b2c259649beb28751e4ec8f56476facc --- /dev/null +++ b/vlm/train/aLE2sEtMNXv/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6155bdff18359d34c2bd6b20dc8fb61e6ad13f80d872ddd16d89f8d02be5961b +size 444536 diff --git a/vlm/train/aLE2sEtMNXv/3.png b/vlm/train/aLE2sEtMNXv/3.png new file mode 100644 index 0000000000000000000000000000000000000000..ec7d779f592ede374d1aac317e85cf399dd60fae --- /dev/null +++ b/vlm/train/aLE2sEtMNXv/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7d283dc6fe856ad6537bd1170541a25b1aed1ed7885fd8c0f53023f68d9b2289 +size 403357 diff --git a/vlm/train/aLE2sEtMNXv/4.png b/vlm/train/aLE2sEtMNXv/4.png new file mode 100644 index 0000000000000000000000000000000000000000..5e3b50012ee21f66ab74dd11c54b2d819d384a0d --- /dev/null +++ b/vlm/train/aLE2sEtMNXv/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f690ee856096a4a554e08071761cfa49021402ab4c8dab102aa7f578a6edebf0 +size 485079 diff --git a/vlm/train/aLE2sEtMNXv/5.png b/vlm/train/aLE2sEtMNXv/5.png new file mode 100644 index 0000000000000000000000000000000000000000..336aebfd7c931d8d0742d15da7f6be39c586dc8c --- /dev/null +++ b/vlm/train/aLE2sEtMNXv/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3475c80eb41cef3ebf958c788b7b2b3a9a4f06735d3e375e52bc39669673f124 +size 489150 diff --git a/vlm/train/aLE2sEtMNXv/6.png b/vlm/train/aLE2sEtMNXv/6.png new file mode 100644 index 0000000000000000000000000000000000000000..af8820e3e3d39863a9cae316ac52c5c2b2400a3c --- /dev/null +++ b/vlm/train/aLE2sEtMNXv/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0cd307e3d12fcf1c0e0381efe435128171f838d4ed4779d6c8804b8c686ecee6 +size 518564 diff --git a/vlm/train/aLE2sEtMNXv/7.png b/vlm/train/aLE2sEtMNXv/7.png new file mode 100644 index 0000000000000000000000000000000000000000..0bce70829e3eb72dd64ce3c25b0a44cbe4f2b474 --- /dev/null +++ b/vlm/train/aLE2sEtMNXv/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6472cadfdc1784db857ef017059f36b561d9d64361fa77b2bca84e635c5f20b8 +size 512107 diff --git a/vlm/train/aLE2sEtMNXv/8.png b/vlm/train/aLE2sEtMNXv/8.png new file mode 100644 index 0000000000000000000000000000000000000000..8c8a20a6afc2bdccd3672f15ab9f6ef8952fb77e --- /dev/null +++ b/vlm/train/aLE2sEtMNXv/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2046eafdcbe79a8f283753ed48fdd6c3220f53f349c7fc38f5ab51b19c075d02 +size 434906 diff --git a/vlm/train/aLE2sEtMNXv/9.png b/vlm/train/aLE2sEtMNXv/9.png new file mode 100644 index 0000000000000000000000000000000000000000..c86c49d37e84eb828e577f9e88b94926f8c5fe20 --- /dev/null +++ b/vlm/train/aLE2sEtMNXv/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a5d76531b075a9351da9e106ae3204906eb8c439b16dc13cc49bc0307cb900cc +size 601149 diff --git a/vlm/train/dvSExzhjG9D/0.png b/vlm/train/dvSExzhjG9D/0.png new file mode 100644 index 0000000000000000000000000000000000000000..98998506a7c35caffa4e72df1d3d3f4e5984f193 --- /dev/null +++ b/vlm/train/dvSExzhjG9D/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8dd3ed98ab6bc1dac926e9eedb8f96590bdb17851b0653b48cc8e13e8b85e99d +size 576790 diff --git a/vlm/train/dvSExzhjG9D/1.png b/vlm/train/dvSExzhjG9D/1.png new file mode 100644 index 0000000000000000000000000000000000000000..03bc26722572e42e56701df630886ad892a21591 --- /dev/null +++ b/vlm/train/dvSExzhjG9D/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f0b910f87066826df6a81412eb536f2fa743dc2c3d9ed754795c7909a8877605 +size 638235 diff --git a/vlm/train/dvSExzhjG9D/10.png b/vlm/train/dvSExzhjG9D/10.png new file mode 100644 index 0000000000000000000000000000000000000000..23f9990c319a055761807f50d4a381450a80106d --- /dev/null +++ b/vlm/train/dvSExzhjG9D/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ef6618f15dc5700263f149881c44af4772436c208df53fb6f62378947110c768 +size 504166 diff --git a/vlm/train/dvSExzhjG9D/11.png b/vlm/train/dvSExzhjG9D/11.png new file mode 100644 index 0000000000000000000000000000000000000000..a5fbe36d55f03bae749f8cc8026d77c84cf61e65 --- /dev/null +++ b/vlm/train/dvSExzhjG9D/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bd8c6dc467b673c7f0c93914bdfe71e15f80390393a3422283e6b525153c6840 +size 490804 diff --git a/vlm/train/dvSExzhjG9D/12.png b/vlm/train/dvSExzhjG9D/12.png new file mode 100644 index 0000000000000000000000000000000000000000..26803016d7dcd234af8c15204c0e1c36395c21e0 --- /dev/null +++ b/vlm/train/dvSExzhjG9D/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cde95b1bf90544630f68c202dc683c7905b61ad4240185615f2f120de7c95b84 +size 137641 diff --git a/vlm/train/dvSExzhjG9D/13.png b/vlm/train/dvSExzhjG9D/13.png new file mode 100644 index 0000000000000000000000000000000000000000..1dbfd7141206eae76db8529aacdddbdb280fdc10 --- /dev/null +++ b/vlm/train/dvSExzhjG9D/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3fe67464bdd75d02ddb23c822916ed955fb8244ca2b2d5c2b13416bc55fa37ea +size 426862 diff --git a/vlm/train/dvSExzhjG9D/14.png b/vlm/train/dvSExzhjG9D/14.png new file mode 100644 index 0000000000000000000000000000000000000000..b26fe9565d217fa665aeccbc137553c513bfa89d --- /dev/null +++ b/vlm/train/dvSExzhjG9D/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f6e1d3961d9893de2574d82ee4cf24f7d1feaf81d86d2d4653c5bd10436d9ae5 +size 371172 diff --git a/vlm/train/dvSExzhjG9D/15.png b/vlm/train/dvSExzhjG9D/15.png new file mode 100644 index 0000000000000000000000000000000000000000..35b2c5ddcc96c52971ead979d241aa15a7d8c85e --- /dev/null +++ b/vlm/train/dvSExzhjG9D/15.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b546718c33069c6dc2603761967de03d3b0ff74b0e4d8df86f789bcfff8cd06b +size 295282 diff --git a/vlm/train/dvSExzhjG9D/16.png b/vlm/train/dvSExzhjG9D/16.png new file mode 100644 index 0000000000000000000000000000000000000000..4b3560730d8dfa14518b6fd9acff2833b5fac846 --- /dev/null +++ b/vlm/train/dvSExzhjG9D/16.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bac24f01650b0eef58d16b2356968b1f4ec09cfb72e27c5458595655216bf688 +size 380087 diff --git a/vlm/train/dvSExzhjG9D/17.png b/vlm/train/dvSExzhjG9D/17.png new file mode 100644 index 0000000000000000000000000000000000000000..6b55f1a75ad772b9f652250c4eade581bed6d08e --- /dev/null +++ b/vlm/train/dvSExzhjG9D/17.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ea00b05f07dfdd5339c22861465d9837da9775ec96d0b6f9dfa49c90487003ed +size 495773 diff --git a/vlm/train/dvSExzhjG9D/18.png b/vlm/train/dvSExzhjG9D/18.png new file mode 100644 index 0000000000000000000000000000000000000000..74b8930d62ef43e88a17679095aed20670faa6ce --- /dev/null +++ b/vlm/train/dvSExzhjG9D/18.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8f63a65e33a38a341113286c48d8a976a02e787fa684c73e9cc01d2eb90595ea +size 835127 diff --git a/vlm/train/dvSExzhjG9D/19.png b/vlm/train/dvSExzhjG9D/19.png new file mode 100644 index 0000000000000000000000000000000000000000..528153e587dd73d22aaf8f3ebb1800bce6684c98 --- /dev/null +++ b/vlm/train/dvSExzhjG9D/19.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f98bf0a3640cce582f1c6d2326e5d512ed810bad32d20dade548e29d6ae0c989 +size 635775 diff --git a/vlm/train/dvSExzhjG9D/2.png b/vlm/train/dvSExzhjG9D/2.png new file mode 100644 index 0000000000000000000000000000000000000000..651cbb1069e2b420e9d8cc4e7130129cfbd91594 --- /dev/null +++ b/vlm/train/dvSExzhjG9D/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:803213a3f24d6c7e92e4fd4827ba9bd4db05632880ce32b4cb1bfff94f206817 +size 568630 diff --git a/vlm/train/dvSExzhjG9D/20.png b/vlm/train/dvSExzhjG9D/20.png new file mode 100644 index 0000000000000000000000000000000000000000..eab5078e431021c0a4a7fa84353e27aee52203ac --- /dev/null +++ b/vlm/train/dvSExzhjG9D/20.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c2d35ae7d380a9e45f48247c42dd5b2d680d1ff4ea2f612f4ed36e95a6555fb1 +size 612266 diff --git a/vlm/train/dvSExzhjG9D/21.png b/vlm/train/dvSExzhjG9D/21.png new file mode 100644 index 0000000000000000000000000000000000000000..2b5e087c4fa8e0d189167cef2cf0d0b7e4c08118 --- /dev/null +++ b/vlm/train/dvSExzhjG9D/21.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:939bbf095f9cd7fbf1992ab0e343cbf04febfaad0f8c1f229d3bca1d3ac1300f +size 582998 diff --git a/vlm/train/dvSExzhjG9D/22.png b/vlm/train/dvSExzhjG9D/22.png new file mode 100644 index 0000000000000000000000000000000000000000..a96ebfceafdfb1d0642c255d49d32d0aeafc7115 --- /dev/null +++ b/vlm/train/dvSExzhjG9D/22.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ac5ae621c46fe8afcf8ebc8f7affabc14639bc065f8fe5caa199e057da7e459a +size 596991 diff --git a/vlm/train/dvSExzhjG9D/23.png b/vlm/train/dvSExzhjG9D/23.png new file mode 100644 index 0000000000000000000000000000000000000000..55beefe16f5f2e60d0784366f8740379b6e5629f --- /dev/null +++ b/vlm/train/dvSExzhjG9D/23.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:26517ccb4c22fba442d6acbc0912bec787f530b08b40528ac4ffb69656f652f3 +size 457734 diff --git a/vlm/train/dvSExzhjG9D/24.png b/vlm/train/dvSExzhjG9D/24.png new file mode 100644 index 0000000000000000000000000000000000000000..487fe04b509b8510bde99162202e5ff58a6cdc4c --- /dev/null +++ b/vlm/train/dvSExzhjG9D/24.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d3f78588a60261947039001cf06f4c6a6349a3117b4d2c5936394891df8487e9 +size 310664 diff --git a/vlm/train/dvSExzhjG9D/3.png b/vlm/train/dvSExzhjG9D/3.png new file mode 100644 index 0000000000000000000000000000000000000000..c3d761a13290d9d74d39d3004d3fa5ae7b432acc --- /dev/null +++ b/vlm/train/dvSExzhjG9D/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4e75540134c302c85679408ea4f4b36dd16a7642dd7abfb93f7d8fed360c28ba +size 578853 diff --git a/vlm/train/dvSExzhjG9D/4.png b/vlm/train/dvSExzhjG9D/4.png new file mode 100644 index 0000000000000000000000000000000000000000..a0a283e640b035d884368b2c77959ff1c01f6f09 --- /dev/null +++ b/vlm/train/dvSExzhjG9D/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:659b25c06403c6b6848e19c7b69a9aaaf30c2dd5604740a7c72b0cac6aa51539 +size 615582 diff --git a/vlm/train/dvSExzhjG9D/5.png b/vlm/train/dvSExzhjG9D/5.png new file mode 100644 index 0000000000000000000000000000000000000000..b58651ea96c9c59524f54f0f2ede5ebc40aed7e2 --- /dev/null +++ b/vlm/train/dvSExzhjG9D/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e6e8e05c5880f1f80fe103dcb8d4880195dee2d0f23e8af1af67e65d1242817a +size 854011 diff --git a/vlm/train/dvSExzhjG9D/6.png b/vlm/train/dvSExzhjG9D/6.png new file mode 100644 index 0000000000000000000000000000000000000000..21c06b9568bddbf913a92f3b77899d89898c124c --- /dev/null +++ b/vlm/train/dvSExzhjG9D/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9878b251f96f28216621cd8cc353f32dfdcc22063db25452c1e4bc486870c665 +size 722251 diff --git a/vlm/train/dvSExzhjG9D/7.png b/vlm/train/dvSExzhjG9D/7.png new file mode 100644 index 0000000000000000000000000000000000000000..ec81eae3b2277963a924e6b2890fe54234e93ae8 --- /dev/null +++ b/vlm/train/dvSExzhjG9D/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ee782e28974228af6d4638fc7cd08906b8b6d05f32cf1963073cab0ed110d2aa +size 670636 diff --git a/vlm/train/dvSExzhjG9D/8.png b/vlm/train/dvSExzhjG9D/8.png new file mode 100644 index 0000000000000000000000000000000000000000..347d24ca2e2044fe4721c17704abbe96b3bdb198 --- /dev/null +++ b/vlm/train/dvSExzhjG9D/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e68fa6cdec1c5c7c853d54de6565c3143a7c1fa745c7318b6a47ccb92d5e5e0a +size 593181 diff --git a/vlm/train/dvSExzhjG9D/9.png b/vlm/train/dvSExzhjG9D/9.png new file mode 100644 index 0000000000000000000000000000000000000000..449a69a465e9c71d79a86364c5ba0165ca1e81f9 --- /dev/null +++ b/vlm/train/dvSExzhjG9D/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a03ad9a4a35fc29e44730cc77e2dbeb1ee573ccdf6069d9d039dd94025ccc17b +size 483390 diff --git a/vlm/train/eNdiU_DbM9/0.png b/vlm/train/eNdiU_DbM9/0.png new file mode 100644 index 0000000000000000000000000000000000000000..4080799eeb8f29fa076c664ab9633eb675e6bf1b --- /dev/null +++ b/vlm/train/eNdiU_DbM9/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:db2b9bd74ec37480e05c142afa62c30c90f624ad394dd4ba8633fa6d61d1dbe1 +size 481189 diff --git a/vlm/train/eNdiU_DbM9/1.png b/vlm/train/eNdiU_DbM9/1.png new file mode 100644 index 0000000000000000000000000000000000000000..2cfec2dcf84224ca3fee25ab1c67a152d9262d52 --- /dev/null +++ b/vlm/train/eNdiU_DbM9/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bdad226f12683b97f8b5129ad5fb5815a71e5d6fd39650a7dd286ad91ddfc27b +size 786441 diff --git a/vlm/train/eNdiU_DbM9/10.png b/vlm/train/eNdiU_DbM9/10.png new file mode 100644 index 0000000000000000000000000000000000000000..eb1f316ea0d39beb9a6b7f009449b2e02620d864 --- /dev/null +++ b/vlm/train/eNdiU_DbM9/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c8e8c1ad980892817242531da33715166e3d5a013fd3a7e9db0a323bba437831 +size 521062 diff --git a/vlm/train/eNdiU_DbM9/11.png b/vlm/train/eNdiU_DbM9/11.png new file mode 100644 index 0000000000000000000000000000000000000000..6b391a350240083ba195e8412eadc0e7a7ae0317 --- /dev/null +++ b/vlm/train/eNdiU_DbM9/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:33d25b5cab350648754d1c7c8aea12131b0dfae4d6bcee1173b0fe97bda24b52 +size 352648 diff --git a/vlm/train/eNdiU_DbM9/12.png b/vlm/train/eNdiU_DbM9/12.png new file mode 100644 index 0000000000000000000000000000000000000000..50c0e5aa598907b14440f1f4ad212157c9e5c216 --- /dev/null +++ b/vlm/train/eNdiU_DbM9/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:09438a20ca2ce22d42a0035873dea96b16b24fbff5e1ecb918bb110644e329cb +size 489110 diff --git a/vlm/train/eNdiU_DbM9/13.png b/vlm/train/eNdiU_DbM9/13.png new file mode 100644 index 0000000000000000000000000000000000000000..276b7fbff62ef429693777ea3c2622362dce7763 --- /dev/null +++ b/vlm/train/eNdiU_DbM9/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:06eb916bc431dd3749b5c9116ace9436e395d3b60e0d495ec12ca43ec2bf3135 +size 432966 diff --git a/vlm/train/eNdiU_DbM9/14.png b/vlm/train/eNdiU_DbM9/14.png new file mode 100644 index 0000000000000000000000000000000000000000..8b093a4e7c912ff380b2e9d9e9ba49a5aaf31ff6 --- /dev/null +++ b/vlm/train/eNdiU_DbM9/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3fe744f574ca7d6fe884c7f82d5d92be17799472ab26e64dfe62b05bc96250ac +size 553599 diff --git a/vlm/train/eNdiU_DbM9/15.png b/vlm/train/eNdiU_DbM9/15.png new file mode 100644 index 0000000000000000000000000000000000000000..f92bfae59cc02a8b6dcf434b0b59d6ca18239f7f --- /dev/null +++ b/vlm/train/eNdiU_DbM9/15.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f9e8a3568b209df5133d6e3901051f716759e3b4f8dfe71588525e76ad09d6e8 +size 145021 diff --git a/vlm/train/eNdiU_DbM9/16.png b/vlm/train/eNdiU_DbM9/16.png new file mode 100644 index 0000000000000000000000000000000000000000..f3dd434dba9519fd0dae7977d02ec6791bed789d --- /dev/null +++ b/vlm/train/eNdiU_DbM9/16.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8c83aefac749cde2750f5b12522d46606c979679efbb1003864451340fc198fa +size 175501 diff --git a/vlm/train/eNdiU_DbM9/2.png b/vlm/train/eNdiU_DbM9/2.png new file mode 100644 index 0000000000000000000000000000000000000000..edb192e7d8558eec8afddc5b5e72965ed167684e --- /dev/null +++ b/vlm/train/eNdiU_DbM9/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0c8c7a0a7a3a1b3f11eebbee2d4929577dd1349980b7955e10a7047bc9c29a43 +size 494937 diff --git a/vlm/train/eNdiU_DbM9/3.png b/vlm/train/eNdiU_DbM9/3.png new file mode 100644 index 0000000000000000000000000000000000000000..a88f2f938594533ff540004658404ab24794fdda --- /dev/null +++ b/vlm/train/eNdiU_DbM9/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8b4b9db9f9022aef0356975652fbe2dfddca85a1bad8d95ca4ae4d107912ef01 +size 510833 diff --git a/vlm/train/eNdiU_DbM9/4.png b/vlm/train/eNdiU_DbM9/4.png new file mode 100644 index 0000000000000000000000000000000000000000..cb36e612d54de774c63b847bea6fde7301e53674 --- /dev/null +++ b/vlm/train/eNdiU_DbM9/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6731ce192485ccb00942af5ddbe94f4ab38b1e9e121d4121752a6fe0107d126c +size 572988 diff --git a/vlm/train/eNdiU_DbM9/5.png b/vlm/train/eNdiU_DbM9/5.png new file mode 100644 index 0000000000000000000000000000000000000000..54bd2e90b01d4a611094979e272b490f6e578c5d --- /dev/null +++ b/vlm/train/eNdiU_DbM9/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:876d143e1baca734b39fab3f8781273f6a44f9fd28d01908a8a28902b33363d7 +size 488382 diff --git a/vlm/train/eNdiU_DbM9/6.png b/vlm/train/eNdiU_DbM9/6.png new file mode 100644 index 0000000000000000000000000000000000000000..eb58b343173e2809a7d6ee82b3fd829f6c9b197b --- /dev/null +++ b/vlm/train/eNdiU_DbM9/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d1822a548582e56b50070a4f7cb6f7fbede0da51049a8986b51369f9b1a1379f +size 600141 diff --git a/vlm/train/eNdiU_DbM9/7.png b/vlm/train/eNdiU_DbM9/7.png new file mode 100644 index 0000000000000000000000000000000000000000..923651314b84aa07df7663a93eba484d06996223 --- /dev/null +++ b/vlm/train/eNdiU_DbM9/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b554fa97ef60e4e6137b7c0f68631714e831e4ccfab3619c9229f3f04909a9ae +size 559496 diff --git a/vlm/train/eNdiU_DbM9/8.png b/vlm/train/eNdiU_DbM9/8.png new file mode 100644 index 0000000000000000000000000000000000000000..35a5f1d1d6d483d2e6fdb10717becefb4015d89a --- /dev/null +++ b/vlm/train/eNdiU_DbM9/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9016f913c6dc03fc2943b8b556a3bf8b0562e031df1b12b0b9819aa6d69864b3 +size 536457 diff --git a/vlm/train/eNdiU_DbM9/9.png b/vlm/train/eNdiU_DbM9/9.png new file mode 100644 index 0000000000000000000000000000000000000000..b8d54a36dba28be7b75891ff5ca3f6db263b55f3 --- /dev/null +++ b/vlm/train/eNdiU_DbM9/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:85d350ee41acd840ca5b094cf9416b28b24550c3fb1fe75a9f4273be59918faa +size 507697 diff --git a/vlm/train/eedamFFTcK5/0.png b/vlm/train/eedamFFTcK5/0.png new file mode 100644 index 0000000000000000000000000000000000000000..f5ebe5b2c9cb6a8a6572886814a46c90f0e69a98 --- /dev/null +++ b/vlm/train/eedamFFTcK5/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:30ccf854e694f17021142fa11fd548bb8f664267fb9be550dcea0c77f9458213 +size 419339 diff --git a/vlm/train/eedamFFTcK5/1.png b/vlm/train/eedamFFTcK5/1.png new file mode 100644 index 0000000000000000000000000000000000000000..64b8f674fbb5400315252d8322865fc14dde818d --- /dev/null +++ b/vlm/train/eedamFFTcK5/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b1d93fb184b6a630800a903062caec83ef3756c4860ada9f13ff82a45ffd3d94 +size 488087 diff --git a/vlm/train/eedamFFTcK5/10.png b/vlm/train/eedamFFTcK5/10.png new file mode 100644 index 0000000000000000000000000000000000000000..ea5ee2299d9e0d83d9fbf7b2cd07a7c611c1eb6a --- /dev/null +++ b/vlm/train/eedamFFTcK5/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec93b40b2a2ec42885f93bf037b1246c0eab9ba3319d37a50e2e14a577464d5d +size 533022 diff --git a/vlm/train/eedamFFTcK5/11.png b/vlm/train/eedamFFTcK5/11.png new file mode 100644 index 0000000000000000000000000000000000000000..41025d29ad3a08c31939dc359cf2b972856dd263 --- /dev/null +++ b/vlm/train/eedamFFTcK5/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:be8bebc6c150a0f1b4dc96620950a59e1e7607924cd36b9e9cacde5122e668d3 +size 120077 diff --git a/vlm/train/eedamFFTcK5/2.png b/vlm/train/eedamFFTcK5/2.png new file mode 100644 index 0000000000000000000000000000000000000000..87295b8098aa305e1c530799d06e70e421c2f414 --- /dev/null +++ b/vlm/train/eedamFFTcK5/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:10524f3a30649617ceeb6f456fcef647ee60f223998d4a15de2c0538e52425d0 +size 577793 diff --git a/vlm/train/eedamFFTcK5/3.png b/vlm/train/eedamFFTcK5/3.png new file mode 100644 index 0000000000000000000000000000000000000000..578a2b5b664d7763e695be235be71630e334924d --- /dev/null +++ b/vlm/train/eedamFFTcK5/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6bed9057d52bc10bcc40e715e4624f47d5066a8ace2b27a32c047f8c20edabf5 +size 489073 diff --git a/vlm/train/eedamFFTcK5/4.png b/vlm/train/eedamFFTcK5/4.png new file mode 100644 index 0000000000000000000000000000000000000000..ae282bc5ce1106a9fc5eb738c114fe921be3aa5e --- /dev/null +++ b/vlm/train/eedamFFTcK5/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:21b493f918f5c3c1868f15eaa79d7501209f564eadfb043756d914f597c46fe9 +size 521125 diff --git a/vlm/train/eedamFFTcK5/5.png b/vlm/train/eedamFFTcK5/5.png new file mode 100644 index 0000000000000000000000000000000000000000..1da6da77231865c6450562c2d62cfd967abc7000 --- /dev/null +++ b/vlm/train/eedamFFTcK5/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:95235e467d1a96881aac87245659510afd1fc729782d79cb67a3f9f322e5a911 +size 458606 diff --git a/vlm/train/eedamFFTcK5/6.png b/vlm/train/eedamFFTcK5/6.png new file mode 100644 index 0000000000000000000000000000000000000000..2885ab30fa0d31684994072ecd5d57fedc1c3d9e --- /dev/null +++ b/vlm/train/eedamFFTcK5/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:94eefaa5d7698cf1f95cabe920cfb517afdd2c1de635577516d4d9d511da1b92 +size 574589 diff --git a/vlm/train/eedamFFTcK5/7.png b/vlm/train/eedamFFTcK5/7.png new file mode 100644 index 0000000000000000000000000000000000000000..4c4ca8b35ef571de934e5000f4c5b6215703a9c7 --- /dev/null +++ b/vlm/train/eedamFFTcK5/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9707d708be072abd6180446276c59b175d2b407e3ebdfa0a4a46045d18f8a8ed +size 534339 diff --git a/vlm/train/eedamFFTcK5/8.png b/vlm/train/eedamFFTcK5/8.png new file mode 100644 index 0000000000000000000000000000000000000000..ac9ef288a3cf09258d2072f3105e7f7d3f015704 --- /dev/null +++ b/vlm/train/eedamFFTcK5/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:85bd340a3b8217c6cd66d6e2270c4002539c724084b08844c15b07604087f19d +size 527161 diff --git a/vlm/train/eedamFFTcK5/9.png b/vlm/train/eedamFFTcK5/9.png new file mode 100644 index 0000000000000000000000000000000000000000..669814d006327f4caa45abf7e04492f634c19de3 --- /dev/null +++ b/vlm/train/eedamFFTcK5/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d470d95de0e2e356ead359939895121948faec08924b438eaa377016dc4eb79f +size 581977 diff --git a/vlm/train/iaO_IH7CnGJ/0.png b/vlm/train/iaO_IH7CnGJ/0.png new file mode 100644 index 0000000000000000000000000000000000000000..bb8b831aa114681ed8fff21642a48e3d80ea3acf --- /dev/null +++ b/vlm/train/iaO_IH7CnGJ/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e6f20b5eade98b3bfc1cb8833f285037d30679d9a463cf1cc8cd5e3d58490f63 +size 472117 diff --git a/vlm/train/iaO_IH7CnGJ/1.png b/vlm/train/iaO_IH7CnGJ/1.png new file mode 100644 index 0000000000000000000000000000000000000000..d145a46a35197d6ac4ce03f0399e62221a3f8aaf --- /dev/null +++ b/vlm/train/iaO_IH7CnGJ/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:85064c71bbc61a89a88ef17a2bf2b182f2bcdc475ac567a8fe714c0a608a4c68 +size 581060 diff --git a/vlm/train/iaO_IH7CnGJ/10.png b/vlm/train/iaO_IH7CnGJ/10.png new file mode 100644 index 0000000000000000000000000000000000000000..23f17d0482cac0bafd8db8115876f04a5268e8d3 --- /dev/null +++ b/vlm/train/iaO_IH7CnGJ/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c5856ca3dfc48e6fb058749c9a1a0f1ca35ea9c16d0e254b39d5fa8bf5755d53 +size 626970 diff --git a/vlm/train/iaO_IH7CnGJ/11.png b/vlm/train/iaO_IH7CnGJ/11.png new file mode 100644 index 0000000000000000000000000000000000000000..6447372bd9601e291be5a7894a0a0e13e2e44911 --- /dev/null +++ b/vlm/train/iaO_IH7CnGJ/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:985b013cefa67ed68a85ae2baa568c9d0edaa4548339916f0303e94a46635757 +size 609090 diff --git a/vlm/train/iaO_IH7CnGJ/12.png b/vlm/train/iaO_IH7CnGJ/12.png new file mode 100644 index 0000000000000000000000000000000000000000..b84789f6b72353f0bd2e740fe908aff8ce8f6cb1 --- /dev/null +++ b/vlm/train/iaO_IH7CnGJ/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b216b9cb566422d8c05ea34b000459468c63c9624b6b305581021c4c841e09f8 +size 596974 diff --git a/vlm/train/iaO_IH7CnGJ/13.png b/vlm/train/iaO_IH7CnGJ/13.png new file mode 100644 index 0000000000000000000000000000000000000000..fef5d6fadaac3c3606a0ea676c9b17954aeec6e4 --- /dev/null +++ b/vlm/train/iaO_IH7CnGJ/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:78fad7bdb83c351c9534a00b4040bb4dff626c155e26ad6346a50d215f58c21b +size 77403 diff --git a/vlm/train/iaO_IH7CnGJ/2.png b/vlm/train/iaO_IH7CnGJ/2.png new file mode 100644 index 0000000000000000000000000000000000000000..271d7cddc5d56ec6f8cfa4669fa59c5f1b527563 --- /dev/null +++ b/vlm/train/iaO_IH7CnGJ/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e09fbe667de9d140b594d38afced749af000f821b116f25f22f7b1d5b1d910f8 +size 592606 diff --git a/vlm/train/iaO_IH7CnGJ/3.png b/vlm/train/iaO_IH7CnGJ/3.png new file mode 100644 index 0000000000000000000000000000000000000000..d8e945e7aa0510847e21764ddf40f9da3c000356 --- /dev/null +++ b/vlm/train/iaO_IH7CnGJ/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:62cf0c1db636d4f4c4bdf4d5695a82aee0480464fb93e89e93d940f4463311b9 +size 536550 diff --git a/vlm/train/iaO_IH7CnGJ/4.png b/vlm/train/iaO_IH7CnGJ/4.png new file mode 100644 index 0000000000000000000000000000000000000000..50bd08dcdabde5b9c4d7a44638e306c95d9523fb --- /dev/null +++ b/vlm/train/iaO_IH7CnGJ/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:159e7b0fe0fc9050524515681a6f80617f94bb41b84ae3616b6dcf13a614cb27 +size 513014 diff --git a/vlm/train/iaO_IH7CnGJ/5.png b/vlm/train/iaO_IH7CnGJ/5.png new file mode 100644 index 0000000000000000000000000000000000000000..e3b5fbd9088af983fc49ed14958e1bd98c17cee8 --- /dev/null +++ b/vlm/train/iaO_IH7CnGJ/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0d209b3fdebfeaaab7a76785d9c59a430c6b4ed272eb1c8819e03c1ba91f4601 +size 541586 diff --git a/vlm/train/iaO_IH7CnGJ/6.png b/vlm/train/iaO_IH7CnGJ/6.png new file mode 100644 index 0000000000000000000000000000000000000000..eea31a9412647d4665801538c2624f35ed046f7f --- /dev/null +++ b/vlm/train/iaO_IH7CnGJ/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f1792b630f053cce34374cc113ec5ead0a0d9fc74ae668240f12de68982c4cde +size 471016 diff --git a/vlm/train/iaO_IH7CnGJ/7.png b/vlm/train/iaO_IH7CnGJ/7.png new file mode 100644 index 0000000000000000000000000000000000000000..f27d638881e229198d4126398d1c70be51e8da9a --- /dev/null +++ b/vlm/train/iaO_IH7CnGJ/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aebe0ecfba455b21b5bf506e1f1414ffbfe888c822318e7fc87c5f4c573f2b6b +size 455427 diff --git a/vlm/train/iaO_IH7CnGJ/8.png b/vlm/train/iaO_IH7CnGJ/8.png new file mode 100644 index 0000000000000000000000000000000000000000..f7cdd49839ee5bed6c966002f0c30efb31b634e3 --- /dev/null +++ b/vlm/train/iaO_IH7CnGJ/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:626af496addf42dfd206e0b6a7db91ce7e55f3e9ca924264522bc0fb7b28abe9 +size 549258 diff --git a/vlm/train/iaO_IH7CnGJ/9.png b/vlm/train/iaO_IH7CnGJ/9.png new file mode 100644 index 0000000000000000000000000000000000000000..908cf6ed58546695628f1330adefe6a3a98e1b3d --- /dev/null +++ b/vlm/train/iaO_IH7CnGJ/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c8f9a4de44f2bc1ff7fba51863ef73eb1e66dc5a078172ca2f6f4090afaa81de +size 563030 diff --git a/vlm/train/jLHWRxwc7_f/0.png b/vlm/train/jLHWRxwc7_f/0.png new file mode 100644 index 0000000000000000000000000000000000000000..bd63fa7a58e7e7a786826c06b2222e63d9016b30 --- /dev/null +++ b/vlm/train/jLHWRxwc7_f/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:233bc014deefc71f7c5a6313b82946b89635f5d3ead553ec0481ab45d3912deb +size 733100 diff --git a/vlm/train/jLHWRxwc7_f/1.png b/vlm/train/jLHWRxwc7_f/1.png new file mode 100644 index 0000000000000000000000000000000000000000..9a1ba2dec2bf08362aa18e87c19758ffd39436c9 --- /dev/null +++ b/vlm/train/jLHWRxwc7_f/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4566ee36dd3443aa1bd325dbddb6e69f71ad2e18bedba2e0e9ac8ea67484aef7 +size 890686 diff --git a/vlm/train/jLHWRxwc7_f/10.png b/vlm/train/jLHWRxwc7_f/10.png new file mode 100644 index 0000000000000000000000000000000000000000..e8bb555548b831c11caa48396cc79a8f14c47980 --- /dev/null +++ b/vlm/train/jLHWRxwc7_f/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0b1888ac65e0758c55cfdeaba1365e61fb4523ea6d19ed90f490187e0140d31e +size 543280 diff --git a/vlm/train/jLHWRxwc7_f/11.png b/vlm/train/jLHWRxwc7_f/11.png new file mode 100644 index 0000000000000000000000000000000000000000..f15af1b9afc17659c91cfad4f5c0177fef20942b --- /dev/null +++ b/vlm/train/jLHWRxwc7_f/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:89f950339d23b58a4900fef352d41ddb7162443b9809ad9477b19629cee3d5aa +size 578181 diff --git a/vlm/train/jLHWRxwc7_f/12.png b/vlm/train/jLHWRxwc7_f/12.png new file mode 100644 index 0000000000000000000000000000000000000000..d3cfec216645e55468e5b60f9a0c3cf27f0f3266 --- /dev/null +++ b/vlm/train/jLHWRxwc7_f/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0316cdd247a5616f8beecc4607fedfb148e22a98ccf50969967738a1f8557977 +size 556634 diff --git a/vlm/train/jLHWRxwc7_f/13.png b/vlm/train/jLHWRxwc7_f/13.png new file mode 100644 index 0000000000000000000000000000000000000000..5d8e57803daad76a5e3d8c1146cc5f3bd81cbac0 --- /dev/null +++ b/vlm/train/jLHWRxwc7_f/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d268c1584da5d00a82fb11738e3b99778c4e665e4a52190ff2dc5b2d4ac4daea +size 566372 diff --git a/vlm/train/jLHWRxwc7_f/14.png b/vlm/train/jLHWRxwc7_f/14.png new file mode 100644 index 0000000000000000000000000000000000000000..1a78bc2113217ba6a75944cd6c927c817eb92362 --- /dev/null +++ b/vlm/train/jLHWRxwc7_f/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0ea217e2a7e0ae02e490a9f57e1589853f81a88f010a62e1bdb08e301de0ebe1 +size 211055 diff --git a/vlm/train/jLHWRxwc7_f/2.png b/vlm/train/jLHWRxwc7_f/2.png new file mode 100644 index 0000000000000000000000000000000000000000..e98f7970338a2e77a75990c0a1dceb06a8e96373 --- /dev/null +++ b/vlm/train/jLHWRxwc7_f/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d490d704453011b779f6c46f409ef915a634c50dbe9f2993ab3355381fc30d8e +size 621981 diff --git a/vlm/train/jLHWRxwc7_f/3.png b/vlm/train/jLHWRxwc7_f/3.png new file mode 100644 index 0000000000000000000000000000000000000000..c2e8a7517d9c342f582e739c36deaeaaa502b6f9 --- /dev/null +++ b/vlm/train/jLHWRxwc7_f/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:be4d68e3c2c59f5d690dad2789d8c49e2b3baf157934644ecb56b39996667cdd +size 710564 diff --git a/vlm/train/jLHWRxwc7_f/4.png b/vlm/train/jLHWRxwc7_f/4.png new file mode 100644 index 0000000000000000000000000000000000000000..2cf8105e765de53fb117220f07feb19d7fe4ef03 --- /dev/null +++ b/vlm/train/jLHWRxwc7_f/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5be4af3262e6930240397426a248bc2a7dc98943cdc31b4af7b11cc7cf68f1de +size 571358 diff --git a/vlm/train/jLHWRxwc7_f/5.png b/vlm/train/jLHWRxwc7_f/5.png new file mode 100644 index 0000000000000000000000000000000000000000..d75c91e6c46a60d0b3c7f942fcec42e8f23ba7bf --- /dev/null +++ b/vlm/train/jLHWRxwc7_f/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:968a2ec4a32cd89e294de765aea78f2ff0924606f347ec37b178a212f4cdc85a +size 1478942 diff --git a/vlm/train/jLHWRxwc7_f/6.png b/vlm/train/jLHWRxwc7_f/6.png new file mode 100644 index 0000000000000000000000000000000000000000..f31b2a524057904df39add064c1751127888a3b3 --- /dev/null +++ b/vlm/train/jLHWRxwc7_f/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8531b12dcd70480bb8943884c41fb85f9f194132bf9a364c3ef27e5ab81a8dfa +size 1431759 diff --git a/vlm/train/jLHWRxwc7_f/7.png b/vlm/train/jLHWRxwc7_f/7.png new file mode 100644 index 0000000000000000000000000000000000000000..cf5388bc9f451a14ac4ed039a625725e3a409218 --- /dev/null +++ b/vlm/train/jLHWRxwc7_f/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:323d4a9513d9cc0edabd9fbb32ea8e101f989535fe674454e803e5870c3cedcd +size 1119314 diff --git a/vlm/train/jLHWRxwc7_f/8.png b/vlm/train/jLHWRxwc7_f/8.png new file mode 100644 index 0000000000000000000000000000000000000000..2ed05a2b03ae432ffa955de6a0921ea0f5e92933 --- /dev/null +++ b/vlm/train/jLHWRxwc7_f/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9c037f1117a53d50b91fd6cfe987120cbe1fa2a859e401626e46951e5e8f6109 +size 1409012 diff --git a/vlm/train/jLHWRxwc7_f/9.png b/vlm/train/jLHWRxwc7_f/9.png new file mode 100644 index 0000000000000000000000000000000000000000..f56fe082157d911fe772d39f626c5f35e3029e34 --- /dev/null +++ b/vlm/train/jLHWRxwc7_f/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:39a7e13560a6c26329a075d30e5be75ea06f0746ce07a5302caaff13c3c6ba98 +size 615276 diff --git a/vlm/train/jQSBcVURlpW/0.png b/vlm/train/jQSBcVURlpW/0.png new file mode 100644 index 0000000000000000000000000000000000000000..067e2728a1be37c771427eba5261216c9cde50ad --- /dev/null +++ b/vlm/train/jQSBcVURlpW/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3a2ba41008aa5acc28c0978ff6b85cb8ef191aca71f67507f4164c2d5cdd6e4f +size 536415 diff --git a/vlm/train/jQSBcVURlpW/1.png b/vlm/train/jQSBcVURlpW/1.png new file mode 100644 index 0000000000000000000000000000000000000000..7b8df8f7a842a78e89c1bb55345f9b5d202f17a6 --- /dev/null +++ b/vlm/train/jQSBcVURlpW/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2c7a99ad1f80fb47a152f1b89783021b4136ee06297f7028bc264d7d1abcf60f +size 658374 diff --git a/vlm/train/jQSBcVURlpW/10.png b/vlm/train/jQSBcVURlpW/10.png new file mode 100644 index 0000000000000000000000000000000000000000..ffc9f8afe6cc730bbc685fc0b94a6492f0dd6022 --- /dev/null +++ b/vlm/train/jQSBcVURlpW/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9c9752034cb06149253e87c3d417559a30d746110d52300acd454adbe8f69b64 +size 497878 diff --git a/vlm/train/jQSBcVURlpW/11.png b/vlm/train/jQSBcVURlpW/11.png new file mode 100644 index 0000000000000000000000000000000000000000..f92c3d58479acb089e76ccb2299ec1812c6b7e27 --- /dev/null +++ b/vlm/train/jQSBcVURlpW/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b615c879be8148c335ad616690913c3d489de81853995530f4d05c21b8e2439c +size 441698 diff --git a/vlm/train/jQSBcVURlpW/12.png b/vlm/train/jQSBcVURlpW/12.png new file mode 100644 index 0000000000000000000000000000000000000000..9b159f6f6316ef7fc365eceddd38e1c08627f466 --- /dev/null +++ b/vlm/train/jQSBcVURlpW/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dc6b5f0a3339dbdfbf2234a64d6276a28b433a30f955ccd8e8acfb95db57b591 +size 409584 diff --git a/vlm/train/jQSBcVURlpW/13.png b/vlm/train/jQSBcVURlpW/13.png new file mode 100644 index 0000000000000000000000000000000000000000..f155eacfd08deb4a78e4dd1c6b33a7f737db086d --- /dev/null +++ b/vlm/train/jQSBcVURlpW/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:277a3019934388b9e0726f940071846ad43ca688633b9513bd7aea21e19d094a +size 450813 diff --git a/vlm/train/jQSBcVURlpW/14.png b/vlm/train/jQSBcVURlpW/14.png new file mode 100644 index 0000000000000000000000000000000000000000..7a370f944171dd94a1f14eacde6997e7187c26de --- /dev/null +++ b/vlm/train/jQSBcVURlpW/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0927e94399e4aef425da5e286e061ba0107b012f0ee045ff4745f6027e1c7278 +size 369709 diff --git a/vlm/train/jQSBcVURlpW/15.png b/vlm/train/jQSBcVURlpW/15.png new file mode 100644 index 0000000000000000000000000000000000000000..44b749229fd3c791572dad9363632d15dfd11bdb --- /dev/null +++ b/vlm/train/jQSBcVURlpW/15.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6ce3589c57192e7c7ec2f3f5f839bb87dae4a79634be4e3b588a265217240161 +size 413836 diff --git a/vlm/train/jQSBcVURlpW/16.png b/vlm/train/jQSBcVURlpW/16.png new file mode 100644 index 0000000000000000000000000000000000000000..924e6063f8890bb717dd79838dbbb0e90843602e --- /dev/null +++ b/vlm/train/jQSBcVURlpW/16.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b8821cb2f25e727e4e996a3f54c5d6cb48ef2658e31865788d5c370248f51823 +size 487179 diff --git a/vlm/train/jQSBcVURlpW/2.png b/vlm/train/jQSBcVURlpW/2.png new file mode 100644 index 0000000000000000000000000000000000000000..476783d8a0acaf3a36ef12308c3323353f4d814e --- /dev/null +++ b/vlm/train/jQSBcVURlpW/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f3f04498f46a1e50aef5f6d350692293c4cb9721e9d8e3bc807c07ef6b74a2fd +size 683823 diff --git a/vlm/train/jQSBcVURlpW/3.png b/vlm/train/jQSBcVURlpW/3.png new file mode 100644 index 0000000000000000000000000000000000000000..b258432a821906396a7ac752e648382de491cbf6 --- /dev/null +++ b/vlm/train/jQSBcVURlpW/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7d2aae595305b968a644fafca719e197c1a2c700292270e3a44d97247c1698a5 +size 592362 diff --git a/vlm/train/jQSBcVURlpW/4.png b/vlm/train/jQSBcVURlpW/4.png new file mode 100644 index 0000000000000000000000000000000000000000..355c81d86551e5702a0bb2d9c19cac03349a1779 --- /dev/null +++ b/vlm/train/jQSBcVURlpW/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:69a32736f716e16d252ed5d3d7971f47a8f84b8b6420773f2ed525f672034588 +size 587279 diff --git a/vlm/train/jQSBcVURlpW/5.png b/vlm/train/jQSBcVURlpW/5.png new file mode 100644 index 0000000000000000000000000000000000000000..11347de7baf99a1f6ac6c182129db781f5116b80 --- /dev/null +++ b/vlm/train/jQSBcVURlpW/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0ff0b9b482f96c411cae339d62d826e12ad39e30982a475eca42b5d8030b2e00 +size 547465 diff --git a/vlm/train/jQSBcVURlpW/6.png b/vlm/train/jQSBcVURlpW/6.png new file mode 100644 index 0000000000000000000000000000000000000000..826f3e87c13f77900705cd7b518a3cfffb17673c --- /dev/null +++ b/vlm/train/jQSBcVURlpW/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d141ba56d09f436b4aff9f2b77b308f59302955c6a91df8619f0c7a310ad3f3b +size 661958 diff --git a/vlm/train/jQSBcVURlpW/7.png b/vlm/train/jQSBcVURlpW/7.png new file mode 100644 index 0000000000000000000000000000000000000000..f3fa826fa361b643bcd542a8c950a82c243dd6b5 --- /dev/null +++ b/vlm/train/jQSBcVURlpW/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d1705be1028b0ebbd874e8dd29b0b0005678997b116aab2902935e4e540c7074 +size 705111 diff --git a/vlm/train/jQSBcVURlpW/8.png b/vlm/train/jQSBcVURlpW/8.png new file mode 100644 index 0000000000000000000000000000000000000000..9ccc26e4b9fdbfdd18cde35dd1610fc60925f1f3 --- /dev/null +++ b/vlm/train/jQSBcVURlpW/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:42cda61539b4a210148b2b952017508d5996511a804736c733e7ac3a0b24e16e +size 497676 diff --git a/vlm/train/jQSBcVURlpW/9.png b/vlm/train/jQSBcVURlpW/9.png new file mode 100644 index 0000000000000000000000000000000000000000..d66163fb6ad10802e3f21f7a354c46dd91310cfb --- /dev/null +++ b/vlm/train/jQSBcVURlpW/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d3932b43ad5e90aff905fa990fde48fd96b7108f9954100c15977b69ff9c150e +size 537136 diff --git a/vlm/train/mEdwVCRJuX4/0.png b/vlm/train/mEdwVCRJuX4/0.png new file mode 100644 index 0000000000000000000000000000000000000000..58079d2a3b0eb87d0b3308799a1204867867cb3b --- /dev/null +++ b/vlm/train/mEdwVCRJuX4/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:526e14a7d2e39d08c8812027876f6baed489be588493ea2b809eb965f973772b +size 496238 diff --git a/vlm/train/mEdwVCRJuX4/1.png b/vlm/train/mEdwVCRJuX4/1.png new file mode 100644 index 0000000000000000000000000000000000000000..0c6196710a2801aac66d24d59735d0a206c95b25 --- /dev/null +++ b/vlm/train/mEdwVCRJuX4/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dba17aa49f632aa45a4202bb865a4de2273fc517b3d568e0d752f2796b1aa9a1 +size 573371 diff --git a/vlm/train/mEdwVCRJuX4/10.png b/vlm/train/mEdwVCRJuX4/10.png new file mode 100644 index 0000000000000000000000000000000000000000..657c713fc2464b6ca03ca8e25e551d496d2113dd --- /dev/null +++ b/vlm/train/mEdwVCRJuX4/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dae3b32da4cdb6ba750e9cb076518e952c5ce321d044afe8664b7c1e786f2c31 +size 564607 diff --git a/vlm/train/mEdwVCRJuX4/11.png b/vlm/train/mEdwVCRJuX4/11.png new file mode 100644 index 0000000000000000000000000000000000000000..63184278dff0667eaed930cee15b66ebf64c9cc6 --- /dev/null +++ b/vlm/train/mEdwVCRJuX4/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5c949280ad3e754cefe80ee846b6151338e7e86d1d4dd6d4e2b6171df37539e8 +size 319368 diff --git a/vlm/train/mEdwVCRJuX4/12.png b/vlm/train/mEdwVCRJuX4/12.png new file mode 100644 index 0000000000000000000000000000000000000000..ae42b0ecca27f86ba6e4704ad44b61a6f936e89b --- /dev/null +++ b/vlm/train/mEdwVCRJuX4/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8fd87b1f6eb2df82d475c51a5544cdb82bcb872436b0d60a2987e985fe4e76d9 +size 450847 diff --git a/vlm/train/mEdwVCRJuX4/13.png b/vlm/train/mEdwVCRJuX4/13.png new file mode 100644 index 0000000000000000000000000000000000000000..713903bb2bceab8ca4b56ae2f0e12da42d28d098 --- /dev/null +++ b/vlm/train/mEdwVCRJuX4/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cec9e8a935a6ce327a2154c86846ac9cde45a1847f9cfa47c0722c2d5916b3bd +size 472383 diff --git a/vlm/train/mEdwVCRJuX4/14.png b/vlm/train/mEdwVCRJuX4/14.png new file mode 100644 index 0000000000000000000000000000000000000000..eb76b843b83c7b70532e4975d0648f28de7cbc3d --- /dev/null +++ b/vlm/train/mEdwVCRJuX4/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aab38db064fe44ea88b01e07a4d5d71c931032c3c7eead7967c77e46d938ecb5 +size 419055 diff --git a/vlm/train/mEdwVCRJuX4/15.png b/vlm/train/mEdwVCRJuX4/15.png new file mode 100644 index 0000000000000000000000000000000000000000..26bdb0d0d2475a6516e65af37c62d251fa1f7d14 --- /dev/null +++ b/vlm/train/mEdwVCRJuX4/15.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0e5c33cf0d676fa2d83b018e665e7db12a20ecfbcbf6948bc271b4ec6feceb65 +size 105868 diff --git a/vlm/train/mEdwVCRJuX4/2.png b/vlm/train/mEdwVCRJuX4/2.png new file mode 100644 index 0000000000000000000000000000000000000000..8a09dd4e026e487e108075ef203a46253c0eca9e --- /dev/null +++ b/vlm/train/mEdwVCRJuX4/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0de99832cd7f66e5c7c53905f7c8e23f1b382006c9ef5c845535e8194ffa5029 +size 522262 diff --git a/vlm/train/mEdwVCRJuX4/3.png b/vlm/train/mEdwVCRJuX4/3.png new file mode 100644 index 0000000000000000000000000000000000000000..d28d712ed3984419d0f532f97067e92bdc08cd02 --- /dev/null +++ b/vlm/train/mEdwVCRJuX4/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:df39e4f0f742c84df95a59153ae7d6200ca8c31c07da52643ecc74e9e2ec1b6b +size 503811 diff --git a/vlm/train/mEdwVCRJuX4/4.png b/vlm/train/mEdwVCRJuX4/4.png new file mode 100644 index 0000000000000000000000000000000000000000..8fb5576540e6d252d9ba57b9ae08e5b3aa49ec9c --- /dev/null +++ b/vlm/train/mEdwVCRJuX4/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ab6fb56d230df22f23e618f6bb0454bfdeefb1a006e718fd43b48fb7e63f2818 +size 534415 diff --git a/vlm/train/mEdwVCRJuX4/5.png b/vlm/train/mEdwVCRJuX4/5.png new file mode 100644 index 0000000000000000000000000000000000000000..1b1f1e0ebe137b5860e7da1f310a1d95d544a111 --- /dev/null +++ b/vlm/train/mEdwVCRJuX4/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0c47d81a87dd52e8398b987604cafa2d299ef5fe6efb22cf3809a58b7d9c6166 +size 664220 diff --git a/vlm/train/mEdwVCRJuX4/6.png b/vlm/train/mEdwVCRJuX4/6.png new file mode 100644 index 0000000000000000000000000000000000000000..2d7f2fd14830168e551cfc95de71f9f930e7aa06 --- /dev/null +++ b/vlm/train/mEdwVCRJuX4/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:934c704232c26f3a84e1402552c8737b3258488627c99a3bb44d352d60372e70 +size 479896 diff --git a/vlm/train/mEdwVCRJuX4/7.png b/vlm/train/mEdwVCRJuX4/7.png new file mode 100644 index 0000000000000000000000000000000000000000..2261d7aa5a2a35aa5fbfb180065ce159d40d6b9b --- /dev/null +++ b/vlm/train/mEdwVCRJuX4/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2de550b7e8696f6767b60fc57e918930c760e3bb625f53b2edf7a5d0bc0f1d73 +size 503546 diff --git a/vlm/train/mEdwVCRJuX4/8.png b/vlm/train/mEdwVCRJuX4/8.png new file mode 100644 index 0000000000000000000000000000000000000000..86e09622ef9621e54b1634cd5921d883033c5e82 --- /dev/null +++ b/vlm/train/mEdwVCRJuX4/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:afc77d25e021ff7544193964bf20e5052064c92f145b1721edece16716b522da +size 501552 diff --git a/vlm/train/mEdwVCRJuX4/9.png b/vlm/train/mEdwVCRJuX4/9.png new file mode 100644 index 0000000000000000000000000000000000000000..8bf079aba312c2a0f685a363c8d3ab6c0dd7ac59 --- /dev/null +++ b/vlm/train/mEdwVCRJuX4/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f38c7a89e96e3b6488801ec590f63a6cec27c5a9a3216c0c8c0e97b07afa6468 +size 546346 diff --git a/vlm/train/r1f0YiCctm/0.png b/vlm/train/r1f0YiCctm/0.png new file mode 100644 index 0000000000000000000000000000000000000000..ea4b413b7943d153942fc9ad8df89371a0bd7cc3 --- /dev/null +++ b/vlm/train/r1f0YiCctm/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e31d21c2da1e250233e3bbbb9434c8cf56ca6327fff2cf2f75f1591e645fe7fd +size 494699 diff --git a/vlm/train/r1f0YiCctm/1.png b/vlm/train/r1f0YiCctm/1.png new file mode 100644 index 0000000000000000000000000000000000000000..bf3309266e4e98bbac8e446cc3d19ccdb3719138 --- /dev/null +++ b/vlm/train/r1f0YiCctm/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:02ff4b4b49363fefce7fa01630853da48f932a4542ed785508e7ceca19e55b6e +size 548534 diff --git a/vlm/train/r1f0YiCctm/10.png b/vlm/train/r1f0YiCctm/10.png new file mode 100644 index 0000000000000000000000000000000000000000..d10919a33cc13ac5a8a842e60daba4c0ee2b0aee --- /dev/null +++ b/vlm/train/r1f0YiCctm/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bced9fd424f91bbe68b1a5314c1278d79caf32d511d7c8cafe60a36d33fa1211 +size 113450 diff --git a/vlm/train/r1f0YiCctm/2.png b/vlm/train/r1f0YiCctm/2.png new file mode 100644 index 0000000000000000000000000000000000000000..1bf65bf7d9fe442e0bc4e5023911b3fd975eb5d8 --- /dev/null +++ b/vlm/train/r1f0YiCctm/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a496bfa3806ecc715a6385aa3729b0ec2be5a61171f399cd2c6a6cce6aeda497 +size 652926 diff --git a/vlm/train/r1f0YiCctm/3.png b/vlm/train/r1f0YiCctm/3.png new file mode 100644 index 0000000000000000000000000000000000000000..cf80817f279c587659c39a704517443d2b949127 --- /dev/null +++ b/vlm/train/r1f0YiCctm/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1cef2fce95ae2fe6ba3b4399ea14a319cbdaaa28728a4effa8c10841faec46b6 +size 579039 diff --git a/vlm/train/r1f0YiCctm/4.png b/vlm/train/r1f0YiCctm/4.png new file mode 100644 index 0000000000000000000000000000000000000000..a59b06280423772f156254ca591b517619e4d289 --- /dev/null +++ b/vlm/train/r1f0YiCctm/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b67c96979c27959b94d3c4cab3446ec3ab46b45dc714294ccf53755e9f4e49aa +size 495967 diff --git a/vlm/train/r1f0YiCctm/5.png b/vlm/train/r1f0YiCctm/5.png new file mode 100644 index 0000000000000000000000000000000000000000..eadc4f3493bc591340d680537faea3eaead82c0e --- /dev/null +++ b/vlm/train/r1f0YiCctm/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:61e76b9e2a1ed77bc6c1f412dec5472a90844733f78e1a823f65f65bc57338b3 +size 446412 diff --git a/vlm/train/r1f0YiCctm/6.png b/vlm/train/r1f0YiCctm/6.png new file mode 100644 index 0000000000000000000000000000000000000000..4230964645acb2fd8dfa7cc3374e259d1da532d6 --- /dev/null +++ b/vlm/train/r1f0YiCctm/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dc754d522edb5df9413aeb136cdec37f1ba7cd3e8d890123217c3fe7cdfce8b8 +size 486233 diff --git a/vlm/train/r1f0YiCctm/7.png b/vlm/train/r1f0YiCctm/7.png new file mode 100644 index 0000000000000000000000000000000000000000..3bf73708bb62b1d3d2ee84232513ba4b9bf9f603 --- /dev/null +++ b/vlm/train/r1f0YiCctm/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:439d480bd8586e858fe30078e30284e603868a4748fd22db58a18ec6d84c0e22 +size 525412 diff --git a/vlm/train/r1f0YiCctm/8.png b/vlm/train/r1f0YiCctm/8.png new file mode 100644 index 0000000000000000000000000000000000000000..ffe8a44fead12cb0f1e9cb8d1de1cd21e6a0f6c3 --- /dev/null +++ b/vlm/train/r1f0YiCctm/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c4cdf76b3063788a2c7182ba097852e483fef83dbc0214e9a26cbd3cff038ed3 +size 492096 diff --git a/vlm/train/r1f0YiCctm/9.png b/vlm/train/r1f0YiCctm/9.png new file mode 100644 index 0000000000000000000000000000000000000000..3331138f7064e9a9343121814d8458e4cec7c4f6 --- /dev/null +++ b/vlm/train/r1f0YiCctm/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:64b60e1100f01c908071cd88cde615510d98189bcb3ae31b5233c81fbaa131f9 +size 355456 diff --git a/vlm/train/r1gR2sC9FX/0.png b/vlm/train/r1gR2sC9FX/0.png new file mode 100644 index 0000000000000000000000000000000000000000..8c7e3600b0991c1bce7f60c0d75ab91d22064f12 --- /dev/null +++ b/vlm/train/r1gR2sC9FX/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bd41a4a056c5b812f566fe3b8f9dbdbc978b1cb88a62b6a72474207bcbeca1d4 +size 462352 diff --git a/vlm/train/r1gR2sC9FX/1.png b/vlm/train/r1gR2sC9FX/1.png new file mode 100644 index 0000000000000000000000000000000000000000..3ce3660be248967a7c540e738222f4e2cbe354ea --- /dev/null +++ b/vlm/train/r1gR2sC9FX/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b4f852e8e0c9bcc0023f208be37a33b530c29bde2926f1a293b102b7bdc516b3 +size 439695 diff --git a/vlm/train/r1gR2sC9FX/10.png b/vlm/train/r1gR2sC9FX/10.png new file mode 100644 index 0000000000000000000000000000000000000000..36f34db0eef1048fd1edbae59fc432b64a108baf --- /dev/null +++ b/vlm/train/r1gR2sC9FX/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d6b561623da1c4a858844612340c2f50b9f9ca58962f7d6572482d25536856b1 +size 534425 diff --git a/vlm/train/r1gR2sC9FX/11.png b/vlm/train/r1gR2sC9FX/11.png new file mode 100644 index 0000000000000000000000000000000000000000..9cb53a82584b2ee8f8f9b57a47a5da2f5465ddb1 --- /dev/null +++ b/vlm/train/r1gR2sC9FX/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b4db23aa40492783a50078cafe11d0d70ca70bf4f881609261b7f89e4e5eb35e +size 480563 diff --git a/vlm/train/r1gR2sC9FX/12.png b/vlm/train/r1gR2sC9FX/12.png new file mode 100644 index 0000000000000000000000000000000000000000..7c587b102baa8f259475d10bba3b70ab485de711 --- /dev/null +++ b/vlm/train/r1gR2sC9FX/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4bada713fa6f5c2a0b5aba3635ca9b68415894b42e2163259b50348042c54ac3 +size 467891 diff --git a/vlm/train/r1gR2sC9FX/13.png b/vlm/train/r1gR2sC9FX/13.png new file mode 100644 index 0000000000000000000000000000000000000000..6f9647b671d62432fa244b44bc3f7d991a91aec5 --- /dev/null +++ b/vlm/train/r1gR2sC9FX/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e73d91d3c5a0e74c16516e59d84961aad444407a47696284534d8616f772dda5 +size 462218 diff --git a/vlm/train/r1gR2sC9FX/14.png b/vlm/train/r1gR2sC9FX/14.png new file mode 100644 index 0000000000000000000000000000000000000000..b2ffbf65668ec768ea721cb4e1c7a8f010ea9c73 --- /dev/null +++ b/vlm/train/r1gR2sC9FX/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7d74ae873a0c8176244938164a7ff1c52877fe0647dd1bfe7b780aea3e6c94b6 +size 389980 diff --git a/vlm/train/r1gR2sC9FX/15.png b/vlm/train/r1gR2sC9FX/15.png new file mode 100644 index 0000000000000000000000000000000000000000..188f857b844d1374b4d2a5f125365dcd89ec76f6 --- /dev/null +++ b/vlm/train/r1gR2sC9FX/15.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:65f6e18cf693b231581b86a638f98d158a7dfc747621925e4aebe9459c1b7689 +size 470619 diff --git a/vlm/train/r1gR2sC9FX/16.png b/vlm/train/r1gR2sC9FX/16.png new file mode 100644 index 0000000000000000000000000000000000000000..c4382da759bb5ee7fb990898ffec92735a2a9f05 --- /dev/null +++ b/vlm/train/r1gR2sC9FX/16.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bd2dc97faa80157c82195d4908c30578bfb587a5f14d77bd770bd5ed4d915fcb +size 931544 diff --git a/vlm/train/r1gR2sC9FX/17.png b/vlm/train/r1gR2sC9FX/17.png new file mode 100644 index 0000000000000000000000000000000000000000..d39d0f68ab34d5438b41da90bb561defe45f691f --- /dev/null +++ b/vlm/train/r1gR2sC9FX/17.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fc859d5b12f9c1d5c32662b42060201ffe434bc889f35faa0928c0bd1d0b7f4e +size 986582 diff --git a/vlm/train/r1gR2sC9FX/18.png b/vlm/train/r1gR2sC9FX/18.png new file mode 100644 index 0000000000000000000000000000000000000000..68117ddfdbaad2c8d1ada0040450403ba58cd540 --- /dev/null +++ b/vlm/train/r1gR2sC9FX/18.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ee8bae630f90393b5516555fdc0d24230d1c0d7474814c41ae6ee0242c40b425 +size 467377 diff --git a/vlm/train/r1gR2sC9FX/19.png b/vlm/train/r1gR2sC9FX/19.png new file mode 100644 index 0000000000000000000000000000000000000000..4ac7704c97aac55b83ceacd53537eace0b3a2e87 --- /dev/null +++ b/vlm/train/r1gR2sC9FX/19.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cc0dfaeb7da508708ce94057f8a41300128eac0507e382a72933d7c7b19b730f +size 462090 diff --git a/vlm/train/r1gR2sC9FX/2.png b/vlm/train/r1gR2sC9FX/2.png new file mode 100644 index 0000000000000000000000000000000000000000..28bc4b65b006444212520f4b3c96d6347480b707 --- /dev/null +++ b/vlm/train/r1gR2sC9FX/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3127b4345d6b8e4603261d78fe353e68f61a69ba18d1af4141f3c283faebd867 +size 489158 diff --git a/vlm/train/r1gR2sC9FX/20.png b/vlm/train/r1gR2sC9FX/20.png new file mode 100644 index 0000000000000000000000000000000000000000..871314404a2751d2b94ce6b4952bc42dc467ce6b --- /dev/null +++ b/vlm/train/r1gR2sC9FX/20.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3d442854ddbf1b2b99b049ee7d156331750bd8b45fa08fc8573aef324b9c4cca +size 443737 diff --git a/vlm/train/r1gR2sC9FX/21.png b/vlm/train/r1gR2sC9FX/21.png new file mode 100644 index 0000000000000000000000000000000000000000..a379b1d8911828177f93472c922c9aa601a38d9f --- /dev/null +++ b/vlm/train/r1gR2sC9FX/21.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:28c5112e565b6a8d30f35fed8523dc2b6fdb6fdb30ebc14e2b9bffe9d6993d28 +size 416453 diff --git a/vlm/train/r1gR2sC9FX/22.png b/vlm/train/r1gR2sC9FX/22.png new file mode 100644 index 0000000000000000000000000000000000000000..6e6314ae1313183efc120a7774355ace6356fd62 --- /dev/null +++ b/vlm/train/r1gR2sC9FX/22.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c2fb6a3ee90d4323448c04f77736fa762cf124aaebf77391f9431cad07921cd9 +size 331869 diff --git a/vlm/train/r1gR2sC9FX/23.png b/vlm/train/r1gR2sC9FX/23.png new file mode 100644 index 0000000000000000000000000000000000000000..3975d89f8053c24c67cc0d2c60ab4ca28b1b201b --- /dev/null +++ b/vlm/train/r1gR2sC9FX/23.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:673f8b86c99e3c475ccd7a1149a24c2b1795be74640dadc0b2785635a09aa0b4 +size 375933 diff --git a/vlm/train/r1gR2sC9FX/3.png b/vlm/train/r1gR2sC9FX/3.png new file mode 100644 index 0000000000000000000000000000000000000000..bccd782dc16e4e04e727e5e4533c0ab441d838cf --- /dev/null +++ b/vlm/train/r1gR2sC9FX/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ee8861d3cc777ac7937107b9cadfec5666dca35b90a27eba933192663cad3fa5 +size 581542 diff --git a/vlm/train/r1gR2sC9FX/4.png b/vlm/train/r1gR2sC9FX/4.png new file mode 100644 index 0000000000000000000000000000000000000000..aced2b94e02b64bb908cbd62e9bdd157d252f0cf --- /dev/null +++ b/vlm/train/r1gR2sC9FX/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f44f8626905e0494612e49478cf0d1dc45ef1b722bbb6f55fa54816e6cb47384 +size 504483 diff --git a/vlm/train/r1gR2sC9FX/5.png b/vlm/train/r1gR2sC9FX/5.png new file mode 100644 index 0000000000000000000000000000000000000000..47f4583cecc0213627f5f61358272cb9aa0d369f --- /dev/null +++ b/vlm/train/r1gR2sC9FX/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:299c82e7fae7846a955a728866551a9c5a9dfa341c3c5e5d27c4cb7aa5a4365d +size 898051 diff --git a/vlm/train/r1gR2sC9FX/6.png b/vlm/train/r1gR2sC9FX/6.png new file mode 100644 index 0000000000000000000000000000000000000000..2a775c1a52ada21f88efbc6541b2f3e889ccdec7 --- /dev/null +++ b/vlm/train/r1gR2sC9FX/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:684a8e114a1d9f7fee0abaeb16727abb6f0b3720767a21ea9bf3fae4887017ae +size 245599 diff --git a/vlm/train/r1gR2sC9FX/7.png b/vlm/train/r1gR2sC9FX/7.png new file mode 100644 index 0000000000000000000000000000000000000000..3a9793db1d47f909d1078e372b6d4b8b9ca06934 --- /dev/null +++ b/vlm/train/r1gR2sC9FX/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d16115b6584a43a892c00d21ddaadeeb988d41a62dc950b2b9885e5f7b5918c4 +size 455342 diff --git a/vlm/train/r1gR2sC9FX/8.png b/vlm/train/r1gR2sC9FX/8.png new file mode 100644 index 0000000000000000000000000000000000000000..699ce70ddd92f6ccb8184081cf12a4c3c15bf530 --- /dev/null +++ b/vlm/train/r1gR2sC9FX/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e052c2a4c75e7d94fd28f5d514575f275964e5cbc761a510c6b561131236175b +size 410767 diff --git a/vlm/train/r1gR2sC9FX/9.png b/vlm/train/r1gR2sC9FX/9.png new file mode 100644 index 0000000000000000000000000000000000000000..467b771e23e04b3b5fdae32c1e4f17d74f439ba0 --- /dev/null +++ b/vlm/train/r1gR2sC9FX/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6c8f9b7f1e37772782add710f0746fc3b8e55983b606399827a6ce5fc84f0d8a +size 562720 diff --git a/vlm/train/r1genAVKPB/0.png b/vlm/train/r1genAVKPB/0.png new file mode 100644 index 0000000000000000000000000000000000000000..ff793ea6468ec212495523a94a085258aa4ad77a --- /dev/null +++ b/vlm/train/r1genAVKPB/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dd12d9bebc934895eebf1edd91b070946b32bbfcf0db0b9bd0111092388d1f0e +size 460560 diff --git a/vlm/train/r1genAVKPB/1.png b/vlm/train/r1genAVKPB/1.png new file mode 100644 index 0000000000000000000000000000000000000000..857af127daa52b176594ee84facba3bc63ee3f98 --- /dev/null +++ b/vlm/train/r1genAVKPB/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:80a997b79fbbdedb71e0e358fa11e304aaa787d268942039a774367f722588e1 +size 562999 diff --git a/vlm/train/r1genAVKPB/10.png b/vlm/train/r1genAVKPB/10.png new file mode 100644 index 0000000000000000000000000000000000000000..8407b18c6fae735402ebef5f540403ffef7cd9ea --- /dev/null +++ b/vlm/train/r1genAVKPB/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:189bd42672d4a8615bcded035c67dc5a7a3229f197b20dc4d2b4b7dd19fe48f7 +size 540214 diff --git a/vlm/train/r1genAVKPB/11.png b/vlm/train/r1genAVKPB/11.png new file mode 100644 index 0000000000000000000000000000000000000000..6a5d87400f58a14b26a3de8b867a6c8ca8098bb2 --- /dev/null +++ b/vlm/train/r1genAVKPB/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a7df1e234d65ce6579b77f999c1160e4d9e92ba71d3a2a54220942807cb4147b +size 589955 diff --git a/vlm/train/r1genAVKPB/12.png b/vlm/train/r1genAVKPB/12.png new file mode 100644 index 0000000000000000000000000000000000000000..69c319932215e6ecfe9b75638055a266bbee1b6b --- /dev/null +++ b/vlm/train/r1genAVKPB/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:41052cf3142cf36ed5aacac8b1781c1e3fb745b447f0439bb8d582a75ef308c0 +size 531023 diff --git a/vlm/train/r1genAVKPB/13.png b/vlm/train/r1genAVKPB/13.png new file mode 100644 index 0000000000000000000000000000000000000000..26af03b075364659099596b52473e9e085f523c9 --- /dev/null +++ b/vlm/train/r1genAVKPB/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ea2e39bd6d719de8b0a20551cceaf4cb4adfe9ed810187170672f5c147abe132 +size 408957 diff --git a/vlm/train/r1genAVKPB/14.png b/vlm/train/r1genAVKPB/14.png new file mode 100644 index 0000000000000000000000000000000000000000..caa61e5c547b9a0c6d2b5172eba4cf5449017e41 --- /dev/null +++ b/vlm/train/r1genAVKPB/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8ac91e7580500e81636fbe368b45bb0f73631d52ac688b5d01fb550fae57707a +size 585250 diff --git a/vlm/train/r1genAVKPB/15.png b/vlm/train/r1genAVKPB/15.png new file mode 100644 index 0000000000000000000000000000000000000000..c5d87dc6e661ba2fc01f479759599d6f2791a3e6 --- /dev/null +++ b/vlm/train/r1genAVKPB/15.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fde4fab1cf89e4c9632d92d229bc3deba8b17843779b978eb1ce18e0f3352da9 +size 121488 diff --git a/vlm/train/r1genAVKPB/16.png b/vlm/train/r1genAVKPB/16.png new file mode 100644 index 0000000000000000000000000000000000000000..ffed5024ca2ce6bcbb0038d579bb21886942b630 --- /dev/null +++ b/vlm/train/r1genAVKPB/16.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2056b7191c6cd509e295b2c39ce3a2c854b56d0481d11425eb038bf0b5840723 +size 548775 diff --git a/vlm/train/r1genAVKPB/17.png b/vlm/train/r1genAVKPB/17.png new file mode 100644 index 0000000000000000000000000000000000000000..e7be509bd576fcf984f0c83bc1d0602ba9865f72 --- /dev/null +++ b/vlm/train/r1genAVKPB/17.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fb7e830e35e3a75c9f9093ee5f210ea96d57a0172077a3e36b719d6a4c1a6fb7 +size 433786 diff --git a/vlm/train/r1genAVKPB/18.png b/vlm/train/r1genAVKPB/18.png new file mode 100644 index 0000000000000000000000000000000000000000..4a4960eefe2d7ccb96be512cbe072451f6c6e546 --- /dev/null +++ b/vlm/train/r1genAVKPB/18.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1a966b92acb36d794de35b906855b21152f89137f6f5ca7f579f411ea70e0bd7 +size 495720 diff --git a/vlm/train/r1genAVKPB/19.png b/vlm/train/r1genAVKPB/19.png new file mode 100644 index 0000000000000000000000000000000000000000..14f4902e47244bc4cc0eb693eaacea3f8719d826 --- /dev/null +++ b/vlm/train/r1genAVKPB/19.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4f5c9193b67ec45e51a292464c5fb0f6a30a804c9bf308c37ccaca254174106a +size 27733 diff --git a/vlm/train/r1genAVKPB/2.png b/vlm/train/r1genAVKPB/2.png new file mode 100644 index 0000000000000000000000000000000000000000..7a5e872c423426ecfcfd94c4924c8035cb541e76 --- /dev/null +++ b/vlm/train/r1genAVKPB/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:58d8a35b0575fd4b5df71420845c763d8f59c9e3435c2f78f226770da270626d +size 508600 diff --git a/vlm/train/r1genAVKPB/3.png b/vlm/train/r1genAVKPB/3.png new file mode 100644 index 0000000000000000000000000000000000000000..02ccc4a233a3bcd317af9578b82b4483a8c96f36 --- /dev/null +++ b/vlm/train/r1genAVKPB/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:21894a0c7f7d39297e12b794fb85d036fea6ef69bdbd71a6aceb715f3860fa9e +size 303532 diff --git a/vlm/train/r1genAVKPB/4.png b/vlm/train/r1genAVKPB/4.png new file mode 100644 index 0000000000000000000000000000000000000000..dc1b1758c92c3e9d8456ad6ec0f47739f179f44e --- /dev/null +++ b/vlm/train/r1genAVKPB/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2105a99876ae16b7b1d78e62673926b36c92fab9620a020d1b72d177bb4b8f47 +size 640592 diff --git a/vlm/train/r1genAVKPB/5.png b/vlm/train/r1genAVKPB/5.png new file mode 100644 index 0000000000000000000000000000000000000000..1319d585dd3a4cc0e37764e30dc53a4e71b810ab --- /dev/null +++ b/vlm/train/r1genAVKPB/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:128a62bfd7706013608f5b3fac8b3ed64e4d6bbb37dc0f258225ce336835afc2 +size 512885 diff --git a/vlm/train/r1genAVKPB/6.png b/vlm/train/r1genAVKPB/6.png new file mode 100644 index 0000000000000000000000000000000000000000..c0481be1b01f62f2834832c49f11ce262224d79c --- /dev/null +++ b/vlm/train/r1genAVKPB/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a1d8c572cb621b4eca73a386e25ea572010e5887018cb0657f77e8e61378a116 +size 573865 diff --git a/vlm/train/r1genAVKPB/7.png b/vlm/train/r1genAVKPB/7.png new file mode 100644 index 0000000000000000000000000000000000000000..b9c2548c455efb6e9350f5d065d177df3455e371 --- /dev/null +++ b/vlm/train/r1genAVKPB/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6ba8ec0535f329d937e949eb24f5103561d57b3fd595f7871120b2312cb0cff6 +size 565237 diff --git a/vlm/train/r1genAVKPB/8.png b/vlm/train/r1genAVKPB/8.png new file mode 100644 index 0000000000000000000000000000000000000000..0de7c2385e3d15cc06ea89ced9c316dbe703eb05 --- /dev/null +++ b/vlm/train/r1genAVKPB/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:544a4c1543edd843d004eeb40689d0b340bc07dbbe35c20ef91daf8db62103a9 +size 644970 diff --git a/vlm/train/r1genAVKPB/9.png b/vlm/train/r1genAVKPB/9.png new file mode 100644 index 0000000000000000000000000000000000000000..ff16d2dcc98e242e7cc9ebdd9400a48d31814c84 --- /dev/null +++ b/vlm/train/r1genAVKPB/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b69e88bff4ab5bbd4902f4459ad4b0b728102cc0dcf3549ea9bba8c1e4775c36 +size 555602 diff --git a/vlm/train/r1xwKoR9Y7/0.png b/vlm/train/r1xwKoR9Y7/0.png new file mode 100644 index 0000000000000000000000000000000000000000..f77712b4dd2e4fed83b7f17a58ab3b7b3d14f737 --- /dev/null +++ b/vlm/train/r1xwKoR9Y7/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ca22ff83703161a9f88f6e161e3b5cb6d2aa64490347a007d8a5702f3b981da1 +size 481388 diff --git a/vlm/train/r1xwKoR9Y7/1.png b/vlm/train/r1xwKoR9Y7/1.png new file mode 100644 index 0000000000000000000000000000000000000000..cf4bed2edc1f71c0a59b600270297e8df1395828 --- /dev/null +++ b/vlm/train/r1xwKoR9Y7/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d335c9583f421a54911d90296218633668d39159d73595463de90179223b55f6 +size 510849 diff --git a/vlm/train/r1xwKoR9Y7/10.png b/vlm/train/r1xwKoR9Y7/10.png new file mode 100644 index 0000000000000000000000000000000000000000..02f0aabde519a6a65e0c77e9fbc178d5f323801d --- /dev/null +++ b/vlm/train/r1xwKoR9Y7/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a7105a40f58be5d6b783f66ac2fd8a192dd797d5a704ca9ccc163d8b8b8084ff +size 542589 diff --git a/vlm/train/r1xwKoR9Y7/11.png b/vlm/train/r1xwKoR9Y7/11.png new file mode 100644 index 0000000000000000000000000000000000000000..38141f5b1e64b2093554aae15d5e15d87be9339c --- /dev/null +++ b/vlm/train/r1xwKoR9Y7/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7b820cfb46d8b77d2c757eeca405216d4ae9e8baf056b15597e7594a43190fb4 +size 109893 diff --git a/vlm/train/r1xwKoR9Y7/2.png b/vlm/train/r1xwKoR9Y7/2.png new file mode 100644 index 0000000000000000000000000000000000000000..b85c0727a131b9c33ae0b9e7415608af04cfa62c --- /dev/null +++ b/vlm/train/r1xwKoR9Y7/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6451191c45e5b25753c738d118bc41fed285f6c7b78b9846e4692a7ced5bafd0 +size 586255 diff --git a/vlm/train/r1xwKoR9Y7/3.png b/vlm/train/r1xwKoR9Y7/3.png new file mode 100644 index 0000000000000000000000000000000000000000..c7487d75615e777ee2e31419a1a7762339a4109e --- /dev/null +++ b/vlm/train/r1xwKoR9Y7/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f3c9521cec69d0409c6f9b01f3a8dac17e9aa230a5dc6269f9e1ea9e0435f783 +size 549574 diff --git a/vlm/train/r1xwKoR9Y7/4.png b/vlm/train/r1xwKoR9Y7/4.png new file mode 100644 index 0000000000000000000000000000000000000000..5fb06dd110ba4224490244ca6c359ee5d68e5e20 --- /dev/null +++ b/vlm/train/r1xwKoR9Y7/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6a375960bb662e450c0df4dc5e16c37370df3a7b2cea15e624da6e3ef82258b9 +size 560034 diff --git a/vlm/train/r1xwKoR9Y7/5.png b/vlm/train/r1xwKoR9Y7/5.png new file mode 100644 index 0000000000000000000000000000000000000000..6cb4df7124a9898cef8bc258ce36ef69dcf16e70 --- /dev/null +++ b/vlm/train/r1xwKoR9Y7/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7a1e6032a55c5ab2be519f93fe8ac8cf854fefb157c6bf55aaab7a725168dda7 +size 514694 diff --git a/vlm/train/r1xwKoR9Y7/6.png b/vlm/train/r1xwKoR9Y7/6.png new file mode 100644 index 0000000000000000000000000000000000000000..f1e03208d51ff5d7dcc71c747ad3e955e3d3b2b2 --- /dev/null +++ b/vlm/train/r1xwKoR9Y7/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:970426adcf983a7b5ab22bcfe9fe5e0ae6e3ded3d455a4c6b0c440f41025b51d +size 530848 diff --git a/vlm/train/r1xwKoR9Y7/7.png b/vlm/train/r1xwKoR9Y7/7.png new file mode 100644 index 0000000000000000000000000000000000000000..50bd0df1769784bed3012f1fd21378eaad4a9787 --- /dev/null +++ b/vlm/train/r1xwKoR9Y7/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:94e9e4efac2faf7855d48cdbd5d6b5b0b7ca5b3370f25d18c7f9bfc6bb14b9b9 +size 526808 diff --git a/vlm/train/r1xwKoR9Y7/8.png b/vlm/train/r1xwKoR9Y7/8.png new file mode 100644 index 0000000000000000000000000000000000000000..e292b01e733adadcf0ec0093fb6dc6e0312e5ce9 --- /dev/null +++ b/vlm/train/r1xwKoR9Y7/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6e002bc3e1f7a600675a7f44f5ecdb719d4f90691f35202f5c90afadeb7cc568 +size 597008 diff --git a/vlm/train/r1xwKoR9Y7/9.png b/vlm/train/r1xwKoR9Y7/9.png new file mode 100644 index 0000000000000000000000000000000000000000..898926b190e19406dda6996ea6c27b72af1d792a --- /dev/null +++ b/vlm/train/r1xwKoR9Y7/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:55218f1725f5356c0c7cfc65f03878f8138e0b24265aed416eac29ebb1cfd68a +size 484192 diff --git a/vlm/train/rJVorjCcKQ/0.png b/vlm/train/rJVorjCcKQ/0.png new file mode 100644 index 0000000000000000000000000000000000000000..13701818bb3d0464f8249ae93bd8cf4cd7189933 --- /dev/null +++ b/vlm/train/rJVorjCcKQ/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:36a2c00fb84631eeae8fe7198951dc379981013a255ed1ea81aaf63bf2f0b772 +size 455463 diff --git a/vlm/train/rJVorjCcKQ/1.png b/vlm/train/rJVorjCcKQ/1.png new file mode 100644 index 0000000000000000000000000000000000000000..993d4d0c9f7aec97cd4280a7fce71a96448fcb9e --- /dev/null +++ b/vlm/train/rJVorjCcKQ/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:abb3309f55c70945a10d5909a36c2ee89916fb7f4f9b2067b09792390990f774 +size 528010 diff --git a/vlm/train/rJVorjCcKQ/10.png b/vlm/train/rJVorjCcKQ/10.png new file mode 100644 index 0000000000000000000000000000000000000000..12ef94471d13dd4be00f6e3109e405f5fe9402db --- /dev/null +++ b/vlm/train/rJVorjCcKQ/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cfaacfe15ba7970d4c33823ed2dc849a0691cc917d8cb8347e953a3c39a0e468 +size 502708 diff --git a/vlm/train/rJVorjCcKQ/11.png b/vlm/train/rJVorjCcKQ/11.png new file mode 100644 index 0000000000000000000000000000000000000000..35c2c221b0749f10ceec0d1a01e640bccd944519 --- /dev/null +++ b/vlm/train/rJVorjCcKQ/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b724ada05b2e68e522a325fb4be276ab9727f4fb64c7748228ff7b254c77fa67 +size 494258 diff --git a/vlm/train/rJVorjCcKQ/12.png b/vlm/train/rJVorjCcKQ/12.png new file mode 100644 index 0000000000000000000000000000000000000000..5588a90d50db9fedc80136eb225b52739cc39e97 --- /dev/null +++ b/vlm/train/rJVorjCcKQ/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:23a6144003fa243d1ad4e2e792734b51718a54c363539fcb0dda0a5138f950ec +size 529169 diff --git a/vlm/train/rJVorjCcKQ/13.png b/vlm/train/rJVorjCcKQ/13.png new file mode 100644 index 0000000000000000000000000000000000000000..5086af43b1e1fabaadbd6785c44b4bfa43c21d3a --- /dev/null +++ b/vlm/train/rJVorjCcKQ/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:099e8a195df28599b4994432ec6c10d589db8e3bcbe93c67a164a24b25d7c89c +size 521391 diff --git a/vlm/train/rJVorjCcKQ/14.png b/vlm/train/rJVorjCcKQ/14.png new file mode 100644 index 0000000000000000000000000000000000000000..f557040bfbae8a0f9bd41d8af19a9c3088681390 --- /dev/null +++ b/vlm/train/rJVorjCcKQ/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f0184d02fa1fc96a96b95dbef16e604609117bc00e5ff39a83a46d932320b3ea +size 523951 diff --git a/vlm/train/rJVorjCcKQ/15.png b/vlm/train/rJVorjCcKQ/15.png new file mode 100644 index 0000000000000000000000000000000000000000..227d07799e722c352f6cc8a41be0cbf2aa95de9f --- /dev/null +++ b/vlm/train/rJVorjCcKQ/15.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ac6aa1c6e7eae13894313db65e0c064b6d5f5c25524fcad0bce64e9faaa63319 +size 417456 diff --git a/vlm/train/rJVorjCcKQ/16.png b/vlm/train/rJVorjCcKQ/16.png new file mode 100644 index 0000000000000000000000000000000000000000..6c23795ce5f68831a04e884aa4f47c7802267ada --- /dev/null +++ b/vlm/train/rJVorjCcKQ/16.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:61afb688af2e48b7cd8da38c5b1bf039a30a1f77efab7c82995edc513a72ca40 +size 486405 diff --git a/vlm/train/rJVorjCcKQ/17.png b/vlm/train/rJVorjCcKQ/17.png new file mode 100644 index 0000000000000000000000000000000000000000..62435f9b404cd584a2313a1e8d4496b44bddaf10 --- /dev/null +++ b/vlm/train/rJVorjCcKQ/17.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7568c9e30b4bc836cf3e00767037c02483808967482bcd26f6ba636df835f035 +size 443129 diff --git a/vlm/train/rJVorjCcKQ/18.png b/vlm/train/rJVorjCcKQ/18.png new file mode 100644 index 0000000000000000000000000000000000000000..332373c6602a4a1149d1fb83b586bdc8e32a835b --- /dev/null +++ b/vlm/train/rJVorjCcKQ/18.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:80cda25878fb15e58189a1196cd458a4dee8c154ee1916362a76106ed50d35b1 +size 123951 diff --git a/vlm/train/rJVorjCcKQ/2.png b/vlm/train/rJVorjCcKQ/2.png new file mode 100644 index 0000000000000000000000000000000000000000..ebc5b49facfc4dfadb5861e4047261f22bbba904 --- /dev/null +++ b/vlm/train/rJVorjCcKQ/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7a8060d27d3b8fcc01ccedb519cf716e2f88b9cee2173b4788757c50c3c43a01 +size 537394 diff --git a/vlm/train/rJVorjCcKQ/3.png b/vlm/train/rJVorjCcKQ/3.png new file mode 100644 index 0000000000000000000000000000000000000000..091bac7affdf68b7948cfb2c74a699bf2093c373 --- /dev/null +++ b/vlm/train/rJVorjCcKQ/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:21d400f5f966b6b87225df0cbf2b0e1d425f1c3030fd50afa6d7d16d3ac631c6 +size 476940 diff --git a/vlm/train/rJVorjCcKQ/4.png b/vlm/train/rJVorjCcKQ/4.png new file mode 100644 index 0000000000000000000000000000000000000000..93daa6a723c86863ef7ab904a3fc270b95d00988 --- /dev/null +++ b/vlm/train/rJVorjCcKQ/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9d1fbab5a0e2c1061eda0796e3e20dd90621a02a4d8c257b2a153d440e390205 +size 466790 diff --git a/vlm/train/rJVorjCcKQ/5.png b/vlm/train/rJVorjCcKQ/5.png new file mode 100644 index 0000000000000000000000000000000000000000..6ba656f96d2b0e8ab904adce9f2881eb18b13572 --- /dev/null +++ b/vlm/train/rJVorjCcKQ/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5e182fd19bacb195c8651afaa9bcaff2f5ed8062fa9c7ae968138fa775027b1a +size 544907 diff --git a/vlm/train/rJVorjCcKQ/6.png b/vlm/train/rJVorjCcKQ/6.png new file mode 100644 index 0000000000000000000000000000000000000000..eeead4cead11dc634fe4633f6232831679d6935e --- /dev/null +++ b/vlm/train/rJVorjCcKQ/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6322cd1841fa80b3da19247bb418fed4f62f0115002315a13fa95e03d31e7690 +size 494753 diff --git a/vlm/train/rJVorjCcKQ/7.png b/vlm/train/rJVorjCcKQ/7.png new file mode 100644 index 0000000000000000000000000000000000000000..a6d60261d0ee035d625b233b942f47c238e19d4a --- /dev/null +++ b/vlm/train/rJVorjCcKQ/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e4a9daf51b33595233667e7a1e4d420c1747555c160130fc6e4920099efff984 +size 551553 diff --git a/vlm/train/rJVorjCcKQ/8.png b/vlm/train/rJVorjCcKQ/8.png new file mode 100644 index 0000000000000000000000000000000000000000..4bf0b2322928492b6b670066cf906a5a7434c14c --- /dev/null +++ b/vlm/train/rJVorjCcKQ/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:83b5f5f7e58fff294fa2ede69976d1aec6987e46bdc476268b4df3fd04733da5 +size 539876 diff --git a/vlm/train/rJVorjCcKQ/9.png b/vlm/train/rJVorjCcKQ/9.png new file mode 100644 index 0000000000000000000000000000000000000000..a17e55984e2f6ba677f9baaaa955f1b2641db04a --- /dev/null +++ b/vlm/train/rJVorjCcKQ/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b8d89bd535ec82e189a7129a02ed1a71327e04fd76ebc150c8d0376a060c7e41 +size 466232 diff --git a/vlm/train/rJbbOLcex/0.png b/vlm/train/rJbbOLcex/0.png new file mode 100644 index 0000000000000000000000000000000000000000..833f9a3c0e82860af506d07ae0a362de6fa7f245 --- /dev/null +++ b/vlm/train/rJbbOLcex/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f61c0d7f5a621201a111dd85f90ade0e2df039d2ede0830d26cdc2adfb06cee4 +size 461109 diff --git a/vlm/train/rJbbOLcex/1.png b/vlm/train/rJbbOLcex/1.png new file mode 100644 index 0000000000000000000000000000000000000000..43b3ef7653d1b9e007acab4f41218f9d5025ad85 --- /dev/null +++ b/vlm/train/rJbbOLcex/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:522e18ef54ea657f25989c01f8c6d97d786df19cfab315cc722ff83dda3c21d8 +size 563972 diff --git a/vlm/train/rJbbOLcex/10.png b/vlm/train/rJbbOLcex/10.png new file mode 100644 index 0000000000000000000000000000000000000000..dd833de5be75d64707d78851966a907d0549b131 --- /dev/null +++ b/vlm/train/rJbbOLcex/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aee964f19e2eb7a7698ab826fe8d190de3aa1a1a7696227dbf7c5992e7f4d1b3 +size 492744 diff --git a/vlm/train/rJbbOLcex/11.png b/vlm/train/rJbbOLcex/11.png new file mode 100644 index 0000000000000000000000000000000000000000..2afbd565805a7071154c5758dd10fd1f77a711cc --- /dev/null +++ b/vlm/train/rJbbOLcex/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7dbc46d81b64bcac12a742aeaa6501a7f28d3db16f9dd00f1937c849647b3881 +size 675441 diff --git a/vlm/train/rJbbOLcex/12.png b/vlm/train/rJbbOLcex/12.png new file mode 100644 index 0000000000000000000000000000000000000000..afccd08b4810cb33b6c721c3a562a57cd0be6f7c --- /dev/null +++ b/vlm/train/rJbbOLcex/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f54e8d456f8eebd57f73684af05907a20546903db1a21a40397844970e17ba15 +size 344010 diff --git a/vlm/train/rJbbOLcex/2.png b/vlm/train/rJbbOLcex/2.png new file mode 100644 index 0000000000000000000000000000000000000000..5dcb436f2856a3c17e71db1b7bf16c3171012368 --- /dev/null +++ b/vlm/train/rJbbOLcex/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d58f5b95df38c9e23214c093d0213fe1d029d249a300f7f090ccf74896bff263 +size 513044 diff --git a/vlm/train/rJbbOLcex/3.png b/vlm/train/rJbbOLcex/3.png new file mode 100644 index 0000000000000000000000000000000000000000..a381f75ff51711aedfe373a9181f4eb05b7907f4 --- /dev/null +++ b/vlm/train/rJbbOLcex/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c61aed31f342292017aa7a42753de0d6d8e9e359a712bc686a5bed80f8fa75ce +size 480105 diff --git a/vlm/train/rJbbOLcex/4.png b/vlm/train/rJbbOLcex/4.png new file mode 100644 index 0000000000000000000000000000000000000000..2014b6578af1fc9ad0f30421a32ccb91c95e0fd7 --- /dev/null +++ b/vlm/train/rJbbOLcex/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3c62058870a9d840de0bda52cf72076005d97787dfcf61037e390320350dd7f1 +size 579989 diff --git a/vlm/train/rJbbOLcex/5.png b/vlm/train/rJbbOLcex/5.png new file mode 100644 index 0000000000000000000000000000000000000000..e56f79564f1accb4daeabbdd8d3ccf5d41ccc2cd --- /dev/null +++ b/vlm/train/rJbbOLcex/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e74f675c4c06067785eacb95dca4b560859065d1f7bc55bff809fee7cc3f1b97 +size 402679 diff --git a/vlm/train/rJbbOLcex/6.png b/vlm/train/rJbbOLcex/6.png new file mode 100644 index 0000000000000000000000000000000000000000..8274af9b17e2aded8f773a52ddc31675c5695c06 --- /dev/null +++ b/vlm/train/rJbbOLcex/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:78932f802c59b9754cfd1e765ff52b705e6de4b42fc2a2cbf3976dfe94227863 +size 522190 diff --git a/vlm/train/rJbbOLcex/7.png b/vlm/train/rJbbOLcex/7.png new file mode 100644 index 0000000000000000000000000000000000000000..c69e35b5983369b3c51dce025412adebf9544f70 --- /dev/null +++ b/vlm/train/rJbbOLcex/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bdaa47c97695545876d9a8ae928c1449c8b9e18e2bd01212d75b058312f56b75 +size 522215 diff --git a/vlm/train/rJbbOLcex/8.png b/vlm/train/rJbbOLcex/8.png new file mode 100644 index 0000000000000000000000000000000000000000..34ba4728ee4c772b940ee9126a8c26b937207816 --- /dev/null +++ b/vlm/train/rJbbOLcex/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3d013ef8663cfcb4ad6a6522c9c653e91cb110a74406637e317c3c853ca45ec6 +size 745949 diff --git a/vlm/train/rJbbOLcex/9.png b/vlm/train/rJbbOLcex/9.png new file mode 100644 index 0000000000000000000000000000000000000000..cc87c7101778dcc32a3d51d265d9d1c0050b1024 --- /dev/null +++ b/vlm/train/rJbbOLcex/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c9d675083d04672b855d6db5facea000f7684b47d2bb9e5c5f3659f73ecc63f1 +size 494425 diff --git a/vlm/train/rkxoh24FPH/13.png b/vlm/train/rkxoh24FPH/13.png new file mode 100644 index 0000000000000000000000000000000000000000..c5b273a16d5890a1e97a85aee8140aa40e4c6917 --- /dev/null +++ b/vlm/train/rkxoh24FPH/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aa98cdabf8f03a2abcbf6739945cd841d2eba51da1dbb297d0701372a9c6de97 +size 433243 diff --git a/vlm/train/rkxoh24FPH/4.png b/vlm/train/rkxoh24FPH/4.png new file mode 100644 index 0000000000000000000000000000000000000000..2c0654f3097f57528c8c311ccd8a82368b246227 --- /dev/null +++ b/vlm/train/rkxoh24FPH/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d82376c7e27d20ce141857cb3243f64b6b29e13743e5dcd6bacf0cecf3e2bc71 +size 599741 diff --git a/vlm/train/rkxoh24FPH/7.png b/vlm/train/rkxoh24FPH/7.png new file mode 100644 index 0000000000000000000000000000000000000000..d44774c0e41a1aa8bf7a35e8edf380b6ba98823d --- /dev/null +++ b/vlm/train/rkxoh24FPH/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d5bef0ae35b08028c50c31367bfcf56fad1f1f7312f590f228c5f9d1bfe22f11 +size 610735 diff --git a/vlm/train/rkxoh24FPH/8.png b/vlm/train/rkxoh24FPH/8.png new file mode 100644 index 0000000000000000000000000000000000000000..e14b84788d1f68be786822e64132152a4b6fce6c --- /dev/null +++ b/vlm/train/rkxoh24FPH/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9382b1ca9a6259aaff7b11bd6f2ebad6042c4d74efe07dbb2ed58204dceababe +size 598532 diff --git a/vlm/train/rkzfuiA9F7/0.png b/vlm/train/rkzfuiA9F7/0.png new file mode 100644 index 0000000000000000000000000000000000000000..0ef8b263bd2ced68bb28d97b613ad5451dcfcb5c --- /dev/null +++ b/vlm/train/rkzfuiA9F7/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b047aee3203b0ca9779bfb5dfd9a3bd49cf9eaf5d66e733f8b0c223ff40ed31d +size 485163 diff --git a/vlm/train/rkzfuiA9F7/1.png b/vlm/train/rkzfuiA9F7/1.png new file mode 100644 index 0000000000000000000000000000000000000000..b2328d80ff2d821104005b6143762d0a554cfa04 --- /dev/null +++ b/vlm/train/rkzfuiA9F7/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:14b3313711f3953011a28b49079dae1bf1ee593a4e33f2580efdc9442c8b11eb +size 479988 diff --git a/vlm/train/rkzfuiA9F7/10.png b/vlm/train/rkzfuiA9F7/10.png new file mode 100644 index 0000000000000000000000000000000000000000..49ec3e43700f1a657171fd965ff49a6020da6d9f --- /dev/null +++ b/vlm/train/rkzfuiA9F7/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:16687495a4581dfb9c433b186678d5786111e163ef5c04dbafcfc56705872172 +size 374905 diff --git a/vlm/train/rkzfuiA9F7/11.png b/vlm/train/rkzfuiA9F7/11.png new file mode 100644 index 0000000000000000000000000000000000000000..cb8cc1d041e0d7799b4935c7d553d387d71ea668 --- /dev/null +++ b/vlm/train/rkzfuiA9F7/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:df2ab1e90f4180ed2721b3d642de645d6fda83ab0ef3b163f6c32489d075cbbe +size 159868 diff --git a/vlm/train/rkzfuiA9F7/2.png b/vlm/train/rkzfuiA9F7/2.png new file mode 100644 index 0000000000000000000000000000000000000000..a0825936df81408ab9e7c68a5855db0e5e832ac4 --- /dev/null +++ b/vlm/train/rkzfuiA9F7/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5a62fb1b5e3fb339c0f512cdcab600a759edd9a700d031b2d5ab3bd9d11cbfd5 +size 422613 diff --git a/vlm/train/rkzfuiA9F7/3.png b/vlm/train/rkzfuiA9F7/3.png new file mode 100644 index 0000000000000000000000000000000000000000..d772b6dbfca2e2ae83eb508afa39761aef11d00f --- /dev/null +++ b/vlm/train/rkzfuiA9F7/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:69a8e18f72323d034169a54238bfbda5175e465c47e7394c55a23fa79c98260c +size 513788 diff --git a/vlm/train/rkzfuiA9F7/4.png b/vlm/train/rkzfuiA9F7/4.png new file mode 100644 index 0000000000000000000000000000000000000000..418ca9b0e9ebb790f71a3c8572ed82ead7fd8f9c --- /dev/null +++ b/vlm/train/rkzfuiA9F7/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7929995871ab7dd5f8d2bcf84f8c8317b6386c2692caf16b0cbb5b0d6cd2c9ee +size 398694 diff --git a/vlm/train/rkzfuiA9F7/5.png b/vlm/train/rkzfuiA9F7/5.png new file mode 100644 index 0000000000000000000000000000000000000000..7fcf4267d19356282dcb47df8d2f2e5bec820589 --- /dev/null +++ b/vlm/train/rkzfuiA9F7/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b0ede3d4b28d9556b590e91e97ef534a7679769f7154655499ad4677742d57cd +size 571755 diff --git a/vlm/train/rkzfuiA9F7/6.png b/vlm/train/rkzfuiA9F7/6.png new file mode 100644 index 0000000000000000000000000000000000000000..47b507fa2217b8b6f2b744cefdf47185231f4603 --- /dev/null +++ b/vlm/train/rkzfuiA9F7/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:604a06bdd67f78ee6058eb612e2fed296da4ca954575ca4ac758506f4b75780a +size 575572 diff --git a/vlm/train/rkzfuiA9F7/7.png b/vlm/train/rkzfuiA9F7/7.png new file mode 100644 index 0000000000000000000000000000000000000000..871b3672ff137061ee0f3f01e38fb0ee43150eab --- /dev/null +++ b/vlm/train/rkzfuiA9F7/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dad6a61025bf21f23b90eb8a9a2f38e1f5e4a409b99f9d10e0f703a479d0017a +size 569139 diff --git a/vlm/train/rkzfuiA9F7/8.png b/vlm/train/rkzfuiA9F7/8.png new file mode 100644 index 0000000000000000000000000000000000000000..bdbb3bd4a874802f7d1c10e21f36c223020e5b88 --- /dev/null +++ b/vlm/train/rkzfuiA9F7/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:33fde669a3c2900990cecda84c99a3958019cc495a8da2f37d6f1a7396357fca +size 534626 diff --git a/vlm/train/rkzfuiA9F7/9.png b/vlm/train/rkzfuiA9F7/9.png new file mode 100644 index 0000000000000000000000000000000000000000..afdfef75f2dd431d47865f6f2954aa3050bbb4b2 --- /dev/null +++ b/vlm/train/rkzfuiA9F7/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7aa7daf64f300e529509056399040598c285befa4b7724bed66c907d7b0687b7 +size 432404 diff --git a/vlm/train/rkzjUoAcFX/0.png b/vlm/train/rkzjUoAcFX/0.png new file mode 100644 index 0000000000000000000000000000000000000000..ce2ed4caba0e64a8885c121687bc16c0c01650ba --- /dev/null +++ b/vlm/train/rkzjUoAcFX/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1900e37853e304d7d7b9a3a39044ab1fc31af2d6d497a491752a48254bee4a30 +size 485406 diff --git a/vlm/train/rkzjUoAcFX/1.png b/vlm/train/rkzjUoAcFX/1.png new file mode 100644 index 0000000000000000000000000000000000000000..8dada4b0f4d68ca58a02fd8b74a8bcb6ffd13bad --- /dev/null +++ b/vlm/train/rkzjUoAcFX/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:88bd363d254ef88670a1f09fa3e7e5c2043b49fa05f3d61b92ae0d19a00e2227 +size 471969 diff --git a/vlm/train/rkzjUoAcFX/10.png b/vlm/train/rkzjUoAcFX/10.png new file mode 100644 index 0000000000000000000000000000000000000000..80dc3d1b992cebdd2265e6df3b270da5e102eef8 --- /dev/null +++ b/vlm/train/rkzjUoAcFX/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:afd4b458d894c83b4ccd25330a2a76a924fb652ed0fc1cecf873f27e95c4039c +size 545212 diff --git a/vlm/train/rkzjUoAcFX/11.png b/vlm/train/rkzjUoAcFX/11.png new file mode 100644 index 0000000000000000000000000000000000000000..b672d58484393014566d0c087e0a875442a8bf94 --- /dev/null +++ b/vlm/train/rkzjUoAcFX/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f983334aa41f4e4b0e93eb23126cb5afdf8f363124b3566de8783c0984d07757 +size 384260 diff --git a/vlm/train/rkzjUoAcFX/12.png b/vlm/train/rkzjUoAcFX/12.png new file mode 100644 index 0000000000000000000000000000000000000000..79fc0cc022041167573b92029d58cc259f9fdfd3 --- /dev/null +++ b/vlm/train/rkzjUoAcFX/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4b1a26a384f95ded9fbeb8b7a9e6d0d7062eb99b3c61ca6a76a8575a3aa7dd1b +size 187812 diff --git a/vlm/train/rkzjUoAcFX/13.png b/vlm/train/rkzjUoAcFX/13.png new file mode 100644 index 0000000000000000000000000000000000000000..d8de76faaf68ee225b7c089f93e82eaee94e170e --- /dev/null +++ b/vlm/train/rkzjUoAcFX/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:40ba330da68cf37d3d0129b2b0bda3790fb32d49584004ede4182c5a20c82eb4 +size 271631 diff --git a/vlm/train/rkzjUoAcFX/14.png b/vlm/train/rkzjUoAcFX/14.png new file mode 100644 index 0000000000000000000000000000000000000000..f95a30d293ae47a03682293e7aa14252ece38f05 --- /dev/null +++ b/vlm/train/rkzjUoAcFX/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7760d12ea49ca32b82e13feac241bad978e75f66ae4932f39f9f6d4a60a02a4e +size 256574 diff --git a/vlm/train/rkzjUoAcFX/2.png b/vlm/train/rkzjUoAcFX/2.png new file mode 100644 index 0000000000000000000000000000000000000000..40ae187ee3c8e7d8e646dc78da8ea6609857a3c7 --- /dev/null +++ b/vlm/train/rkzjUoAcFX/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2c77915f223d4dfd7270d65a81e6a3a496031f108dcae4b8667d7aa3843e74d0 +size 490226 diff --git a/vlm/train/rkzjUoAcFX/3.png b/vlm/train/rkzjUoAcFX/3.png new file mode 100644 index 0000000000000000000000000000000000000000..ef6bc275ef9545c925ab8a630ed962de47d4210f --- /dev/null +++ b/vlm/train/rkzjUoAcFX/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1b9d5f9db32b5352e991d91bf0d2c9a034baff713e88dc9fcaab11a5d2ff46a1 +size 505535 diff --git a/vlm/train/rkzjUoAcFX/4.png b/vlm/train/rkzjUoAcFX/4.png new file mode 100644 index 0000000000000000000000000000000000000000..4e7c3a1c3a232a4fd415f9809aa95684c677701d --- /dev/null +++ b/vlm/train/rkzjUoAcFX/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5ac9d5a5aedd87d5bda227606418cc4832f770da0ead09380d92e9734b8d7e0b +size 610966 diff --git a/vlm/train/rkzjUoAcFX/5.png b/vlm/train/rkzjUoAcFX/5.png new file mode 100644 index 0000000000000000000000000000000000000000..0734dd30853df752218fd4ba26b2826edce0dfde --- /dev/null +++ b/vlm/train/rkzjUoAcFX/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8593defee7275b64297ad1459ff16ee2bdc6b1e208dc45f041bfe9eeb8e94e8c +size 557811 diff --git a/vlm/train/rkzjUoAcFX/6.png b/vlm/train/rkzjUoAcFX/6.png new file mode 100644 index 0000000000000000000000000000000000000000..84abc57e70f22baafdd7859b2a7bee22cd26a4de --- /dev/null +++ b/vlm/train/rkzjUoAcFX/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7917f196047cb7fb86934eb1f9f9d0ac93d37506a7fdc71b8e7fc9c1fa89e30f +size 504241 diff --git a/vlm/train/rkzjUoAcFX/7.png b/vlm/train/rkzjUoAcFX/7.png new file mode 100644 index 0000000000000000000000000000000000000000..baaf30ba49e8c40e52b321b95e776399814a8644 --- /dev/null +++ b/vlm/train/rkzjUoAcFX/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f4118c3fc5a273da91bd769d45c63b998b4642843f7b0b59338cd0105b79ab67 +size 401926 diff --git a/vlm/train/rkzjUoAcFX/8.png b/vlm/train/rkzjUoAcFX/8.png new file mode 100644 index 0000000000000000000000000000000000000000..c41bb730d0290f1f60ce17fbe6689a1f0f1a6d5e --- /dev/null +++ b/vlm/train/rkzjUoAcFX/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b0d1354c281fd6a79a7affb257f8a432f1fbe54bcae666fabe35579706bccc2a +size 470922 diff --git a/vlm/train/rkzjUoAcFX/9.png b/vlm/train/rkzjUoAcFX/9.png new file mode 100644 index 0000000000000000000000000000000000000000..ea983e6bf507506f47e3b203eb761eca0891dcfa --- /dev/null +++ b/vlm/train/rkzjUoAcFX/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fbe3c910f0cd2f19fc12a7cfc4bf272a369e19c67d967209e53880108e99e0f7 +size 479172 diff --git a/vlm/train/rqfq0CYIekd/0.png b/vlm/train/rqfq0CYIekd/0.png new file mode 100644 index 0000000000000000000000000000000000000000..1887033d0f9066a1530f890e17d6d603e27a5070 --- /dev/null +++ b/vlm/train/rqfq0CYIekd/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e659147d9d0f11f495cf187f7a219832f363f2a2d264410ddeebf137d94b3bdd +size 449240 diff --git a/vlm/train/rqfq0CYIekd/1.png b/vlm/train/rqfq0CYIekd/1.png new file mode 100644 index 0000000000000000000000000000000000000000..d5e8f4604efafe33e41fd6d4b0ce83ded0c89288 --- /dev/null +++ b/vlm/train/rqfq0CYIekd/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d9f4bb385ba714c9f09932b0e559534b5b271a023757cf326cb64d72532b1e63 +size 572625 diff --git a/vlm/train/rqfq0CYIekd/10.png b/vlm/train/rqfq0CYIekd/10.png new file mode 100644 index 0000000000000000000000000000000000000000..0574c37e2b81c2c64191f0e6acbcb3690ac892de --- /dev/null +++ b/vlm/train/rqfq0CYIekd/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f8798175e8d02be24ad4cfabb6f342155f9b375f625bf5f87b8eaa876783f77e +size 539986 diff --git a/vlm/train/rqfq0CYIekd/11.png b/vlm/train/rqfq0CYIekd/11.png new file mode 100644 index 0000000000000000000000000000000000000000..87cb30efdd4a406ec13ad66dcad76958a9a9b5ea --- /dev/null +++ b/vlm/train/rqfq0CYIekd/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:13ced499de10f9888d3be8b45261487fe5eb16b7cded9c13e57ccecfe6871a25 +size 529638 diff --git a/vlm/train/rqfq0CYIekd/12.png b/vlm/train/rqfq0CYIekd/12.png new file mode 100644 index 0000000000000000000000000000000000000000..954d2941f3db8f9443296e062bd3d5bfdecc86bb --- /dev/null +++ b/vlm/train/rqfq0CYIekd/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:972e0dc2e3b9b3cf82df61f6ea05f67ad8a547ef9f08c515ba656b6d35eef5f3 +size 564829 diff --git a/vlm/train/rqfq0CYIekd/13.png b/vlm/train/rqfq0CYIekd/13.png new file mode 100644 index 0000000000000000000000000000000000000000..958b27ba529a75e4c3540b14f6ffd0259f09e2f8 --- /dev/null +++ b/vlm/train/rqfq0CYIekd/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c3654ef86366cc2e3e71cb921d5b96662efd0939973b31da0e8f6772bebd64d8 +size 515736 diff --git a/vlm/train/rqfq0CYIekd/2.png b/vlm/train/rqfq0CYIekd/2.png new file mode 100644 index 0000000000000000000000000000000000000000..4a66dc5da3b73085c3d8d6a7e604c1abc9e65d46 --- /dev/null +++ b/vlm/train/rqfq0CYIekd/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b957021a2d1190c47cc598514e6aa2f6691e9fd3bd2b23ddb2e8c0e22e3d1ea9 +size 518121 diff --git a/vlm/train/rqfq0CYIekd/3.png b/vlm/train/rqfq0CYIekd/3.png new file mode 100644 index 0000000000000000000000000000000000000000..e150128b3be6692fd71f78ea7855dad5dab1fc43 --- /dev/null +++ b/vlm/train/rqfq0CYIekd/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3677f8ee82cdb8d673e1b600a4a84713da632c293f0304319037f2ffad8e1ea2 +size 549711 diff --git a/vlm/train/rqfq0CYIekd/4.png b/vlm/train/rqfq0CYIekd/4.png new file mode 100644 index 0000000000000000000000000000000000000000..f84bb5af4f5a70d7340bf95501aa733c48cf25d5 --- /dev/null +++ b/vlm/train/rqfq0CYIekd/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8061f97fe601f5580e297f8f061677696ea37d93ba1c5d8ee2d112f9d69957aa +size 551331 diff --git a/vlm/train/rqfq0CYIekd/5.png b/vlm/train/rqfq0CYIekd/5.png new file mode 100644 index 0000000000000000000000000000000000000000..506018441897bee83c6ba1a1f14c7e05f0678cde --- /dev/null +++ b/vlm/train/rqfq0CYIekd/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9046a9c9f36a9b4265374757afde6ebb0c8f4059ff2fd57e2c45af9814bdf1b0 +size 496922 diff --git a/vlm/train/rqfq0CYIekd/6.png b/vlm/train/rqfq0CYIekd/6.png new file mode 100644 index 0000000000000000000000000000000000000000..9a68240e22905f3f9fc43e5828e5dcd2475a9e74 --- /dev/null +++ b/vlm/train/rqfq0CYIekd/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:70668ed80b79f71c3cf66be67ffff76a74dd1a2a03e91cc45d9df94e59b3dd99 +size 604554 diff --git a/vlm/train/rqfq0CYIekd/7.png b/vlm/train/rqfq0CYIekd/7.png new file mode 100644 index 0000000000000000000000000000000000000000..722262b6e0063466b813b006f0ac738fbe5703e4 --- /dev/null +++ b/vlm/train/rqfq0CYIekd/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b0345cdffb6ddf5696d314931f420da04bc930473e411f6b76544860495ae3aa +size 523535 diff --git a/vlm/train/rqfq0CYIekd/8.png b/vlm/train/rqfq0CYIekd/8.png new file mode 100644 index 0000000000000000000000000000000000000000..d973f2cb72a388b322bd2334a14283257b451a31 --- /dev/null +++ b/vlm/train/rqfq0CYIekd/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5a4588d6eef5e2d6ebc71dce5104a5c844c297dd790d24d5c7ea3be5a8d732bc +size 592273 diff --git a/vlm/train/rqfq0CYIekd/9.png b/vlm/train/rqfq0CYIekd/9.png new file mode 100644 index 0000000000000000000000000000000000000000..61c0501961eaf0894f479a4b92a02ad004bfa1b7 --- /dev/null +++ b/vlm/train/rqfq0CYIekd/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cb47f9be7dbc828199c016acc0ad4c66ccff44f39400978491e44817cfea273c +size 567714 diff --git a/vlm/train/uxpzitPEooJ/12.png b/vlm/train/uxpzitPEooJ/12.png new file mode 100644 index 0000000000000000000000000000000000000000..643d10948f044994e927ada70b5ba310fdd071a4 --- /dev/null +++ b/vlm/train/uxpzitPEooJ/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3d1b954cda993b41422a1c6d88ab95161cd14bb8aad22fcc546dca4fd951c7c3 +size 483892 diff --git a/vlm/train/uxpzitPEooJ/13.png b/vlm/train/uxpzitPEooJ/13.png new file mode 100644 index 0000000000000000000000000000000000000000..2c8c2d65482196f419492f9297044ca17a72cde5 --- /dev/null +++ b/vlm/train/uxpzitPEooJ/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ef8b6ef4cdfaf11fdfbed29fcabcb5489506e78de5bd4ffa365e5a43c3f0f669 +size 304720 diff --git a/vlm/train/uxpzitPEooJ/16.png b/vlm/train/uxpzitPEooJ/16.png new file mode 100644 index 0000000000000000000000000000000000000000..787282ea3472f703b0d0439a030d67b6d434094a --- /dev/null +++ b/vlm/train/uxpzitPEooJ/16.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:906d4a1e540a36e339b1fe342b5b5a54b6647ea5ed84fedae8b2c1ef874a072e +size 518007 diff --git a/vlm/train/uxpzitPEooJ/18.png b/vlm/train/uxpzitPEooJ/18.png new file mode 100644 index 0000000000000000000000000000000000000000..b49f231f8b6a5883747d30a3608a68d48d231959 --- /dev/null +++ b/vlm/train/uxpzitPEooJ/18.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1c53dae87a711ddae0236582940a8157b62fa242e2bd3e2952aaffb9b77efb22 +size 344976 diff --git a/vlm/train/uxpzitPEooJ/19.png b/vlm/train/uxpzitPEooJ/19.png new file mode 100644 index 0000000000000000000000000000000000000000..8413dacf4536051a1a42ade724d35d5e7a44c1e9 --- /dev/null +++ b/vlm/train/uxpzitPEooJ/19.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f85515ac71298832333edfc2fa706f37441d4c2e43b2a5605f7602a2d89494ef +size 336606 diff --git a/vlm/train/uxpzitPEooJ/20.png b/vlm/train/uxpzitPEooJ/20.png new file mode 100644 index 0000000000000000000000000000000000000000..dd7d8944bd219751af957de87a557f7597f8713b --- /dev/null +++ b/vlm/train/uxpzitPEooJ/20.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cf3ba895d20156a5913179b25688885b696f9ec832805c69d4dbd940bd564dda +size 358936 diff --git a/vlm/train/uxpzitPEooJ/22.png b/vlm/train/uxpzitPEooJ/22.png new file mode 100644 index 0000000000000000000000000000000000000000..844a8698ca403f48b7f8368d844ed5b1cb7154b8 --- /dev/null +++ b/vlm/train/uxpzitPEooJ/22.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8d41caeaeaa7641186e7293e0bee14a369d56523b8707f969324e2dd670a3bcf +size 337112 diff --git a/vlm/train/uxpzitPEooJ/26.png b/vlm/train/uxpzitPEooJ/26.png new file mode 100644 index 0000000000000000000000000000000000000000..bbee7d77c0cb5c0e11e5a448e844707d53d04192 --- /dev/null +++ b/vlm/train/uxpzitPEooJ/26.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b86f51155e21ab00f60c0074188f8bcc4796a69f32a0af2837764406ce48f8c4 +size 559811 diff --git a/vlm/train/uxpzitPEooJ/4.png b/vlm/train/uxpzitPEooJ/4.png new file mode 100644 index 0000000000000000000000000000000000000000..94448f64e9aa31c4b57201995af94c87e92d5a2f --- /dev/null +++ b/vlm/train/uxpzitPEooJ/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2f6a2893f0702390f4cbc5174b7444e250ab61411f5a9c7bbe6b44bf293d9476 +size 570389 diff --git a/vlm/train/uxpzitPEooJ/7.png b/vlm/train/uxpzitPEooJ/7.png new file mode 100644 index 0000000000000000000000000000000000000000..b8c451af4b08c71f383bde269ad11888aff35bb1 --- /dev/null +++ b/vlm/train/uxpzitPEooJ/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4a50bfb8944e3d0d1385c3c42c6a97a3652203b355ed2c0bf850a62a4e7bfa2b +size 545155 diff --git a/vlm/train/uxpzitPEooJ/8.png b/vlm/train/uxpzitPEooJ/8.png new file mode 100644 index 0000000000000000000000000000000000000000..99c2fbe3b1c716e04cb93b616622a45c8160ecb0 --- /dev/null +++ b/vlm/train/uxpzitPEooJ/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:036d1386968f379797c3b6a3e9b70c55d022dab4a28ede8bd13e62de6f77ebb8 +size 504167 diff --git a/vlm/train/vQTYEUtSUr/0.png b/vlm/train/vQTYEUtSUr/0.png new file mode 100644 index 0000000000000000000000000000000000000000..c41fb3ad9eb38b9564a0b58d402170146a9047c2 --- /dev/null +++ b/vlm/train/vQTYEUtSUr/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b77167d859dc1b36b86be1bc4fd2843e0b66eb28d9fe9c267cda15acbfd52098 +size 449333 diff --git a/vlm/train/vQTYEUtSUr/1.png b/vlm/train/vQTYEUtSUr/1.png new file mode 100644 index 0000000000000000000000000000000000000000..a90a0790264cabcf04e3c54e1b6d06d40ca45ae1 --- /dev/null +++ b/vlm/train/vQTYEUtSUr/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ffcd0227f6632982a322650239807d22993119d52dc0a551044b9a13e60f071c +size 807717 diff --git a/vlm/train/vQTYEUtSUr/10.png b/vlm/train/vQTYEUtSUr/10.png new file mode 100644 index 0000000000000000000000000000000000000000..199b76f247bf9f9721f8e524204ab4eed27b145a --- /dev/null +++ b/vlm/train/vQTYEUtSUr/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:11ccd7a4268b61dada2b2516c46c28b55220b25118b668236f69d186a614b24c +size 477800 diff --git a/vlm/train/vQTYEUtSUr/11.png b/vlm/train/vQTYEUtSUr/11.png new file mode 100644 index 0000000000000000000000000000000000000000..aa9fa6fd5bb238a550a847ad169db3bdc889f489 --- /dev/null +++ b/vlm/train/vQTYEUtSUr/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ddbdfe4d5a7d5d203d5f6c1ddd0713e2a99476c9a75619df9c7c75fef0dd615c +size 364643 diff --git a/vlm/train/vQTYEUtSUr/2.png b/vlm/train/vQTYEUtSUr/2.png new file mode 100644 index 0000000000000000000000000000000000000000..8a28ccad777b2a9e547d03c49e6c7df65432ee26 --- /dev/null +++ b/vlm/train/vQTYEUtSUr/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4f78d7f1d26e3a343b8b6b59940827ba28a50817de9ade97e38a2ac8dcb6352b +size 671014 diff --git a/vlm/train/vQTYEUtSUr/3.png b/vlm/train/vQTYEUtSUr/3.png new file mode 100644 index 0000000000000000000000000000000000000000..a3454e3a6c88f92b5007c91e861b002c2bc82520 --- /dev/null +++ b/vlm/train/vQTYEUtSUr/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:55be9903f647fa5303f56470e9349688d2a8bdfc9c58bdd603d3097d18429f7b +size 631814 diff --git a/vlm/train/vQTYEUtSUr/4.png b/vlm/train/vQTYEUtSUr/4.png new file mode 100644 index 0000000000000000000000000000000000000000..2a9d257bb2863a0d4aa35e9ba6eeca4e15c03071 --- /dev/null +++ b/vlm/train/vQTYEUtSUr/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:92e61cf0a2806b0512697bb027ff72eaa41001c8fd932f1d18c71f0ab3bc47ef +size 667779 diff --git a/vlm/train/vQTYEUtSUr/5.png b/vlm/train/vQTYEUtSUr/5.png new file mode 100644 index 0000000000000000000000000000000000000000..e82912c66a8723c128563eeef4e13385add3294c --- /dev/null +++ b/vlm/train/vQTYEUtSUr/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:18c722ac9cb1b89984c61e38ecc6fbea206d9c7df91fc96aa29d3ffc1785130a +size 598846 diff --git a/vlm/train/vQTYEUtSUr/6.png b/vlm/train/vQTYEUtSUr/6.png new file mode 100644 index 0000000000000000000000000000000000000000..29b931fef6f6d0694f08c93c2e9b2f068b63dd6c --- /dev/null +++ b/vlm/train/vQTYEUtSUr/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9058f9672f4a1a4b598f321ce538b6b32bdfb38c272d4c263e46c13c18d51fe8 +size 593822 diff --git a/vlm/train/vQTYEUtSUr/7.png b/vlm/train/vQTYEUtSUr/7.png new file mode 100644 index 0000000000000000000000000000000000000000..c4c7af1bc13cc070f2801d2c56f3abf7dcf295a2 --- /dev/null +++ b/vlm/train/vQTYEUtSUr/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:87433b124f46e7c57ec1cc105fb90627e5d5339f5c73de1a3088224facff39f7 +size 1024295 diff --git a/vlm/train/vQTYEUtSUr/8.png b/vlm/train/vQTYEUtSUr/8.png new file mode 100644 index 0000000000000000000000000000000000000000..f5b16b65b9c95073d08c5ad88458fb459bc83171 --- /dev/null +++ b/vlm/train/vQTYEUtSUr/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:253be0f8072476d70e3742bcc17f03d001160f18184be3f2920d9d0a0fa4f03d +size 643405 diff --git a/vlm/train/vQTYEUtSUr/9.png b/vlm/train/vQTYEUtSUr/9.png new file mode 100644 index 0000000000000000000000000000000000000000..632b8036fe27b030fe1fa86e50c9adc717332a91 --- /dev/null +++ b/vlm/train/vQTYEUtSUr/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:61bcbfeffb5c56d778b61e33004f4d745707324f81659d62aa9e0739bf576db6 +size 567479 diff --git a/vlm/train/vujTf_I8Kmc/0.png b/vlm/train/vujTf_I8Kmc/0.png new file mode 100644 index 0000000000000000000000000000000000000000..724f681f3beea9d8e8557cbac7ba891a24683224 --- /dev/null +++ b/vlm/train/vujTf_I8Kmc/0.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9dfdfc6a239510e35c89d8dbdb19911f8db10e7b0c70071e87a8cca179bb47c1 +size 520170 diff --git a/vlm/train/vujTf_I8Kmc/1.png b/vlm/train/vujTf_I8Kmc/1.png new file mode 100644 index 0000000000000000000000000000000000000000..5775b4e13ac002cab194841d162123002888958d --- /dev/null +++ b/vlm/train/vujTf_I8Kmc/1.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a5ab8b183b1fabb1b2511823c1a7d952ed11d0be516e307e5deda02732141e62 +size 686257 diff --git a/vlm/train/vujTf_I8Kmc/10.png b/vlm/train/vujTf_I8Kmc/10.png new file mode 100644 index 0000000000000000000000000000000000000000..fe84040982817544cf5b882dc8ea02681be99fb9 --- /dev/null +++ b/vlm/train/vujTf_I8Kmc/10.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0fa9379b7eecc6bd8abe9a63e0aeebcf6120322a2085f52ebe208fbea0ae02c4 +size 526466 diff --git a/vlm/train/vujTf_I8Kmc/11.png b/vlm/train/vujTf_I8Kmc/11.png new file mode 100644 index 0000000000000000000000000000000000000000..493f488eeb35b348333cac8e145c9c8b5fd5a9ac --- /dev/null +++ b/vlm/train/vujTf_I8Kmc/11.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a4dd7857bfd2c6677098f708e73e9297c93606a8ab453bf22ccf63d88c2aa856 +size 497623 diff --git a/vlm/train/vujTf_I8Kmc/12.png b/vlm/train/vujTf_I8Kmc/12.png new file mode 100644 index 0000000000000000000000000000000000000000..c9b74efb64a5e53b9b2d3910c932f7a7a67d9a39 --- /dev/null +++ b/vlm/train/vujTf_I8Kmc/12.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:669f68e005b936f786de87d4c7fcb8c994a689ae8e74171c34764208d93a64c6 +size 484471 diff --git a/vlm/train/vujTf_I8Kmc/13.png b/vlm/train/vujTf_I8Kmc/13.png new file mode 100644 index 0000000000000000000000000000000000000000..2733b688a008d0acf236d5620b0a638911702fc8 --- /dev/null +++ b/vlm/train/vujTf_I8Kmc/13.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fd20f2ded3470907d22805202e5cf71ee60e914218c218375f1a9392ba3ecdab +size 521216 diff --git a/vlm/train/vujTf_I8Kmc/14.png b/vlm/train/vujTf_I8Kmc/14.png new file mode 100644 index 0000000000000000000000000000000000000000..7f2c3e9c7c868ee82b014795fdc0516cf5f83d7e --- /dev/null +++ b/vlm/train/vujTf_I8Kmc/14.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:90b5e8630d1fed18089885b4c4f1eea0626e834485830d5b2b485fe4e500da5e +size 458867 diff --git a/vlm/train/vujTf_I8Kmc/15.png b/vlm/train/vujTf_I8Kmc/15.png new file mode 100644 index 0000000000000000000000000000000000000000..28bf6b7cc26ed59e4a923c1ba8299d7475d74cf7 --- /dev/null +++ b/vlm/train/vujTf_I8Kmc/15.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:790f45b55f80574eb26f752e93de52d80ba883a3d7887df588b178f9de9dc262 +size 181497 diff --git a/vlm/train/vujTf_I8Kmc/2.png b/vlm/train/vujTf_I8Kmc/2.png new file mode 100644 index 0000000000000000000000000000000000000000..02a2b2b41e50b0c7feeff0984a630591a3167fba --- /dev/null +++ b/vlm/train/vujTf_I8Kmc/2.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cf93e9b252e32ac1fe865502ec9ee2cf4af20b29d2c1accbf60e4f946662eb2d +size 537520 diff --git a/vlm/train/vujTf_I8Kmc/3.png b/vlm/train/vujTf_I8Kmc/3.png new file mode 100644 index 0000000000000000000000000000000000000000..4bc0b7f72a7b4e8141f0cba2be961993ca752af1 --- /dev/null +++ b/vlm/train/vujTf_I8Kmc/3.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e29f2e89be4b920a40752cb774e64089e06e67d9e62e1f0e7799166f130fdc07 +size 542972 diff --git a/vlm/train/vujTf_I8Kmc/4.png b/vlm/train/vujTf_I8Kmc/4.png new file mode 100644 index 0000000000000000000000000000000000000000..a5afd987be973baca3801b25e8e5b9c3c0ff58c0 --- /dev/null +++ b/vlm/train/vujTf_I8Kmc/4.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c50535ec43c192001d89a18f8c5358cbc5ed46c4934f2c3ff790afab6c9934a3 +size 573091 diff --git a/vlm/train/vujTf_I8Kmc/5.png b/vlm/train/vujTf_I8Kmc/5.png new file mode 100644 index 0000000000000000000000000000000000000000..d37044500e9cac3f1206d9a4dc67361b667c51c5 --- /dev/null +++ b/vlm/train/vujTf_I8Kmc/5.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f0c9090863fad48beee322848e4f86372e998980c61e8900625718277d1c3ccd +size 612559 diff --git a/vlm/train/vujTf_I8Kmc/6.png b/vlm/train/vujTf_I8Kmc/6.png new file mode 100644 index 0000000000000000000000000000000000000000..e290574549ed9ad6433eba028e7045b7df3f5f81 --- /dev/null +++ b/vlm/train/vujTf_I8Kmc/6.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c16e2f17c060ebd87c01fc8d537df1e69e3eea12fcee6bb33171dd634bab5ed5 +size 616109 diff --git a/vlm/train/vujTf_I8Kmc/7.png b/vlm/train/vujTf_I8Kmc/7.png new file mode 100644 index 0000000000000000000000000000000000000000..c33a6cb27c595a5d3483108e344144b25db278ff --- /dev/null +++ b/vlm/train/vujTf_I8Kmc/7.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:77f0baf3fba94b44000f7193944e37bd46dc2938b1d4efc0a4f281f530ad8f43 +size 592983 diff --git a/vlm/train/vujTf_I8Kmc/8.png b/vlm/train/vujTf_I8Kmc/8.png new file mode 100644 index 0000000000000000000000000000000000000000..2a4a8cde78f5fef5c6c37e68aff34e771bd6cf6b --- /dev/null +++ b/vlm/train/vujTf_I8Kmc/8.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:29710e9edac885d46de367c15d7d706c83dd4067d9b5427a13374f4a1b30ac92 +size 1276405 diff --git a/vlm/train/vujTf_I8Kmc/9.png b/vlm/train/vujTf_I8Kmc/9.png new file mode 100644 index 0000000000000000000000000000000000000000..c9be22fa5d029003ea6882e5fcf98db5cd1c1b8e --- /dev/null +++ b/vlm/train/vujTf_I8Kmc/9.png @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:00022b1548b2444e7bd240b71f1fa3e953b653e44c5640a82e737fec712989af +size 523684